xref: /openbmc/linux/fs/btrfs/volumes.c (revision 33ac9dba)
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 static 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->alloc_list);
78 	INIT_LIST_HEAD(&fs_devs->list);
79 
80 	return fs_devs;
81 }
82 
83 /**
84  * alloc_fs_devices - allocate struct btrfs_fs_devices
85  * @fsid:	a pointer to UUID for this FS.  If NULL a new UUID is
86  *		generated.
87  *
88  * Return: a pointer to a new &struct btrfs_fs_devices on success;
89  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
90  * can be destroyed with kfree() right away.
91  */
92 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
93 {
94 	struct btrfs_fs_devices *fs_devs;
95 
96 	fs_devs = __alloc_fs_devices();
97 	if (IS_ERR(fs_devs))
98 		return fs_devs;
99 
100 	if (fsid)
101 		memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
102 	else
103 		generate_random_uuid(fs_devs->fsid);
104 
105 	return fs_devs;
106 }
107 
108 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
109 {
110 	struct btrfs_device *device;
111 	WARN_ON(fs_devices->opened);
112 	while (!list_empty(&fs_devices->devices)) {
113 		device = list_entry(fs_devices->devices.next,
114 				    struct btrfs_device, dev_list);
115 		list_del(&device->dev_list);
116 		rcu_string_free(device->name);
117 		kfree(device);
118 	}
119 	kfree(fs_devices);
120 }
121 
122 static void btrfs_kobject_uevent(struct block_device *bdev,
123 				 enum kobject_action action)
124 {
125 	int ret;
126 
127 	ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
128 	if (ret)
129 		pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
130 			action,
131 			kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
132 			&disk_to_dev(bdev->bd_disk)->kobj);
133 }
134 
135 void btrfs_cleanup_fs_uuids(void)
136 {
137 	struct btrfs_fs_devices *fs_devices;
138 
139 	while (!list_empty(&fs_uuids)) {
140 		fs_devices = list_entry(fs_uuids.next,
141 					struct btrfs_fs_devices, list);
142 		list_del(&fs_devices->list);
143 		free_fs_devices(fs_devices);
144 	}
145 }
146 
147 static struct btrfs_device *__alloc_device(void)
148 {
149 	struct btrfs_device *dev;
150 
151 	dev = kzalloc(sizeof(*dev), GFP_NOFS);
152 	if (!dev)
153 		return ERR_PTR(-ENOMEM);
154 
155 	INIT_LIST_HEAD(&dev->dev_list);
156 	INIT_LIST_HEAD(&dev->dev_alloc_list);
157 
158 	spin_lock_init(&dev->io_lock);
159 
160 	spin_lock_init(&dev->reada_lock);
161 	atomic_set(&dev->reada_in_flight, 0);
162 	INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
163 	INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
164 
165 	return dev;
166 }
167 
168 static noinline struct btrfs_device *__find_device(struct list_head *head,
169 						   u64 devid, u8 *uuid)
170 {
171 	struct btrfs_device *dev;
172 
173 	list_for_each_entry(dev, head, dev_list) {
174 		if (dev->devid == devid &&
175 		    (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
176 			return dev;
177 		}
178 	}
179 	return NULL;
180 }
181 
182 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
183 {
184 	struct btrfs_fs_devices *fs_devices;
185 
186 	list_for_each_entry(fs_devices, &fs_uuids, list) {
187 		if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
188 			return fs_devices;
189 	}
190 	return NULL;
191 }
192 
193 static int
194 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
195 		      int flush, struct block_device **bdev,
196 		      struct buffer_head **bh)
197 {
198 	int ret;
199 
200 	*bdev = blkdev_get_by_path(device_path, flags, holder);
201 
202 	if (IS_ERR(*bdev)) {
203 		ret = PTR_ERR(*bdev);
204 		printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
205 		goto error;
206 	}
207 
208 	if (flush)
209 		filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
210 	ret = set_blocksize(*bdev, 4096);
211 	if (ret) {
212 		blkdev_put(*bdev, flags);
213 		goto error;
214 	}
215 	invalidate_bdev(*bdev);
216 	*bh = btrfs_read_dev_super(*bdev);
217 	if (!*bh) {
218 		ret = -EINVAL;
219 		blkdev_put(*bdev, flags);
220 		goto error;
221 	}
222 
223 	return 0;
224 
225 error:
226 	*bdev = NULL;
227 	*bh = NULL;
228 	return ret;
229 }
230 
231 static void requeue_list(struct btrfs_pending_bios *pending_bios,
232 			struct bio *head, struct bio *tail)
233 {
234 
235 	struct bio *old_head;
236 
237 	old_head = pending_bios->head;
238 	pending_bios->head = head;
239 	if (pending_bios->tail)
240 		tail->bi_next = old_head;
241 	else
242 		pending_bios->tail = tail;
243 }
244 
245 /*
246  * we try to collect pending bios for a device so we don't get a large
247  * number of procs sending bios down to the same device.  This greatly
248  * improves the schedulers ability to collect and merge the bios.
249  *
250  * But, it also turns into a long list of bios to process and that is sure
251  * to eventually make the worker thread block.  The solution here is to
252  * make some progress and then put this work struct back at the end of
253  * the list if the block device is congested.  This way, multiple devices
254  * can make progress from a single worker thread.
255  */
256 static noinline void run_scheduled_bios(struct btrfs_device *device)
257 {
258 	struct bio *pending;
259 	struct backing_dev_info *bdi;
260 	struct btrfs_fs_info *fs_info;
261 	struct btrfs_pending_bios *pending_bios;
262 	struct bio *tail;
263 	struct bio *cur;
264 	int again = 0;
265 	unsigned long num_run;
266 	unsigned long batch_run = 0;
267 	unsigned long limit;
268 	unsigned long last_waited = 0;
269 	int force_reg = 0;
270 	int sync_pending = 0;
271 	struct blk_plug plug;
272 
273 	/*
274 	 * this function runs all the bios we've collected for
275 	 * a particular device.  We don't want to wander off to
276 	 * another device without first sending all of these down.
277 	 * So, setup a plug here and finish it off before we return
278 	 */
279 	blk_start_plug(&plug);
280 
281 	bdi = blk_get_backing_dev_info(device->bdev);
282 	fs_info = device->dev_root->fs_info;
283 	limit = btrfs_async_submit_limit(fs_info);
284 	limit = limit * 2 / 3;
285 
286 loop:
287 	spin_lock(&device->io_lock);
288 
289 loop_lock:
290 	num_run = 0;
291 
292 	/* take all the bios off the list at once and process them
293 	 * later on (without the lock held).  But, remember the
294 	 * tail and other pointers so the bios can be properly reinserted
295 	 * into the list if we hit congestion
296 	 */
297 	if (!force_reg && device->pending_sync_bios.head) {
298 		pending_bios = &device->pending_sync_bios;
299 		force_reg = 1;
300 	} else {
301 		pending_bios = &device->pending_bios;
302 		force_reg = 0;
303 	}
304 
305 	pending = pending_bios->head;
306 	tail = pending_bios->tail;
307 	WARN_ON(pending && !tail);
308 
309 	/*
310 	 * if pending was null this time around, no bios need processing
311 	 * at all and we can stop.  Otherwise it'll loop back up again
312 	 * and do an additional check so no bios are missed.
313 	 *
314 	 * device->running_pending is used to synchronize with the
315 	 * schedule_bio code.
316 	 */
317 	if (device->pending_sync_bios.head == NULL &&
318 	    device->pending_bios.head == NULL) {
319 		again = 0;
320 		device->running_pending = 0;
321 	} else {
322 		again = 1;
323 		device->running_pending = 1;
324 	}
325 
326 	pending_bios->head = NULL;
327 	pending_bios->tail = NULL;
328 
329 	spin_unlock(&device->io_lock);
330 
331 	while (pending) {
332 
333 		rmb();
334 		/* we want to work on both lists, but do more bios on the
335 		 * sync list than the regular list
336 		 */
337 		if ((num_run > 32 &&
338 		    pending_bios != &device->pending_sync_bios &&
339 		    device->pending_sync_bios.head) ||
340 		   (num_run > 64 && pending_bios == &device->pending_sync_bios &&
341 		    device->pending_bios.head)) {
342 			spin_lock(&device->io_lock);
343 			requeue_list(pending_bios, pending, tail);
344 			goto loop_lock;
345 		}
346 
347 		cur = pending;
348 		pending = pending->bi_next;
349 		cur->bi_next = NULL;
350 
351 		if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
352 		    waitqueue_active(&fs_info->async_submit_wait))
353 			wake_up(&fs_info->async_submit_wait);
354 
355 		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
356 
357 		/*
358 		 * if we're doing the sync list, record that our
359 		 * plug has some sync requests on it
360 		 *
361 		 * If we're doing the regular list and there are
362 		 * sync requests sitting around, unplug before
363 		 * we add more
364 		 */
365 		if (pending_bios == &device->pending_sync_bios) {
366 			sync_pending = 1;
367 		} else if (sync_pending) {
368 			blk_finish_plug(&plug);
369 			blk_start_plug(&plug);
370 			sync_pending = 0;
371 		}
372 
373 		btrfsic_submit_bio(cur->bi_rw, cur);
374 		num_run++;
375 		batch_run++;
376 		if (need_resched())
377 			cond_resched();
378 
379 		/*
380 		 * we made progress, there is more work to do and the bdi
381 		 * is now congested.  Back off and let other work structs
382 		 * run instead
383 		 */
384 		if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
385 		    fs_info->fs_devices->open_devices > 1) {
386 			struct io_context *ioc;
387 
388 			ioc = current->io_context;
389 
390 			/*
391 			 * the main goal here is that we don't want to
392 			 * block if we're going to be able to submit
393 			 * more requests without blocking.
394 			 *
395 			 * This code does two great things, it pokes into
396 			 * the elevator code from a filesystem _and_
397 			 * it makes assumptions about how batching works.
398 			 */
399 			if (ioc && ioc->nr_batch_requests > 0 &&
400 			    time_before(jiffies, ioc->last_waited + HZ/50UL) &&
401 			    (last_waited == 0 ||
402 			     ioc->last_waited == last_waited)) {
403 				/*
404 				 * we want to go through our batch of
405 				 * requests and stop.  So, we copy out
406 				 * the ioc->last_waited time and test
407 				 * against it before looping
408 				 */
409 				last_waited = ioc->last_waited;
410 				if (need_resched())
411 					cond_resched();
412 				continue;
413 			}
414 			spin_lock(&device->io_lock);
415 			requeue_list(pending_bios, pending, tail);
416 			device->running_pending = 1;
417 
418 			spin_unlock(&device->io_lock);
419 			btrfs_queue_work(fs_info->submit_workers,
420 					 &device->work);
421 			goto done;
422 		}
423 		/* unplug every 64 requests just for good measure */
424 		if (batch_run % 64 == 0) {
425 			blk_finish_plug(&plug);
426 			blk_start_plug(&plug);
427 			sync_pending = 0;
428 		}
429 	}
430 
431 	cond_resched();
432 	if (again)
433 		goto loop;
434 
435 	spin_lock(&device->io_lock);
436 	if (device->pending_bios.head || device->pending_sync_bios.head)
437 		goto loop_lock;
438 	spin_unlock(&device->io_lock);
439 
440 done:
441 	blk_finish_plug(&plug);
442 }
443 
444 static void pending_bios_fn(struct btrfs_work *work)
445 {
446 	struct btrfs_device *device;
447 
448 	device = container_of(work, struct btrfs_device, work);
449 	run_scheduled_bios(device);
450 }
451 
452 /*
453  * Add new device to list of registered devices
454  *
455  * Returns:
456  * 1   - first time device is seen
457  * 0   - device already known
458  * < 0 - error
459  */
460 static noinline int device_list_add(const char *path,
461 			   struct btrfs_super_block *disk_super,
462 			   u64 devid, struct btrfs_fs_devices **fs_devices_ret)
463 {
464 	struct btrfs_device *device;
465 	struct btrfs_fs_devices *fs_devices;
466 	struct rcu_string *name;
467 	int ret = 0;
468 	u64 found_transid = btrfs_super_generation(disk_super);
469 
470 	fs_devices = find_fsid(disk_super->fsid);
471 	if (!fs_devices) {
472 		fs_devices = alloc_fs_devices(disk_super->fsid);
473 		if (IS_ERR(fs_devices))
474 			return PTR_ERR(fs_devices);
475 
476 		list_add(&fs_devices->list, &fs_uuids);
477 		fs_devices->latest_devid = devid;
478 		fs_devices->latest_trans = found_transid;
479 
480 		device = NULL;
481 	} else {
482 		device = __find_device(&fs_devices->devices, devid,
483 				       disk_super->dev_item.uuid);
484 	}
485 	if (!device) {
486 		if (fs_devices->opened)
487 			return -EBUSY;
488 
489 		device = btrfs_alloc_device(NULL, &devid,
490 					    disk_super->dev_item.uuid);
491 		if (IS_ERR(device)) {
492 			/* we can safely leave the fs_devices entry around */
493 			return PTR_ERR(device);
494 		}
495 
496 		name = rcu_string_strdup(path, GFP_NOFS);
497 		if (!name) {
498 			kfree(device);
499 			return -ENOMEM;
500 		}
501 		rcu_assign_pointer(device->name, name);
502 
503 		mutex_lock(&fs_devices->device_list_mutex);
504 		list_add_rcu(&device->dev_list, &fs_devices->devices);
505 		fs_devices->num_devices++;
506 		mutex_unlock(&fs_devices->device_list_mutex);
507 
508 		ret = 1;
509 		device->fs_devices = fs_devices;
510 	} else if (!device->name || strcmp(device->name->str, path)) {
511 		name = rcu_string_strdup(path, GFP_NOFS);
512 		if (!name)
513 			return -ENOMEM;
514 		rcu_string_free(device->name);
515 		rcu_assign_pointer(device->name, name);
516 		if (device->missing) {
517 			fs_devices->missing_devices--;
518 			device->missing = 0;
519 		}
520 	}
521 
522 	if (found_transid > fs_devices->latest_trans) {
523 		fs_devices->latest_devid = devid;
524 		fs_devices->latest_trans = found_transid;
525 	}
526 	*fs_devices_ret = fs_devices;
527 
528 	return ret;
529 }
530 
531 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
532 {
533 	struct btrfs_fs_devices *fs_devices;
534 	struct btrfs_device *device;
535 	struct btrfs_device *orig_dev;
536 
537 	fs_devices = alloc_fs_devices(orig->fsid);
538 	if (IS_ERR(fs_devices))
539 		return fs_devices;
540 
541 	fs_devices->latest_devid = orig->latest_devid;
542 	fs_devices->latest_trans = orig->latest_trans;
543 	fs_devices->total_devices = orig->total_devices;
544 
545 	/* We have held the volume lock, it is safe to get the devices. */
546 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
547 		struct rcu_string *name;
548 
549 		device = btrfs_alloc_device(NULL, &orig_dev->devid,
550 					    orig_dev->uuid);
551 		if (IS_ERR(device))
552 			goto error;
553 
554 		/*
555 		 * This is ok to do without rcu read locked because we hold the
556 		 * uuid mutex so nothing we touch in here is going to disappear.
557 		 */
558 		if (orig_dev->name) {
559 			name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
560 			if (!name) {
561 				kfree(device);
562 				goto error;
563 			}
564 			rcu_assign_pointer(device->name, name);
565 		}
566 
567 		list_add(&device->dev_list, &fs_devices->devices);
568 		device->fs_devices = fs_devices;
569 		fs_devices->num_devices++;
570 	}
571 	return fs_devices;
572 error:
573 	free_fs_devices(fs_devices);
574 	return ERR_PTR(-ENOMEM);
575 }
576 
577 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
578 			       struct btrfs_fs_devices *fs_devices, int step)
579 {
580 	struct btrfs_device *device, *next;
581 
582 	struct block_device *latest_bdev = NULL;
583 	u64 latest_devid = 0;
584 	u64 latest_transid = 0;
585 
586 	mutex_lock(&uuid_mutex);
587 again:
588 	/* This is the initialized path, it is safe to release the devices. */
589 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
590 		if (device->in_fs_metadata) {
591 			if (!device->is_tgtdev_for_dev_replace &&
592 			    (!latest_transid ||
593 			     device->generation > latest_transid)) {
594 				latest_devid = device->devid;
595 				latest_transid = device->generation;
596 				latest_bdev = device->bdev;
597 			}
598 			continue;
599 		}
600 
601 		if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
602 			/*
603 			 * In the first step, keep the device which has
604 			 * the correct fsid and the devid that is used
605 			 * for the dev_replace procedure.
606 			 * In the second step, the dev_replace state is
607 			 * read from the device tree and it is known
608 			 * whether the procedure is really active or
609 			 * not, which means whether this device is
610 			 * used or whether it should be removed.
611 			 */
612 			if (step == 0 || device->is_tgtdev_for_dev_replace) {
613 				continue;
614 			}
615 		}
616 		if (device->bdev) {
617 			blkdev_put(device->bdev, device->mode);
618 			device->bdev = NULL;
619 			fs_devices->open_devices--;
620 		}
621 		if (device->writeable) {
622 			list_del_init(&device->dev_alloc_list);
623 			device->writeable = 0;
624 			if (!device->is_tgtdev_for_dev_replace)
625 				fs_devices->rw_devices--;
626 		}
627 		list_del_init(&device->dev_list);
628 		fs_devices->num_devices--;
629 		rcu_string_free(device->name);
630 		kfree(device);
631 	}
632 
633 	if (fs_devices->seed) {
634 		fs_devices = fs_devices->seed;
635 		goto again;
636 	}
637 
638 	fs_devices->latest_bdev = latest_bdev;
639 	fs_devices->latest_devid = latest_devid;
640 	fs_devices->latest_trans = latest_transid;
641 
642 	mutex_unlock(&uuid_mutex);
643 }
644 
645 static void __free_device(struct work_struct *work)
646 {
647 	struct btrfs_device *device;
648 
649 	device = container_of(work, struct btrfs_device, rcu_work);
650 
651 	if (device->bdev)
652 		blkdev_put(device->bdev, device->mode);
653 
654 	rcu_string_free(device->name);
655 	kfree(device);
656 }
657 
658 static void free_device(struct rcu_head *head)
659 {
660 	struct btrfs_device *device;
661 
662 	device = container_of(head, struct btrfs_device, rcu);
663 
664 	INIT_WORK(&device->rcu_work, __free_device);
665 	schedule_work(&device->rcu_work);
666 }
667 
668 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
669 {
670 	struct btrfs_device *device;
671 
672 	if (--fs_devices->opened > 0)
673 		return 0;
674 
675 	mutex_lock(&fs_devices->device_list_mutex);
676 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
677 		struct btrfs_device *new_device;
678 		struct rcu_string *name;
679 
680 		if (device->bdev)
681 			fs_devices->open_devices--;
682 
683 		if (device->writeable &&
684 		    device->devid != BTRFS_DEV_REPLACE_DEVID) {
685 			list_del_init(&device->dev_alloc_list);
686 			fs_devices->rw_devices--;
687 		}
688 
689 		if (device->can_discard)
690 			fs_devices->num_can_discard--;
691 		if (device->missing)
692 			fs_devices->missing_devices--;
693 
694 		new_device = btrfs_alloc_device(NULL, &device->devid,
695 						device->uuid);
696 		BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
697 
698 		/* Safe because we are under uuid_mutex */
699 		if (device->name) {
700 			name = rcu_string_strdup(device->name->str, GFP_NOFS);
701 			BUG_ON(!name); /* -ENOMEM */
702 			rcu_assign_pointer(new_device->name, name);
703 		}
704 
705 		list_replace_rcu(&device->dev_list, &new_device->dev_list);
706 		new_device->fs_devices = device->fs_devices;
707 
708 		call_rcu(&device->rcu, free_device);
709 	}
710 	mutex_unlock(&fs_devices->device_list_mutex);
711 
712 	WARN_ON(fs_devices->open_devices);
713 	WARN_ON(fs_devices->rw_devices);
714 	fs_devices->opened = 0;
715 	fs_devices->seeding = 0;
716 
717 	return 0;
718 }
719 
720 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
721 {
722 	struct btrfs_fs_devices *seed_devices = NULL;
723 	int ret;
724 
725 	mutex_lock(&uuid_mutex);
726 	ret = __btrfs_close_devices(fs_devices);
727 	if (!fs_devices->opened) {
728 		seed_devices = fs_devices->seed;
729 		fs_devices->seed = NULL;
730 	}
731 	mutex_unlock(&uuid_mutex);
732 
733 	while (seed_devices) {
734 		fs_devices = seed_devices;
735 		seed_devices = fs_devices->seed;
736 		__btrfs_close_devices(fs_devices);
737 		free_fs_devices(fs_devices);
738 	}
739 	/*
740 	 * Wait for rcu kworkers under __btrfs_close_devices
741 	 * to finish all blkdev_puts so device is really
742 	 * free when umount is done.
743 	 */
744 	rcu_barrier();
745 	return ret;
746 }
747 
748 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
749 				fmode_t flags, void *holder)
750 {
751 	struct request_queue *q;
752 	struct block_device *bdev;
753 	struct list_head *head = &fs_devices->devices;
754 	struct btrfs_device *device;
755 	struct block_device *latest_bdev = NULL;
756 	struct buffer_head *bh;
757 	struct btrfs_super_block *disk_super;
758 	u64 latest_devid = 0;
759 	u64 latest_transid = 0;
760 	u64 devid;
761 	int seeding = 1;
762 	int ret = 0;
763 
764 	flags |= FMODE_EXCL;
765 
766 	list_for_each_entry(device, head, dev_list) {
767 		if (device->bdev)
768 			continue;
769 		if (!device->name)
770 			continue;
771 
772 		/* Just open everything we can; ignore failures here */
773 		if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
774 					    &bdev, &bh))
775 			continue;
776 
777 		disk_super = (struct btrfs_super_block *)bh->b_data;
778 		devid = btrfs_stack_device_id(&disk_super->dev_item);
779 		if (devid != device->devid)
780 			goto error_brelse;
781 
782 		if (memcmp(device->uuid, disk_super->dev_item.uuid,
783 			   BTRFS_UUID_SIZE))
784 			goto error_brelse;
785 
786 		device->generation = btrfs_super_generation(disk_super);
787 		if (!latest_transid || device->generation > latest_transid) {
788 			latest_devid = devid;
789 			latest_transid = device->generation;
790 			latest_bdev = bdev;
791 		}
792 
793 		if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
794 			device->writeable = 0;
795 		} else {
796 			device->writeable = !bdev_read_only(bdev);
797 			seeding = 0;
798 		}
799 
800 		q = bdev_get_queue(bdev);
801 		if (blk_queue_discard(q)) {
802 			device->can_discard = 1;
803 			fs_devices->num_can_discard++;
804 		}
805 
806 		device->bdev = bdev;
807 		device->in_fs_metadata = 0;
808 		device->mode = flags;
809 
810 		if (!blk_queue_nonrot(bdev_get_queue(bdev)))
811 			fs_devices->rotating = 1;
812 
813 		fs_devices->open_devices++;
814 		if (device->writeable &&
815 		    device->devid != BTRFS_DEV_REPLACE_DEVID) {
816 			fs_devices->rw_devices++;
817 			list_add(&device->dev_alloc_list,
818 				 &fs_devices->alloc_list);
819 		}
820 		brelse(bh);
821 		continue;
822 
823 error_brelse:
824 		brelse(bh);
825 		blkdev_put(bdev, flags);
826 		continue;
827 	}
828 	if (fs_devices->open_devices == 0) {
829 		ret = -EINVAL;
830 		goto out;
831 	}
832 	fs_devices->seeding = seeding;
833 	fs_devices->opened = 1;
834 	fs_devices->latest_bdev = latest_bdev;
835 	fs_devices->latest_devid = latest_devid;
836 	fs_devices->latest_trans = latest_transid;
837 	fs_devices->total_rw_bytes = 0;
838 out:
839 	return ret;
840 }
841 
842 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
843 		       fmode_t flags, void *holder)
844 {
845 	int ret;
846 
847 	mutex_lock(&uuid_mutex);
848 	if (fs_devices->opened) {
849 		fs_devices->opened++;
850 		ret = 0;
851 	} else {
852 		ret = __btrfs_open_devices(fs_devices, flags, holder);
853 	}
854 	mutex_unlock(&uuid_mutex);
855 	return ret;
856 }
857 
858 /*
859  * Look for a btrfs signature on a device. This may be called out of the mount path
860  * and we are not allowed to call set_blocksize during the scan. The superblock
861  * is read via pagecache
862  */
863 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
864 			  struct btrfs_fs_devices **fs_devices_ret)
865 {
866 	struct btrfs_super_block *disk_super;
867 	struct block_device *bdev;
868 	struct page *page;
869 	void *p;
870 	int ret = -EINVAL;
871 	u64 devid;
872 	u64 transid;
873 	u64 total_devices;
874 	u64 bytenr;
875 	pgoff_t index;
876 
877 	/*
878 	 * we would like to check all the supers, but that would make
879 	 * a btrfs mount succeed after a mkfs from a different FS.
880 	 * So, we need to add a special mount option to scan for
881 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
882 	 */
883 	bytenr = btrfs_sb_offset(0);
884 	flags |= FMODE_EXCL;
885 	mutex_lock(&uuid_mutex);
886 
887 	bdev = blkdev_get_by_path(path, flags, holder);
888 
889 	if (IS_ERR(bdev)) {
890 		ret = PTR_ERR(bdev);
891 		goto error;
892 	}
893 
894 	/* make sure our super fits in the device */
895 	if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
896 		goto error_bdev_put;
897 
898 	/* make sure our super fits in the page */
899 	if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
900 		goto error_bdev_put;
901 
902 	/* make sure our super doesn't straddle pages on disk */
903 	index = bytenr >> PAGE_CACHE_SHIFT;
904 	if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
905 		goto error_bdev_put;
906 
907 	/* pull in the page with our super */
908 	page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
909 				   index, GFP_NOFS);
910 
911 	if (IS_ERR_OR_NULL(page))
912 		goto error_bdev_put;
913 
914 	p = kmap(page);
915 
916 	/* align our pointer to the offset of the super block */
917 	disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
918 
919 	if (btrfs_super_bytenr(disk_super) != bytenr ||
920 	    btrfs_super_magic(disk_super) != BTRFS_MAGIC)
921 		goto error_unmap;
922 
923 	devid = btrfs_stack_device_id(&disk_super->dev_item);
924 	transid = btrfs_super_generation(disk_super);
925 	total_devices = btrfs_super_num_devices(disk_super);
926 
927 	ret = device_list_add(path, disk_super, devid, fs_devices_ret);
928 	if (ret > 0) {
929 		if (disk_super->label[0]) {
930 			if (disk_super->label[BTRFS_LABEL_SIZE - 1])
931 				disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
932 			printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
933 		} else {
934 			printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
935 		}
936 
937 		printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
938 		ret = 0;
939 	}
940 	if (!ret && fs_devices_ret)
941 		(*fs_devices_ret)->total_devices = total_devices;
942 
943 error_unmap:
944 	kunmap(page);
945 	page_cache_release(page);
946 
947 error_bdev_put:
948 	blkdev_put(bdev, flags);
949 error:
950 	mutex_unlock(&uuid_mutex);
951 	return ret;
952 }
953 
954 /* helper to account the used device space in the range */
955 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
956 				   u64 end, u64 *length)
957 {
958 	struct btrfs_key key;
959 	struct btrfs_root *root = device->dev_root;
960 	struct btrfs_dev_extent *dev_extent;
961 	struct btrfs_path *path;
962 	u64 extent_end;
963 	int ret;
964 	int slot;
965 	struct extent_buffer *l;
966 
967 	*length = 0;
968 
969 	if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
970 		return 0;
971 
972 	path = btrfs_alloc_path();
973 	if (!path)
974 		return -ENOMEM;
975 	path->reada = 2;
976 
977 	key.objectid = device->devid;
978 	key.offset = start;
979 	key.type = BTRFS_DEV_EXTENT_KEY;
980 
981 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
982 	if (ret < 0)
983 		goto out;
984 	if (ret > 0) {
985 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
986 		if (ret < 0)
987 			goto out;
988 	}
989 
990 	while (1) {
991 		l = path->nodes[0];
992 		slot = path->slots[0];
993 		if (slot >= btrfs_header_nritems(l)) {
994 			ret = btrfs_next_leaf(root, path);
995 			if (ret == 0)
996 				continue;
997 			if (ret < 0)
998 				goto out;
999 
1000 			break;
1001 		}
1002 		btrfs_item_key_to_cpu(l, &key, slot);
1003 
1004 		if (key.objectid < device->devid)
1005 			goto next;
1006 
1007 		if (key.objectid > device->devid)
1008 			break;
1009 
1010 		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1011 			goto next;
1012 
1013 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1014 		extent_end = key.offset + btrfs_dev_extent_length(l,
1015 								  dev_extent);
1016 		if (key.offset <= start && extent_end > end) {
1017 			*length = end - start + 1;
1018 			break;
1019 		} else if (key.offset <= start && extent_end > start)
1020 			*length += extent_end - start;
1021 		else if (key.offset > start && extent_end <= end)
1022 			*length += extent_end - key.offset;
1023 		else if (key.offset > start && key.offset <= end) {
1024 			*length += end - key.offset + 1;
1025 			break;
1026 		} else if (key.offset > end)
1027 			break;
1028 
1029 next:
1030 		path->slots[0]++;
1031 	}
1032 	ret = 0;
1033 out:
1034 	btrfs_free_path(path);
1035 	return ret;
1036 }
1037 
1038 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1039 				   struct btrfs_device *device,
1040 				   u64 *start, u64 len)
1041 {
1042 	struct extent_map *em;
1043 	int ret = 0;
1044 
1045 	list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1046 		struct map_lookup *map;
1047 		int i;
1048 
1049 		map = (struct map_lookup *)em->bdev;
1050 		for (i = 0; i < map->num_stripes; i++) {
1051 			if (map->stripes[i].dev != device)
1052 				continue;
1053 			if (map->stripes[i].physical >= *start + len ||
1054 			    map->stripes[i].physical + em->orig_block_len <=
1055 			    *start)
1056 				continue;
1057 			*start = map->stripes[i].physical +
1058 				em->orig_block_len;
1059 			ret = 1;
1060 		}
1061 	}
1062 
1063 	return ret;
1064 }
1065 
1066 
1067 /*
1068  * find_free_dev_extent - find free space in the specified device
1069  * @device:	the device which we search the free space in
1070  * @num_bytes:	the size of the free space that we need
1071  * @start:	store the start of the free space.
1072  * @len:	the size of the free space. that we find, or the size of the max
1073  * 		free space if we don't find suitable free space
1074  *
1075  * this uses a pretty simple search, the expectation is that it is
1076  * called very infrequently and that a given device has a small number
1077  * of extents
1078  *
1079  * @start is used to store the start of the free space if we find. But if we
1080  * don't find suitable free space, it will be used to store the start position
1081  * of the max free space.
1082  *
1083  * @len is used to store the size of the free space that we find.
1084  * But if we don't find suitable free space, it is used to store the size of
1085  * the max free space.
1086  */
1087 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1088 			 struct btrfs_device *device, u64 num_bytes,
1089 			 u64 *start, u64 *len)
1090 {
1091 	struct btrfs_key key;
1092 	struct btrfs_root *root = device->dev_root;
1093 	struct btrfs_dev_extent *dev_extent;
1094 	struct btrfs_path *path;
1095 	u64 hole_size;
1096 	u64 max_hole_start;
1097 	u64 max_hole_size;
1098 	u64 extent_end;
1099 	u64 search_start;
1100 	u64 search_end = device->total_bytes;
1101 	int ret;
1102 	int slot;
1103 	struct extent_buffer *l;
1104 
1105 	/* FIXME use last free of some kind */
1106 
1107 	/* we don't want to overwrite the superblock on the drive,
1108 	 * so we make sure to start at an offset of at least 1MB
1109 	 */
1110 	search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1111 
1112 	path = btrfs_alloc_path();
1113 	if (!path)
1114 		return -ENOMEM;
1115 again:
1116 	max_hole_start = search_start;
1117 	max_hole_size = 0;
1118 	hole_size = 0;
1119 
1120 	if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1121 		ret = -ENOSPC;
1122 		goto out;
1123 	}
1124 
1125 	path->reada = 2;
1126 	path->search_commit_root = 1;
1127 	path->skip_locking = 1;
1128 
1129 	key.objectid = device->devid;
1130 	key.offset = search_start;
1131 	key.type = BTRFS_DEV_EXTENT_KEY;
1132 
1133 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1134 	if (ret < 0)
1135 		goto out;
1136 	if (ret > 0) {
1137 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
1138 		if (ret < 0)
1139 			goto out;
1140 	}
1141 
1142 	while (1) {
1143 		l = path->nodes[0];
1144 		slot = path->slots[0];
1145 		if (slot >= btrfs_header_nritems(l)) {
1146 			ret = btrfs_next_leaf(root, path);
1147 			if (ret == 0)
1148 				continue;
1149 			if (ret < 0)
1150 				goto out;
1151 
1152 			break;
1153 		}
1154 		btrfs_item_key_to_cpu(l, &key, slot);
1155 
1156 		if (key.objectid < device->devid)
1157 			goto next;
1158 
1159 		if (key.objectid > device->devid)
1160 			break;
1161 
1162 		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1163 			goto next;
1164 
1165 		if (key.offset > search_start) {
1166 			hole_size = key.offset - search_start;
1167 
1168 			/*
1169 			 * Have to check before we set max_hole_start, otherwise
1170 			 * we could end up sending back this offset anyway.
1171 			 */
1172 			if (contains_pending_extent(trans, device,
1173 						    &search_start,
1174 						    hole_size))
1175 				hole_size = 0;
1176 
1177 			if (hole_size > max_hole_size) {
1178 				max_hole_start = search_start;
1179 				max_hole_size = hole_size;
1180 			}
1181 
1182 			/*
1183 			 * If this free space is greater than which we need,
1184 			 * it must be the max free space that we have found
1185 			 * until now, so max_hole_start must point to the start
1186 			 * of this free space and the length of this free space
1187 			 * is stored in max_hole_size. Thus, we return
1188 			 * max_hole_start and max_hole_size and go back to the
1189 			 * caller.
1190 			 */
1191 			if (hole_size >= num_bytes) {
1192 				ret = 0;
1193 				goto out;
1194 			}
1195 		}
1196 
1197 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1198 		extent_end = key.offset + btrfs_dev_extent_length(l,
1199 								  dev_extent);
1200 		if (extent_end > search_start)
1201 			search_start = extent_end;
1202 next:
1203 		path->slots[0]++;
1204 		cond_resched();
1205 	}
1206 
1207 	/*
1208 	 * At this point, search_start should be the end of
1209 	 * allocated dev extents, and when shrinking the device,
1210 	 * search_end may be smaller than search_start.
1211 	 */
1212 	if (search_end > search_start)
1213 		hole_size = search_end - search_start;
1214 
1215 	if (hole_size > max_hole_size) {
1216 		max_hole_start = search_start;
1217 		max_hole_size = hole_size;
1218 	}
1219 
1220 	if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1221 		btrfs_release_path(path);
1222 		goto again;
1223 	}
1224 
1225 	/* See above. */
1226 	if (hole_size < num_bytes)
1227 		ret = -ENOSPC;
1228 	else
1229 		ret = 0;
1230 
1231 out:
1232 	btrfs_free_path(path);
1233 	*start = max_hole_start;
1234 	if (len)
1235 		*len = max_hole_size;
1236 	return ret;
1237 }
1238 
1239 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1240 			  struct btrfs_device *device,
1241 			  u64 start)
1242 {
1243 	int ret;
1244 	struct btrfs_path *path;
1245 	struct btrfs_root *root = device->dev_root;
1246 	struct btrfs_key key;
1247 	struct btrfs_key found_key;
1248 	struct extent_buffer *leaf = NULL;
1249 	struct btrfs_dev_extent *extent = NULL;
1250 
1251 	path = btrfs_alloc_path();
1252 	if (!path)
1253 		return -ENOMEM;
1254 
1255 	key.objectid = device->devid;
1256 	key.offset = start;
1257 	key.type = BTRFS_DEV_EXTENT_KEY;
1258 again:
1259 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1260 	if (ret > 0) {
1261 		ret = btrfs_previous_item(root, path, key.objectid,
1262 					  BTRFS_DEV_EXTENT_KEY);
1263 		if (ret)
1264 			goto out;
1265 		leaf = path->nodes[0];
1266 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1267 		extent = btrfs_item_ptr(leaf, path->slots[0],
1268 					struct btrfs_dev_extent);
1269 		BUG_ON(found_key.offset > start || found_key.offset +
1270 		       btrfs_dev_extent_length(leaf, extent) < start);
1271 		key = found_key;
1272 		btrfs_release_path(path);
1273 		goto again;
1274 	} else if (ret == 0) {
1275 		leaf = path->nodes[0];
1276 		extent = btrfs_item_ptr(leaf, path->slots[0],
1277 					struct btrfs_dev_extent);
1278 	} else {
1279 		btrfs_error(root->fs_info, ret, "Slot search failed");
1280 		goto out;
1281 	}
1282 
1283 	if (device->bytes_used > 0) {
1284 		u64 len = btrfs_dev_extent_length(leaf, extent);
1285 		device->bytes_used -= len;
1286 		spin_lock(&root->fs_info->free_chunk_lock);
1287 		root->fs_info->free_chunk_space += len;
1288 		spin_unlock(&root->fs_info->free_chunk_lock);
1289 	}
1290 	ret = btrfs_del_item(trans, root, path);
1291 	if (ret) {
1292 		btrfs_error(root->fs_info, ret,
1293 			    "Failed to remove dev extent item");
1294 	}
1295 out:
1296 	btrfs_free_path(path);
1297 	return ret;
1298 }
1299 
1300 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1301 				  struct btrfs_device *device,
1302 				  u64 chunk_tree, u64 chunk_objectid,
1303 				  u64 chunk_offset, u64 start, u64 num_bytes)
1304 {
1305 	int ret;
1306 	struct btrfs_path *path;
1307 	struct btrfs_root *root = device->dev_root;
1308 	struct btrfs_dev_extent *extent;
1309 	struct extent_buffer *leaf;
1310 	struct btrfs_key key;
1311 
1312 	WARN_ON(!device->in_fs_metadata);
1313 	WARN_ON(device->is_tgtdev_for_dev_replace);
1314 	path = btrfs_alloc_path();
1315 	if (!path)
1316 		return -ENOMEM;
1317 
1318 	key.objectid = device->devid;
1319 	key.offset = start;
1320 	key.type = BTRFS_DEV_EXTENT_KEY;
1321 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1322 				      sizeof(*extent));
1323 	if (ret)
1324 		goto out;
1325 
1326 	leaf = path->nodes[0];
1327 	extent = btrfs_item_ptr(leaf, path->slots[0],
1328 				struct btrfs_dev_extent);
1329 	btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1330 	btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1331 	btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1332 
1333 	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1334 		    btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1335 
1336 	btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1337 	btrfs_mark_buffer_dirty(leaf);
1338 out:
1339 	btrfs_free_path(path);
1340 	return ret;
1341 }
1342 
1343 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1344 {
1345 	struct extent_map_tree *em_tree;
1346 	struct extent_map *em;
1347 	struct rb_node *n;
1348 	u64 ret = 0;
1349 
1350 	em_tree = &fs_info->mapping_tree.map_tree;
1351 	read_lock(&em_tree->lock);
1352 	n = rb_last(&em_tree->map);
1353 	if (n) {
1354 		em = rb_entry(n, struct extent_map, rb_node);
1355 		ret = em->start + em->len;
1356 	}
1357 	read_unlock(&em_tree->lock);
1358 
1359 	return ret;
1360 }
1361 
1362 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1363 				    u64 *devid_ret)
1364 {
1365 	int ret;
1366 	struct btrfs_key key;
1367 	struct btrfs_key found_key;
1368 	struct btrfs_path *path;
1369 
1370 	path = btrfs_alloc_path();
1371 	if (!path)
1372 		return -ENOMEM;
1373 
1374 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1375 	key.type = BTRFS_DEV_ITEM_KEY;
1376 	key.offset = (u64)-1;
1377 
1378 	ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1379 	if (ret < 0)
1380 		goto error;
1381 
1382 	BUG_ON(ret == 0); /* Corruption */
1383 
1384 	ret = btrfs_previous_item(fs_info->chunk_root, path,
1385 				  BTRFS_DEV_ITEMS_OBJECTID,
1386 				  BTRFS_DEV_ITEM_KEY);
1387 	if (ret) {
1388 		*devid_ret = 1;
1389 	} else {
1390 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1391 				      path->slots[0]);
1392 		*devid_ret = found_key.offset + 1;
1393 	}
1394 	ret = 0;
1395 error:
1396 	btrfs_free_path(path);
1397 	return ret;
1398 }
1399 
1400 /*
1401  * the device information is stored in the chunk root
1402  * the btrfs_device struct should be fully filled in
1403  */
1404 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1405 			    struct btrfs_root *root,
1406 			    struct btrfs_device *device)
1407 {
1408 	int ret;
1409 	struct btrfs_path *path;
1410 	struct btrfs_dev_item *dev_item;
1411 	struct extent_buffer *leaf;
1412 	struct btrfs_key key;
1413 	unsigned long ptr;
1414 
1415 	root = root->fs_info->chunk_root;
1416 
1417 	path = btrfs_alloc_path();
1418 	if (!path)
1419 		return -ENOMEM;
1420 
1421 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1422 	key.type = BTRFS_DEV_ITEM_KEY;
1423 	key.offset = device->devid;
1424 
1425 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1426 				      sizeof(*dev_item));
1427 	if (ret)
1428 		goto out;
1429 
1430 	leaf = path->nodes[0];
1431 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1432 
1433 	btrfs_set_device_id(leaf, dev_item, device->devid);
1434 	btrfs_set_device_generation(leaf, dev_item, 0);
1435 	btrfs_set_device_type(leaf, dev_item, device->type);
1436 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1437 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1438 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1439 	btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1440 	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1441 	btrfs_set_device_group(leaf, dev_item, 0);
1442 	btrfs_set_device_seek_speed(leaf, dev_item, 0);
1443 	btrfs_set_device_bandwidth(leaf, dev_item, 0);
1444 	btrfs_set_device_start_offset(leaf, dev_item, 0);
1445 
1446 	ptr = btrfs_device_uuid(dev_item);
1447 	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1448 	ptr = btrfs_device_fsid(dev_item);
1449 	write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1450 	btrfs_mark_buffer_dirty(leaf);
1451 
1452 	ret = 0;
1453 out:
1454 	btrfs_free_path(path);
1455 	return ret;
1456 }
1457 
1458 /*
1459  * Function to update ctime/mtime for a given device path.
1460  * Mainly used for ctime/mtime based probe like libblkid.
1461  */
1462 static void update_dev_time(char *path_name)
1463 {
1464 	struct file *filp;
1465 
1466 	filp = filp_open(path_name, O_RDWR, 0);
1467 	if (!filp)
1468 		return;
1469 	file_update_time(filp);
1470 	filp_close(filp, NULL);
1471 	return;
1472 }
1473 
1474 static int btrfs_rm_dev_item(struct btrfs_root *root,
1475 			     struct btrfs_device *device)
1476 {
1477 	int ret;
1478 	struct btrfs_path *path;
1479 	struct btrfs_key key;
1480 	struct btrfs_trans_handle *trans;
1481 
1482 	root = root->fs_info->chunk_root;
1483 
1484 	path = btrfs_alloc_path();
1485 	if (!path)
1486 		return -ENOMEM;
1487 
1488 	trans = btrfs_start_transaction(root, 0);
1489 	if (IS_ERR(trans)) {
1490 		btrfs_free_path(path);
1491 		return PTR_ERR(trans);
1492 	}
1493 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1494 	key.type = BTRFS_DEV_ITEM_KEY;
1495 	key.offset = device->devid;
1496 	lock_chunks(root);
1497 
1498 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1499 	if (ret < 0)
1500 		goto out;
1501 
1502 	if (ret > 0) {
1503 		ret = -ENOENT;
1504 		goto out;
1505 	}
1506 
1507 	ret = btrfs_del_item(trans, root, path);
1508 	if (ret)
1509 		goto out;
1510 out:
1511 	btrfs_free_path(path);
1512 	unlock_chunks(root);
1513 	btrfs_commit_transaction(trans, root);
1514 	return ret;
1515 }
1516 
1517 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1518 {
1519 	struct btrfs_device *device;
1520 	struct btrfs_device *next_device;
1521 	struct block_device *bdev;
1522 	struct buffer_head *bh = NULL;
1523 	struct btrfs_super_block *disk_super;
1524 	struct btrfs_fs_devices *cur_devices;
1525 	u64 all_avail;
1526 	u64 devid;
1527 	u64 num_devices;
1528 	u8 *dev_uuid;
1529 	unsigned seq;
1530 	int ret = 0;
1531 	bool clear_super = false;
1532 
1533 	mutex_lock(&uuid_mutex);
1534 
1535 	do {
1536 		seq = read_seqbegin(&root->fs_info->profiles_lock);
1537 
1538 		all_avail = root->fs_info->avail_data_alloc_bits |
1539 			    root->fs_info->avail_system_alloc_bits |
1540 			    root->fs_info->avail_metadata_alloc_bits;
1541 	} while (read_seqretry(&root->fs_info->profiles_lock, seq));
1542 
1543 	num_devices = root->fs_info->fs_devices->num_devices;
1544 	btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1545 	if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1546 		WARN_ON(num_devices < 1);
1547 		num_devices--;
1548 	}
1549 	btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1550 
1551 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1552 		ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1553 		goto out;
1554 	}
1555 
1556 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1557 		ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1558 		goto out;
1559 	}
1560 
1561 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1562 	    root->fs_info->fs_devices->rw_devices <= 2) {
1563 		ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1564 		goto out;
1565 	}
1566 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1567 	    root->fs_info->fs_devices->rw_devices <= 3) {
1568 		ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1569 		goto out;
1570 	}
1571 
1572 	if (strcmp(device_path, "missing") == 0) {
1573 		struct list_head *devices;
1574 		struct btrfs_device *tmp;
1575 
1576 		device = NULL;
1577 		devices = &root->fs_info->fs_devices->devices;
1578 		/*
1579 		 * It is safe to read the devices since the volume_mutex
1580 		 * is held.
1581 		 */
1582 		list_for_each_entry(tmp, devices, dev_list) {
1583 			if (tmp->in_fs_metadata &&
1584 			    !tmp->is_tgtdev_for_dev_replace &&
1585 			    !tmp->bdev) {
1586 				device = tmp;
1587 				break;
1588 			}
1589 		}
1590 		bdev = NULL;
1591 		bh = NULL;
1592 		disk_super = NULL;
1593 		if (!device) {
1594 			ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1595 			goto out;
1596 		}
1597 	} else {
1598 		ret = btrfs_get_bdev_and_sb(device_path,
1599 					    FMODE_WRITE | FMODE_EXCL,
1600 					    root->fs_info->bdev_holder, 0,
1601 					    &bdev, &bh);
1602 		if (ret)
1603 			goto out;
1604 		disk_super = (struct btrfs_super_block *)bh->b_data;
1605 		devid = btrfs_stack_device_id(&disk_super->dev_item);
1606 		dev_uuid = disk_super->dev_item.uuid;
1607 		device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1608 					   disk_super->fsid);
1609 		if (!device) {
1610 			ret = -ENOENT;
1611 			goto error_brelse;
1612 		}
1613 	}
1614 
1615 	if (device->is_tgtdev_for_dev_replace) {
1616 		ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1617 		goto error_brelse;
1618 	}
1619 
1620 	if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1621 		ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1622 		goto error_brelse;
1623 	}
1624 
1625 	if (device->writeable) {
1626 		lock_chunks(root);
1627 		list_del_init(&device->dev_alloc_list);
1628 		unlock_chunks(root);
1629 		root->fs_info->fs_devices->rw_devices--;
1630 		clear_super = true;
1631 	}
1632 
1633 	mutex_unlock(&uuid_mutex);
1634 	ret = btrfs_shrink_device(device, 0);
1635 	mutex_lock(&uuid_mutex);
1636 	if (ret)
1637 		goto error_undo;
1638 
1639 	/*
1640 	 * TODO: the superblock still includes this device in its num_devices
1641 	 * counter although write_all_supers() is not locked out. This
1642 	 * could give a filesystem state which requires a degraded mount.
1643 	 */
1644 	ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1645 	if (ret)
1646 		goto error_undo;
1647 
1648 	spin_lock(&root->fs_info->free_chunk_lock);
1649 	root->fs_info->free_chunk_space = device->total_bytes -
1650 		device->bytes_used;
1651 	spin_unlock(&root->fs_info->free_chunk_lock);
1652 
1653 	device->in_fs_metadata = 0;
1654 	btrfs_scrub_cancel_dev(root->fs_info, device);
1655 
1656 	/*
1657 	 * the device list mutex makes sure that we don't change
1658 	 * the device list while someone else is writing out all
1659 	 * the device supers. Whoever is writing all supers, should
1660 	 * lock the device list mutex before getting the number of
1661 	 * devices in the super block (super_copy). Conversely,
1662 	 * whoever updates the number of devices in the super block
1663 	 * (super_copy) should hold the device list mutex.
1664 	 */
1665 
1666 	cur_devices = device->fs_devices;
1667 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1668 	list_del_rcu(&device->dev_list);
1669 
1670 	device->fs_devices->num_devices--;
1671 	device->fs_devices->total_devices--;
1672 
1673 	if (device->missing)
1674 		root->fs_info->fs_devices->missing_devices--;
1675 
1676 	next_device = list_entry(root->fs_info->fs_devices->devices.next,
1677 				 struct btrfs_device, dev_list);
1678 	if (device->bdev == root->fs_info->sb->s_bdev)
1679 		root->fs_info->sb->s_bdev = next_device->bdev;
1680 	if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1681 		root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1682 
1683 	if (device->bdev) {
1684 		device->fs_devices->open_devices--;
1685 		/* remove sysfs entry */
1686 		btrfs_kobj_rm_device(root->fs_info, device);
1687 	}
1688 
1689 	call_rcu(&device->rcu, free_device);
1690 
1691 	num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1692 	btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1693 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1694 
1695 	if (cur_devices->open_devices == 0) {
1696 		struct btrfs_fs_devices *fs_devices;
1697 		fs_devices = root->fs_info->fs_devices;
1698 		while (fs_devices) {
1699 			if (fs_devices->seed == cur_devices) {
1700 				fs_devices->seed = cur_devices->seed;
1701 				break;
1702 			}
1703 			fs_devices = fs_devices->seed;
1704 		}
1705 		cur_devices->seed = NULL;
1706 		lock_chunks(root);
1707 		__btrfs_close_devices(cur_devices);
1708 		unlock_chunks(root);
1709 		free_fs_devices(cur_devices);
1710 	}
1711 
1712 	root->fs_info->num_tolerated_disk_barrier_failures =
1713 		btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1714 
1715 	/*
1716 	 * at this point, the device is zero sized.  We want to
1717 	 * remove it from the devices list and zero out the old super
1718 	 */
1719 	if (clear_super && disk_super) {
1720 		u64 bytenr;
1721 		int i;
1722 
1723 		/* make sure this device isn't detected as part of
1724 		 * the FS anymore
1725 		 */
1726 		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1727 		set_buffer_dirty(bh);
1728 		sync_dirty_buffer(bh);
1729 
1730 		/* clear the mirror copies of super block on the disk
1731 		 * being removed, 0th copy is been taken care above and
1732 		 * the below would take of the rest
1733 		 */
1734 		for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1735 			bytenr = btrfs_sb_offset(i);
1736 			if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1737 					i_size_read(bdev->bd_inode))
1738 				break;
1739 
1740 			brelse(bh);
1741 			bh = __bread(bdev, bytenr / 4096,
1742 					BTRFS_SUPER_INFO_SIZE);
1743 			if (!bh)
1744 				continue;
1745 
1746 			disk_super = (struct btrfs_super_block *)bh->b_data;
1747 
1748 			if (btrfs_super_bytenr(disk_super) != bytenr ||
1749 				btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1750 				continue;
1751 			}
1752 			memset(&disk_super->magic, 0,
1753 						sizeof(disk_super->magic));
1754 			set_buffer_dirty(bh);
1755 			sync_dirty_buffer(bh);
1756 		}
1757 	}
1758 
1759 	ret = 0;
1760 
1761 	if (bdev) {
1762 		/* Notify udev that device has changed */
1763 		btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1764 
1765 		/* Update ctime/mtime for device path for libblkid */
1766 		update_dev_time(device_path);
1767 	}
1768 
1769 error_brelse:
1770 	brelse(bh);
1771 	if (bdev)
1772 		blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1773 out:
1774 	mutex_unlock(&uuid_mutex);
1775 	return ret;
1776 error_undo:
1777 	if (device->writeable) {
1778 		lock_chunks(root);
1779 		list_add(&device->dev_alloc_list,
1780 			 &root->fs_info->fs_devices->alloc_list);
1781 		unlock_chunks(root);
1782 		root->fs_info->fs_devices->rw_devices++;
1783 	}
1784 	goto error_brelse;
1785 }
1786 
1787 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1788 				 struct btrfs_device *srcdev)
1789 {
1790 	WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1791 
1792 	list_del_rcu(&srcdev->dev_list);
1793 	list_del_rcu(&srcdev->dev_alloc_list);
1794 	fs_info->fs_devices->num_devices--;
1795 	if (srcdev->missing) {
1796 		fs_info->fs_devices->missing_devices--;
1797 		fs_info->fs_devices->rw_devices++;
1798 	}
1799 	if (srcdev->can_discard)
1800 		fs_info->fs_devices->num_can_discard--;
1801 	if (srcdev->bdev) {
1802 		fs_info->fs_devices->open_devices--;
1803 
1804 		/* zero out the old super */
1805 		btrfs_scratch_superblock(srcdev);
1806 	}
1807 
1808 	call_rcu(&srcdev->rcu, free_device);
1809 }
1810 
1811 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1812 				      struct btrfs_device *tgtdev)
1813 {
1814 	struct btrfs_device *next_device;
1815 
1816 	WARN_ON(!tgtdev);
1817 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
1818 	if (tgtdev->bdev) {
1819 		btrfs_scratch_superblock(tgtdev);
1820 		fs_info->fs_devices->open_devices--;
1821 	}
1822 	fs_info->fs_devices->num_devices--;
1823 	if (tgtdev->can_discard)
1824 		fs_info->fs_devices->num_can_discard++;
1825 
1826 	next_device = list_entry(fs_info->fs_devices->devices.next,
1827 				 struct btrfs_device, dev_list);
1828 	if (tgtdev->bdev == fs_info->sb->s_bdev)
1829 		fs_info->sb->s_bdev = next_device->bdev;
1830 	if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1831 		fs_info->fs_devices->latest_bdev = next_device->bdev;
1832 	list_del_rcu(&tgtdev->dev_list);
1833 
1834 	call_rcu(&tgtdev->rcu, free_device);
1835 
1836 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1837 }
1838 
1839 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1840 				     struct btrfs_device **device)
1841 {
1842 	int ret = 0;
1843 	struct btrfs_super_block *disk_super;
1844 	u64 devid;
1845 	u8 *dev_uuid;
1846 	struct block_device *bdev;
1847 	struct buffer_head *bh;
1848 
1849 	*device = NULL;
1850 	ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1851 				    root->fs_info->bdev_holder, 0, &bdev, &bh);
1852 	if (ret)
1853 		return ret;
1854 	disk_super = (struct btrfs_super_block *)bh->b_data;
1855 	devid = btrfs_stack_device_id(&disk_super->dev_item);
1856 	dev_uuid = disk_super->dev_item.uuid;
1857 	*device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1858 				    disk_super->fsid);
1859 	brelse(bh);
1860 	if (!*device)
1861 		ret = -ENOENT;
1862 	blkdev_put(bdev, FMODE_READ);
1863 	return ret;
1864 }
1865 
1866 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1867 					 char *device_path,
1868 					 struct btrfs_device **device)
1869 {
1870 	*device = NULL;
1871 	if (strcmp(device_path, "missing") == 0) {
1872 		struct list_head *devices;
1873 		struct btrfs_device *tmp;
1874 
1875 		devices = &root->fs_info->fs_devices->devices;
1876 		/*
1877 		 * It is safe to read the devices since the volume_mutex
1878 		 * is held by the caller.
1879 		 */
1880 		list_for_each_entry(tmp, devices, dev_list) {
1881 			if (tmp->in_fs_metadata && !tmp->bdev) {
1882 				*device = tmp;
1883 				break;
1884 			}
1885 		}
1886 
1887 		if (!*device) {
1888 			btrfs_err(root->fs_info, "no missing device found");
1889 			return -ENOENT;
1890 		}
1891 
1892 		return 0;
1893 	} else {
1894 		return btrfs_find_device_by_path(root, device_path, device);
1895 	}
1896 }
1897 
1898 /*
1899  * does all the dirty work required for changing file system's UUID.
1900  */
1901 static int btrfs_prepare_sprout(struct btrfs_root *root)
1902 {
1903 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1904 	struct btrfs_fs_devices *old_devices;
1905 	struct btrfs_fs_devices *seed_devices;
1906 	struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1907 	struct btrfs_device *device;
1908 	u64 super_flags;
1909 
1910 	BUG_ON(!mutex_is_locked(&uuid_mutex));
1911 	if (!fs_devices->seeding)
1912 		return -EINVAL;
1913 
1914 	seed_devices = __alloc_fs_devices();
1915 	if (IS_ERR(seed_devices))
1916 		return PTR_ERR(seed_devices);
1917 
1918 	old_devices = clone_fs_devices(fs_devices);
1919 	if (IS_ERR(old_devices)) {
1920 		kfree(seed_devices);
1921 		return PTR_ERR(old_devices);
1922 	}
1923 
1924 	list_add(&old_devices->list, &fs_uuids);
1925 
1926 	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1927 	seed_devices->opened = 1;
1928 	INIT_LIST_HEAD(&seed_devices->devices);
1929 	INIT_LIST_HEAD(&seed_devices->alloc_list);
1930 	mutex_init(&seed_devices->device_list_mutex);
1931 
1932 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1933 	list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1934 			      synchronize_rcu);
1935 
1936 	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1937 	list_for_each_entry(device, &seed_devices->devices, dev_list) {
1938 		device->fs_devices = seed_devices;
1939 	}
1940 
1941 	fs_devices->seeding = 0;
1942 	fs_devices->num_devices = 0;
1943 	fs_devices->open_devices = 0;
1944 	fs_devices->seed = seed_devices;
1945 
1946 	generate_random_uuid(fs_devices->fsid);
1947 	memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1948 	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1949 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1950 
1951 	super_flags = btrfs_super_flags(disk_super) &
1952 		      ~BTRFS_SUPER_FLAG_SEEDING;
1953 	btrfs_set_super_flags(disk_super, super_flags);
1954 
1955 	return 0;
1956 }
1957 
1958 /*
1959  * strore the expected generation for seed devices in device items.
1960  */
1961 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1962 			       struct btrfs_root *root)
1963 {
1964 	struct btrfs_path *path;
1965 	struct extent_buffer *leaf;
1966 	struct btrfs_dev_item *dev_item;
1967 	struct btrfs_device *device;
1968 	struct btrfs_key key;
1969 	u8 fs_uuid[BTRFS_UUID_SIZE];
1970 	u8 dev_uuid[BTRFS_UUID_SIZE];
1971 	u64 devid;
1972 	int ret;
1973 
1974 	path = btrfs_alloc_path();
1975 	if (!path)
1976 		return -ENOMEM;
1977 
1978 	root = root->fs_info->chunk_root;
1979 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1980 	key.offset = 0;
1981 	key.type = BTRFS_DEV_ITEM_KEY;
1982 
1983 	while (1) {
1984 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1985 		if (ret < 0)
1986 			goto error;
1987 
1988 		leaf = path->nodes[0];
1989 next_slot:
1990 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1991 			ret = btrfs_next_leaf(root, path);
1992 			if (ret > 0)
1993 				break;
1994 			if (ret < 0)
1995 				goto error;
1996 			leaf = path->nodes[0];
1997 			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1998 			btrfs_release_path(path);
1999 			continue;
2000 		}
2001 
2002 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2003 		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2004 		    key.type != BTRFS_DEV_ITEM_KEY)
2005 			break;
2006 
2007 		dev_item = btrfs_item_ptr(leaf, path->slots[0],
2008 					  struct btrfs_dev_item);
2009 		devid = btrfs_device_id(leaf, dev_item);
2010 		read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2011 				   BTRFS_UUID_SIZE);
2012 		read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2013 				   BTRFS_UUID_SIZE);
2014 		device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2015 					   fs_uuid);
2016 		BUG_ON(!device); /* Logic error */
2017 
2018 		if (device->fs_devices->seeding) {
2019 			btrfs_set_device_generation(leaf, dev_item,
2020 						    device->generation);
2021 			btrfs_mark_buffer_dirty(leaf);
2022 		}
2023 
2024 		path->slots[0]++;
2025 		goto next_slot;
2026 	}
2027 	ret = 0;
2028 error:
2029 	btrfs_free_path(path);
2030 	return ret;
2031 }
2032 
2033 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2034 {
2035 	struct request_queue *q;
2036 	struct btrfs_trans_handle *trans;
2037 	struct btrfs_device *device;
2038 	struct block_device *bdev;
2039 	struct list_head *devices;
2040 	struct super_block *sb = root->fs_info->sb;
2041 	struct rcu_string *name;
2042 	u64 total_bytes;
2043 	int seeding_dev = 0;
2044 	int ret = 0;
2045 
2046 	if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2047 		return -EROFS;
2048 
2049 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2050 				  root->fs_info->bdev_holder);
2051 	if (IS_ERR(bdev))
2052 		return PTR_ERR(bdev);
2053 
2054 	if (root->fs_info->fs_devices->seeding) {
2055 		seeding_dev = 1;
2056 		down_write(&sb->s_umount);
2057 		mutex_lock(&uuid_mutex);
2058 	}
2059 
2060 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
2061 
2062 	devices = &root->fs_info->fs_devices->devices;
2063 
2064 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2065 	list_for_each_entry(device, devices, dev_list) {
2066 		if (device->bdev == bdev) {
2067 			ret = -EEXIST;
2068 			mutex_unlock(
2069 				&root->fs_info->fs_devices->device_list_mutex);
2070 			goto error;
2071 		}
2072 	}
2073 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2074 
2075 	device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2076 	if (IS_ERR(device)) {
2077 		/* we can safely leave the fs_devices entry around */
2078 		ret = PTR_ERR(device);
2079 		goto error;
2080 	}
2081 
2082 	name = rcu_string_strdup(device_path, GFP_NOFS);
2083 	if (!name) {
2084 		kfree(device);
2085 		ret = -ENOMEM;
2086 		goto error;
2087 	}
2088 	rcu_assign_pointer(device->name, name);
2089 
2090 	trans = btrfs_start_transaction(root, 0);
2091 	if (IS_ERR(trans)) {
2092 		rcu_string_free(device->name);
2093 		kfree(device);
2094 		ret = PTR_ERR(trans);
2095 		goto error;
2096 	}
2097 
2098 	lock_chunks(root);
2099 
2100 	q = bdev_get_queue(bdev);
2101 	if (blk_queue_discard(q))
2102 		device->can_discard = 1;
2103 	device->writeable = 1;
2104 	device->generation = trans->transid;
2105 	device->io_width = root->sectorsize;
2106 	device->io_align = root->sectorsize;
2107 	device->sector_size = root->sectorsize;
2108 	device->total_bytes = i_size_read(bdev->bd_inode);
2109 	device->disk_total_bytes = device->total_bytes;
2110 	device->dev_root = root->fs_info->dev_root;
2111 	device->bdev = bdev;
2112 	device->in_fs_metadata = 1;
2113 	device->is_tgtdev_for_dev_replace = 0;
2114 	device->mode = FMODE_EXCL;
2115 	device->dev_stats_valid = 1;
2116 	set_blocksize(device->bdev, 4096);
2117 
2118 	if (seeding_dev) {
2119 		sb->s_flags &= ~MS_RDONLY;
2120 		ret = btrfs_prepare_sprout(root);
2121 		BUG_ON(ret); /* -ENOMEM */
2122 	}
2123 
2124 	device->fs_devices = root->fs_info->fs_devices;
2125 
2126 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2127 	list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2128 	list_add(&device->dev_alloc_list,
2129 		 &root->fs_info->fs_devices->alloc_list);
2130 	root->fs_info->fs_devices->num_devices++;
2131 	root->fs_info->fs_devices->open_devices++;
2132 	root->fs_info->fs_devices->rw_devices++;
2133 	root->fs_info->fs_devices->total_devices++;
2134 	if (device->can_discard)
2135 		root->fs_info->fs_devices->num_can_discard++;
2136 	root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2137 
2138 	spin_lock(&root->fs_info->free_chunk_lock);
2139 	root->fs_info->free_chunk_space += device->total_bytes;
2140 	spin_unlock(&root->fs_info->free_chunk_lock);
2141 
2142 	if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2143 		root->fs_info->fs_devices->rotating = 1;
2144 
2145 	total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2146 	btrfs_set_super_total_bytes(root->fs_info->super_copy,
2147 				    total_bytes + device->total_bytes);
2148 
2149 	total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2150 	btrfs_set_super_num_devices(root->fs_info->super_copy,
2151 				    total_bytes + 1);
2152 
2153 	/* add sysfs device entry */
2154 	btrfs_kobj_add_device(root->fs_info, device);
2155 
2156 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2157 
2158 	if (seeding_dev) {
2159 		char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2160 		ret = init_first_rw_device(trans, root, device);
2161 		if (ret) {
2162 			btrfs_abort_transaction(trans, root, ret);
2163 			goto error_trans;
2164 		}
2165 		ret = btrfs_finish_sprout(trans, root);
2166 		if (ret) {
2167 			btrfs_abort_transaction(trans, root, ret);
2168 			goto error_trans;
2169 		}
2170 
2171 		/* Sprouting would change fsid of the mounted root,
2172 		 * so rename the fsid on the sysfs
2173 		 */
2174 		snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2175 						root->fs_info->fsid);
2176 		if (kobject_rename(&root->fs_info->super_kobj, fsid_buf))
2177 			goto error_trans;
2178 	} else {
2179 		ret = btrfs_add_device(trans, root, device);
2180 		if (ret) {
2181 			btrfs_abort_transaction(trans, root, ret);
2182 			goto error_trans;
2183 		}
2184 	}
2185 
2186 	/*
2187 	 * we've got more storage, clear any full flags on the space
2188 	 * infos
2189 	 */
2190 	btrfs_clear_space_info_full(root->fs_info);
2191 
2192 	unlock_chunks(root);
2193 	root->fs_info->num_tolerated_disk_barrier_failures =
2194 		btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2195 	ret = btrfs_commit_transaction(trans, root);
2196 
2197 	if (seeding_dev) {
2198 		mutex_unlock(&uuid_mutex);
2199 		up_write(&sb->s_umount);
2200 
2201 		if (ret) /* transaction commit */
2202 			return ret;
2203 
2204 		ret = btrfs_relocate_sys_chunks(root);
2205 		if (ret < 0)
2206 			btrfs_error(root->fs_info, ret,
2207 				    "Failed to relocate sys chunks after "
2208 				    "device initialization. This can be fixed "
2209 				    "using the \"btrfs balance\" command.");
2210 		trans = btrfs_attach_transaction(root);
2211 		if (IS_ERR(trans)) {
2212 			if (PTR_ERR(trans) == -ENOENT)
2213 				return 0;
2214 			return PTR_ERR(trans);
2215 		}
2216 		ret = btrfs_commit_transaction(trans, root);
2217 	}
2218 
2219 	/* Update ctime/mtime for libblkid */
2220 	update_dev_time(device_path);
2221 	return ret;
2222 
2223 error_trans:
2224 	unlock_chunks(root);
2225 	btrfs_end_transaction(trans, root);
2226 	rcu_string_free(device->name);
2227 	btrfs_kobj_rm_device(root->fs_info, device);
2228 	kfree(device);
2229 error:
2230 	blkdev_put(bdev, FMODE_EXCL);
2231 	if (seeding_dev) {
2232 		mutex_unlock(&uuid_mutex);
2233 		up_write(&sb->s_umount);
2234 	}
2235 	return ret;
2236 }
2237 
2238 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2239 				  struct btrfs_device **device_out)
2240 {
2241 	struct request_queue *q;
2242 	struct btrfs_device *device;
2243 	struct block_device *bdev;
2244 	struct btrfs_fs_info *fs_info = root->fs_info;
2245 	struct list_head *devices;
2246 	struct rcu_string *name;
2247 	u64 devid = BTRFS_DEV_REPLACE_DEVID;
2248 	int ret = 0;
2249 
2250 	*device_out = NULL;
2251 	if (fs_info->fs_devices->seeding)
2252 		return -EINVAL;
2253 
2254 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2255 				  fs_info->bdev_holder);
2256 	if (IS_ERR(bdev))
2257 		return PTR_ERR(bdev);
2258 
2259 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
2260 
2261 	devices = &fs_info->fs_devices->devices;
2262 	list_for_each_entry(device, devices, dev_list) {
2263 		if (device->bdev == bdev) {
2264 			ret = -EEXIST;
2265 			goto error;
2266 		}
2267 	}
2268 
2269 	device = btrfs_alloc_device(NULL, &devid, NULL);
2270 	if (IS_ERR(device)) {
2271 		ret = PTR_ERR(device);
2272 		goto error;
2273 	}
2274 
2275 	name = rcu_string_strdup(device_path, GFP_NOFS);
2276 	if (!name) {
2277 		kfree(device);
2278 		ret = -ENOMEM;
2279 		goto error;
2280 	}
2281 	rcu_assign_pointer(device->name, name);
2282 
2283 	q = bdev_get_queue(bdev);
2284 	if (blk_queue_discard(q))
2285 		device->can_discard = 1;
2286 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2287 	device->writeable = 1;
2288 	device->generation = 0;
2289 	device->io_width = root->sectorsize;
2290 	device->io_align = root->sectorsize;
2291 	device->sector_size = root->sectorsize;
2292 	device->total_bytes = i_size_read(bdev->bd_inode);
2293 	device->disk_total_bytes = device->total_bytes;
2294 	device->dev_root = fs_info->dev_root;
2295 	device->bdev = bdev;
2296 	device->in_fs_metadata = 1;
2297 	device->is_tgtdev_for_dev_replace = 1;
2298 	device->mode = FMODE_EXCL;
2299 	device->dev_stats_valid = 1;
2300 	set_blocksize(device->bdev, 4096);
2301 	device->fs_devices = fs_info->fs_devices;
2302 	list_add(&device->dev_list, &fs_info->fs_devices->devices);
2303 	fs_info->fs_devices->num_devices++;
2304 	fs_info->fs_devices->open_devices++;
2305 	if (device->can_discard)
2306 		fs_info->fs_devices->num_can_discard++;
2307 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2308 
2309 	*device_out = device;
2310 	return ret;
2311 
2312 error:
2313 	blkdev_put(bdev, FMODE_EXCL);
2314 	return ret;
2315 }
2316 
2317 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2318 					      struct btrfs_device *tgtdev)
2319 {
2320 	WARN_ON(fs_info->fs_devices->rw_devices == 0);
2321 	tgtdev->io_width = fs_info->dev_root->sectorsize;
2322 	tgtdev->io_align = fs_info->dev_root->sectorsize;
2323 	tgtdev->sector_size = fs_info->dev_root->sectorsize;
2324 	tgtdev->dev_root = fs_info->dev_root;
2325 	tgtdev->in_fs_metadata = 1;
2326 }
2327 
2328 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2329 					struct btrfs_device *device)
2330 {
2331 	int ret;
2332 	struct btrfs_path *path;
2333 	struct btrfs_root *root;
2334 	struct btrfs_dev_item *dev_item;
2335 	struct extent_buffer *leaf;
2336 	struct btrfs_key key;
2337 
2338 	root = device->dev_root->fs_info->chunk_root;
2339 
2340 	path = btrfs_alloc_path();
2341 	if (!path)
2342 		return -ENOMEM;
2343 
2344 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2345 	key.type = BTRFS_DEV_ITEM_KEY;
2346 	key.offset = device->devid;
2347 
2348 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2349 	if (ret < 0)
2350 		goto out;
2351 
2352 	if (ret > 0) {
2353 		ret = -ENOENT;
2354 		goto out;
2355 	}
2356 
2357 	leaf = path->nodes[0];
2358 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2359 
2360 	btrfs_set_device_id(leaf, dev_item, device->devid);
2361 	btrfs_set_device_type(leaf, dev_item, device->type);
2362 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2363 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2364 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2365 	btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2366 	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2367 	btrfs_mark_buffer_dirty(leaf);
2368 
2369 out:
2370 	btrfs_free_path(path);
2371 	return ret;
2372 }
2373 
2374 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2375 		      struct btrfs_device *device, u64 new_size)
2376 {
2377 	struct btrfs_super_block *super_copy =
2378 		device->dev_root->fs_info->super_copy;
2379 	u64 old_total = btrfs_super_total_bytes(super_copy);
2380 	u64 diff = new_size - device->total_bytes;
2381 
2382 	if (!device->writeable)
2383 		return -EACCES;
2384 	if (new_size <= device->total_bytes ||
2385 	    device->is_tgtdev_for_dev_replace)
2386 		return -EINVAL;
2387 
2388 	btrfs_set_super_total_bytes(super_copy, old_total + diff);
2389 	device->fs_devices->total_rw_bytes += diff;
2390 
2391 	device->total_bytes = new_size;
2392 	device->disk_total_bytes = new_size;
2393 	btrfs_clear_space_info_full(device->dev_root->fs_info);
2394 
2395 	return btrfs_update_device(trans, device);
2396 }
2397 
2398 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2399 		      struct btrfs_device *device, u64 new_size)
2400 {
2401 	int ret;
2402 	lock_chunks(device->dev_root);
2403 	ret = __btrfs_grow_device(trans, device, new_size);
2404 	unlock_chunks(device->dev_root);
2405 	return ret;
2406 }
2407 
2408 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2409 			    struct btrfs_root *root,
2410 			    u64 chunk_tree, u64 chunk_objectid,
2411 			    u64 chunk_offset)
2412 {
2413 	int ret;
2414 	struct btrfs_path *path;
2415 	struct btrfs_key key;
2416 
2417 	root = root->fs_info->chunk_root;
2418 	path = btrfs_alloc_path();
2419 	if (!path)
2420 		return -ENOMEM;
2421 
2422 	key.objectid = chunk_objectid;
2423 	key.offset = chunk_offset;
2424 	key.type = BTRFS_CHUNK_ITEM_KEY;
2425 
2426 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2427 	if (ret < 0)
2428 		goto out;
2429 	else if (ret > 0) { /* Logic error or corruption */
2430 		btrfs_error(root->fs_info, -ENOENT,
2431 			    "Failed lookup while freeing chunk.");
2432 		ret = -ENOENT;
2433 		goto out;
2434 	}
2435 
2436 	ret = btrfs_del_item(trans, root, path);
2437 	if (ret < 0)
2438 		btrfs_error(root->fs_info, ret,
2439 			    "Failed to delete chunk item.");
2440 out:
2441 	btrfs_free_path(path);
2442 	return ret;
2443 }
2444 
2445 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2446 			chunk_offset)
2447 {
2448 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2449 	struct btrfs_disk_key *disk_key;
2450 	struct btrfs_chunk *chunk;
2451 	u8 *ptr;
2452 	int ret = 0;
2453 	u32 num_stripes;
2454 	u32 array_size;
2455 	u32 len = 0;
2456 	u32 cur;
2457 	struct btrfs_key key;
2458 
2459 	array_size = btrfs_super_sys_array_size(super_copy);
2460 
2461 	ptr = super_copy->sys_chunk_array;
2462 	cur = 0;
2463 
2464 	while (cur < array_size) {
2465 		disk_key = (struct btrfs_disk_key *)ptr;
2466 		btrfs_disk_key_to_cpu(&key, disk_key);
2467 
2468 		len = sizeof(*disk_key);
2469 
2470 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2471 			chunk = (struct btrfs_chunk *)(ptr + len);
2472 			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2473 			len += btrfs_chunk_item_size(num_stripes);
2474 		} else {
2475 			ret = -EIO;
2476 			break;
2477 		}
2478 		if (key.objectid == chunk_objectid &&
2479 		    key.offset == chunk_offset) {
2480 			memmove(ptr, ptr + len, array_size - (cur + len));
2481 			array_size -= len;
2482 			btrfs_set_super_sys_array_size(super_copy, array_size);
2483 		} else {
2484 			ptr += len;
2485 			cur += len;
2486 		}
2487 	}
2488 	return ret;
2489 }
2490 
2491 static int btrfs_relocate_chunk(struct btrfs_root *root,
2492 			 u64 chunk_tree, u64 chunk_objectid,
2493 			 u64 chunk_offset)
2494 {
2495 	struct extent_map_tree *em_tree;
2496 	struct btrfs_root *extent_root;
2497 	struct btrfs_trans_handle *trans;
2498 	struct extent_map *em;
2499 	struct map_lookup *map;
2500 	int ret;
2501 	int i;
2502 
2503 	root = root->fs_info->chunk_root;
2504 	extent_root = root->fs_info->extent_root;
2505 	em_tree = &root->fs_info->mapping_tree.map_tree;
2506 
2507 	ret = btrfs_can_relocate(extent_root, chunk_offset);
2508 	if (ret)
2509 		return -ENOSPC;
2510 
2511 	/* step one, relocate all the extents inside this chunk */
2512 	ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2513 	if (ret)
2514 		return ret;
2515 
2516 	trans = btrfs_start_transaction(root, 0);
2517 	if (IS_ERR(trans)) {
2518 		ret = PTR_ERR(trans);
2519 		btrfs_std_error(root->fs_info, ret);
2520 		return ret;
2521 	}
2522 
2523 	lock_chunks(root);
2524 
2525 	/*
2526 	 * step two, delete the device extents and the
2527 	 * chunk tree entries
2528 	 */
2529 	read_lock(&em_tree->lock);
2530 	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2531 	read_unlock(&em_tree->lock);
2532 
2533 	BUG_ON(!em || em->start > chunk_offset ||
2534 	       em->start + em->len < chunk_offset);
2535 	map = (struct map_lookup *)em->bdev;
2536 
2537 	for (i = 0; i < map->num_stripes; i++) {
2538 		ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2539 					    map->stripes[i].physical);
2540 		BUG_ON(ret);
2541 
2542 		if (map->stripes[i].dev) {
2543 			ret = btrfs_update_device(trans, map->stripes[i].dev);
2544 			BUG_ON(ret);
2545 		}
2546 	}
2547 	ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2548 			       chunk_offset);
2549 
2550 	BUG_ON(ret);
2551 
2552 	trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2553 
2554 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2555 		ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2556 		BUG_ON(ret);
2557 	}
2558 
2559 	ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2560 	BUG_ON(ret);
2561 
2562 	write_lock(&em_tree->lock);
2563 	remove_extent_mapping(em_tree, em);
2564 	write_unlock(&em_tree->lock);
2565 
2566 	/* once for the tree */
2567 	free_extent_map(em);
2568 	/* once for us */
2569 	free_extent_map(em);
2570 
2571 	unlock_chunks(root);
2572 	btrfs_end_transaction(trans, root);
2573 	return 0;
2574 }
2575 
2576 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2577 {
2578 	struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2579 	struct btrfs_path *path;
2580 	struct extent_buffer *leaf;
2581 	struct btrfs_chunk *chunk;
2582 	struct btrfs_key key;
2583 	struct btrfs_key found_key;
2584 	u64 chunk_tree = chunk_root->root_key.objectid;
2585 	u64 chunk_type;
2586 	bool retried = false;
2587 	int failed = 0;
2588 	int ret;
2589 
2590 	path = btrfs_alloc_path();
2591 	if (!path)
2592 		return -ENOMEM;
2593 
2594 again:
2595 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2596 	key.offset = (u64)-1;
2597 	key.type = BTRFS_CHUNK_ITEM_KEY;
2598 
2599 	while (1) {
2600 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2601 		if (ret < 0)
2602 			goto error;
2603 		BUG_ON(ret == 0); /* Corruption */
2604 
2605 		ret = btrfs_previous_item(chunk_root, path, key.objectid,
2606 					  key.type);
2607 		if (ret < 0)
2608 			goto error;
2609 		if (ret > 0)
2610 			break;
2611 
2612 		leaf = path->nodes[0];
2613 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2614 
2615 		chunk = btrfs_item_ptr(leaf, path->slots[0],
2616 				       struct btrfs_chunk);
2617 		chunk_type = btrfs_chunk_type(leaf, chunk);
2618 		btrfs_release_path(path);
2619 
2620 		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2621 			ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2622 						   found_key.objectid,
2623 						   found_key.offset);
2624 			if (ret == -ENOSPC)
2625 				failed++;
2626 			else if (ret)
2627 				BUG();
2628 		}
2629 
2630 		if (found_key.offset == 0)
2631 			break;
2632 		key.offset = found_key.offset - 1;
2633 	}
2634 	ret = 0;
2635 	if (failed && !retried) {
2636 		failed = 0;
2637 		retried = true;
2638 		goto again;
2639 	} else if (WARN_ON(failed && retried)) {
2640 		ret = -ENOSPC;
2641 	}
2642 error:
2643 	btrfs_free_path(path);
2644 	return ret;
2645 }
2646 
2647 static int insert_balance_item(struct btrfs_root *root,
2648 			       struct btrfs_balance_control *bctl)
2649 {
2650 	struct btrfs_trans_handle *trans;
2651 	struct btrfs_balance_item *item;
2652 	struct btrfs_disk_balance_args disk_bargs;
2653 	struct btrfs_path *path;
2654 	struct extent_buffer *leaf;
2655 	struct btrfs_key key;
2656 	int ret, err;
2657 
2658 	path = btrfs_alloc_path();
2659 	if (!path)
2660 		return -ENOMEM;
2661 
2662 	trans = btrfs_start_transaction(root, 0);
2663 	if (IS_ERR(trans)) {
2664 		btrfs_free_path(path);
2665 		return PTR_ERR(trans);
2666 	}
2667 
2668 	key.objectid = BTRFS_BALANCE_OBJECTID;
2669 	key.type = BTRFS_BALANCE_ITEM_KEY;
2670 	key.offset = 0;
2671 
2672 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2673 				      sizeof(*item));
2674 	if (ret)
2675 		goto out;
2676 
2677 	leaf = path->nodes[0];
2678 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2679 
2680 	memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2681 
2682 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2683 	btrfs_set_balance_data(leaf, item, &disk_bargs);
2684 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2685 	btrfs_set_balance_meta(leaf, item, &disk_bargs);
2686 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2687 	btrfs_set_balance_sys(leaf, item, &disk_bargs);
2688 
2689 	btrfs_set_balance_flags(leaf, item, bctl->flags);
2690 
2691 	btrfs_mark_buffer_dirty(leaf);
2692 out:
2693 	btrfs_free_path(path);
2694 	err = btrfs_commit_transaction(trans, root);
2695 	if (err && !ret)
2696 		ret = err;
2697 	return ret;
2698 }
2699 
2700 static int del_balance_item(struct btrfs_root *root)
2701 {
2702 	struct btrfs_trans_handle *trans;
2703 	struct btrfs_path *path;
2704 	struct btrfs_key key;
2705 	int ret, err;
2706 
2707 	path = btrfs_alloc_path();
2708 	if (!path)
2709 		return -ENOMEM;
2710 
2711 	trans = btrfs_start_transaction(root, 0);
2712 	if (IS_ERR(trans)) {
2713 		btrfs_free_path(path);
2714 		return PTR_ERR(trans);
2715 	}
2716 
2717 	key.objectid = BTRFS_BALANCE_OBJECTID;
2718 	key.type = BTRFS_BALANCE_ITEM_KEY;
2719 	key.offset = 0;
2720 
2721 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2722 	if (ret < 0)
2723 		goto out;
2724 	if (ret > 0) {
2725 		ret = -ENOENT;
2726 		goto out;
2727 	}
2728 
2729 	ret = btrfs_del_item(trans, root, path);
2730 out:
2731 	btrfs_free_path(path);
2732 	err = btrfs_commit_transaction(trans, root);
2733 	if (err && !ret)
2734 		ret = err;
2735 	return ret;
2736 }
2737 
2738 /*
2739  * This is a heuristic used to reduce the number of chunks balanced on
2740  * resume after balance was interrupted.
2741  */
2742 static void update_balance_args(struct btrfs_balance_control *bctl)
2743 {
2744 	/*
2745 	 * Turn on soft mode for chunk types that were being converted.
2746 	 */
2747 	if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2748 		bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2749 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2750 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2751 	if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2752 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2753 
2754 	/*
2755 	 * Turn on usage filter if is not already used.  The idea is
2756 	 * that chunks that we have already balanced should be
2757 	 * reasonably full.  Don't do it for chunks that are being
2758 	 * converted - that will keep us from relocating unconverted
2759 	 * (albeit full) chunks.
2760 	 */
2761 	if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2762 	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2763 		bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2764 		bctl->data.usage = 90;
2765 	}
2766 	if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2767 	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2768 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2769 		bctl->sys.usage = 90;
2770 	}
2771 	if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2772 	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2773 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2774 		bctl->meta.usage = 90;
2775 	}
2776 }
2777 
2778 /*
2779  * Should be called with both balance and volume mutexes held to
2780  * serialize other volume operations (add_dev/rm_dev/resize) with
2781  * restriper.  Same goes for unset_balance_control.
2782  */
2783 static void set_balance_control(struct btrfs_balance_control *bctl)
2784 {
2785 	struct btrfs_fs_info *fs_info = bctl->fs_info;
2786 
2787 	BUG_ON(fs_info->balance_ctl);
2788 
2789 	spin_lock(&fs_info->balance_lock);
2790 	fs_info->balance_ctl = bctl;
2791 	spin_unlock(&fs_info->balance_lock);
2792 }
2793 
2794 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2795 {
2796 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2797 
2798 	BUG_ON(!fs_info->balance_ctl);
2799 
2800 	spin_lock(&fs_info->balance_lock);
2801 	fs_info->balance_ctl = NULL;
2802 	spin_unlock(&fs_info->balance_lock);
2803 
2804 	kfree(bctl);
2805 }
2806 
2807 /*
2808  * Balance filters.  Return 1 if chunk should be filtered out
2809  * (should not be balanced).
2810  */
2811 static int chunk_profiles_filter(u64 chunk_type,
2812 				 struct btrfs_balance_args *bargs)
2813 {
2814 	chunk_type = chunk_to_extended(chunk_type) &
2815 				BTRFS_EXTENDED_PROFILE_MASK;
2816 
2817 	if (bargs->profiles & chunk_type)
2818 		return 0;
2819 
2820 	return 1;
2821 }
2822 
2823 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2824 			      struct btrfs_balance_args *bargs)
2825 {
2826 	struct btrfs_block_group_cache *cache;
2827 	u64 chunk_used, user_thresh;
2828 	int ret = 1;
2829 
2830 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2831 	chunk_used = btrfs_block_group_used(&cache->item);
2832 
2833 	if (bargs->usage == 0)
2834 		user_thresh = 1;
2835 	else if (bargs->usage > 100)
2836 		user_thresh = cache->key.offset;
2837 	else
2838 		user_thresh = div_factor_fine(cache->key.offset,
2839 					      bargs->usage);
2840 
2841 	if (chunk_used < user_thresh)
2842 		ret = 0;
2843 
2844 	btrfs_put_block_group(cache);
2845 	return ret;
2846 }
2847 
2848 static int chunk_devid_filter(struct extent_buffer *leaf,
2849 			      struct btrfs_chunk *chunk,
2850 			      struct btrfs_balance_args *bargs)
2851 {
2852 	struct btrfs_stripe *stripe;
2853 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2854 	int i;
2855 
2856 	for (i = 0; i < num_stripes; i++) {
2857 		stripe = btrfs_stripe_nr(chunk, i);
2858 		if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2859 			return 0;
2860 	}
2861 
2862 	return 1;
2863 }
2864 
2865 /* [pstart, pend) */
2866 static int chunk_drange_filter(struct extent_buffer *leaf,
2867 			       struct btrfs_chunk *chunk,
2868 			       u64 chunk_offset,
2869 			       struct btrfs_balance_args *bargs)
2870 {
2871 	struct btrfs_stripe *stripe;
2872 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2873 	u64 stripe_offset;
2874 	u64 stripe_length;
2875 	int factor;
2876 	int i;
2877 
2878 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2879 		return 0;
2880 
2881 	if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2882 	     BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2883 		factor = num_stripes / 2;
2884 	} else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2885 		factor = num_stripes - 1;
2886 	} else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2887 		factor = num_stripes - 2;
2888 	} else {
2889 		factor = num_stripes;
2890 	}
2891 
2892 	for (i = 0; i < num_stripes; i++) {
2893 		stripe = btrfs_stripe_nr(chunk, i);
2894 		if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2895 			continue;
2896 
2897 		stripe_offset = btrfs_stripe_offset(leaf, stripe);
2898 		stripe_length = btrfs_chunk_length(leaf, chunk);
2899 		do_div(stripe_length, factor);
2900 
2901 		if (stripe_offset < bargs->pend &&
2902 		    stripe_offset + stripe_length > bargs->pstart)
2903 			return 0;
2904 	}
2905 
2906 	return 1;
2907 }
2908 
2909 /* [vstart, vend) */
2910 static int chunk_vrange_filter(struct extent_buffer *leaf,
2911 			       struct btrfs_chunk *chunk,
2912 			       u64 chunk_offset,
2913 			       struct btrfs_balance_args *bargs)
2914 {
2915 	if (chunk_offset < bargs->vend &&
2916 	    chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2917 		/* at least part of the chunk is inside this vrange */
2918 		return 0;
2919 
2920 	return 1;
2921 }
2922 
2923 static int chunk_soft_convert_filter(u64 chunk_type,
2924 				     struct btrfs_balance_args *bargs)
2925 {
2926 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2927 		return 0;
2928 
2929 	chunk_type = chunk_to_extended(chunk_type) &
2930 				BTRFS_EXTENDED_PROFILE_MASK;
2931 
2932 	if (bargs->target == chunk_type)
2933 		return 1;
2934 
2935 	return 0;
2936 }
2937 
2938 static int should_balance_chunk(struct btrfs_root *root,
2939 				struct extent_buffer *leaf,
2940 				struct btrfs_chunk *chunk, u64 chunk_offset)
2941 {
2942 	struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2943 	struct btrfs_balance_args *bargs = NULL;
2944 	u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2945 
2946 	/* type filter */
2947 	if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2948 	      (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2949 		return 0;
2950 	}
2951 
2952 	if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2953 		bargs = &bctl->data;
2954 	else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2955 		bargs = &bctl->sys;
2956 	else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2957 		bargs = &bctl->meta;
2958 
2959 	/* profiles filter */
2960 	if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2961 	    chunk_profiles_filter(chunk_type, bargs)) {
2962 		return 0;
2963 	}
2964 
2965 	/* usage filter */
2966 	if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2967 	    chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2968 		return 0;
2969 	}
2970 
2971 	/* devid filter */
2972 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2973 	    chunk_devid_filter(leaf, chunk, bargs)) {
2974 		return 0;
2975 	}
2976 
2977 	/* drange filter, makes sense only with devid filter */
2978 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2979 	    chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2980 		return 0;
2981 	}
2982 
2983 	/* vrange filter */
2984 	if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2985 	    chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2986 		return 0;
2987 	}
2988 
2989 	/* soft profile changing mode */
2990 	if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2991 	    chunk_soft_convert_filter(chunk_type, bargs)) {
2992 		return 0;
2993 	}
2994 
2995 	/*
2996 	 * limited by count, must be the last filter
2997 	 */
2998 	if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
2999 		if (bargs->limit == 0)
3000 			return 0;
3001 		else
3002 			bargs->limit--;
3003 	}
3004 
3005 	return 1;
3006 }
3007 
3008 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3009 {
3010 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3011 	struct btrfs_root *chunk_root = fs_info->chunk_root;
3012 	struct btrfs_root *dev_root = fs_info->dev_root;
3013 	struct list_head *devices;
3014 	struct btrfs_device *device;
3015 	u64 old_size;
3016 	u64 size_to_free;
3017 	struct btrfs_chunk *chunk;
3018 	struct btrfs_path *path;
3019 	struct btrfs_key key;
3020 	struct btrfs_key found_key;
3021 	struct btrfs_trans_handle *trans;
3022 	struct extent_buffer *leaf;
3023 	int slot;
3024 	int ret;
3025 	int enospc_errors = 0;
3026 	bool counting = true;
3027 	u64 limit_data = bctl->data.limit;
3028 	u64 limit_meta = bctl->meta.limit;
3029 	u64 limit_sys = bctl->sys.limit;
3030 
3031 	/* step one make some room on all the devices */
3032 	devices = &fs_info->fs_devices->devices;
3033 	list_for_each_entry(device, devices, dev_list) {
3034 		old_size = device->total_bytes;
3035 		size_to_free = div_factor(old_size, 1);
3036 		size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3037 		if (!device->writeable ||
3038 		    device->total_bytes - device->bytes_used > size_to_free ||
3039 		    device->is_tgtdev_for_dev_replace)
3040 			continue;
3041 
3042 		ret = btrfs_shrink_device(device, old_size - size_to_free);
3043 		if (ret == -ENOSPC)
3044 			break;
3045 		BUG_ON(ret);
3046 
3047 		trans = btrfs_start_transaction(dev_root, 0);
3048 		BUG_ON(IS_ERR(trans));
3049 
3050 		ret = btrfs_grow_device(trans, device, old_size);
3051 		BUG_ON(ret);
3052 
3053 		btrfs_end_transaction(trans, dev_root);
3054 	}
3055 
3056 	/* step two, relocate all the chunks */
3057 	path = btrfs_alloc_path();
3058 	if (!path) {
3059 		ret = -ENOMEM;
3060 		goto error;
3061 	}
3062 
3063 	/* zero out stat counters */
3064 	spin_lock(&fs_info->balance_lock);
3065 	memset(&bctl->stat, 0, sizeof(bctl->stat));
3066 	spin_unlock(&fs_info->balance_lock);
3067 again:
3068 	if (!counting) {
3069 		bctl->data.limit = limit_data;
3070 		bctl->meta.limit = limit_meta;
3071 		bctl->sys.limit = limit_sys;
3072 	}
3073 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3074 	key.offset = (u64)-1;
3075 	key.type = BTRFS_CHUNK_ITEM_KEY;
3076 
3077 	while (1) {
3078 		if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3079 		    atomic_read(&fs_info->balance_cancel_req)) {
3080 			ret = -ECANCELED;
3081 			goto error;
3082 		}
3083 
3084 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3085 		if (ret < 0)
3086 			goto error;
3087 
3088 		/*
3089 		 * this shouldn't happen, it means the last relocate
3090 		 * failed
3091 		 */
3092 		if (ret == 0)
3093 			BUG(); /* FIXME break ? */
3094 
3095 		ret = btrfs_previous_item(chunk_root, path, 0,
3096 					  BTRFS_CHUNK_ITEM_KEY);
3097 		if (ret) {
3098 			ret = 0;
3099 			break;
3100 		}
3101 
3102 		leaf = path->nodes[0];
3103 		slot = path->slots[0];
3104 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3105 
3106 		if (found_key.objectid != key.objectid)
3107 			break;
3108 
3109 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3110 
3111 		if (!counting) {
3112 			spin_lock(&fs_info->balance_lock);
3113 			bctl->stat.considered++;
3114 			spin_unlock(&fs_info->balance_lock);
3115 		}
3116 
3117 		ret = should_balance_chunk(chunk_root, leaf, chunk,
3118 					   found_key.offset);
3119 		btrfs_release_path(path);
3120 		if (!ret)
3121 			goto loop;
3122 
3123 		if (counting) {
3124 			spin_lock(&fs_info->balance_lock);
3125 			bctl->stat.expected++;
3126 			spin_unlock(&fs_info->balance_lock);
3127 			goto loop;
3128 		}
3129 
3130 		ret = btrfs_relocate_chunk(chunk_root,
3131 					   chunk_root->root_key.objectid,
3132 					   found_key.objectid,
3133 					   found_key.offset);
3134 		if (ret && ret != -ENOSPC)
3135 			goto error;
3136 		if (ret == -ENOSPC) {
3137 			enospc_errors++;
3138 		} else {
3139 			spin_lock(&fs_info->balance_lock);
3140 			bctl->stat.completed++;
3141 			spin_unlock(&fs_info->balance_lock);
3142 		}
3143 loop:
3144 		if (found_key.offset == 0)
3145 			break;
3146 		key.offset = found_key.offset - 1;
3147 	}
3148 
3149 	if (counting) {
3150 		btrfs_release_path(path);
3151 		counting = false;
3152 		goto again;
3153 	}
3154 error:
3155 	btrfs_free_path(path);
3156 	if (enospc_errors) {
3157 		btrfs_info(fs_info, "%d enospc errors during balance",
3158 		       enospc_errors);
3159 		if (!ret)
3160 			ret = -ENOSPC;
3161 	}
3162 
3163 	return ret;
3164 }
3165 
3166 /**
3167  * alloc_profile_is_valid - see if a given profile is valid and reduced
3168  * @flags: profile to validate
3169  * @extended: if true @flags is treated as an extended profile
3170  */
3171 static int alloc_profile_is_valid(u64 flags, int extended)
3172 {
3173 	u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3174 			       BTRFS_BLOCK_GROUP_PROFILE_MASK);
3175 
3176 	flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3177 
3178 	/* 1) check that all other bits are zeroed */
3179 	if (flags & ~mask)
3180 		return 0;
3181 
3182 	/* 2) see if profile is reduced */
3183 	if (flags == 0)
3184 		return !extended; /* "0" is valid for usual profiles */
3185 
3186 	/* true if exactly one bit set */
3187 	return (flags & (flags - 1)) == 0;
3188 }
3189 
3190 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3191 {
3192 	/* cancel requested || normal exit path */
3193 	return atomic_read(&fs_info->balance_cancel_req) ||
3194 		(atomic_read(&fs_info->balance_pause_req) == 0 &&
3195 		 atomic_read(&fs_info->balance_cancel_req) == 0);
3196 }
3197 
3198 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3199 {
3200 	int ret;
3201 
3202 	unset_balance_control(fs_info);
3203 	ret = del_balance_item(fs_info->tree_root);
3204 	if (ret)
3205 		btrfs_std_error(fs_info, ret);
3206 
3207 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3208 }
3209 
3210 /*
3211  * Should be called with both balance and volume mutexes held
3212  */
3213 int btrfs_balance(struct btrfs_balance_control *bctl,
3214 		  struct btrfs_ioctl_balance_args *bargs)
3215 {
3216 	struct btrfs_fs_info *fs_info = bctl->fs_info;
3217 	u64 allowed;
3218 	int mixed = 0;
3219 	int ret;
3220 	u64 num_devices;
3221 	unsigned seq;
3222 
3223 	if (btrfs_fs_closing(fs_info) ||
3224 	    atomic_read(&fs_info->balance_pause_req) ||
3225 	    atomic_read(&fs_info->balance_cancel_req)) {
3226 		ret = -EINVAL;
3227 		goto out;
3228 	}
3229 
3230 	allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3231 	if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3232 		mixed = 1;
3233 
3234 	/*
3235 	 * In case of mixed groups both data and meta should be picked,
3236 	 * and identical options should be given for both of them.
3237 	 */
3238 	allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3239 	if (mixed && (bctl->flags & allowed)) {
3240 		if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3241 		    !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3242 		    memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3243 			btrfs_err(fs_info, "with mixed groups data and "
3244 				   "metadata balance options must be the same");
3245 			ret = -EINVAL;
3246 			goto out;
3247 		}
3248 	}
3249 
3250 	num_devices = fs_info->fs_devices->num_devices;
3251 	btrfs_dev_replace_lock(&fs_info->dev_replace);
3252 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3253 		BUG_ON(num_devices < 1);
3254 		num_devices--;
3255 	}
3256 	btrfs_dev_replace_unlock(&fs_info->dev_replace);
3257 	allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3258 	if (num_devices == 1)
3259 		allowed |= BTRFS_BLOCK_GROUP_DUP;
3260 	else if (num_devices > 1)
3261 		allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3262 	if (num_devices > 2)
3263 		allowed |= BTRFS_BLOCK_GROUP_RAID5;
3264 	if (num_devices > 3)
3265 		allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3266 			    BTRFS_BLOCK_GROUP_RAID6);
3267 	if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3268 	    (!alloc_profile_is_valid(bctl->data.target, 1) ||
3269 	     (bctl->data.target & ~allowed))) {
3270 		btrfs_err(fs_info, "unable to start balance with target "
3271 			   "data profile %llu",
3272 		       bctl->data.target);
3273 		ret = -EINVAL;
3274 		goto out;
3275 	}
3276 	if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3277 	    (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3278 	     (bctl->meta.target & ~allowed))) {
3279 		btrfs_err(fs_info,
3280 			   "unable to start balance with target metadata profile %llu",
3281 		       bctl->meta.target);
3282 		ret = -EINVAL;
3283 		goto out;
3284 	}
3285 	if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3286 	    (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3287 	     (bctl->sys.target & ~allowed))) {
3288 		btrfs_err(fs_info,
3289 			   "unable to start balance with target system profile %llu",
3290 		       bctl->sys.target);
3291 		ret = -EINVAL;
3292 		goto out;
3293 	}
3294 
3295 	/* allow dup'ed data chunks only in mixed mode */
3296 	if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3297 	    (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3298 		btrfs_err(fs_info, "dup for data is not allowed");
3299 		ret = -EINVAL;
3300 		goto out;
3301 	}
3302 
3303 	/* allow to reduce meta or sys integrity only if force set */
3304 	allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3305 			BTRFS_BLOCK_GROUP_RAID10 |
3306 			BTRFS_BLOCK_GROUP_RAID5 |
3307 			BTRFS_BLOCK_GROUP_RAID6;
3308 	do {
3309 		seq = read_seqbegin(&fs_info->profiles_lock);
3310 
3311 		if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3312 		     (fs_info->avail_system_alloc_bits & allowed) &&
3313 		     !(bctl->sys.target & allowed)) ||
3314 		    ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3315 		     (fs_info->avail_metadata_alloc_bits & allowed) &&
3316 		     !(bctl->meta.target & allowed))) {
3317 			if (bctl->flags & BTRFS_BALANCE_FORCE) {
3318 				btrfs_info(fs_info, "force reducing metadata integrity");
3319 			} else {
3320 				btrfs_err(fs_info, "balance will reduce metadata "
3321 					   "integrity, use force if you want this");
3322 				ret = -EINVAL;
3323 				goto out;
3324 			}
3325 		}
3326 	} while (read_seqretry(&fs_info->profiles_lock, seq));
3327 
3328 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3329 		int num_tolerated_disk_barrier_failures;
3330 		u64 target = bctl->sys.target;
3331 
3332 		num_tolerated_disk_barrier_failures =
3333 			btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3334 		if (num_tolerated_disk_barrier_failures > 0 &&
3335 		    (target &
3336 		     (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3337 		      BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3338 			num_tolerated_disk_barrier_failures = 0;
3339 		else if (num_tolerated_disk_barrier_failures > 1 &&
3340 			 (target &
3341 			  (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3342 			num_tolerated_disk_barrier_failures = 1;
3343 
3344 		fs_info->num_tolerated_disk_barrier_failures =
3345 			num_tolerated_disk_barrier_failures;
3346 	}
3347 
3348 	ret = insert_balance_item(fs_info->tree_root, bctl);
3349 	if (ret && ret != -EEXIST)
3350 		goto out;
3351 
3352 	if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3353 		BUG_ON(ret == -EEXIST);
3354 		set_balance_control(bctl);
3355 	} else {
3356 		BUG_ON(ret != -EEXIST);
3357 		spin_lock(&fs_info->balance_lock);
3358 		update_balance_args(bctl);
3359 		spin_unlock(&fs_info->balance_lock);
3360 	}
3361 
3362 	atomic_inc(&fs_info->balance_running);
3363 	mutex_unlock(&fs_info->balance_mutex);
3364 
3365 	ret = __btrfs_balance(fs_info);
3366 
3367 	mutex_lock(&fs_info->balance_mutex);
3368 	atomic_dec(&fs_info->balance_running);
3369 
3370 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3371 		fs_info->num_tolerated_disk_barrier_failures =
3372 			btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3373 	}
3374 
3375 	if (bargs) {
3376 		memset(bargs, 0, sizeof(*bargs));
3377 		update_ioctl_balance_args(fs_info, 0, bargs);
3378 	}
3379 
3380 	if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3381 	    balance_need_close(fs_info)) {
3382 		__cancel_balance(fs_info);
3383 	}
3384 
3385 	wake_up(&fs_info->balance_wait_q);
3386 
3387 	return ret;
3388 out:
3389 	if (bctl->flags & BTRFS_BALANCE_RESUME)
3390 		__cancel_balance(fs_info);
3391 	else {
3392 		kfree(bctl);
3393 		atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3394 	}
3395 	return ret;
3396 }
3397 
3398 static int balance_kthread(void *data)
3399 {
3400 	struct btrfs_fs_info *fs_info = data;
3401 	int ret = 0;
3402 
3403 	mutex_lock(&fs_info->volume_mutex);
3404 	mutex_lock(&fs_info->balance_mutex);
3405 
3406 	if (fs_info->balance_ctl) {
3407 		btrfs_info(fs_info, "continuing balance");
3408 		ret = btrfs_balance(fs_info->balance_ctl, NULL);
3409 	}
3410 
3411 	mutex_unlock(&fs_info->balance_mutex);
3412 	mutex_unlock(&fs_info->volume_mutex);
3413 
3414 	return ret;
3415 }
3416 
3417 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3418 {
3419 	struct task_struct *tsk;
3420 
3421 	spin_lock(&fs_info->balance_lock);
3422 	if (!fs_info->balance_ctl) {
3423 		spin_unlock(&fs_info->balance_lock);
3424 		return 0;
3425 	}
3426 	spin_unlock(&fs_info->balance_lock);
3427 
3428 	if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3429 		btrfs_info(fs_info, "force skipping balance");
3430 		return 0;
3431 	}
3432 
3433 	tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3434 	return PTR_ERR_OR_ZERO(tsk);
3435 }
3436 
3437 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3438 {
3439 	struct btrfs_balance_control *bctl;
3440 	struct btrfs_balance_item *item;
3441 	struct btrfs_disk_balance_args disk_bargs;
3442 	struct btrfs_path *path;
3443 	struct extent_buffer *leaf;
3444 	struct btrfs_key key;
3445 	int ret;
3446 
3447 	path = btrfs_alloc_path();
3448 	if (!path)
3449 		return -ENOMEM;
3450 
3451 	key.objectid = BTRFS_BALANCE_OBJECTID;
3452 	key.type = BTRFS_BALANCE_ITEM_KEY;
3453 	key.offset = 0;
3454 
3455 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3456 	if (ret < 0)
3457 		goto out;
3458 	if (ret > 0) { /* ret = -ENOENT; */
3459 		ret = 0;
3460 		goto out;
3461 	}
3462 
3463 	bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3464 	if (!bctl) {
3465 		ret = -ENOMEM;
3466 		goto out;
3467 	}
3468 
3469 	leaf = path->nodes[0];
3470 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3471 
3472 	bctl->fs_info = fs_info;
3473 	bctl->flags = btrfs_balance_flags(leaf, item);
3474 	bctl->flags |= BTRFS_BALANCE_RESUME;
3475 
3476 	btrfs_balance_data(leaf, item, &disk_bargs);
3477 	btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3478 	btrfs_balance_meta(leaf, item, &disk_bargs);
3479 	btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3480 	btrfs_balance_sys(leaf, item, &disk_bargs);
3481 	btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3482 
3483 	WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3484 
3485 	mutex_lock(&fs_info->volume_mutex);
3486 	mutex_lock(&fs_info->balance_mutex);
3487 
3488 	set_balance_control(bctl);
3489 
3490 	mutex_unlock(&fs_info->balance_mutex);
3491 	mutex_unlock(&fs_info->volume_mutex);
3492 out:
3493 	btrfs_free_path(path);
3494 	return ret;
3495 }
3496 
3497 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3498 {
3499 	int ret = 0;
3500 
3501 	mutex_lock(&fs_info->balance_mutex);
3502 	if (!fs_info->balance_ctl) {
3503 		mutex_unlock(&fs_info->balance_mutex);
3504 		return -ENOTCONN;
3505 	}
3506 
3507 	if (atomic_read(&fs_info->balance_running)) {
3508 		atomic_inc(&fs_info->balance_pause_req);
3509 		mutex_unlock(&fs_info->balance_mutex);
3510 
3511 		wait_event(fs_info->balance_wait_q,
3512 			   atomic_read(&fs_info->balance_running) == 0);
3513 
3514 		mutex_lock(&fs_info->balance_mutex);
3515 		/* we are good with balance_ctl ripped off from under us */
3516 		BUG_ON(atomic_read(&fs_info->balance_running));
3517 		atomic_dec(&fs_info->balance_pause_req);
3518 	} else {
3519 		ret = -ENOTCONN;
3520 	}
3521 
3522 	mutex_unlock(&fs_info->balance_mutex);
3523 	return ret;
3524 }
3525 
3526 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3527 {
3528 	if (fs_info->sb->s_flags & MS_RDONLY)
3529 		return -EROFS;
3530 
3531 	mutex_lock(&fs_info->balance_mutex);
3532 	if (!fs_info->balance_ctl) {
3533 		mutex_unlock(&fs_info->balance_mutex);
3534 		return -ENOTCONN;
3535 	}
3536 
3537 	atomic_inc(&fs_info->balance_cancel_req);
3538 	/*
3539 	 * if we are running just wait and return, balance item is
3540 	 * deleted in btrfs_balance in this case
3541 	 */
3542 	if (atomic_read(&fs_info->balance_running)) {
3543 		mutex_unlock(&fs_info->balance_mutex);
3544 		wait_event(fs_info->balance_wait_q,
3545 			   atomic_read(&fs_info->balance_running) == 0);
3546 		mutex_lock(&fs_info->balance_mutex);
3547 	} else {
3548 		/* __cancel_balance needs volume_mutex */
3549 		mutex_unlock(&fs_info->balance_mutex);
3550 		mutex_lock(&fs_info->volume_mutex);
3551 		mutex_lock(&fs_info->balance_mutex);
3552 
3553 		if (fs_info->balance_ctl)
3554 			__cancel_balance(fs_info);
3555 
3556 		mutex_unlock(&fs_info->volume_mutex);
3557 	}
3558 
3559 	BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3560 	atomic_dec(&fs_info->balance_cancel_req);
3561 	mutex_unlock(&fs_info->balance_mutex);
3562 	return 0;
3563 }
3564 
3565 static int btrfs_uuid_scan_kthread(void *data)
3566 {
3567 	struct btrfs_fs_info *fs_info = data;
3568 	struct btrfs_root *root = fs_info->tree_root;
3569 	struct btrfs_key key;
3570 	struct btrfs_key max_key;
3571 	struct btrfs_path *path = NULL;
3572 	int ret = 0;
3573 	struct extent_buffer *eb;
3574 	int slot;
3575 	struct btrfs_root_item root_item;
3576 	u32 item_size;
3577 	struct btrfs_trans_handle *trans = NULL;
3578 
3579 	path = btrfs_alloc_path();
3580 	if (!path) {
3581 		ret = -ENOMEM;
3582 		goto out;
3583 	}
3584 
3585 	key.objectid = 0;
3586 	key.type = BTRFS_ROOT_ITEM_KEY;
3587 	key.offset = 0;
3588 
3589 	max_key.objectid = (u64)-1;
3590 	max_key.type = BTRFS_ROOT_ITEM_KEY;
3591 	max_key.offset = (u64)-1;
3592 
3593 	path->keep_locks = 1;
3594 
3595 	while (1) {
3596 		ret = btrfs_search_forward(root, &key, path, 0);
3597 		if (ret) {
3598 			if (ret > 0)
3599 				ret = 0;
3600 			break;
3601 		}
3602 
3603 		if (key.type != BTRFS_ROOT_ITEM_KEY ||
3604 		    (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3605 		     key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3606 		    key.objectid > BTRFS_LAST_FREE_OBJECTID)
3607 			goto skip;
3608 
3609 		eb = path->nodes[0];
3610 		slot = path->slots[0];
3611 		item_size = btrfs_item_size_nr(eb, slot);
3612 		if (item_size < sizeof(root_item))
3613 			goto skip;
3614 
3615 		read_extent_buffer(eb, &root_item,
3616 				   btrfs_item_ptr_offset(eb, slot),
3617 				   (int)sizeof(root_item));
3618 		if (btrfs_root_refs(&root_item) == 0)
3619 			goto skip;
3620 
3621 		if (!btrfs_is_empty_uuid(root_item.uuid) ||
3622 		    !btrfs_is_empty_uuid(root_item.received_uuid)) {
3623 			if (trans)
3624 				goto update_tree;
3625 
3626 			btrfs_release_path(path);
3627 			/*
3628 			 * 1 - subvol uuid item
3629 			 * 1 - received_subvol uuid item
3630 			 */
3631 			trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3632 			if (IS_ERR(trans)) {
3633 				ret = PTR_ERR(trans);
3634 				break;
3635 			}
3636 			continue;
3637 		} else {
3638 			goto skip;
3639 		}
3640 update_tree:
3641 		if (!btrfs_is_empty_uuid(root_item.uuid)) {
3642 			ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3643 						  root_item.uuid,
3644 						  BTRFS_UUID_KEY_SUBVOL,
3645 						  key.objectid);
3646 			if (ret < 0) {
3647 				btrfs_warn(fs_info, "uuid_tree_add failed %d",
3648 					ret);
3649 				break;
3650 			}
3651 		}
3652 
3653 		if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3654 			ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3655 						  root_item.received_uuid,
3656 						 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3657 						  key.objectid);
3658 			if (ret < 0) {
3659 				btrfs_warn(fs_info, "uuid_tree_add failed %d",
3660 					ret);
3661 				break;
3662 			}
3663 		}
3664 
3665 skip:
3666 		if (trans) {
3667 			ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3668 			trans = NULL;
3669 			if (ret)
3670 				break;
3671 		}
3672 
3673 		btrfs_release_path(path);
3674 		if (key.offset < (u64)-1) {
3675 			key.offset++;
3676 		} else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3677 			key.offset = 0;
3678 			key.type = BTRFS_ROOT_ITEM_KEY;
3679 		} else if (key.objectid < (u64)-1) {
3680 			key.offset = 0;
3681 			key.type = BTRFS_ROOT_ITEM_KEY;
3682 			key.objectid++;
3683 		} else {
3684 			break;
3685 		}
3686 		cond_resched();
3687 	}
3688 
3689 out:
3690 	btrfs_free_path(path);
3691 	if (trans && !IS_ERR(trans))
3692 		btrfs_end_transaction(trans, fs_info->uuid_root);
3693 	if (ret)
3694 		btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3695 	else
3696 		fs_info->update_uuid_tree_gen = 1;
3697 	up(&fs_info->uuid_tree_rescan_sem);
3698 	return 0;
3699 }
3700 
3701 /*
3702  * Callback for btrfs_uuid_tree_iterate().
3703  * returns:
3704  * 0	check succeeded, the entry is not outdated.
3705  * < 0	if an error occured.
3706  * > 0	if the check failed, which means the caller shall remove the entry.
3707  */
3708 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3709 				       u8 *uuid, u8 type, u64 subid)
3710 {
3711 	struct btrfs_key key;
3712 	int ret = 0;
3713 	struct btrfs_root *subvol_root;
3714 
3715 	if (type != BTRFS_UUID_KEY_SUBVOL &&
3716 	    type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3717 		goto out;
3718 
3719 	key.objectid = subid;
3720 	key.type = BTRFS_ROOT_ITEM_KEY;
3721 	key.offset = (u64)-1;
3722 	subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3723 	if (IS_ERR(subvol_root)) {
3724 		ret = PTR_ERR(subvol_root);
3725 		if (ret == -ENOENT)
3726 			ret = 1;
3727 		goto out;
3728 	}
3729 
3730 	switch (type) {
3731 	case BTRFS_UUID_KEY_SUBVOL:
3732 		if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3733 			ret = 1;
3734 		break;
3735 	case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3736 		if (memcmp(uuid, subvol_root->root_item.received_uuid,
3737 			   BTRFS_UUID_SIZE))
3738 			ret = 1;
3739 		break;
3740 	}
3741 
3742 out:
3743 	return ret;
3744 }
3745 
3746 static int btrfs_uuid_rescan_kthread(void *data)
3747 {
3748 	struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3749 	int ret;
3750 
3751 	/*
3752 	 * 1st step is to iterate through the existing UUID tree and
3753 	 * to delete all entries that contain outdated data.
3754 	 * 2nd step is to add all missing entries to the UUID tree.
3755 	 */
3756 	ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3757 	if (ret < 0) {
3758 		btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3759 		up(&fs_info->uuid_tree_rescan_sem);
3760 		return ret;
3761 	}
3762 	return btrfs_uuid_scan_kthread(data);
3763 }
3764 
3765 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3766 {
3767 	struct btrfs_trans_handle *trans;
3768 	struct btrfs_root *tree_root = fs_info->tree_root;
3769 	struct btrfs_root *uuid_root;
3770 	struct task_struct *task;
3771 	int ret;
3772 
3773 	/*
3774 	 * 1 - root node
3775 	 * 1 - root item
3776 	 */
3777 	trans = btrfs_start_transaction(tree_root, 2);
3778 	if (IS_ERR(trans))
3779 		return PTR_ERR(trans);
3780 
3781 	uuid_root = btrfs_create_tree(trans, fs_info,
3782 				      BTRFS_UUID_TREE_OBJECTID);
3783 	if (IS_ERR(uuid_root)) {
3784 		btrfs_abort_transaction(trans, tree_root,
3785 					PTR_ERR(uuid_root));
3786 		return PTR_ERR(uuid_root);
3787 	}
3788 
3789 	fs_info->uuid_root = uuid_root;
3790 
3791 	ret = btrfs_commit_transaction(trans, tree_root);
3792 	if (ret)
3793 		return ret;
3794 
3795 	down(&fs_info->uuid_tree_rescan_sem);
3796 	task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3797 	if (IS_ERR(task)) {
3798 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
3799 		btrfs_warn(fs_info, "failed to start uuid_scan task");
3800 		up(&fs_info->uuid_tree_rescan_sem);
3801 		return PTR_ERR(task);
3802 	}
3803 
3804 	return 0;
3805 }
3806 
3807 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3808 {
3809 	struct task_struct *task;
3810 
3811 	down(&fs_info->uuid_tree_rescan_sem);
3812 	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3813 	if (IS_ERR(task)) {
3814 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
3815 		btrfs_warn(fs_info, "failed to start uuid_rescan task");
3816 		up(&fs_info->uuid_tree_rescan_sem);
3817 		return PTR_ERR(task);
3818 	}
3819 
3820 	return 0;
3821 }
3822 
3823 /*
3824  * shrinking a device means finding all of the device extents past
3825  * the new size, and then following the back refs to the chunks.
3826  * The chunk relocation code actually frees the device extent
3827  */
3828 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3829 {
3830 	struct btrfs_trans_handle *trans;
3831 	struct btrfs_root *root = device->dev_root;
3832 	struct btrfs_dev_extent *dev_extent = NULL;
3833 	struct btrfs_path *path;
3834 	u64 length;
3835 	u64 chunk_tree;
3836 	u64 chunk_objectid;
3837 	u64 chunk_offset;
3838 	int ret;
3839 	int slot;
3840 	int failed = 0;
3841 	bool retried = false;
3842 	struct extent_buffer *l;
3843 	struct btrfs_key key;
3844 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3845 	u64 old_total = btrfs_super_total_bytes(super_copy);
3846 	u64 old_size = device->total_bytes;
3847 	u64 diff = device->total_bytes - new_size;
3848 
3849 	if (device->is_tgtdev_for_dev_replace)
3850 		return -EINVAL;
3851 
3852 	path = btrfs_alloc_path();
3853 	if (!path)
3854 		return -ENOMEM;
3855 
3856 	path->reada = 2;
3857 
3858 	lock_chunks(root);
3859 
3860 	device->total_bytes = new_size;
3861 	if (device->writeable) {
3862 		device->fs_devices->total_rw_bytes -= diff;
3863 		spin_lock(&root->fs_info->free_chunk_lock);
3864 		root->fs_info->free_chunk_space -= diff;
3865 		spin_unlock(&root->fs_info->free_chunk_lock);
3866 	}
3867 	unlock_chunks(root);
3868 
3869 again:
3870 	key.objectid = device->devid;
3871 	key.offset = (u64)-1;
3872 	key.type = BTRFS_DEV_EXTENT_KEY;
3873 
3874 	do {
3875 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3876 		if (ret < 0)
3877 			goto done;
3878 
3879 		ret = btrfs_previous_item(root, path, 0, key.type);
3880 		if (ret < 0)
3881 			goto done;
3882 		if (ret) {
3883 			ret = 0;
3884 			btrfs_release_path(path);
3885 			break;
3886 		}
3887 
3888 		l = path->nodes[0];
3889 		slot = path->slots[0];
3890 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3891 
3892 		if (key.objectid != device->devid) {
3893 			btrfs_release_path(path);
3894 			break;
3895 		}
3896 
3897 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3898 		length = btrfs_dev_extent_length(l, dev_extent);
3899 
3900 		if (key.offset + length <= new_size) {
3901 			btrfs_release_path(path);
3902 			break;
3903 		}
3904 
3905 		chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3906 		chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3907 		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3908 		btrfs_release_path(path);
3909 
3910 		ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3911 					   chunk_offset);
3912 		if (ret && ret != -ENOSPC)
3913 			goto done;
3914 		if (ret == -ENOSPC)
3915 			failed++;
3916 	} while (key.offset-- > 0);
3917 
3918 	if (failed && !retried) {
3919 		failed = 0;
3920 		retried = true;
3921 		goto again;
3922 	} else if (failed && retried) {
3923 		ret = -ENOSPC;
3924 		lock_chunks(root);
3925 
3926 		device->total_bytes = old_size;
3927 		if (device->writeable)
3928 			device->fs_devices->total_rw_bytes += diff;
3929 		spin_lock(&root->fs_info->free_chunk_lock);
3930 		root->fs_info->free_chunk_space += diff;
3931 		spin_unlock(&root->fs_info->free_chunk_lock);
3932 		unlock_chunks(root);
3933 		goto done;
3934 	}
3935 
3936 	/* Shrinking succeeded, else we would be at "done". */
3937 	trans = btrfs_start_transaction(root, 0);
3938 	if (IS_ERR(trans)) {
3939 		ret = PTR_ERR(trans);
3940 		goto done;
3941 	}
3942 
3943 	lock_chunks(root);
3944 
3945 	device->disk_total_bytes = new_size;
3946 	/* Now btrfs_update_device() will change the on-disk size. */
3947 	ret = btrfs_update_device(trans, device);
3948 	if (ret) {
3949 		unlock_chunks(root);
3950 		btrfs_end_transaction(trans, root);
3951 		goto done;
3952 	}
3953 	WARN_ON(diff > old_total);
3954 	btrfs_set_super_total_bytes(super_copy, old_total - diff);
3955 	unlock_chunks(root);
3956 	btrfs_end_transaction(trans, root);
3957 done:
3958 	btrfs_free_path(path);
3959 	return ret;
3960 }
3961 
3962 static int btrfs_add_system_chunk(struct btrfs_root *root,
3963 			   struct btrfs_key *key,
3964 			   struct btrfs_chunk *chunk, int item_size)
3965 {
3966 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3967 	struct btrfs_disk_key disk_key;
3968 	u32 array_size;
3969 	u8 *ptr;
3970 
3971 	array_size = btrfs_super_sys_array_size(super_copy);
3972 	if (array_size + item_size + sizeof(disk_key)
3973 			> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3974 		return -EFBIG;
3975 
3976 	ptr = super_copy->sys_chunk_array + array_size;
3977 	btrfs_cpu_key_to_disk(&disk_key, key);
3978 	memcpy(ptr, &disk_key, sizeof(disk_key));
3979 	ptr += sizeof(disk_key);
3980 	memcpy(ptr, chunk, item_size);
3981 	item_size += sizeof(disk_key);
3982 	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3983 	return 0;
3984 }
3985 
3986 /*
3987  * sort the devices in descending order by max_avail, total_avail
3988  */
3989 static int btrfs_cmp_device_info(const void *a, const void *b)
3990 {
3991 	const struct btrfs_device_info *di_a = a;
3992 	const struct btrfs_device_info *di_b = b;
3993 
3994 	if (di_a->max_avail > di_b->max_avail)
3995 		return -1;
3996 	if (di_a->max_avail < di_b->max_avail)
3997 		return 1;
3998 	if (di_a->total_avail > di_b->total_avail)
3999 		return -1;
4000 	if (di_a->total_avail < di_b->total_avail)
4001 		return 1;
4002 	return 0;
4003 }
4004 
4005 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
4006 	[BTRFS_RAID_RAID10] = {
4007 		.sub_stripes	= 2,
4008 		.dev_stripes	= 1,
4009 		.devs_max	= 0,	/* 0 == as many as possible */
4010 		.devs_min	= 4,
4011 		.devs_increment	= 2,
4012 		.ncopies	= 2,
4013 	},
4014 	[BTRFS_RAID_RAID1] = {
4015 		.sub_stripes	= 1,
4016 		.dev_stripes	= 1,
4017 		.devs_max	= 2,
4018 		.devs_min	= 2,
4019 		.devs_increment	= 2,
4020 		.ncopies	= 2,
4021 	},
4022 	[BTRFS_RAID_DUP] = {
4023 		.sub_stripes	= 1,
4024 		.dev_stripes	= 2,
4025 		.devs_max	= 1,
4026 		.devs_min	= 1,
4027 		.devs_increment	= 1,
4028 		.ncopies	= 2,
4029 	},
4030 	[BTRFS_RAID_RAID0] = {
4031 		.sub_stripes	= 1,
4032 		.dev_stripes	= 1,
4033 		.devs_max	= 0,
4034 		.devs_min	= 2,
4035 		.devs_increment	= 1,
4036 		.ncopies	= 1,
4037 	},
4038 	[BTRFS_RAID_SINGLE] = {
4039 		.sub_stripes	= 1,
4040 		.dev_stripes	= 1,
4041 		.devs_max	= 1,
4042 		.devs_min	= 1,
4043 		.devs_increment	= 1,
4044 		.ncopies	= 1,
4045 	},
4046 	[BTRFS_RAID_RAID5] = {
4047 		.sub_stripes	= 1,
4048 		.dev_stripes	= 1,
4049 		.devs_max	= 0,
4050 		.devs_min	= 2,
4051 		.devs_increment	= 1,
4052 		.ncopies	= 2,
4053 	},
4054 	[BTRFS_RAID_RAID6] = {
4055 		.sub_stripes	= 1,
4056 		.dev_stripes	= 1,
4057 		.devs_max	= 0,
4058 		.devs_min	= 3,
4059 		.devs_increment	= 1,
4060 		.ncopies	= 3,
4061 	},
4062 };
4063 
4064 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4065 {
4066 	/* TODO allow them to set a preferred stripe size */
4067 	return 64 * 1024;
4068 }
4069 
4070 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4071 {
4072 	if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
4073 		return;
4074 
4075 	btrfs_set_fs_incompat(info, RAID56);
4076 }
4077 
4078 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)		\
4079 			- sizeof(struct btrfs_item)		\
4080 			- sizeof(struct btrfs_chunk))		\
4081 			/ sizeof(struct btrfs_stripe) + 1)
4082 
4083 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
4084 				- 2 * sizeof(struct btrfs_disk_key)	\
4085 				- 2 * sizeof(struct btrfs_chunk))	\
4086 				/ sizeof(struct btrfs_stripe) + 1)
4087 
4088 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4089 			       struct btrfs_root *extent_root, u64 start,
4090 			       u64 type)
4091 {
4092 	struct btrfs_fs_info *info = extent_root->fs_info;
4093 	struct btrfs_fs_devices *fs_devices = info->fs_devices;
4094 	struct list_head *cur;
4095 	struct map_lookup *map = NULL;
4096 	struct extent_map_tree *em_tree;
4097 	struct extent_map *em;
4098 	struct btrfs_device_info *devices_info = NULL;
4099 	u64 total_avail;
4100 	int num_stripes;	/* total number of stripes to allocate */
4101 	int data_stripes;	/* number of stripes that count for
4102 				   block group size */
4103 	int sub_stripes;	/* sub_stripes info for map */
4104 	int dev_stripes;	/* stripes per dev */
4105 	int devs_max;		/* max devs to use */
4106 	int devs_min;		/* min devs needed */
4107 	int devs_increment;	/* ndevs has to be a multiple of this */
4108 	int ncopies;		/* how many copies to data has */
4109 	int ret;
4110 	u64 max_stripe_size;
4111 	u64 max_chunk_size;
4112 	u64 stripe_size;
4113 	u64 num_bytes;
4114 	u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4115 	int ndevs;
4116 	int i;
4117 	int j;
4118 	int index;
4119 
4120 	BUG_ON(!alloc_profile_is_valid(type, 0));
4121 
4122 	if (list_empty(&fs_devices->alloc_list))
4123 		return -ENOSPC;
4124 
4125 	index = __get_raid_index(type);
4126 
4127 	sub_stripes = btrfs_raid_array[index].sub_stripes;
4128 	dev_stripes = btrfs_raid_array[index].dev_stripes;
4129 	devs_max = btrfs_raid_array[index].devs_max;
4130 	devs_min = btrfs_raid_array[index].devs_min;
4131 	devs_increment = btrfs_raid_array[index].devs_increment;
4132 	ncopies = btrfs_raid_array[index].ncopies;
4133 
4134 	if (type & BTRFS_BLOCK_GROUP_DATA) {
4135 		max_stripe_size = 1024 * 1024 * 1024;
4136 		max_chunk_size = 10 * max_stripe_size;
4137 		if (!devs_max)
4138 			devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4139 	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4140 		/* for larger filesystems, use larger metadata chunks */
4141 		if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4142 			max_stripe_size = 1024 * 1024 * 1024;
4143 		else
4144 			max_stripe_size = 256 * 1024 * 1024;
4145 		max_chunk_size = max_stripe_size;
4146 		if (!devs_max)
4147 			devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4148 	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4149 		max_stripe_size = 32 * 1024 * 1024;
4150 		max_chunk_size = 2 * max_stripe_size;
4151 		if (!devs_max)
4152 			devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4153 	} else {
4154 		btrfs_err(info, "invalid chunk type 0x%llx requested",
4155 		       type);
4156 		BUG_ON(1);
4157 	}
4158 
4159 	/* we don't want a chunk larger than 10% of writeable space */
4160 	max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4161 			     max_chunk_size);
4162 
4163 	devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4164 			       GFP_NOFS);
4165 	if (!devices_info)
4166 		return -ENOMEM;
4167 
4168 	cur = fs_devices->alloc_list.next;
4169 
4170 	/*
4171 	 * in the first pass through the devices list, we gather information
4172 	 * about the available holes on each device.
4173 	 */
4174 	ndevs = 0;
4175 	while (cur != &fs_devices->alloc_list) {
4176 		struct btrfs_device *device;
4177 		u64 max_avail;
4178 		u64 dev_offset;
4179 
4180 		device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4181 
4182 		cur = cur->next;
4183 
4184 		if (!device->writeable) {
4185 			WARN(1, KERN_ERR
4186 			       "BTRFS: read-only device in alloc_list\n");
4187 			continue;
4188 		}
4189 
4190 		if (!device->in_fs_metadata ||
4191 		    device->is_tgtdev_for_dev_replace)
4192 			continue;
4193 
4194 		if (device->total_bytes > device->bytes_used)
4195 			total_avail = device->total_bytes - device->bytes_used;
4196 		else
4197 			total_avail = 0;
4198 
4199 		/* If there is no space on this device, skip it. */
4200 		if (total_avail == 0)
4201 			continue;
4202 
4203 		ret = find_free_dev_extent(trans, device,
4204 					   max_stripe_size * dev_stripes,
4205 					   &dev_offset, &max_avail);
4206 		if (ret && ret != -ENOSPC)
4207 			goto error;
4208 
4209 		if (ret == 0)
4210 			max_avail = max_stripe_size * dev_stripes;
4211 
4212 		if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4213 			continue;
4214 
4215 		if (ndevs == fs_devices->rw_devices) {
4216 			WARN(1, "%s: found more than %llu devices\n",
4217 			     __func__, fs_devices->rw_devices);
4218 			break;
4219 		}
4220 		devices_info[ndevs].dev_offset = dev_offset;
4221 		devices_info[ndevs].max_avail = max_avail;
4222 		devices_info[ndevs].total_avail = total_avail;
4223 		devices_info[ndevs].dev = device;
4224 		++ndevs;
4225 	}
4226 
4227 	/*
4228 	 * now sort the devices by hole size / available space
4229 	 */
4230 	sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4231 	     btrfs_cmp_device_info, NULL);
4232 
4233 	/* round down to number of usable stripes */
4234 	ndevs -= ndevs % devs_increment;
4235 
4236 	if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4237 		ret = -ENOSPC;
4238 		goto error;
4239 	}
4240 
4241 	if (devs_max && ndevs > devs_max)
4242 		ndevs = devs_max;
4243 	/*
4244 	 * the primary goal is to maximize the number of stripes, so use as many
4245 	 * devices as possible, even if the stripes are not maximum sized.
4246 	 */
4247 	stripe_size = devices_info[ndevs-1].max_avail;
4248 	num_stripes = ndevs * dev_stripes;
4249 
4250 	/*
4251 	 * this will have to be fixed for RAID1 and RAID10 over
4252 	 * more drives
4253 	 */
4254 	data_stripes = num_stripes / ncopies;
4255 
4256 	if (type & BTRFS_BLOCK_GROUP_RAID5) {
4257 		raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4258 				 btrfs_super_stripesize(info->super_copy));
4259 		data_stripes = num_stripes - 1;
4260 	}
4261 	if (type & BTRFS_BLOCK_GROUP_RAID6) {
4262 		raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4263 				 btrfs_super_stripesize(info->super_copy));
4264 		data_stripes = num_stripes - 2;
4265 	}
4266 
4267 	/*
4268 	 * Use the number of data stripes to figure out how big this chunk
4269 	 * is really going to be in terms of logical address space,
4270 	 * and compare that answer with the max chunk size
4271 	 */
4272 	if (stripe_size * data_stripes > max_chunk_size) {
4273 		u64 mask = (1ULL << 24) - 1;
4274 		stripe_size = max_chunk_size;
4275 		do_div(stripe_size, data_stripes);
4276 
4277 		/* bump the answer up to a 16MB boundary */
4278 		stripe_size = (stripe_size + mask) & ~mask;
4279 
4280 		/* but don't go higher than the limits we found
4281 		 * while searching for free extents
4282 		 */
4283 		if (stripe_size > devices_info[ndevs-1].max_avail)
4284 			stripe_size = devices_info[ndevs-1].max_avail;
4285 	}
4286 
4287 	do_div(stripe_size, dev_stripes);
4288 
4289 	/* align to BTRFS_STRIPE_LEN */
4290 	do_div(stripe_size, raid_stripe_len);
4291 	stripe_size *= raid_stripe_len;
4292 
4293 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4294 	if (!map) {
4295 		ret = -ENOMEM;
4296 		goto error;
4297 	}
4298 	map->num_stripes = num_stripes;
4299 
4300 	for (i = 0; i < ndevs; ++i) {
4301 		for (j = 0; j < dev_stripes; ++j) {
4302 			int s = i * dev_stripes + j;
4303 			map->stripes[s].dev = devices_info[i].dev;
4304 			map->stripes[s].physical = devices_info[i].dev_offset +
4305 						   j * stripe_size;
4306 		}
4307 	}
4308 	map->sector_size = extent_root->sectorsize;
4309 	map->stripe_len = raid_stripe_len;
4310 	map->io_align = raid_stripe_len;
4311 	map->io_width = raid_stripe_len;
4312 	map->type = type;
4313 	map->sub_stripes = sub_stripes;
4314 
4315 	num_bytes = stripe_size * data_stripes;
4316 
4317 	trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4318 
4319 	em = alloc_extent_map();
4320 	if (!em) {
4321 		kfree(map);
4322 		ret = -ENOMEM;
4323 		goto error;
4324 	}
4325 	set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4326 	em->bdev = (struct block_device *)map;
4327 	em->start = start;
4328 	em->len = num_bytes;
4329 	em->block_start = 0;
4330 	em->block_len = em->len;
4331 	em->orig_block_len = stripe_size;
4332 
4333 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4334 	write_lock(&em_tree->lock);
4335 	ret = add_extent_mapping(em_tree, em, 0);
4336 	if (!ret) {
4337 		list_add_tail(&em->list, &trans->transaction->pending_chunks);
4338 		atomic_inc(&em->refs);
4339 	}
4340 	write_unlock(&em_tree->lock);
4341 	if (ret) {
4342 		free_extent_map(em);
4343 		goto error;
4344 	}
4345 
4346 	ret = btrfs_make_block_group(trans, extent_root, 0, type,
4347 				     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4348 				     start, num_bytes);
4349 	if (ret)
4350 		goto error_del_extent;
4351 
4352 	free_extent_map(em);
4353 	check_raid56_incompat_flag(extent_root->fs_info, type);
4354 
4355 	kfree(devices_info);
4356 	return 0;
4357 
4358 error_del_extent:
4359 	write_lock(&em_tree->lock);
4360 	remove_extent_mapping(em_tree, em);
4361 	write_unlock(&em_tree->lock);
4362 
4363 	/* One for our allocation */
4364 	free_extent_map(em);
4365 	/* One for the tree reference */
4366 	free_extent_map(em);
4367 error:
4368 	kfree(devices_info);
4369 	return ret;
4370 }
4371 
4372 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4373 				struct btrfs_root *extent_root,
4374 				u64 chunk_offset, u64 chunk_size)
4375 {
4376 	struct btrfs_key key;
4377 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4378 	struct btrfs_device *device;
4379 	struct btrfs_chunk *chunk;
4380 	struct btrfs_stripe *stripe;
4381 	struct extent_map_tree *em_tree;
4382 	struct extent_map *em;
4383 	struct map_lookup *map;
4384 	size_t item_size;
4385 	u64 dev_offset;
4386 	u64 stripe_size;
4387 	int i = 0;
4388 	int ret;
4389 
4390 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4391 	read_lock(&em_tree->lock);
4392 	em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4393 	read_unlock(&em_tree->lock);
4394 
4395 	if (!em) {
4396 		btrfs_crit(extent_root->fs_info, "unable to find logical "
4397 			   "%Lu len %Lu", chunk_offset, chunk_size);
4398 		return -EINVAL;
4399 	}
4400 
4401 	if (em->start != chunk_offset || em->len != chunk_size) {
4402 		btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4403 			  " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4404 			  chunk_size, em->start, em->len);
4405 		free_extent_map(em);
4406 		return -EINVAL;
4407 	}
4408 
4409 	map = (struct map_lookup *)em->bdev;
4410 	item_size = btrfs_chunk_item_size(map->num_stripes);
4411 	stripe_size = em->orig_block_len;
4412 
4413 	chunk = kzalloc(item_size, GFP_NOFS);
4414 	if (!chunk) {
4415 		ret = -ENOMEM;
4416 		goto out;
4417 	}
4418 
4419 	for (i = 0; i < map->num_stripes; i++) {
4420 		device = map->stripes[i].dev;
4421 		dev_offset = map->stripes[i].physical;
4422 
4423 		device->bytes_used += stripe_size;
4424 		ret = btrfs_update_device(trans, device);
4425 		if (ret)
4426 			goto out;
4427 		ret = btrfs_alloc_dev_extent(trans, device,
4428 					     chunk_root->root_key.objectid,
4429 					     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4430 					     chunk_offset, dev_offset,
4431 					     stripe_size);
4432 		if (ret)
4433 			goto out;
4434 	}
4435 
4436 	spin_lock(&extent_root->fs_info->free_chunk_lock);
4437 	extent_root->fs_info->free_chunk_space -= (stripe_size *
4438 						   map->num_stripes);
4439 	spin_unlock(&extent_root->fs_info->free_chunk_lock);
4440 
4441 	stripe = &chunk->stripe;
4442 	for (i = 0; i < map->num_stripes; i++) {
4443 		device = map->stripes[i].dev;
4444 		dev_offset = map->stripes[i].physical;
4445 
4446 		btrfs_set_stack_stripe_devid(stripe, device->devid);
4447 		btrfs_set_stack_stripe_offset(stripe, dev_offset);
4448 		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4449 		stripe++;
4450 	}
4451 
4452 	btrfs_set_stack_chunk_length(chunk, chunk_size);
4453 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4454 	btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4455 	btrfs_set_stack_chunk_type(chunk, map->type);
4456 	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4457 	btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4458 	btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4459 	btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4460 	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4461 
4462 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4463 	key.type = BTRFS_CHUNK_ITEM_KEY;
4464 	key.offset = chunk_offset;
4465 
4466 	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4467 	if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4468 		/*
4469 		 * TODO: Cleanup of inserted chunk root in case of
4470 		 * failure.
4471 		 */
4472 		ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4473 					     item_size);
4474 	}
4475 
4476 out:
4477 	kfree(chunk);
4478 	free_extent_map(em);
4479 	return ret;
4480 }
4481 
4482 /*
4483  * Chunk allocation falls into two parts. The first part does works
4484  * that make the new allocated chunk useable, but not do any operation
4485  * that modifies the chunk tree. The second part does the works that
4486  * require modifying the chunk tree. This division is important for the
4487  * bootstrap process of adding storage to a seed btrfs.
4488  */
4489 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4490 		      struct btrfs_root *extent_root, u64 type)
4491 {
4492 	u64 chunk_offset;
4493 
4494 	chunk_offset = find_next_chunk(extent_root->fs_info);
4495 	return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4496 }
4497 
4498 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4499 					 struct btrfs_root *root,
4500 					 struct btrfs_device *device)
4501 {
4502 	u64 chunk_offset;
4503 	u64 sys_chunk_offset;
4504 	u64 alloc_profile;
4505 	struct btrfs_fs_info *fs_info = root->fs_info;
4506 	struct btrfs_root *extent_root = fs_info->extent_root;
4507 	int ret;
4508 
4509 	chunk_offset = find_next_chunk(fs_info);
4510 	alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4511 	ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4512 				  alloc_profile);
4513 	if (ret)
4514 		return ret;
4515 
4516 	sys_chunk_offset = find_next_chunk(root->fs_info);
4517 	alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4518 	ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4519 				  alloc_profile);
4520 	if (ret) {
4521 		btrfs_abort_transaction(trans, root, ret);
4522 		goto out;
4523 	}
4524 
4525 	ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4526 	if (ret)
4527 		btrfs_abort_transaction(trans, root, ret);
4528 out:
4529 	return ret;
4530 }
4531 
4532 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4533 {
4534 	struct extent_map *em;
4535 	struct map_lookup *map;
4536 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4537 	int readonly = 0;
4538 	int i;
4539 
4540 	read_lock(&map_tree->map_tree.lock);
4541 	em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4542 	read_unlock(&map_tree->map_tree.lock);
4543 	if (!em)
4544 		return 1;
4545 
4546 	if (btrfs_test_opt(root, DEGRADED)) {
4547 		free_extent_map(em);
4548 		return 0;
4549 	}
4550 
4551 	map = (struct map_lookup *)em->bdev;
4552 	for (i = 0; i < map->num_stripes; i++) {
4553 		if (!map->stripes[i].dev->writeable) {
4554 			readonly = 1;
4555 			break;
4556 		}
4557 	}
4558 	free_extent_map(em);
4559 	return readonly;
4560 }
4561 
4562 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4563 {
4564 	extent_map_tree_init(&tree->map_tree);
4565 }
4566 
4567 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4568 {
4569 	struct extent_map *em;
4570 
4571 	while (1) {
4572 		write_lock(&tree->map_tree.lock);
4573 		em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4574 		if (em)
4575 			remove_extent_mapping(&tree->map_tree, em);
4576 		write_unlock(&tree->map_tree.lock);
4577 		if (!em)
4578 			break;
4579 		/* once for us */
4580 		free_extent_map(em);
4581 		/* once for the tree */
4582 		free_extent_map(em);
4583 	}
4584 }
4585 
4586 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4587 {
4588 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4589 	struct extent_map *em;
4590 	struct map_lookup *map;
4591 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4592 	int ret;
4593 
4594 	read_lock(&em_tree->lock);
4595 	em = lookup_extent_mapping(em_tree, logical, len);
4596 	read_unlock(&em_tree->lock);
4597 
4598 	/*
4599 	 * We could return errors for these cases, but that could get ugly and
4600 	 * we'd probably do the same thing which is just not do anything else
4601 	 * and exit, so return 1 so the callers don't try to use other copies.
4602 	 */
4603 	if (!em) {
4604 		btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4605 			    logical+len);
4606 		return 1;
4607 	}
4608 
4609 	if (em->start > logical || em->start + em->len < logical) {
4610 		btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4611 			    "%Lu-%Lu", logical, logical+len, em->start,
4612 			    em->start + em->len);
4613 		free_extent_map(em);
4614 		return 1;
4615 	}
4616 
4617 	map = (struct map_lookup *)em->bdev;
4618 	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4619 		ret = map->num_stripes;
4620 	else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4621 		ret = map->sub_stripes;
4622 	else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4623 		ret = 2;
4624 	else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4625 		ret = 3;
4626 	else
4627 		ret = 1;
4628 	free_extent_map(em);
4629 
4630 	btrfs_dev_replace_lock(&fs_info->dev_replace);
4631 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4632 		ret++;
4633 	btrfs_dev_replace_unlock(&fs_info->dev_replace);
4634 
4635 	return ret;
4636 }
4637 
4638 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4639 				    struct btrfs_mapping_tree *map_tree,
4640 				    u64 logical)
4641 {
4642 	struct extent_map *em;
4643 	struct map_lookup *map;
4644 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4645 	unsigned long len = root->sectorsize;
4646 
4647 	read_lock(&em_tree->lock);
4648 	em = lookup_extent_mapping(em_tree, logical, len);
4649 	read_unlock(&em_tree->lock);
4650 	BUG_ON(!em);
4651 
4652 	BUG_ON(em->start > logical || em->start + em->len < logical);
4653 	map = (struct map_lookup *)em->bdev;
4654 	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4655 			 BTRFS_BLOCK_GROUP_RAID6)) {
4656 		len = map->stripe_len * nr_data_stripes(map);
4657 	}
4658 	free_extent_map(em);
4659 	return len;
4660 }
4661 
4662 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4663 			   u64 logical, u64 len, int mirror_num)
4664 {
4665 	struct extent_map *em;
4666 	struct map_lookup *map;
4667 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4668 	int ret = 0;
4669 
4670 	read_lock(&em_tree->lock);
4671 	em = lookup_extent_mapping(em_tree, logical, len);
4672 	read_unlock(&em_tree->lock);
4673 	BUG_ON(!em);
4674 
4675 	BUG_ON(em->start > logical || em->start + em->len < logical);
4676 	map = (struct map_lookup *)em->bdev;
4677 	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4678 			 BTRFS_BLOCK_GROUP_RAID6))
4679 		ret = 1;
4680 	free_extent_map(em);
4681 	return ret;
4682 }
4683 
4684 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4685 			    struct map_lookup *map, int first, int num,
4686 			    int optimal, int dev_replace_is_ongoing)
4687 {
4688 	int i;
4689 	int tolerance;
4690 	struct btrfs_device *srcdev;
4691 
4692 	if (dev_replace_is_ongoing &&
4693 	    fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4694 	     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4695 		srcdev = fs_info->dev_replace.srcdev;
4696 	else
4697 		srcdev = NULL;
4698 
4699 	/*
4700 	 * try to avoid the drive that is the source drive for a
4701 	 * dev-replace procedure, only choose it if no other non-missing
4702 	 * mirror is available
4703 	 */
4704 	for (tolerance = 0; tolerance < 2; tolerance++) {
4705 		if (map->stripes[optimal].dev->bdev &&
4706 		    (tolerance || map->stripes[optimal].dev != srcdev))
4707 			return optimal;
4708 		for (i = first; i < first + num; i++) {
4709 			if (map->stripes[i].dev->bdev &&
4710 			    (tolerance || map->stripes[i].dev != srcdev))
4711 				return i;
4712 		}
4713 	}
4714 
4715 	/* we couldn't find one that doesn't fail.  Just return something
4716 	 * and the io error handling code will clean up eventually
4717 	 */
4718 	return optimal;
4719 }
4720 
4721 static inline int parity_smaller(u64 a, u64 b)
4722 {
4723 	return a > b;
4724 }
4725 
4726 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4727 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4728 {
4729 	struct btrfs_bio_stripe s;
4730 	int i;
4731 	u64 l;
4732 	int again = 1;
4733 
4734 	while (again) {
4735 		again = 0;
4736 		for (i = 0; i < bbio->num_stripes - 1; i++) {
4737 			if (parity_smaller(raid_map[i], raid_map[i+1])) {
4738 				s = bbio->stripes[i];
4739 				l = raid_map[i];
4740 				bbio->stripes[i] = bbio->stripes[i+1];
4741 				raid_map[i] = raid_map[i+1];
4742 				bbio->stripes[i+1] = s;
4743 				raid_map[i+1] = l;
4744 				again = 1;
4745 			}
4746 		}
4747 	}
4748 }
4749 
4750 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4751 			     u64 logical, u64 *length,
4752 			     struct btrfs_bio **bbio_ret,
4753 			     int mirror_num, u64 **raid_map_ret)
4754 {
4755 	struct extent_map *em;
4756 	struct map_lookup *map;
4757 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4758 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4759 	u64 offset;
4760 	u64 stripe_offset;
4761 	u64 stripe_end_offset;
4762 	u64 stripe_nr;
4763 	u64 stripe_nr_orig;
4764 	u64 stripe_nr_end;
4765 	u64 stripe_len;
4766 	u64 *raid_map = NULL;
4767 	int stripe_index;
4768 	int i;
4769 	int ret = 0;
4770 	int num_stripes;
4771 	int max_errors = 0;
4772 	struct btrfs_bio *bbio = NULL;
4773 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4774 	int dev_replace_is_ongoing = 0;
4775 	int num_alloc_stripes;
4776 	int patch_the_first_stripe_for_dev_replace = 0;
4777 	u64 physical_to_patch_in_first_stripe = 0;
4778 	u64 raid56_full_stripe_start = (u64)-1;
4779 
4780 	read_lock(&em_tree->lock);
4781 	em = lookup_extent_mapping(em_tree, logical, *length);
4782 	read_unlock(&em_tree->lock);
4783 
4784 	if (!em) {
4785 		btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4786 			logical, *length);
4787 		return -EINVAL;
4788 	}
4789 
4790 	if (em->start > logical || em->start + em->len < logical) {
4791 		btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4792 			   "found %Lu-%Lu", logical, em->start,
4793 			   em->start + em->len);
4794 		free_extent_map(em);
4795 		return -EINVAL;
4796 	}
4797 
4798 	map = (struct map_lookup *)em->bdev;
4799 	offset = logical - em->start;
4800 
4801 	stripe_len = map->stripe_len;
4802 	stripe_nr = offset;
4803 	/*
4804 	 * stripe_nr counts the total number of stripes we have to stride
4805 	 * to get to this block
4806 	 */
4807 	do_div(stripe_nr, stripe_len);
4808 
4809 	stripe_offset = stripe_nr * stripe_len;
4810 	BUG_ON(offset < stripe_offset);
4811 
4812 	/* stripe_offset is the offset of this block in its stripe*/
4813 	stripe_offset = offset - stripe_offset;
4814 
4815 	/* if we're here for raid56, we need to know the stripe aligned start */
4816 	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4817 		unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4818 		raid56_full_stripe_start = offset;
4819 
4820 		/* allow a write of a full stripe, but make sure we don't
4821 		 * allow straddling of stripes
4822 		 */
4823 		do_div(raid56_full_stripe_start, full_stripe_len);
4824 		raid56_full_stripe_start *= full_stripe_len;
4825 	}
4826 
4827 	if (rw & REQ_DISCARD) {
4828 		/* we don't discard raid56 yet */
4829 		if (map->type &
4830 		    (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4831 			ret = -EOPNOTSUPP;
4832 			goto out;
4833 		}
4834 		*length = min_t(u64, em->len - offset, *length);
4835 	} else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4836 		u64 max_len;
4837 		/* For writes to RAID[56], allow a full stripeset across all disks.
4838 		   For other RAID types and for RAID[56] reads, just allow a single
4839 		   stripe (on a single disk). */
4840 		if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4841 		    (rw & REQ_WRITE)) {
4842 			max_len = stripe_len * nr_data_stripes(map) -
4843 				(offset - raid56_full_stripe_start);
4844 		} else {
4845 			/* we limit the length of each bio to what fits in a stripe */
4846 			max_len = stripe_len - stripe_offset;
4847 		}
4848 		*length = min_t(u64, em->len - offset, max_len);
4849 	} else {
4850 		*length = em->len - offset;
4851 	}
4852 
4853 	/* This is for when we're called from btrfs_merge_bio_hook() and all
4854 	   it cares about is the length */
4855 	if (!bbio_ret)
4856 		goto out;
4857 
4858 	btrfs_dev_replace_lock(dev_replace);
4859 	dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4860 	if (!dev_replace_is_ongoing)
4861 		btrfs_dev_replace_unlock(dev_replace);
4862 
4863 	if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4864 	    !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4865 	    dev_replace->tgtdev != NULL) {
4866 		/*
4867 		 * in dev-replace case, for repair case (that's the only
4868 		 * case where the mirror is selected explicitly when
4869 		 * calling btrfs_map_block), blocks left of the left cursor
4870 		 * can also be read from the target drive.
4871 		 * For REQ_GET_READ_MIRRORS, the target drive is added as
4872 		 * the last one to the array of stripes. For READ, it also
4873 		 * needs to be supported using the same mirror number.
4874 		 * If the requested block is not left of the left cursor,
4875 		 * EIO is returned. This can happen because btrfs_num_copies()
4876 		 * returns one more in the dev-replace case.
4877 		 */
4878 		u64 tmp_length = *length;
4879 		struct btrfs_bio *tmp_bbio = NULL;
4880 		int tmp_num_stripes;
4881 		u64 srcdev_devid = dev_replace->srcdev->devid;
4882 		int index_srcdev = 0;
4883 		int found = 0;
4884 		u64 physical_of_found = 0;
4885 
4886 		ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4887 			     logical, &tmp_length, &tmp_bbio, 0, NULL);
4888 		if (ret) {
4889 			WARN_ON(tmp_bbio != NULL);
4890 			goto out;
4891 		}
4892 
4893 		tmp_num_stripes = tmp_bbio->num_stripes;
4894 		if (mirror_num > tmp_num_stripes) {
4895 			/*
4896 			 * REQ_GET_READ_MIRRORS does not contain this
4897 			 * mirror, that means that the requested area
4898 			 * is not left of the left cursor
4899 			 */
4900 			ret = -EIO;
4901 			kfree(tmp_bbio);
4902 			goto out;
4903 		}
4904 
4905 		/*
4906 		 * process the rest of the function using the mirror_num
4907 		 * of the source drive. Therefore look it up first.
4908 		 * At the end, patch the device pointer to the one of the
4909 		 * target drive.
4910 		 */
4911 		for (i = 0; i < tmp_num_stripes; i++) {
4912 			if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4913 				/*
4914 				 * In case of DUP, in order to keep it
4915 				 * simple, only add the mirror with the
4916 				 * lowest physical address
4917 				 */
4918 				if (found &&
4919 				    physical_of_found <=
4920 				     tmp_bbio->stripes[i].physical)
4921 					continue;
4922 				index_srcdev = i;
4923 				found = 1;
4924 				physical_of_found =
4925 					tmp_bbio->stripes[i].physical;
4926 			}
4927 		}
4928 
4929 		if (found) {
4930 			mirror_num = index_srcdev + 1;
4931 			patch_the_first_stripe_for_dev_replace = 1;
4932 			physical_to_patch_in_first_stripe = physical_of_found;
4933 		} else {
4934 			WARN_ON(1);
4935 			ret = -EIO;
4936 			kfree(tmp_bbio);
4937 			goto out;
4938 		}
4939 
4940 		kfree(tmp_bbio);
4941 	} else if (mirror_num > map->num_stripes) {
4942 		mirror_num = 0;
4943 	}
4944 
4945 	num_stripes = 1;
4946 	stripe_index = 0;
4947 	stripe_nr_orig = stripe_nr;
4948 	stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4949 	do_div(stripe_nr_end, map->stripe_len);
4950 	stripe_end_offset = stripe_nr_end * map->stripe_len -
4951 			    (offset + *length);
4952 
4953 	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4954 		if (rw & REQ_DISCARD)
4955 			num_stripes = min_t(u64, map->num_stripes,
4956 					    stripe_nr_end - stripe_nr_orig);
4957 		stripe_index = do_div(stripe_nr, map->num_stripes);
4958 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4959 		if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4960 			num_stripes = map->num_stripes;
4961 		else if (mirror_num)
4962 			stripe_index = mirror_num - 1;
4963 		else {
4964 			stripe_index = find_live_mirror(fs_info, map, 0,
4965 					    map->num_stripes,
4966 					    current->pid % map->num_stripes,
4967 					    dev_replace_is_ongoing);
4968 			mirror_num = stripe_index + 1;
4969 		}
4970 
4971 	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4972 		if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4973 			num_stripes = map->num_stripes;
4974 		} else if (mirror_num) {
4975 			stripe_index = mirror_num - 1;
4976 		} else {
4977 			mirror_num = 1;
4978 		}
4979 
4980 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4981 		int factor = map->num_stripes / map->sub_stripes;
4982 
4983 		stripe_index = do_div(stripe_nr, factor);
4984 		stripe_index *= map->sub_stripes;
4985 
4986 		if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4987 			num_stripes = map->sub_stripes;
4988 		else if (rw & REQ_DISCARD)
4989 			num_stripes = min_t(u64, map->sub_stripes *
4990 					    (stripe_nr_end - stripe_nr_orig),
4991 					    map->num_stripes);
4992 		else if (mirror_num)
4993 			stripe_index += mirror_num - 1;
4994 		else {
4995 			int old_stripe_index = stripe_index;
4996 			stripe_index = find_live_mirror(fs_info, map,
4997 					      stripe_index,
4998 					      map->sub_stripes, stripe_index +
4999 					      current->pid % map->sub_stripes,
5000 					      dev_replace_is_ongoing);
5001 			mirror_num = stripe_index - old_stripe_index + 1;
5002 		}
5003 
5004 	} else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5005 				BTRFS_BLOCK_GROUP_RAID6)) {
5006 		u64 tmp;
5007 
5008 		if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
5009 		    && raid_map_ret) {
5010 			int i, rot;
5011 
5012 			/* push stripe_nr back to the start of the full stripe */
5013 			stripe_nr = raid56_full_stripe_start;
5014 			do_div(stripe_nr, stripe_len);
5015 
5016 			stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5017 
5018 			/* RAID[56] write or recovery. Return all stripes */
5019 			num_stripes = map->num_stripes;
5020 			max_errors = nr_parity_stripes(map);
5021 
5022 			raid_map = kmalloc_array(num_stripes, sizeof(u64),
5023 					   GFP_NOFS);
5024 			if (!raid_map) {
5025 				ret = -ENOMEM;
5026 				goto out;
5027 			}
5028 
5029 			/* Work out the disk rotation on this stripe-set */
5030 			tmp = stripe_nr;
5031 			rot = do_div(tmp, num_stripes);
5032 
5033 			/* Fill in the logical address of each stripe */
5034 			tmp = stripe_nr * nr_data_stripes(map);
5035 			for (i = 0; i < nr_data_stripes(map); i++)
5036 				raid_map[(i+rot) % num_stripes] =
5037 					em->start + (tmp + i) * map->stripe_len;
5038 
5039 			raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5040 			if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5041 				raid_map[(i+rot+1) % num_stripes] =
5042 					RAID6_Q_STRIPE;
5043 
5044 			*length = map->stripe_len;
5045 			stripe_index = 0;
5046 			stripe_offset = 0;
5047 		} else {
5048 			/*
5049 			 * Mirror #0 or #1 means the original data block.
5050 			 * Mirror #2 is RAID5 parity block.
5051 			 * Mirror #3 is RAID6 Q block.
5052 			 */
5053 			stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5054 			if (mirror_num > 1)
5055 				stripe_index = nr_data_stripes(map) +
5056 						mirror_num - 2;
5057 
5058 			/* We distribute the parity blocks across stripes */
5059 			tmp = stripe_nr + stripe_index;
5060 			stripe_index = do_div(tmp, map->num_stripes);
5061 		}
5062 	} else {
5063 		/*
5064 		 * after this do_div call, stripe_nr is the number of stripes
5065 		 * on this device we have to walk to find the data, and
5066 		 * stripe_index is the number of our device in the stripe array
5067 		 */
5068 		stripe_index = do_div(stripe_nr, map->num_stripes);
5069 		mirror_num = stripe_index + 1;
5070 	}
5071 	BUG_ON(stripe_index >= map->num_stripes);
5072 
5073 	num_alloc_stripes = num_stripes;
5074 	if (dev_replace_is_ongoing) {
5075 		if (rw & (REQ_WRITE | REQ_DISCARD))
5076 			num_alloc_stripes <<= 1;
5077 		if (rw & REQ_GET_READ_MIRRORS)
5078 			num_alloc_stripes++;
5079 	}
5080 	bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
5081 	if (!bbio) {
5082 		kfree(raid_map);
5083 		ret = -ENOMEM;
5084 		goto out;
5085 	}
5086 	atomic_set(&bbio->error, 0);
5087 
5088 	if (rw & REQ_DISCARD) {
5089 		int factor = 0;
5090 		int sub_stripes = 0;
5091 		u64 stripes_per_dev = 0;
5092 		u32 remaining_stripes = 0;
5093 		u32 last_stripe = 0;
5094 
5095 		if (map->type &
5096 		    (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5097 			if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5098 				sub_stripes = 1;
5099 			else
5100 				sub_stripes = map->sub_stripes;
5101 
5102 			factor = map->num_stripes / sub_stripes;
5103 			stripes_per_dev = div_u64_rem(stripe_nr_end -
5104 						      stripe_nr_orig,
5105 						      factor,
5106 						      &remaining_stripes);
5107 			div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5108 			last_stripe *= sub_stripes;
5109 		}
5110 
5111 		for (i = 0; i < num_stripes; i++) {
5112 			bbio->stripes[i].physical =
5113 				map->stripes[stripe_index].physical +
5114 				stripe_offset + stripe_nr * map->stripe_len;
5115 			bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5116 
5117 			if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5118 					 BTRFS_BLOCK_GROUP_RAID10)) {
5119 				bbio->stripes[i].length = stripes_per_dev *
5120 							  map->stripe_len;
5121 
5122 				if (i / sub_stripes < remaining_stripes)
5123 					bbio->stripes[i].length +=
5124 						map->stripe_len;
5125 
5126 				/*
5127 				 * Special for the first stripe and
5128 				 * the last stripe:
5129 				 *
5130 				 * |-------|...|-------|
5131 				 *     |----------|
5132 				 *    off     end_off
5133 				 */
5134 				if (i < sub_stripes)
5135 					bbio->stripes[i].length -=
5136 						stripe_offset;
5137 
5138 				if (stripe_index >= last_stripe &&
5139 				    stripe_index <= (last_stripe +
5140 						     sub_stripes - 1))
5141 					bbio->stripes[i].length -=
5142 						stripe_end_offset;
5143 
5144 				if (i == sub_stripes - 1)
5145 					stripe_offset = 0;
5146 			} else
5147 				bbio->stripes[i].length = *length;
5148 
5149 			stripe_index++;
5150 			if (stripe_index == map->num_stripes) {
5151 				/* This could only happen for RAID0/10 */
5152 				stripe_index = 0;
5153 				stripe_nr++;
5154 			}
5155 		}
5156 	} else {
5157 		for (i = 0; i < num_stripes; i++) {
5158 			bbio->stripes[i].physical =
5159 				map->stripes[stripe_index].physical +
5160 				stripe_offset +
5161 				stripe_nr * map->stripe_len;
5162 			bbio->stripes[i].dev =
5163 				map->stripes[stripe_index].dev;
5164 			stripe_index++;
5165 		}
5166 	}
5167 
5168 	if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
5169 		if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5170 				 BTRFS_BLOCK_GROUP_RAID10 |
5171 				 BTRFS_BLOCK_GROUP_RAID5 |
5172 				 BTRFS_BLOCK_GROUP_DUP)) {
5173 			max_errors = 1;
5174 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5175 			max_errors = 2;
5176 		}
5177 	}
5178 
5179 	if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5180 	    dev_replace->tgtdev != NULL) {
5181 		int index_where_to_add;
5182 		u64 srcdev_devid = dev_replace->srcdev->devid;
5183 
5184 		/*
5185 		 * duplicate the write operations while the dev replace
5186 		 * procedure is running. Since the copying of the old disk
5187 		 * to the new disk takes place at run time while the
5188 		 * filesystem is mounted writable, the regular write
5189 		 * operations to the old disk have to be duplicated to go
5190 		 * to the new disk as well.
5191 		 * Note that device->missing is handled by the caller, and
5192 		 * that the write to the old disk is already set up in the
5193 		 * stripes array.
5194 		 */
5195 		index_where_to_add = num_stripes;
5196 		for (i = 0; i < num_stripes; i++) {
5197 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
5198 				/* write to new disk, too */
5199 				struct btrfs_bio_stripe *new =
5200 					bbio->stripes + index_where_to_add;
5201 				struct btrfs_bio_stripe *old =
5202 					bbio->stripes + i;
5203 
5204 				new->physical = old->physical;
5205 				new->length = old->length;
5206 				new->dev = dev_replace->tgtdev;
5207 				index_where_to_add++;
5208 				max_errors++;
5209 			}
5210 		}
5211 		num_stripes = index_where_to_add;
5212 	} else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5213 		   dev_replace->tgtdev != NULL) {
5214 		u64 srcdev_devid = dev_replace->srcdev->devid;
5215 		int index_srcdev = 0;
5216 		int found = 0;
5217 		u64 physical_of_found = 0;
5218 
5219 		/*
5220 		 * During the dev-replace procedure, the target drive can
5221 		 * also be used to read data in case it is needed to repair
5222 		 * a corrupt block elsewhere. This is possible if the
5223 		 * requested area is left of the left cursor. In this area,
5224 		 * the target drive is a full copy of the source drive.
5225 		 */
5226 		for (i = 0; i < num_stripes; i++) {
5227 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
5228 				/*
5229 				 * In case of DUP, in order to keep it
5230 				 * simple, only add the mirror with the
5231 				 * lowest physical address
5232 				 */
5233 				if (found &&
5234 				    physical_of_found <=
5235 				     bbio->stripes[i].physical)
5236 					continue;
5237 				index_srcdev = i;
5238 				found = 1;
5239 				physical_of_found = bbio->stripes[i].physical;
5240 			}
5241 		}
5242 		if (found) {
5243 			u64 length = map->stripe_len;
5244 
5245 			if (physical_of_found + length <=
5246 			    dev_replace->cursor_left) {
5247 				struct btrfs_bio_stripe *tgtdev_stripe =
5248 					bbio->stripes + num_stripes;
5249 
5250 				tgtdev_stripe->physical = physical_of_found;
5251 				tgtdev_stripe->length =
5252 					bbio->stripes[index_srcdev].length;
5253 				tgtdev_stripe->dev = dev_replace->tgtdev;
5254 
5255 				num_stripes++;
5256 			}
5257 		}
5258 	}
5259 
5260 	*bbio_ret = bbio;
5261 	bbio->num_stripes = num_stripes;
5262 	bbio->max_errors = max_errors;
5263 	bbio->mirror_num = mirror_num;
5264 
5265 	/*
5266 	 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5267 	 * mirror_num == num_stripes + 1 && dev_replace target drive is
5268 	 * available as a mirror
5269 	 */
5270 	if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5271 		WARN_ON(num_stripes > 1);
5272 		bbio->stripes[0].dev = dev_replace->tgtdev;
5273 		bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5274 		bbio->mirror_num = map->num_stripes + 1;
5275 	}
5276 	if (raid_map) {
5277 		sort_parity_stripes(bbio, raid_map);
5278 		*raid_map_ret = raid_map;
5279 	}
5280 out:
5281 	if (dev_replace_is_ongoing)
5282 		btrfs_dev_replace_unlock(dev_replace);
5283 	free_extent_map(em);
5284 	return ret;
5285 }
5286 
5287 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5288 		      u64 logical, u64 *length,
5289 		      struct btrfs_bio **bbio_ret, int mirror_num)
5290 {
5291 	return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5292 				 mirror_num, NULL);
5293 }
5294 
5295 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5296 		     u64 chunk_start, u64 physical, u64 devid,
5297 		     u64 **logical, int *naddrs, int *stripe_len)
5298 {
5299 	struct extent_map_tree *em_tree = &map_tree->map_tree;
5300 	struct extent_map *em;
5301 	struct map_lookup *map;
5302 	u64 *buf;
5303 	u64 bytenr;
5304 	u64 length;
5305 	u64 stripe_nr;
5306 	u64 rmap_len;
5307 	int i, j, nr = 0;
5308 
5309 	read_lock(&em_tree->lock);
5310 	em = lookup_extent_mapping(em_tree, chunk_start, 1);
5311 	read_unlock(&em_tree->lock);
5312 
5313 	if (!em) {
5314 		printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5315 		       chunk_start);
5316 		return -EIO;
5317 	}
5318 
5319 	if (em->start != chunk_start) {
5320 		printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5321 		       em->start, chunk_start);
5322 		free_extent_map(em);
5323 		return -EIO;
5324 	}
5325 	map = (struct map_lookup *)em->bdev;
5326 
5327 	length = em->len;
5328 	rmap_len = map->stripe_len;
5329 
5330 	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5331 		do_div(length, map->num_stripes / map->sub_stripes);
5332 	else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5333 		do_div(length, map->num_stripes);
5334 	else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5335 			      BTRFS_BLOCK_GROUP_RAID6)) {
5336 		do_div(length, nr_data_stripes(map));
5337 		rmap_len = map->stripe_len * nr_data_stripes(map);
5338 	}
5339 
5340 	buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5341 	BUG_ON(!buf); /* -ENOMEM */
5342 
5343 	for (i = 0; i < map->num_stripes; i++) {
5344 		if (devid && map->stripes[i].dev->devid != devid)
5345 			continue;
5346 		if (map->stripes[i].physical > physical ||
5347 		    map->stripes[i].physical + length <= physical)
5348 			continue;
5349 
5350 		stripe_nr = physical - map->stripes[i].physical;
5351 		do_div(stripe_nr, map->stripe_len);
5352 
5353 		if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5354 			stripe_nr = stripe_nr * map->num_stripes + i;
5355 			do_div(stripe_nr, map->sub_stripes);
5356 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5357 			stripe_nr = stripe_nr * map->num_stripes + i;
5358 		} /* else if RAID[56], multiply by nr_data_stripes().
5359 		   * Alternatively, just use rmap_len below instead of
5360 		   * map->stripe_len */
5361 
5362 		bytenr = chunk_start + stripe_nr * rmap_len;
5363 		WARN_ON(nr >= map->num_stripes);
5364 		for (j = 0; j < nr; j++) {
5365 			if (buf[j] == bytenr)
5366 				break;
5367 		}
5368 		if (j == nr) {
5369 			WARN_ON(nr >= map->num_stripes);
5370 			buf[nr++] = bytenr;
5371 		}
5372 	}
5373 
5374 	*logical = buf;
5375 	*naddrs = nr;
5376 	*stripe_len = rmap_len;
5377 
5378 	free_extent_map(em);
5379 	return 0;
5380 }
5381 
5382 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err)
5383 {
5384 	if (likely(bbio->flags & BTRFS_BIO_ORIG_BIO_SUBMITTED))
5385 		bio_endio_nodec(bio, err);
5386 	else
5387 		bio_endio(bio, err);
5388 	kfree(bbio);
5389 }
5390 
5391 static void btrfs_end_bio(struct bio *bio, int err)
5392 {
5393 	struct btrfs_bio *bbio = bio->bi_private;
5394 	struct btrfs_device *dev = bbio->stripes[0].dev;
5395 	int is_orig_bio = 0;
5396 
5397 	if (err) {
5398 		atomic_inc(&bbio->error);
5399 		if (err == -EIO || err == -EREMOTEIO) {
5400 			unsigned int stripe_index =
5401 				btrfs_io_bio(bio)->stripe_index;
5402 
5403 			BUG_ON(stripe_index >= bbio->num_stripes);
5404 			dev = bbio->stripes[stripe_index].dev;
5405 			if (dev->bdev) {
5406 				if (bio->bi_rw & WRITE)
5407 					btrfs_dev_stat_inc(dev,
5408 						BTRFS_DEV_STAT_WRITE_ERRS);
5409 				else
5410 					btrfs_dev_stat_inc(dev,
5411 						BTRFS_DEV_STAT_READ_ERRS);
5412 				if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5413 					btrfs_dev_stat_inc(dev,
5414 						BTRFS_DEV_STAT_FLUSH_ERRS);
5415 				btrfs_dev_stat_print_on_error(dev);
5416 			}
5417 		}
5418 	}
5419 
5420 	if (bio == bbio->orig_bio)
5421 		is_orig_bio = 1;
5422 
5423 	btrfs_bio_counter_dec(bbio->fs_info);
5424 
5425 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
5426 		if (!is_orig_bio) {
5427 			bio_put(bio);
5428 			bio = bbio->orig_bio;
5429 		}
5430 
5431 		bio->bi_private = bbio->private;
5432 		bio->bi_end_io = bbio->end_io;
5433 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5434 		/* only send an error to the higher layers if it is
5435 		 * beyond the tolerance of the btrfs bio
5436 		 */
5437 		if (atomic_read(&bbio->error) > bbio->max_errors) {
5438 			err = -EIO;
5439 		} else {
5440 			/*
5441 			 * this bio is actually up to date, we didn't
5442 			 * go over the max number of errors
5443 			 */
5444 			set_bit(BIO_UPTODATE, &bio->bi_flags);
5445 			err = 0;
5446 		}
5447 
5448 		btrfs_end_bbio(bbio, bio, err);
5449 	} else if (!is_orig_bio) {
5450 		bio_put(bio);
5451 	}
5452 }
5453 
5454 /*
5455  * see run_scheduled_bios for a description of why bios are collected for
5456  * async submit.
5457  *
5458  * This will add one bio to the pending list for a device and make sure
5459  * the work struct is scheduled.
5460  */
5461 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5462 					struct btrfs_device *device,
5463 					int rw, struct bio *bio)
5464 {
5465 	int should_queue = 1;
5466 	struct btrfs_pending_bios *pending_bios;
5467 
5468 	if (device->missing || !device->bdev) {
5469 		bio_endio(bio, -EIO);
5470 		return;
5471 	}
5472 
5473 	/* don't bother with additional async steps for reads, right now */
5474 	if (!(rw & REQ_WRITE)) {
5475 		bio_get(bio);
5476 		btrfsic_submit_bio(rw, bio);
5477 		bio_put(bio);
5478 		return;
5479 	}
5480 
5481 	/*
5482 	 * nr_async_bios allows us to reliably return congestion to the
5483 	 * higher layers.  Otherwise, the async bio makes it appear we have
5484 	 * made progress against dirty pages when we've really just put it
5485 	 * on a queue for later
5486 	 */
5487 	atomic_inc(&root->fs_info->nr_async_bios);
5488 	WARN_ON(bio->bi_next);
5489 	bio->bi_next = NULL;
5490 	bio->bi_rw |= rw;
5491 
5492 	spin_lock(&device->io_lock);
5493 	if (bio->bi_rw & REQ_SYNC)
5494 		pending_bios = &device->pending_sync_bios;
5495 	else
5496 		pending_bios = &device->pending_bios;
5497 
5498 	if (pending_bios->tail)
5499 		pending_bios->tail->bi_next = bio;
5500 
5501 	pending_bios->tail = bio;
5502 	if (!pending_bios->head)
5503 		pending_bios->head = bio;
5504 	if (device->running_pending)
5505 		should_queue = 0;
5506 
5507 	spin_unlock(&device->io_lock);
5508 
5509 	if (should_queue)
5510 		btrfs_queue_work(root->fs_info->submit_workers,
5511 				 &device->work);
5512 }
5513 
5514 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5515 		       sector_t sector)
5516 {
5517 	struct bio_vec *prev;
5518 	struct request_queue *q = bdev_get_queue(bdev);
5519 	unsigned int max_sectors = queue_max_sectors(q);
5520 	struct bvec_merge_data bvm = {
5521 		.bi_bdev = bdev,
5522 		.bi_sector = sector,
5523 		.bi_rw = bio->bi_rw,
5524 	};
5525 
5526 	if (WARN_ON(bio->bi_vcnt == 0))
5527 		return 1;
5528 
5529 	prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5530 	if (bio_sectors(bio) > max_sectors)
5531 		return 0;
5532 
5533 	if (!q->merge_bvec_fn)
5534 		return 1;
5535 
5536 	bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5537 	if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5538 		return 0;
5539 	return 1;
5540 }
5541 
5542 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5543 			      struct bio *bio, u64 physical, int dev_nr,
5544 			      int rw, int async)
5545 {
5546 	struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5547 
5548 	bio->bi_private = bbio;
5549 	btrfs_io_bio(bio)->stripe_index = dev_nr;
5550 	bio->bi_end_io = btrfs_end_bio;
5551 	bio->bi_iter.bi_sector = physical >> 9;
5552 #ifdef DEBUG
5553 	{
5554 		struct rcu_string *name;
5555 
5556 		rcu_read_lock();
5557 		name = rcu_dereference(dev->name);
5558 		pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5559 			 "(%s id %llu), size=%u\n", rw,
5560 			 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5561 			 name->str, dev->devid, bio->bi_size);
5562 		rcu_read_unlock();
5563 	}
5564 #endif
5565 	bio->bi_bdev = dev->bdev;
5566 
5567 	btrfs_bio_counter_inc_noblocked(root->fs_info);
5568 
5569 	if (async)
5570 		btrfs_schedule_bio(root, dev, rw, bio);
5571 	else
5572 		btrfsic_submit_bio(rw, bio);
5573 }
5574 
5575 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5576 			      struct bio *first_bio, struct btrfs_device *dev,
5577 			      int dev_nr, int rw, int async)
5578 {
5579 	struct bio_vec *bvec = first_bio->bi_io_vec;
5580 	struct bio *bio;
5581 	int nr_vecs = bio_get_nr_vecs(dev->bdev);
5582 	u64 physical = bbio->stripes[dev_nr].physical;
5583 
5584 again:
5585 	bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5586 	if (!bio)
5587 		return -ENOMEM;
5588 
5589 	while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5590 		if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5591 				 bvec->bv_offset) < bvec->bv_len) {
5592 			u64 len = bio->bi_iter.bi_size;
5593 
5594 			atomic_inc(&bbio->stripes_pending);
5595 			submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5596 					  rw, async);
5597 			physical += len;
5598 			goto again;
5599 		}
5600 		bvec++;
5601 	}
5602 
5603 	submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5604 	return 0;
5605 }
5606 
5607 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5608 {
5609 	atomic_inc(&bbio->error);
5610 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
5611 		/* Shoud be the original bio. */
5612 		WARN_ON(bio != bbio->orig_bio);
5613 
5614 		bio->bi_private = bbio->private;
5615 		bio->bi_end_io = bbio->end_io;
5616 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5617 		bio->bi_iter.bi_sector = logical >> 9;
5618 
5619 		btrfs_end_bbio(bbio, bio, -EIO);
5620 	}
5621 }
5622 
5623 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5624 		  int mirror_num, int async_submit)
5625 {
5626 	struct btrfs_device *dev;
5627 	struct bio *first_bio = bio;
5628 	u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5629 	u64 length = 0;
5630 	u64 map_length;
5631 	u64 *raid_map = NULL;
5632 	int ret;
5633 	int dev_nr = 0;
5634 	int total_devs = 1;
5635 	struct btrfs_bio *bbio = NULL;
5636 
5637 	length = bio->bi_iter.bi_size;
5638 	map_length = length;
5639 
5640 	btrfs_bio_counter_inc_blocked(root->fs_info);
5641 	ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5642 			      mirror_num, &raid_map);
5643 	if (ret) {
5644 		btrfs_bio_counter_dec(root->fs_info);
5645 		return ret;
5646 	}
5647 
5648 	total_devs = bbio->num_stripes;
5649 	bbio->orig_bio = first_bio;
5650 	bbio->private = first_bio->bi_private;
5651 	bbio->end_io = first_bio->bi_end_io;
5652 	bbio->fs_info = root->fs_info;
5653 	atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5654 
5655 	if (raid_map) {
5656 		/* In this case, map_length has been set to the length of
5657 		   a single stripe; not the whole write */
5658 		if (rw & WRITE) {
5659 			ret = raid56_parity_write(root, bio, bbio,
5660 						  raid_map, map_length);
5661 		} else {
5662 			ret = raid56_parity_recover(root, bio, bbio,
5663 						    raid_map, map_length,
5664 						    mirror_num);
5665 		}
5666 		/*
5667 		 * FIXME, replace dosen't support raid56 yet, please fix
5668 		 * it in the future.
5669 		 */
5670 		btrfs_bio_counter_dec(root->fs_info);
5671 		return ret;
5672 	}
5673 
5674 	if (map_length < length) {
5675 		btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5676 			logical, length, map_length);
5677 		BUG();
5678 	}
5679 
5680 	while (dev_nr < total_devs) {
5681 		dev = bbio->stripes[dev_nr].dev;
5682 		if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5683 			bbio_error(bbio, first_bio, logical);
5684 			dev_nr++;
5685 			continue;
5686 		}
5687 
5688 		/*
5689 		 * Check and see if we're ok with this bio based on it's size
5690 		 * and offset with the given device.
5691 		 */
5692 		if (!bio_size_ok(dev->bdev, first_bio,
5693 				 bbio->stripes[dev_nr].physical >> 9)) {
5694 			ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5695 						 dev_nr, rw, async_submit);
5696 			BUG_ON(ret);
5697 			dev_nr++;
5698 			continue;
5699 		}
5700 
5701 		if (dev_nr < total_devs - 1) {
5702 			bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5703 			BUG_ON(!bio); /* -ENOMEM */
5704 		} else {
5705 			bio = first_bio;
5706 			bbio->flags |= BTRFS_BIO_ORIG_BIO_SUBMITTED;
5707 		}
5708 
5709 		submit_stripe_bio(root, bbio, bio,
5710 				  bbio->stripes[dev_nr].physical, dev_nr, rw,
5711 				  async_submit);
5712 		dev_nr++;
5713 	}
5714 	btrfs_bio_counter_dec(root->fs_info);
5715 	return 0;
5716 }
5717 
5718 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5719 				       u8 *uuid, u8 *fsid)
5720 {
5721 	struct btrfs_device *device;
5722 	struct btrfs_fs_devices *cur_devices;
5723 
5724 	cur_devices = fs_info->fs_devices;
5725 	while (cur_devices) {
5726 		if (!fsid ||
5727 		    !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5728 			device = __find_device(&cur_devices->devices,
5729 					       devid, uuid);
5730 			if (device)
5731 				return device;
5732 		}
5733 		cur_devices = cur_devices->seed;
5734 	}
5735 	return NULL;
5736 }
5737 
5738 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5739 					    u64 devid, u8 *dev_uuid)
5740 {
5741 	struct btrfs_device *device;
5742 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5743 
5744 	device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5745 	if (IS_ERR(device))
5746 		return NULL;
5747 
5748 	list_add(&device->dev_list, &fs_devices->devices);
5749 	device->fs_devices = fs_devices;
5750 	fs_devices->num_devices++;
5751 
5752 	device->missing = 1;
5753 	fs_devices->missing_devices++;
5754 
5755 	return device;
5756 }
5757 
5758 /**
5759  * btrfs_alloc_device - allocate struct btrfs_device
5760  * @fs_info:	used only for generating a new devid, can be NULL if
5761  *		devid is provided (i.e. @devid != NULL).
5762  * @devid:	a pointer to devid for this device.  If NULL a new devid
5763  *		is generated.
5764  * @uuid:	a pointer to UUID for this device.  If NULL a new UUID
5765  *		is generated.
5766  *
5767  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5768  * on error.  Returned struct is not linked onto any lists and can be
5769  * destroyed with kfree() right away.
5770  */
5771 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5772 					const u64 *devid,
5773 					const u8 *uuid)
5774 {
5775 	struct btrfs_device *dev;
5776 	u64 tmp;
5777 
5778 	if (WARN_ON(!devid && !fs_info))
5779 		return ERR_PTR(-EINVAL);
5780 
5781 	dev = __alloc_device();
5782 	if (IS_ERR(dev))
5783 		return dev;
5784 
5785 	if (devid)
5786 		tmp = *devid;
5787 	else {
5788 		int ret;
5789 
5790 		ret = find_next_devid(fs_info, &tmp);
5791 		if (ret) {
5792 			kfree(dev);
5793 			return ERR_PTR(ret);
5794 		}
5795 	}
5796 	dev->devid = tmp;
5797 
5798 	if (uuid)
5799 		memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5800 	else
5801 		generate_random_uuid(dev->uuid);
5802 
5803 	btrfs_init_work(&dev->work, pending_bios_fn, NULL, NULL);
5804 
5805 	return dev;
5806 }
5807 
5808 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5809 			  struct extent_buffer *leaf,
5810 			  struct btrfs_chunk *chunk)
5811 {
5812 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5813 	struct map_lookup *map;
5814 	struct extent_map *em;
5815 	u64 logical;
5816 	u64 length;
5817 	u64 devid;
5818 	u8 uuid[BTRFS_UUID_SIZE];
5819 	int num_stripes;
5820 	int ret;
5821 	int i;
5822 
5823 	logical = key->offset;
5824 	length = btrfs_chunk_length(leaf, chunk);
5825 
5826 	read_lock(&map_tree->map_tree.lock);
5827 	em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5828 	read_unlock(&map_tree->map_tree.lock);
5829 
5830 	/* already mapped? */
5831 	if (em && em->start <= logical && em->start + em->len > logical) {
5832 		free_extent_map(em);
5833 		return 0;
5834 	} else if (em) {
5835 		free_extent_map(em);
5836 	}
5837 
5838 	em = alloc_extent_map();
5839 	if (!em)
5840 		return -ENOMEM;
5841 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5842 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5843 	if (!map) {
5844 		free_extent_map(em);
5845 		return -ENOMEM;
5846 	}
5847 
5848 	set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5849 	em->bdev = (struct block_device *)map;
5850 	em->start = logical;
5851 	em->len = length;
5852 	em->orig_start = 0;
5853 	em->block_start = 0;
5854 	em->block_len = em->len;
5855 
5856 	map->num_stripes = num_stripes;
5857 	map->io_width = btrfs_chunk_io_width(leaf, chunk);
5858 	map->io_align = btrfs_chunk_io_align(leaf, chunk);
5859 	map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5860 	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5861 	map->type = btrfs_chunk_type(leaf, chunk);
5862 	map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5863 	for (i = 0; i < num_stripes; i++) {
5864 		map->stripes[i].physical =
5865 			btrfs_stripe_offset_nr(leaf, chunk, i);
5866 		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5867 		read_extent_buffer(leaf, uuid, (unsigned long)
5868 				   btrfs_stripe_dev_uuid_nr(chunk, i),
5869 				   BTRFS_UUID_SIZE);
5870 		map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5871 							uuid, NULL);
5872 		if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5873 			free_extent_map(em);
5874 			return -EIO;
5875 		}
5876 		if (!map->stripes[i].dev) {
5877 			map->stripes[i].dev =
5878 				add_missing_dev(root, devid, uuid);
5879 			if (!map->stripes[i].dev) {
5880 				free_extent_map(em);
5881 				return -EIO;
5882 			}
5883 		}
5884 		map->stripes[i].dev->in_fs_metadata = 1;
5885 	}
5886 
5887 	write_lock(&map_tree->map_tree.lock);
5888 	ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5889 	write_unlock(&map_tree->map_tree.lock);
5890 	BUG_ON(ret); /* Tree corruption */
5891 	free_extent_map(em);
5892 
5893 	return 0;
5894 }
5895 
5896 static void fill_device_from_item(struct extent_buffer *leaf,
5897 				 struct btrfs_dev_item *dev_item,
5898 				 struct btrfs_device *device)
5899 {
5900 	unsigned long ptr;
5901 
5902 	device->devid = btrfs_device_id(leaf, dev_item);
5903 	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5904 	device->total_bytes = device->disk_total_bytes;
5905 	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5906 	device->type = btrfs_device_type(leaf, dev_item);
5907 	device->io_align = btrfs_device_io_align(leaf, dev_item);
5908 	device->io_width = btrfs_device_io_width(leaf, dev_item);
5909 	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5910 	WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5911 	device->is_tgtdev_for_dev_replace = 0;
5912 
5913 	ptr = btrfs_device_uuid(dev_item);
5914 	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5915 }
5916 
5917 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5918 {
5919 	struct btrfs_fs_devices *fs_devices;
5920 	int ret;
5921 
5922 	BUG_ON(!mutex_is_locked(&uuid_mutex));
5923 
5924 	fs_devices = root->fs_info->fs_devices->seed;
5925 	while (fs_devices) {
5926 		if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5927 			ret = 0;
5928 			goto out;
5929 		}
5930 		fs_devices = fs_devices->seed;
5931 	}
5932 
5933 	fs_devices = find_fsid(fsid);
5934 	if (!fs_devices) {
5935 		ret = -ENOENT;
5936 		goto out;
5937 	}
5938 
5939 	fs_devices = clone_fs_devices(fs_devices);
5940 	if (IS_ERR(fs_devices)) {
5941 		ret = PTR_ERR(fs_devices);
5942 		goto out;
5943 	}
5944 
5945 	ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5946 				   root->fs_info->bdev_holder);
5947 	if (ret) {
5948 		free_fs_devices(fs_devices);
5949 		goto out;
5950 	}
5951 
5952 	if (!fs_devices->seeding) {
5953 		__btrfs_close_devices(fs_devices);
5954 		free_fs_devices(fs_devices);
5955 		ret = -EINVAL;
5956 		goto out;
5957 	}
5958 
5959 	fs_devices->seed = root->fs_info->fs_devices->seed;
5960 	root->fs_info->fs_devices->seed = fs_devices;
5961 out:
5962 	return ret;
5963 }
5964 
5965 static int read_one_dev(struct btrfs_root *root,
5966 			struct extent_buffer *leaf,
5967 			struct btrfs_dev_item *dev_item)
5968 {
5969 	struct btrfs_device *device;
5970 	u64 devid;
5971 	int ret;
5972 	u8 fs_uuid[BTRFS_UUID_SIZE];
5973 	u8 dev_uuid[BTRFS_UUID_SIZE];
5974 
5975 	devid = btrfs_device_id(leaf, dev_item);
5976 	read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
5977 			   BTRFS_UUID_SIZE);
5978 	read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
5979 			   BTRFS_UUID_SIZE);
5980 
5981 	if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5982 		ret = open_seed_devices(root, fs_uuid);
5983 		if (ret && !btrfs_test_opt(root, DEGRADED))
5984 			return ret;
5985 	}
5986 
5987 	device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5988 	if (!device || !device->bdev) {
5989 		if (!btrfs_test_opt(root, DEGRADED))
5990 			return -EIO;
5991 
5992 		if (!device) {
5993 			btrfs_warn(root->fs_info, "devid %llu missing", devid);
5994 			device = add_missing_dev(root, devid, dev_uuid);
5995 			if (!device)
5996 				return -ENOMEM;
5997 		} else if (!device->missing) {
5998 			/*
5999 			 * this happens when a device that was properly setup
6000 			 * in the device info lists suddenly goes bad.
6001 			 * device->bdev is NULL, and so we have to set
6002 			 * device->missing to one here
6003 			 */
6004 			root->fs_info->fs_devices->missing_devices++;
6005 			device->missing = 1;
6006 		}
6007 	}
6008 
6009 	if (device->fs_devices != root->fs_info->fs_devices) {
6010 		BUG_ON(device->writeable);
6011 		if (device->generation !=
6012 		    btrfs_device_generation(leaf, dev_item))
6013 			return -EINVAL;
6014 	}
6015 
6016 	fill_device_from_item(leaf, dev_item, device);
6017 	device->in_fs_metadata = 1;
6018 	if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6019 		device->fs_devices->total_rw_bytes += device->total_bytes;
6020 		spin_lock(&root->fs_info->free_chunk_lock);
6021 		root->fs_info->free_chunk_space += device->total_bytes -
6022 			device->bytes_used;
6023 		spin_unlock(&root->fs_info->free_chunk_lock);
6024 	}
6025 	ret = 0;
6026 	return ret;
6027 }
6028 
6029 int btrfs_read_sys_array(struct btrfs_root *root)
6030 {
6031 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6032 	struct extent_buffer *sb;
6033 	struct btrfs_disk_key *disk_key;
6034 	struct btrfs_chunk *chunk;
6035 	u8 *ptr;
6036 	unsigned long sb_ptr;
6037 	int ret = 0;
6038 	u32 num_stripes;
6039 	u32 array_size;
6040 	u32 len = 0;
6041 	u32 cur;
6042 	struct btrfs_key key;
6043 
6044 	sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
6045 					  BTRFS_SUPER_INFO_SIZE);
6046 	if (!sb)
6047 		return -ENOMEM;
6048 	btrfs_set_buffer_uptodate(sb);
6049 	btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6050 	/*
6051 	 * The sb extent buffer is artifical and just used to read the system array.
6052 	 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6053 	 * pages up-to-date when the page is larger: extent does not cover the
6054 	 * whole page and consequently check_page_uptodate does not find all
6055 	 * the page's extents up-to-date (the hole beyond sb),
6056 	 * write_extent_buffer then triggers a WARN_ON.
6057 	 *
6058 	 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6059 	 * but sb spans only this function. Add an explicit SetPageUptodate call
6060 	 * to silence the warning eg. on PowerPC 64.
6061 	 */
6062 	if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6063 		SetPageUptodate(sb->pages[0]);
6064 
6065 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6066 	array_size = btrfs_super_sys_array_size(super_copy);
6067 
6068 	ptr = super_copy->sys_chunk_array;
6069 	sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
6070 	cur = 0;
6071 
6072 	while (cur < array_size) {
6073 		disk_key = (struct btrfs_disk_key *)ptr;
6074 		btrfs_disk_key_to_cpu(&key, disk_key);
6075 
6076 		len = sizeof(*disk_key); ptr += len;
6077 		sb_ptr += len;
6078 		cur += len;
6079 
6080 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6081 			chunk = (struct btrfs_chunk *)sb_ptr;
6082 			ret = read_one_chunk(root, &key, sb, chunk);
6083 			if (ret)
6084 				break;
6085 			num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6086 			len = btrfs_chunk_item_size(num_stripes);
6087 		} else {
6088 			ret = -EIO;
6089 			break;
6090 		}
6091 		ptr += len;
6092 		sb_ptr += len;
6093 		cur += len;
6094 	}
6095 	free_extent_buffer(sb);
6096 	return ret;
6097 }
6098 
6099 int btrfs_read_chunk_tree(struct btrfs_root *root)
6100 {
6101 	struct btrfs_path *path;
6102 	struct extent_buffer *leaf;
6103 	struct btrfs_key key;
6104 	struct btrfs_key found_key;
6105 	int ret;
6106 	int slot;
6107 
6108 	root = root->fs_info->chunk_root;
6109 
6110 	path = btrfs_alloc_path();
6111 	if (!path)
6112 		return -ENOMEM;
6113 
6114 	mutex_lock(&uuid_mutex);
6115 	lock_chunks(root);
6116 
6117 	/*
6118 	 * Read all device items, and then all the chunk items. All
6119 	 * device items are found before any chunk item (their object id
6120 	 * is smaller than the lowest possible object id for a chunk
6121 	 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6122 	 */
6123 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6124 	key.offset = 0;
6125 	key.type = 0;
6126 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6127 	if (ret < 0)
6128 		goto error;
6129 	while (1) {
6130 		leaf = path->nodes[0];
6131 		slot = path->slots[0];
6132 		if (slot >= btrfs_header_nritems(leaf)) {
6133 			ret = btrfs_next_leaf(root, path);
6134 			if (ret == 0)
6135 				continue;
6136 			if (ret < 0)
6137 				goto error;
6138 			break;
6139 		}
6140 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
6141 		if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6142 			struct btrfs_dev_item *dev_item;
6143 			dev_item = btrfs_item_ptr(leaf, slot,
6144 						  struct btrfs_dev_item);
6145 			ret = read_one_dev(root, leaf, dev_item);
6146 			if (ret)
6147 				goto error;
6148 		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6149 			struct btrfs_chunk *chunk;
6150 			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6151 			ret = read_one_chunk(root, &found_key, leaf, chunk);
6152 			if (ret)
6153 				goto error;
6154 		}
6155 		path->slots[0]++;
6156 	}
6157 	ret = 0;
6158 error:
6159 	unlock_chunks(root);
6160 	mutex_unlock(&uuid_mutex);
6161 
6162 	btrfs_free_path(path);
6163 	return ret;
6164 }
6165 
6166 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6167 {
6168 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6169 	struct btrfs_device *device;
6170 
6171 	while (fs_devices) {
6172 		mutex_lock(&fs_devices->device_list_mutex);
6173 		list_for_each_entry(device, &fs_devices->devices, dev_list)
6174 			device->dev_root = fs_info->dev_root;
6175 		mutex_unlock(&fs_devices->device_list_mutex);
6176 
6177 		fs_devices = fs_devices->seed;
6178 	}
6179 }
6180 
6181 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6182 {
6183 	int i;
6184 
6185 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6186 		btrfs_dev_stat_reset(dev, i);
6187 }
6188 
6189 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6190 {
6191 	struct btrfs_key key;
6192 	struct btrfs_key found_key;
6193 	struct btrfs_root *dev_root = fs_info->dev_root;
6194 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6195 	struct extent_buffer *eb;
6196 	int slot;
6197 	int ret = 0;
6198 	struct btrfs_device *device;
6199 	struct btrfs_path *path = NULL;
6200 	int i;
6201 
6202 	path = btrfs_alloc_path();
6203 	if (!path) {
6204 		ret = -ENOMEM;
6205 		goto out;
6206 	}
6207 
6208 	mutex_lock(&fs_devices->device_list_mutex);
6209 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
6210 		int item_size;
6211 		struct btrfs_dev_stats_item *ptr;
6212 
6213 		key.objectid = 0;
6214 		key.type = BTRFS_DEV_STATS_KEY;
6215 		key.offset = device->devid;
6216 		ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6217 		if (ret) {
6218 			__btrfs_reset_dev_stats(device);
6219 			device->dev_stats_valid = 1;
6220 			btrfs_release_path(path);
6221 			continue;
6222 		}
6223 		slot = path->slots[0];
6224 		eb = path->nodes[0];
6225 		btrfs_item_key_to_cpu(eb, &found_key, slot);
6226 		item_size = btrfs_item_size_nr(eb, slot);
6227 
6228 		ptr = btrfs_item_ptr(eb, slot,
6229 				     struct btrfs_dev_stats_item);
6230 
6231 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6232 			if (item_size >= (1 + i) * sizeof(__le64))
6233 				btrfs_dev_stat_set(device, i,
6234 					btrfs_dev_stats_value(eb, ptr, i));
6235 			else
6236 				btrfs_dev_stat_reset(device, i);
6237 		}
6238 
6239 		device->dev_stats_valid = 1;
6240 		btrfs_dev_stat_print_on_load(device);
6241 		btrfs_release_path(path);
6242 	}
6243 	mutex_unlock(&fs_devices->device_list_mutex);
6244 
6245 out:
6246 	btrfs_free_path(path);
6247 	return ret < 0 ? ret : 0;
6248 }
6249 
6250 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6251 				struct btrfs_root *dev_root,
6252 				struct btrfs_device *device)
6253 {
6254 	struct btrfs_path *path;
6255 	struct btrfs_key key;
6256 	struct extent_buffer *eb;
6257 	struct btrfs_dev_stats_item *ptr;
6258 	int ret;
6259 	int i;
6260 
6261 	key.objectid = 0;
6262 	key.type = BTRFS_DEV_STATS_KEY;
6263 	key.offset = device->devid;
6264 
6265 	path = btrfs_alloc_path();
6266 	BUG_ON(!path);
6267 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6268 	if (ret < 0) {
6269 		printk_in_rcu(KERN_WARNING "BTRFS: "
6270 			"error %d while searching for dev_stats item for device %s!\n",
6271 			      ret, rcu_str_deref(device->name));
6272 		goto out;
6273 	}
6274 
6275 	if (ret == 0 &&
6276 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6277 		/* need to delete old one and insert a new one */
6278 		ret = btrfs_del_item(trans, dev_root, path);
6279 		if (ret != 0) {
6280 			printk_in_rcu(KERN_WARNING "BTRFS: "
6281 				"delete too small dev_stats item for device %s failed %d!\n",
6282 				      rcu_str_deref(device->name), ret);
6283 			goto out;
6284 		}
6285 		ret = 1;
6286 	}
6287 
6288 	if (ret == 1) {
6289 		/* need to insert a new item */
6290 		btrfs_release_path(path);
6291 		ret = btrfs_insert_empty_item(trans, dev_root, path,
6292 					      &key, sizeof(*ptr));
6293 		if (ret < 0) {
6294 			printk_in_rcu(KERN_WARNING "BTRFS: "
6295 					  "insert dev_stats item for device %s failed %d!\n",
6296 				      rcu_str_deref(device->name), ret);
6297 			goto out;
6298 		}
6299 	}
6300 
6301 	eb = path->nodes[0];
6302 	ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6303 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6304 		btrfs_set_dev_stats_value(eb, ptr, i,
6305 					  btrfs_dev_stat_read(device, i));
6306 	btrfs_mark_buffer_dirty(eb);
6307 
6308 out:
6309 	btrfs_free_path(path);
6310 	return ret;
6311 }
6312 
6313 /*
6314  * called from commit_transaction. Writes all changed device stats to disk.
6315  */
6316 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6317 			struct btrfs_fs_info *fs_info)
6318 {
6319 	struct btrfs_root *dev_root = fs_info->dev_root;
6320 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6321 	struct btrfs_device *device;
6322 	int ret = 0;
6323 
6324 	mutex_lock(&fs_devices->device_list_mutex);
6325 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
6326 		if (!device->dev_stats_valid || !device->dev_stats_dirty)
6327 			continue;
6328 
6329 		ret = update_dev_stat_item(trans, dev_root, device);
6330 		if (!ret)
6331 			device->dev_stats_dirty = 0;
6332 	}
6333 	mutex_unlock(&fs_devices->device_list_mutex);
6334 
6335 	return ret;
6336 }
6337 
6338 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6339 {
6340 	btrfs_dev_stat_inc(dev, index);
6341 	btrfs_dev_stat_print_on_error(dev);
6342 }
6343 
6344 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6345 {
6346 	if (!dev->dev_stats_valid)
6347 		return;
6348 	printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6349 			   "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6350 			   rcu_str_deref(dev->name),
6351 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6352 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6353 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6354 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6355 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6356 }
6357 
6358 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6359 {
6360 	int i;
6361 
6362 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6363 		if (btrfs_dev_stat_read(dev, i) != 0)
6364 			break;
6365 	if (i == BTRFS_DEV_STAT_VALUES_MAX)
6366 		return; /* all values == 0, suppress message */
6367 
6368 	printk_in_rcu(KERN_INFO "BTRFS: "
6369 		   "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6370 	       rcu_str_deref(dev->name),
6371 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6372 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6373 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6374 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6375 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6376 }
6377 
6378 int btrfs_get_dev_stats(struct btrfs_root *root,
6379 			struct btrfs_ioctl_get_dev_stats *stats)
6380 {
6381 	struct btrfs_device *dev;
6382 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6383 	int i;
6384 
6385 	mutex_lock(&fs_devices->device_list_mutex);
6386 	dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6387 	mutex_unlock(&fs_devices->device_list_mutex);
6388 
6389 	if (!dev) {
6390 		btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6391 		return -ENODEV;
6392 	} else if (!dev->dev_stats_valid) {
6393 		btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6394 		return -ENODEV;
6395 	} else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6396 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6397 			if (stats->nr_items > i)
6398 				stats->values[i] =
6399 					btrfs_dev_stat_read_and_reset(dev, i);
6400 			else
6401 				btrfs_dev_stat_reset(dev, i);
6402 		}
6403 	} else {
6404 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6405 			if (stats->nr_items > i)
6406 				stats->values[i] = btrfs_dev_stat_read(dev, i);
6407 	}
6408 	if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6409 		stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6410 	return 0;
6411 }
6412 
6413 int btrfs_scratch_superblock(struct btrfs_device *device)
6414 {
6415 	struct buffer_head *bh;
6416 	struct btrfs_super_block *disk_super;
6417 
6418 	bh = btrfs_read_dev_super(device->bdev);
6419 	if (!bh)
6420 		return -EINVAL;
6421 	disk_super = (struct btrfs_super_block *)bh->b_data;
6422 
6423 	memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6424 	set_buffer_dirty(bh);
6425 	sync_dirty_buffer(bh);
6426 	brelse(bh);
6427 
6428 	return 0;
6429 }
6430