xref: /openbmc/linux/fs/btrfs/volumes.c (revision cdfce539)
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 <asm/div64.h>
30 #include "compat.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 
44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
45 				struct btrfs_root *root,
46 				struct btrfs_device *device);
47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
48 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
49 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
50 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
51 
52 static DEFINE_MUTEX(uuid_mutex);
53 static LIST_HEAD(fs_uuids);
54 
55 static void lock_chunks(struct btrfs_root *root)
56 {
57 	mutex_lock(&root->fs_info->chunk_mutex);
58 }
59 
60 static void unlock_chunks(struct btrfs_root *root)
61 {
62 	mutex_unlock(&root->fs_info->chunk_mutex);
63 }
64 
65 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
66 {
67 	struct btrfs_device *device;
68 	WARN_ON(fs_devices->opened);
69 	while (!list_empty(&fs_devices->devices)) {
70 		device = list_entry(fs_devices->devices.next,
71 				    struct btrfs_device, dev_list);
72 		list_del(&device->dev_list);
73 		rcu_string_free(device->name);
74 		kfree(device);
75 	}
76 	kfree(fs_devices);
77 }
78 
79 static void btrfs_kobject_uevent(struct block_device *bdev,
80 				 enum kobject_action action)
81 {
82 	int ret;
83 
84 	ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
85 	if (ret)
86 		pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
87 			action,
88 			kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
89 			&disk_to_dev(bdev->bd_disk)->kobj);
90 }
91 
92 void btrfs_cleanup_fs_uuids(void)
93 {
94 	struct btrfs_fs_devices *fs_devices;
95 
96 	while (!list_empty(&fs_uuids)) {
97 		fs_devices = list_entry(fs_uuids.next,
98 					struct btrfs_fs_devices, list);
99 		list_del(&fs_devices->list);
100 		free_fs_devices(fs_devices);
101 	}
102 }
103 
104 static noinline struct btrfs_device *__find_device(struct list_head *head,
105 						   u64 devid, u8 *uuid)
106 {
107 	struct btrfs_device *dev;
108 
109 	list_for_each_entry(dev, head, dev_list) {
110 		if (dev->devid == devid &&
111 		    (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
112 			return dev;
113 		}
114 	}
115 	return NULL;
116 }
117 
118 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
119 {
120 	struct btrfs_fs_devices *fs_devices;
121 
122 	list_for_each_entry(fs_devices, &fs_uuids, list) {
123 		if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
124 			return fs_devices;
125 	}
126 	return NULL;
127 }
128 
129 static int
130 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
131 		      int flush, struct block_device **bdev,
132 		      struct buffer_head **bh)
133 {
134 	int ret;
135 
136 	*bdev = blkdev_get_by_path(device_path, flags, holder);
137 
138 	if (IS_ERR(*bdev)) {
139 		ret = PTR_ERR(*bdev);
140 		printk(KERN_INFO "btrfs: open %s failed\n", device_path);
141 		goto error;
142 	}
143 
144 	if (flush)
145 		filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
146 	ret = set_blocksize(*bdev, 4096);
147 	if (ret) {
148 		blkdev_put(*bdev, flags);
149 		goto error;
150 	}
151 	invalidate_bdev(*bdev);
152 	*bh = btrfs_read_dev_super(*bdev);
153 	if (!*bh) {
154 		ret = -EINVAL;
155 		blkdev_put(*bdev, flags);
156 		goto error;
157 	}
158 
159 	return 0;
160 
161 error:
162 	*bdev = NULL;
163 	*bh = NULL;
164 	return ret;
165 }
166 
167 static void requeue_list(struct btrfs_pending_bios *pending_bios,
168 			struct bio *head, struct bio *tail)
169 {
170 
171 	struct bio *old_head;
172 
173 	old_head = pending_bios->head;
174 	pending_bios->head = head;
175 	if (pending_bios->tail)
176 		tail->bi_next = old_head;
177 	else
178 		pending_bios->tail = tail;
179 }
180 
181 /*
182  * we try to collect pending bios for a device so we don't get a large
183  * number of procs sending bios down to the same device.  This greatly
184  * improves the schedulers ability to collect and merge the bios.
185  *
186  * But, it also turns into a long list of bios to process and that is sure
187  * to eventually make the worker thread block.  The solution here is to
188  * make some progress and then put this work struct back at the end of
189  * the list if the block device is congested.  This way, multiple devices
190  * can make progress from a single worker thread.
191  */
192 static noinline void run_scheduled_bios(struct btrfs_device *device)
193 {
194 	struct bio *pending;
195 	struct backing_dev_info *bdi;
196 	struct btrfs_fs_info *fs_info;
197 	struct btrfs_pending_bios *pending_bios;
198 	struct bio *tail;
199 	struct bio *cur;
200 	int again = 0;
201 	unsigned long num_run;
202 	unsigned long batch_run = 0;
203 	unsigned long limit;
204 	unsigned long last_waited = 0;
205 	int force_reg = 0;
206 	int sync_pending = 0;
207 	struct blk_plug plug;
208 
209 	/*
210 	 * this function runs all the bios we've collected for
211 	 * a particular device.  We don't want to wander off to
212 	 * another device without first sending all of these down.
213 	 * So, setup a plug here and finish it off before we return
214 	 */
215 	blk_start_plug(&plug);
216 
217 	bdi = blk_get_backing_dev_info(device->bdev);
218 	fs_info = device->dev_root->fs_info;
219 	limit = btrfs_async_submit_limit(fs_info);
220 	limit = limit * 2 / 3;
221 
222 loop:
223 	spin_lock(&device->io_lock);
224 
225 loop_lock:
226 	num_run = 0;
227 
228 	/* take all the bios off the list at once and process them
229 	 * later on (without the lock held).  But, remember the
230 	 * tail and other pointers so the bios can be properly reinserted
231 	 * into the list if we hit congestion
232 	 */
233 	if (!force_reg && device->pending_sync_bios.head) {
234 		pending_bios = &device->pending_sync_bios;
235 		force_reg = 1;
236 	} else {
237 		pending_bios = &device->pending_bios;
238 		force_reg = 0;
239 	}
240 
241 	pending = pending_bios->head;
242 	tail = pending_bios->tail;
243 	WARN_ON(pending && !tail);
244 
245 	/*
246 	 * if pending was null this time around, no bios need processing
247 	 * at all and we can stop.  Otherwise it'll loop back up again
248 	 * and do an additional check so no bios are missed.
249 	 *
250 	 * device->running_pending is used to synchronize with the
251 	 * schedule_bio code.
252 	 */
253 	if (device->pending_sync_bios.head == NULL &&
254 	    device->pending_bios.head == NULL) {
255 		again = 0;
256 		device->running_pending = 0;
257 	} else {
258 		again = 1;
259 		device->running_pending = 1;
260 	}
261 
262 	pending_bios->head = NULL;
263 	pending_bios->tail = NULL;
264 
265 	spin_unlock(&device->io_lock);
266 
267 	while (pending) {
268 
269 		rmb();
270 		/* we want to work on both lists, but do more bios on the
271 		 * sync list than the regular list
272 		 */
273 		if ((num_run > 32 &&
274 		    pending_bios != &device->pending_sync_bios &&
275 		    device->pending_sync_bios.head) ||
276 		   (num_run > 64 && pending_bios == &device->pending_sync_bios &&
277 		    device->pending_bios.head)) {
278 			spin_lock(&device->io_lock);
279 			requeue_list(pending_bios, pending, tail);
280 			goto loop_lock;
281 		}
282 
283 		cur = pending;
284 		pending = pending->bi_next;
285 		cur->bi_next = NULL;
286 
287 		if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
288 		    waitqueue_active(&fs_info->async_submit_wait))
289 			wake_up(&fs_info->async_submit_wait);
290 
291 		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
292 
293 		/*
294 		 * if we're doing the sync list, record that our
295 		 * plug has some sync requests on it
296 		 *
297 		 * If we're doing the regular list and there are
298 		 * sync requests sitting around, unplug before
299 		 * we add more
300 		 */
301 		if (pending_bios == &device->pending_sync_bios) {
302 			sync_pending = 1;
303 		} else if (sync_pending) {
304 			blk_finish_plug(&plug);
305 			blk_start_plug(&plug);
306 			sync_pending = 0;
307 		}
308 
309 		btrfsic_submit_bio(cur->bi_rw, cur);
310 		num_run++;
311 		batch_run++;
312 		if (need_resched())
313 			cond_resched();
314 
315 		/*
316 		 * we made progress, there is more work to do and the bdi
317 		 * is now congested.  Back off and let other work structs
318 		 * run instead
319 		 */
320 		if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
321 		    fs_info->fs_devices->open_devices > 1) {
322 			struct io_context *ioc;
323 
324 			ioc = current->io_context;
325 
326 			/*
327 			 * the main goal here is that we don't want to
328 			 * block if we're going to be able to submit
329 			 * more requests without blocking.
330 			 *
331 			 * This code does two great things, it pokes into
332 			 * the elevator code from a filesystem _and_
333 			 * it makes assumptions about how batching works.
334 			 */
335 			if (ioc && ioc->nr_batch_requests > 0 &&
336 			    time_before(jiffies, ioc->last_waited + HZ/50UL) &&
337 			    (last_waited == 0 ||
338 			     ioc->last_waited == last_waited)) {
339 				/*
340 				 * we want to go through our batch of
341 				 * requests and stop.  So, we copy out
342 				 * the ioc->last_waited time and test
343 				 * against it before looping
344 				 */
345 				last_waited = ioc->last_waited;
346 				if (need_resched())
347 					cond_resched();
348 				continue;
349 			}
350 			spin_lock(&device->io_lock);
351 			requeue_list(pending_bios, pending, tail);
352 			device->running_pending = 1;
353 
354 			spin_unlock(&device->io_lock);
355 			btrfs_requeue_work(&device->work);
356 			goto done;
357 		}
358 		/* unplug every 64 requests just for good measure */
359 		if (batch_run % 64 == 0) {
360 			blk_finish_plug(&plug);
361 			blk_start_plug(&plug);
362 			sync_pending = 0;
363 		}
364 	}
365 
366 	cond_resched();
367 	if (again)
368 		goto loop;
369 
370 	spin_lock(&device->io_lock);
371 	if (device->pending_bios.head || device->pending_sync_bios.head)
372 		goto loop_lock;
373 	spin_unlock(&device->io_lock);
374 
375 done:
376 	blk_finish_plug(&plug);
377 }
378 
379 static void pending_bios_fn(struct btrfs_work *work)
380 {
381 	struct btrfs_device *device;
382 
383 	device = container_of(work, struct btrfs_device, work);
384 	run_scheduled_bios(device);
385 }
386 
387 static noinline int device_list_add(const char *path,
388 			   struct btrfs_super_block *disk_super,
389 			   u64 devid, struct btrfs_fs_devices **fs_devices_ret)
390 {
391 	struct btrfs_device *device;
392 	struct btrfs_fs_devices *fs_devices;
393 	struct rcu_string *name;
394 	u64 found_transid = btrfs_super_generation(disk_super);
395 
396 	fs_devices = find_fsid(disk_super->fsid);
397 	if (!fs_devices) {
398 		fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
399 		if (!fs_devices)
400 			return -ENOMEM;
401 		INIT_LIST_HEAD(&fs_devices->devices);
402 		INIT_LIST_HEAD(&fs_devices->alloc_list);
403 		list_add(&fs_devices->list, &fs_uuids);
404 		memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
405 		fs_devices->latest_devid = devid;
406 		fs_devices->latest_trans = found_transid;
407 		mutex_init(&fs_devices->device_list_mutex);
408 		device = NULL;
409 	} else {
410 		device = __find_device(&fs_devices->devices, devid,
411 				       disk_super->dev_item.uuid);
412 	}
413 	if (!device) {
414 		if (fs_devices->opened)
415 			return -EBUSY;
416 
417 		device = kzalloc(sizeof(*device), GFP_NOFS);
418 		if (!device) {
419 			/* we can safely leave the fs_devices entry around */
420 			return -ENOMEM;
421 		}
422 		device->devid = devid;
423 		device->dev_stats_valid = 0;
424 		device->work.func = pending_bios_fn;
425 		memcpy(device->uuid, disk_super->dev_item.uuid,
426 		       BTRFS_UUID_SIZE);
427 		spin_lock_init(&device->io_lock);
428 
429 		name = rcu_string_strdup(path, GFP_NOFS);
430 		if (!name) {
431 			kfree(device);
432 			return -ENOMEM;
433 		}
434 		rcu_assign_pointer(device->name, name);
435 		INIT_LIST_HEAD(&device->dev_alloc_list);
436 
437 		/* init readahead state */
438 		spin_lock_init(&device->reada_lock);
439 		device->reada_curr_zone = NULL;
440 		atomic_set(&device->reada_in_flight, 0);
441 		device->reada_next = 0;
442 		INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
443 		INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
444 
445 		mutex_lock(&fs_devices->device_list_mutex);
446 		list_add_rcu(&device->dev_list, &fs_devices->devices);
447 		mutex_unlock(&fs_devices->device_list_mutex);
448 
449 		device->fs_devices = fs_devices;
450 		fs_devices->num_devices++;
451 	} else if (!device->name || strcmp(device->name->str, path)) {
452 		name = rcu_string_strdup(path, GFP_NOFS);
453 		if (!name)
454 			return -ENOMEM;
455 		rcu_string_free(device->name);
456 		rcu_assign_pointer(device->name, name);
457 		if (device->missing) {
458 			fs_devices->missing_devices--;
459 			device->missing = 0;
460 		}
461 	}
462 
463 	if (found_transid > fs_devices->latest_trans) {
464 		fs_devices->latest_devid = devid;
465 		fs_devices->latest_trans = found_transid;
466 	}
467 	*fs_devices_ret = fs_devices;
468 	return 0;
469 }
470 
471 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
472 {
473 	struct btrfs_fs_devices *fs_devices;
474 	struct btrfs_device *device;
475 	struct btrfs_device *orig_dev;
476 
477 	fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
478 	if (!fs_devices)
479 		return ERR_PTR(-ENOMEM);
480 
481 	INIT_LIST_HEAD(&fs_devices->devices);
482 	INIT_LIST_HEAD(&fs_devices->alloc_list);
483 	INIT_LIST_HEAD(&fs_devices->list);
484 	mutex_init(&fs_devices->device_list_mutex);
485 	fs_devices->latest_devid = orig->latest_devid;
486 	fs_devices->latest_trans = orig->latest_trans;
487 	fs_devices->total_devices = orig->total_devices;
488 	memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
489 
490 	/* We have held the volume lock, it is safe to get the devices. */
491 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
492 		struct rcu_string *name;
493 
494 		device = kzalloc(sizeof(*device), GFP_NOFS);
495 		if (!device)
496 			goto error;
497 
498 		/*
499 		 * This is ok to do without rcu read locked because we hold the
500 		 * uuid mutex so nothing we touch in here is going to disappear.
501 		 */
502 		name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
503 		if (!name) {
504 			kfree(device);
505 			goto error;
506 		}
507 		rcu_assign_pointer(device->name, name);
508 
509 		device->devid = orig_dev->devid;
510 		device->work.func = pending_bios_fn;
511 		memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
512 		spin_lock_init(&device->io_lock);
513 		INIT_LIST_HEAD(&device->dev_list);
514 		INIT_LIST_HEAD(&device->dev_alloc_list);
515 
516 		list_add(&device->dev_list, &fs_devices->devices);
517 		device->fs_devices = fs_devices;
518 		fs_devices->num_devices++;
519 	}
520 	return fs_devices;
521 error:
522 	free_fs_devices(fs_devices);
523 	return ERR_PTR(-ENOMEM);
524 }
525 
526 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
527 			       struct btrfs_fs_devices *fs_devices, int step)
528 {
529 	struct btrfs_device *device, *next;
530 
531 	struct block_device *latest_bdev = NULL;
532 	u64 latest_devid = 0;
533 	u64 latest_transid = 0;
534 
535 	mutex_lock(&uuid_mutex);
536 again:
537 	/* This is the initialized path, it is safe to release the devices. */
538 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
539 		if (device->in_fs_metadata) {
540 			if (!device->is_tgtdev_for_dev_replace &&
541 			    (!latest_transid ||
542 			     device->generation > latest_transid)) {
543 				latest_devid = device->devid;
544 				latest_transid = device->generation;
545 				latest_bdev = device->bdev;
546 			}
547 			continue;
548 		}
549 
550 		if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
551 			/*
552 			 * In the first step, keep the device which has
553 			 * the correct fsid and the devid that is used
554 			 * for the dev_replace procedure.
555 			 * In the second step, the dev_replace state is
556 			 * read from the device tree and it is known
557 			 * whether the procedure is really active or
558 			 * not, which means whether this device is
559 			 * used or whether it should be removed.
560 			 */
561 			if (step == 0 || device->is_tgtdev_for_dev_replace) {
562 				continue;
563 			}
564 		}
565 		if (device->bdev) {
566 			blkdev_put(device->bdev, device->mode);
567 			device->bdev = NULL;
568 			fs_devices->open_devices--;
569 		}
570 		if (device->writeable) {
571 			list_del_init(&device->dev_alloc_list);
572 			device->writeable = 0;
573 			if (!device->is_tgtdev_for_dev_replace)
574 				fs_devices->rw_devices--;
575 		}
576 		list_del_init(&device->dev_list);
577 		fs_devices->num_devices--;
578 		rcu_string_free(device->name);
579 		kfree(device);
580 	}
581 
582 	if (fs_devices->seed) {
583 		fs_devices = fs_devices->seed;
584 		goto again;
585 	}
586 
587 	fs_devices->latest_bdev = latest_bdev;
588 	fs_devices->latest_devid = latest_devid;
589 	fs_devices->latest_trans = latest_transid;
590 
591 	mutex_unlock(&uuid_mutex);
592 }
593 
594 static void __free_device(struct work_struct *work)
595 {
596 	struct btrfs_device *device;
597 
598 	device = container_of(work, struct btrfs_device, rcu_work);
599 
600 	if (device->bdev)
601 		blkdev_put(device->bdev, device->mode);
602 
603 	rcu_string_free(device->name);
604 	kfree(device);
605 }
606 
607 static void free_device(struct rcu_head *head)
608 {
609 	struct btrfs_device *device;
610 
611 	device = container_of(head, struct btrfs_device, rcu);
612 
613 	INIT_WORK(&device->rcu_work, __free_device);
614 	schedule_work(&device->rcu_work);
615 }
616 
617 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
618 {
619 	struct btrfs_device *device;
620 
621 	if (--fs_devices->opened > 0)
622 		return 0;
623 
624 	mutex_lock(&fs_devices->device_list_mutex);
625 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
626 		struct btrfs_device *new_device;
627 		struct rcu_string *name;
628 
629 		if (device->bdev)
630 			fs_devices->open_devices--;
631 
632 		if (device->writeable && !device->is_tgtdev_for_dev_replace) {
633 			list_del_init(&device->dev_alloc_list);
634 			fs_devices->rw_devices--;
635 		}
636 
637 		if (device->can_discard)
638 			fs_devices->num_can_discard--;
639 
640 		new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
641 		BUG_ON(!new_device); /* -ENOMEM */
642 		memcpy(new_device, device, sizeof(*new_device));
643 
644 		/* Safe because we are under uuid_mutex */
645 		if (device->name) {
646 			name = rcu_string_strdup(device->name->str, GFP_NOFS);
647 			BUG_ON(device->name && !name); /* -ENOMEM */
648 			rcu_assign_pointer(new_device->name, name);
649 		}
650 		new_device->bdev = NULL;
651 		new_device->writeable = 0;
652 		new_device->in_fs_metadata = 0;
653 		new_device->can_discard = 0;
654 		spin_lock_init(&new_device->io_lock);
655 		list_replace_rcu(&device->dev_list, &new_device->dev_list);
656 
657 		call_rcu(&device->rcu, free_device);
658 	}
659 	mutex_unlock(&fs_devices->device_list_mutex);
660 
661 	WARN_ON(fs_devices->open_devices);
662 	WARN_ON(fs_devices->rw_devices);
663 	fs_devices->opened = 0;
664 	fs_devices->seeding = 0;
665 
666 	return 0;
667 }
668 
669 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
670 {
671 	struct btrfs_fs_devices *seed_devices = NULL;
672 	int ret;
673 
674 	mutex_lock(&uuid_mutex);
675 	ret = __btrfs_close_devices(fs_devices);
676 	if (!fs_devices->opened) {
677 		seed_devices = fs_devices->seed;
678 		fs_devices->seed = NULL;
679 	}
680 	mutex_unlock(&uuid_mutex);
681 
682 	while (seed_devices) {
683 		fs_devices = seed_devices;
684 		seed_devices = fs_devices->seed;
685 		__btrfs_close_devices(fs_devices);
686 		free_fs_devices(fs_devices);
687 	}
688 	/*
689 	 * Wait for rcu kworkers under __btrfs_close_devices
690 	 * to finish all blkdev_puts so device is really
691 	 * free when umount is done.
692 	 */
693 	rcu_barrier();
694 	return ret;
695 }
696 
697 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
698 				fmode_t flags, void *holder)
699 {
700 	struct request_queue *q;
701 	struct block_device *bdev;
702 	struct list_head *head = &fs_devices->devices;
703 	struct btrfs_device *device;
704 	struct block_device *latest_bdev = NULL;
705 	struct buffer_head *bh;
706 	struct btrfs_super_block *disk_super;
707 	u64 latest_devid = 0;
708 	u64 latest_transid = 0;
709 	u64 devid;
710 	int seeding = 1;
711 	int ret = 0;
712 
713 	flags |= FMODE_EXCL;
714 
715 	list_for_each_entry(device, head, dev_list) {
716 		if (device->bdev)
717 			continue;
718 		if (!device->name)
719 			continue;
720 
721 		/* Just open everything we can; ignore failures here */
722 		if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
723 					    &bdev, &bh))
724 			continue;
725 
726 		disk_super = (struct btrfs_super_block *)bh->b_data;
727 		devid = btrfs_stack_device_id(&disk_super->dev_item);
728 		if (devid != device->devid)
729 			goto error_brelse;
730 
731 		if (memcmp(device->uuid, disk_super->dev_item.uuid,
732 			   BTRFS_UUID_SIZE))
733 			goto error_brelse;
734 
735 		device->generation = btrfs_super_generation(disk_super);
736 		if (!latest_transid || device->generation > latest_transid) {
737 			latest_devid = devid;
738 			latest_transid = device->generation;
739 			latest_bdev = bdev;
740 		}
741 
742 		if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
743 			device->writeable = 0;
744 		} else {
745 			device->writeable = !bdev_read_only(bdev);
746 			seeding = 0;
747 		}
748 
749 		q = bdev_get_queue(bdev);
750 		if (blk_queue_discard(q)) {
751 			device->can_discard = 1;
752 			fs_devices->num_can_discard++;
753 		}
754 
755 		device->bdev = bdev;
756 		device->in_fs_metadata = 0;
757 		device->mode = flags;
758 
759 		if (!blk_queue_nonrot(bdev_get_queue(bdev)))
760 			fs_devices->rotating = 1;
761 
762 		fs_devices->open_devices++;
763 		if (device->writeable && !device->is_tgtdev_for_dev_replace) {
764 			fs_devices->rw_devices++;
765 			list_add(&device->dev_alloc_list,
766 				 &fs_devices->alloc_list);
767 		}
768 		brelse(bh);
769 		continue;
770 
771 error_brelse:
772 		brelse(bh);
773 		blkdev_put(bdev, flags);
774 		continue;
775 	}
776 	if (fs_devices->open_devices == 0) {
777 		ret = -EINVAL;
778 		goto out;
779 	}
780 	fs_devices->seeding = seeding;
781 	fs_devices->opened = 1;
782 	fs_devices->latest_bdev = latest_bdev;
783 	fs_devices->latest_devid = latest_devid;
784 	fs_devices->latest_trans = latest_transid;
785 	fs_devices->total_rw_bytes = 0;
786 out:
787 	return ret;
788 }
789 
790 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
791 		       fmode_t flags, void *holder)
792 {
793 	int ret;
794 
795 	mutex_lock(&uuid_mutex);
796 	if (fs_devices->opened) {
797 		fs_devices->opened++;
798 		ret = 0;
799 	} else {
800 		ret = __btrfs_open_devices(fs_devices, flags, holder);
801 	}
802 	mutex_unlock(&uuid_mutex);
803 	return ret;
804 }
805 
806 /*
807  * Look for a btrfs signature on a device. This may be called out of the mount path
808  * and we are not allowed to call set_blocksize during the scan. The superblock
809  * is read via pagecache
810  */
811 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
812 			  struct btrfs_fs_devices **fs_devices_ret)
813 {
814 	struct btrfs_super_block *disk_super;
815 	struct block_device *bdev;
816 	struct page *page;
817 	void *p;
818 	int ret = -EINVAL;
819 	u64 devid;
820 	u64 transid;
821 	u64 total_devices;
822 	u64 bytenr;
823 	pgoff_t index;
824 
825 	/*
826 	 * we would like to check all the supers, but that would make
827 	 * a btrfs mount succeed after a mkfs from a different FS.
828 	 * So, we need to add a special mount option to scan for
829 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
830 	 */
831 	bytenr = btrfs_sb_offset(0);
832 	flags |= FMODE_EXCL;
833 	mutex_lock(&uuid_mutex);
834 
835 	bdev = blkdev_get_by_path(path, flags, holder);
836 
837 	if (IS_ERR(bdev)) {
838 		ret = PTR_ERR(bdev);
839 		goto error;
840 	}
841 
842 	/* make sure our super fits in the device */
843 	if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
844 		goto error_bdev_put;
845 
846 	/* make sure our super fits in the page */
847 	if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
848 		goto error_bdev_put;
849 
850 	/* make sure our super doesn't straddle pages on disk */
851 	index = bytenr >> PAGE_CACHE_SHIFT;
852 	if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
853 		goto error_bdev_put;
854 
855 	/* pull in the page with our super */
856 	page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
857 				   index, GFP_NOFS);
858 
859 	if (IS_ERR_OR_NULL(page))
860 		goto error_bdev_put;
861 
862 	p = kmap(page);
863 
864 	/* align our pointer to the offset of the super block */
865 	disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
866 
867 	if (btrfs_super_bytenr(disk_super) != bytenr ||
868 	    disk_super->magic != cpu_to_le64(BTRFS_MAGIC))
869 		goto error_unmap;
870 
871 	devid = btrfs_stack_device_id(&disk_super->dev_item);
872 	transid = btrfs_super_generation(disk_super);
873 	total_devices = btrfs_super_num_devices(disk_super);
874 
875 	if (disk_super->label[0]) {
876 		if (disk_super->label[BTRFS_LABEL_SIZE - 1])
877 			disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
878 		printk(KERN_INFO "device label %s ", disk_super->label);
879 	} else {
880 		printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
881 	}
882 
883 	printk(KERN_CONT "devid %llu transid %llu %s\n",
884 	       (unsigned long long)devid, (unsigned long long)transid, path);
885 
886 	ret = device_list_add(path, disk_super, devid, fs_devices_ret);
887 	if (!ret && fs_devices_ret)
888 		(*fs_devices_ret)->total_devices = total_devices;
889 
890 error_unmap:
891 	kunmap(page);
892 	page_cache_release(page);
893 
894 error_bdev_put:
895 	blkdev_put(bdev, flags);
896 error:
897 	mutex_unlock(&uuid_mutex);
898 	return ret;
899 }
900 
901 /* helper to account the used device space in the range */
902 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
903 				   u64 end, u64 *length)
904 {
905 	struct btrfs_key key;
906 	struct btrfs_root *root = device->dev_root;
907 	struct btrfs_dev_extent *dev_extent;
908 	struct btrfs_path *path;
909 	u64 extent_end;
910 	int ret;
911 	int slot;
912 	struct extent_buffer *l;
913 
914 	*length = 0;
915 
916 	if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
917 		return 0;
918 
919 	path = btrfs_alloc_path();
920 	if (!path)
921 		return -ENOMEM;
922 	path->reada = 2;
923 
924 	key.objectid = device->devid;
925 	key.offset = start;
926 	key.type = BTRFS_DEV_EXTENT_KEY;
927 
928 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
929 	if (ret < 0)
930 		goto out;
931 	if (ret > 0) {
932 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
933 		if (ret < 0)
934 			goto out;
935 	}
936 
937 	while (1) {
938 		l = path->nodes[0];
939 		slot = path->slots[0];
940 		if (slot >= btrfs_header_nritems(l)) {
941 			ret = btrfs_next_leaf(root, path);
942 			if (ret == 0)
943 				continue;
944 			if (ret < 0)
945 				goto out;
946 
947 			break;
948 		}
949 		btrfs_item_key_to_cpu(l, &key, slot);
950 
951 		if (key.objectid < device->devid)
952 			goto next;
953 
954 		if (key.objectid > device->devid)
955 			break;
956 
957 		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
958 			goto next;
959 
960 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
961 		extent_end = key.offset + btrfs_dev_extent_length(l,
962 								  dev_extent);
963 		if (key.offset <= start && extent_end > end) {
964 			*length = end - start + 1;
965 			break;
966 		} else if (key.offset <= start && extent_end > start)
967 			*length += extent_end - start;
968 		else if (key.offset > start && extent_end <= end)
969 			*length += extent_end - key.offset;
970 		else if (key.offset > start && key.offset <= end) {
971 			*length += end - key.offset + 1;
972 			break;
973 		} else if (key.offset > end)
974 			break;
975 
976 next:
977 		path->slots[0]++;
978 	}
979 	ret = 0;
980 out:
981 	btrfs_free_path(path);
982 	return ret;
983 }
984 
985 /*
986  * find_free_dev_extent - find free space in the specified device
987  * @device:	the device which we search the free space in
988  * @num_bytes:	the size of the free space that we need
989  * @start:	store the start of the free space.
990  * @len:	the size of the free space. that we find, or the size of the max
991  * 		free space if we don't find suitable free space
992  *
993  * this uses a pretty simple search, the expectation is that it is
994  * called very infrequently and that a given device has a small number
995  * of extents
996  *
997  * @start is used to store the start of the free space if we find. But if we
998  * don't find suitable free space, it will be used to store the start position
999  * of the max free space.
1000  *
1001  * @len is used to store the size of the free space that we find.
1002  * But if we don't find suitable free space, it is used to store the size of
1003  * the max free space.
1004  */
1005 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1006 			 u64 *start, u64 *len)
1007 {
1008 	struct btrfs_key key;
1009 	struct btrfs_root *root = device->dev_root;
1010 	struct btrfs_dev_extent *dev_extent;
1011 	struct btrfs_path *path;
1012 	u64 hole_size;
1013 	u64 max_hole_start;
1014 	u64 max_hole_size;
1015 	u64 extent_end;
1016 	u64 search_start;
1017 	u64 search_end = device->total_bytes;
1018 	int ret;
1019 	int slot;
1020 	struct extent_buffer *l;
1021 
1022 	/* FIXME use last free of some kind */
1023 
1024 	/* we don't want to overwrite the superblock on the drive,
1025 	 * so we make sure to start at an offset of at least 1MB
1026 	 */
1027 	search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1028 
1029 	max_hole_start = search_start;
1030 	max_hole_size = 0;
1031 	hole_size = 0;
1032 
1033 	if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1034 		ret = -ENOSPC;
1035 		goto error;
1036 	}
1037 
1038 	path = btrfs_alloc_path();
1039 	if (!path) {
1040 		ret = -ENOMEM;
1041 		goto error;
1042 	}
1043 	path->reada = 2;
1044 
1045 	key.objectid = device->devid;
1046 	key.offset = search_start;
1047 	key.type = BTRFS_DEV_EXTENT_KEY;
1048 
1049 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1050 	if (ret < 0)
1051 		goto out;
1052 	if (ret > 0) {
1053 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
1054 		if (ret < 0)
1055 			goto out;
1056 	}
1057 
1058 	while (1) {
1059 		l = path->nodes[0];
1060 		slot = path->slots[0];
1061 		if (slot >= btrfs_header_nritems(l)) {
1062 			ret = btrfs_next_leaf(root, path);
1063 			if (ret == 0)
1064 				continue;
1065 			if (ret < 0)
1066 				goto out;
1067 
1068 			break;
1069 		}
1070 		btrfs_item_key_to_cpu(l, &key, slot);
1071 
1072 		if (key.objectid < device->devid)
1073 			goto next;
1074 
1075 		if (key.objectid > device->devid)
1076 			break;
1077 
1078 		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1079 			goto next;
1080 
1081 		if (key.offset > search_start) {
1082 			hole_size = key.offset - search_start;
1083 
1084 			if (hole_size > max_hole_size) {
1085 				max_hole_start = search_start;
1086 				max_hole_size = hole_size;
1087 			}
1088 
1089 			/*
1090 			 * If this free space is greater than which we need,
1091 			 * it must be the max free space that we have found
1092 			 * until now, so max_hole_start must point to the start
1093 			 * of this free space and the length of this free space
1094 			 * is stored in max_hole_size. Thus, we return
1095 			 * max_hole_start and max_hole_size and go back to the
1096 			 * caller.
1097 			 */
1098 			if (hole_size >= num_bytes) {
1099 				ret = 0;
1100 				goto out;
1101 			}
1102 		}
1103 
1104 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1105 		extent_end = key.offset + btrfs_dev_extent_length(l,
1106 								  dev_extent);
1107 		if (extent_end > search_start)
1108 			search_start = extent_end;
1109 next:
1110 		path->slots[0]++;
1111 		cond_resched();
1112 	}
1113 
1114 	/*
1115 	 * At this point, search_start should be the end of
1116 	 * allocated dev extents, and when shrinking the device,
1117 	 * search_end may be smaller than search_start.
1118 	 */
1119 	if (search_end > search_start)
1120 		hole_size = search_end - search_start;
1121 
1122 	if (hole_size > max_hole_size) {
1123 		max_hole_start = search_start;
1124 		max_hole_size = hole_size;
1125 	}
1126 
1127 	/* See above. */
1128 	if (hole_size < num_bytes)
1129 		ret = -ENOSPC;
1130 	else
1131 		ret = 0;
1132 
1133 out:
1134 	btrfs_free_path(path);
1135 error:
1136 	*start = max_hole_start;
1137 	if (len)
1138 		*len = max_hole_size;
1139 	return ret;
1140 }
1141 
1142 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1143 			  struct btrfs_device *device,
1144 			  u64 start)
1145 {
1146 	int ret;
1147 	struct btrfs_path *path;
1148 	struct btrfs_root *root = device->dev_root;
1149 	struct btrfs_key key;
1150 	struct btrfs_key found_key;
1151 	struct extent_buffer *leaf = NULL;
1152 	struct btrfs_dev_extent *extent = NULL;
1153 
1154 	path = btrfs_alloc_path();
1155 	if (!path)
1156 		return -ENOMEM;
1157 
1158 	key.objectid = device->devid;
1159 	key.offset = start;
1160 	key.type = BTRFS_DEV_EXTENT_KEY;
1161 again:
1162 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1163 	if (ret > 0) {
1164 		ret = btrfs_previous_item(root, path, key.objectid,
1165 					  BTRFS_DEV_EXTENT_KEY);
1166 		if (ret)
1167 			goto out;
1168 		leaf = path->nodes[0];
1169 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1170 		extent = btrfs_item_ptr(leaf, path->slots[0],
1171 					struct btrfs_dev_extent);
1172 		BUG_ON(found_key.offset > start || found_key.offset +
1173 		       btrfs_dev_extent_length(leaf, extent) < start);
1174 		key = found_key;
1175 		btrfs_release_path(path);
1176 		goto again;
1177 	} else if (ret == 0) {
1178 		leaf = path->nodes[0];
1179 		extent = btrfs_item_ptr(leaf, path->slots[0],
1180 					struct btrfs_dev_extent);
1181 	} else {
1182 		btrfs_error(root->fs_info, ret, "Slot search failed");
1183 		goto out;
1184 	}
1185 
1186 	if (device->bytes_used > 0) {
1187 		u64 len = btrfs_dev_extent_length(leaf, extent);
1188 		device->bytes_used -= len;
1189 		spin_lock(&root->fs_info->free_chunk_lock);
1190 		root->fs_info->free_chunk_space += len;
1191 		spin_unlock(&root->fs_info->free_chunk_lock);
1192 	}
1193 	ret = btrfs_del_item(trans, root, path);
1194 	if (ret) {
1195 		btrfs_error(root->fs_info, ret,
1196 			    "Failed to remove dev extent item");
1197 	}
1198 out:
1199 	btrfs_free_path(path);
1200 	return ret;
1201 }
1202 
1203 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1204 				  struct btrfs_device *device,
1205 				  u64 chunk_tree, u64 chunk_objectid,
1206 				  u64 chunk_offset, u64 start, u64 num_bytes)
1207 {
1208 	int ret;
1209 	struct btrfs_path *path;
1210 	struct btrfs_root *root = device->dev_root;
1211 	struct btrfs_dev_extent *extent;
1212 	struct extent_buffer *leaf;
1213 	struct btrfs_key key;
1214 
1215 	WARN_ON(!device->in_fs_metadata);
1216 	WARN_ON(device->is_tgtdev_for_dev_replace);
1217 	path = btrfs_alloc_path();
1218 	if (!path)
1219 		return -ENOMEM;
1220 
1221 	key.objectid = device->devid;
1222 	key.offset = start;
1223 	key.type = BTRFS_DEV_EXTENT_KEY;
1224 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1225 				      sizeof(*extent));
1226 	if (ret)
1227 		goto out;
1228 
1229 	leaf = path->nodes[0];
1230 	extent = btrfs_item_ptr(leaf, path->slots[0],
1231 				struct btrfs_dev_extent);
1232 	btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1233 	btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1234 	btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1235 
1236 	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1237 		    (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1238 		    BTRFS_UUID_SIZE);
1239 
1240 	btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1241 	btrfs_mark_buffer_dirty(leaf);
1242 out:
1243 	btrfs_free_path(path);
1244 	return ret;
1245 }
1246 
1247 static noinline int find_next_chunk(struct btrfs_root *root,
1248 				    u64 objectid, u64 *offset)
1249 {
1250 	struct btrfs_path *path;
1251 	int ret;
1252 	struct btrfs_key key;
1253 	struct btrfs_chunk *chunk;
1254 	struct btrfs_key found_key;
1255 
1256 	path = btrfs_alloc_path();
1257 	if (!path)
1258 		return -ENOMEM;
1259 
1260 	key.objectid = objectid;
1261 	key.offset = (u64)-1;
1262 	key.type = BTRFS_CHUNK_ITEM_KEY;
1263 
1264 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1265 	if (ret < 0)
1266 		goto error;
1267 
1268 	BUG_ON(ret == 0); /* Corruption */
1269 
1270 	ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1271 	if (ret) {
1272 		*offset = 0;
1273 	} else {
1274 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1275 				      path->slots[0]);
1276 		if (found_key.objectid != objectid)
1277 			*offset = 0;
1278 		else {
1279 			chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1280 					       struct btrfs_chunk);
1281 			*offset = found_key.offset +
1282 				btrfs_chunk_length(path->nodes[0], chunk);
1283 		}
1284 	}
1285 	ret = 0;
1286 error:
1287 	btrfs_free_path(path);
1288 	return ret;
1289 }
1290 
1291 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1292 {
1293 	int ret;
1294 	struct btrfs_key key;
1295 	struct btrfs_key found_key;
1296 	struct btrfs_path *path;
1297 
1298 	root = root->fs_info->chunk_root;
1299 
1300 	path = btrfs_alloc_path();
1301 	if (!path)
1302 		return -ENOMEM;
1303 
1304 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1305 	key.type = BTRFS_DEV_ITEM_KEY;
1306 	key.offset = (u64)-1;
1307 
1308 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1309 	if (ret < 0)
1310 		goto error;
1311 
1312 	BUG_ON(ret == 0); /* Corruption */
1313 
1314 	ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1315 				  BTRFS_DEV_ITEM_KEY);
1316 	if (ret) {
1317 		*objectid = 1;
1318 	} else {
1319 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1320 				      path->slots[0]);
1321 		*objectid = found_key.offset + 1;
1322 	}
1323 	ret = 0;
1324 error:
1325 	btrfs_free_path(path);
1326 	return ret;
1327 }
1328 
1329 /*
1330  * the device information is stored in the chunk root
1331  * the btrfs_device struct should be fully filled in
1332  */
1333 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1334 			    struct btrfs_root *root,
1335 			    struct btrfs_device *device)
1336 {
1337 	int ret;
1338 	struct btrfs_path *path;
1339 	struct btrfs_dev_item *dev_item;
1340 	struct extent_buffer *leaf;
1341 	struct btrfs_key key;
1342 	unsigned long ptr;
1343 
1344 	root = root->fs_info->chunk_root;
1345 
1346 	path = btrfs_alloc_path();
1347 	if (!path)
1348 		return -ENOMEM;
1349 
1350 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1351 	key.type = BTRFS_DEV_ITEM_KEY;
1352 	key.offset = device->devid;
1353 
1354 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1355 				      sizeof(*dev_item));
1356 	if (ret)
1357 		goto out;
1358 
1359 	leaf = path->nodes[0];
1360 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1361 
1362 	btrfs_set_device_id(leaf, dev_item, device->devid);
1363 	btrfs_set_device_generation(leaf, dev_item, 0);
1364 	btrfs_set_device_type(leaf, dev_item, device->type);
1365 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1366 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1367 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1368 	btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1369 	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1370 	btrfs_set_device_group(leaf, dev_item, 0);
1371 	btrfs_set_device_seek_speed(leaf, dev_item, 0);
1372 	btrfs_set_device_bandwidth(leaf, dev_item, 0);
1373 	btrfs_set_device_start_offset(leaf, dev_item, 0);
1374 
1375 	ptr = (unsigned long)btrfs_device_uuid(dev_item);
1376 	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1377 	ptr = (unsigned long)btrfs_device_fsid(dev_item);
1378 	write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1379 	btrfs_mark_buffer_dirty(leaf);
1380 
1381 	ret = 0;
1382 out:
1383 	btrfs_free_path(path);
1384 	return ret;
1385 }
1386 
1387 static int btrfs_rm_dev_item(struct btrfs_root *root,
1388 			     struct btrfs_device *device)
1389 {
1390 	int ret;
1391 	struct btrfs_path *path;
1392 	struct btrfs_key key;
1393 	struct btrfs_trans_handle *trans;
1394 
1395 	root = root->fs_info->chunk_root;
1396 
1397 	path = btrfs_alloc_path();
1398 	if (!path)
1399 		return -ENOMEM;
1400 
1401 	trans = btrfs_start_transaction(root, 0);
1402 	if (IS_ERR(trans)) {
1403 		btrfs_free_path(path);
1404 		return PTR_ERR(trans);
1405 	}
1406 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1407 	key.type = BTRFS_DEV_ITEM_KEY;
1408 	key.offset = device->devid;
1409 	lock_chunks(root);
1410 
1411 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1412 	if (ret < 0)
1413 		goto out;
1414 
1415 	if (ret > 0) {
1416 		ret = -ENOENT;
1417 		goto out;
1418 	}
1419 
1420 	ret = btrfs_del_item(trans, root, path);
1421 	if (ret)
1422 		goto out;
1423 out:
1424 	btrfs_free_path(path);
1425 	unlock_chunks(root);
1426 	btrfs_commit_transaction(trans, root);
1427 	return ret;
1428 }
1429 
1430 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1431 {
1432 	struct btrfs_device *device;
1433 	struct btrfs_device *next_device;
1434 	struct block_device *bdev;
1435 	struct buffer_head *bh = NULL;
1436 	struct btrfs_super_block *disk_super;
1437 	struct btrfs_fs_devices *cur_devices;
1438 	u64 all_avail;
1439 	u64 devid;
1440 	u64 num_devices;
1441 	u8 *dev_uuid;
1442 	unsigned seq;
1443 	int ret = 0;
1444 	bool clear_super = false;
1445 
1446 	mutex_lock(&uuid_mutex);
1447 
1448 	do {
1449 		seq = read_seqbegin(&root->fs_info->profiles_lock);
1450 
1451 		all_avail = root->fs_info->avail_data_alloc_bits |
1452 			    root->fs_info->avail_system_alloc_bits |
1453 			    root->fs_info->avail_metadata_alloc_bits;
1454 	} while (read_seqretry(&root->fs_info->profiles_lock, seq));
1455 
1456 	num_devices = root->fs_info->fs_devices->num_devices;
1457 	btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1458 	if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1459 		WARN_ON(num_devices < 1);
1460 		num_devices--;
1461 	}
1462 	btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1463 
1464 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1465 		printk(KERN_ERR "btrfs: unable to go below four devices "
1466 		       "on raid10\n");
1467 		ret = -EINVAL;
1468 		goto out;
1469 	}
1470 
1471 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1472 		printk(KERN_ERR "btrfs: unable to go below two "
1473 		       "devices on raid1\n");
1474 		ret = -EINVAL;
1475 		goto out;
1476 	}
1477 
1478 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1479 	    root->fs_info->fs_devices->rw_devices <= 2) {
1480 		printk(KERN_ERR "btrfs: unable to go below two "
1481 		       "devices on raid5\n");
1482 		ret = -EINVAL;
1483 		goto out;
1484 	}
1485 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1486 	    root->fs_info->fs_devices->rw_devices <= 3) {
1487 		printk(KERN_ERR "btrfs: unable to go below three "
1488 		       "devices on raid6\n");
1489 		ret = -EINVAL;
1490 		goto out;
1491 	}
1492 
1493 	if (strcmp(device_path, "missing") == 0) {
1494 		struct list_head *devices;
1495 		struct btrfs_device *tmp;
1496 
1497 		device = NULL;
1498 		devices = &root->fs_info->fs_devices->devices;
1499 		/*
1500 		 * It is safe to read the devices since the volume_mutex
1501 		 * is held.
1502 		 */
1503 		list_for_each_entry(tmp, devices, dev_list) {
1504 			if (tmp->in_fs_metadata &&
1505 			    !tmp->is_tgtdev_for_dev_replace &&
1506 			    !tmp->bdev) {
1507 				device = tmp;
1508 				break;
1509 			}
1510 		}
1511 		bdev = NULL;
1512 		bh = NULL;
1513 		disk_super = NULL;
1514 		if (!device) {
1515 			printk(KERN_ERR "btrfs: no missing devices found to "
1516 			       "remove\n");
1517 			goto out;
1518 		}
1519 	} else {
1520 		ret = btrfs_get_bdev_and_sb(device_path,
1521 					    FMODE_WRITE | FMODE_EXCL,
1522 					    root->fs_info->bdev_holder, 0,
1523 					    &bdev, &bh);
1524 		if (ret)
1525 			goto out;
1526 		disk_super = (struct btrfs_super_block *)bh->b_data;
1527 		devid = btrfs_stack_device_id(&disk_super->dev_item);
1528 		dev_uuid = disk_super->dev_item.uuid;
1529 		device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1530 					   disk_super->fsid);
1531 		if (!device) {
1532 			ret = -ENOENT;
1533 			goto error_brelse;
1534 		}
1535 	}
1536 
1537 	if (device->is_tgtdev_for_dev_replace) {
1538 		pr_err("btrfs: unable to remove the dev_replace target dev\n");
1539 		ret = -EINVAL;
1540 		goto error_brelse;
1541 	}
1542 
1543 	if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1544 		printk(KERN_ERR "btrfs: unable to remove the only writeable "
1545 		       "device\n");
1546 		ret = -EINVAL;
1547 		goto error_brelse;
1548 	}
1549 
1550 	if (device->writeable) {
1551 		lock_chunks(root);
1552 		list_del_init(&device->dev_alloc_list);
1553 		unlock_chunks(root);
1554 		root->fs_info->fs_devices->rw_devices--;
1555 		clear_super = true;
1556 	}
1557 
1558 	ret = btrfs_shrink_device(device, 0);
1559 	if (ret)
1560 		goto error_undo;
1561 
1562 	/*
1563 	 * TODO: the superblock still includes this device in its num_devices
1564 	 * counter although write_all_supers() is not locked out. This
1565 	 * could give a filesystem state which requires a degraded mount.
1566 	 */
1567 	ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1568 	if (ret)
1569 		goto error_undo;
1570 
1571 	spin_lock(&root->fs_info->free_chunk_lock);
1572 	root->fs_info->free_chunk_space = device->total_bytes -
1573 		device->bytes_used;
1574 	spin_unlock(&root->fs_info->free_chunk_lock);
1575 
1576 	device->in_fs_metadata = 0;
1577 	btrfs_scrub_cancel_dev(root->fs_info, device);
1578 
1579 	/*
1580 	 * the device list mutex makes sure that we don't change
1581 	 * the device list while someone else is writing out all
1582 	 * the device supers.
1583 	 */
1584 
1585 	cur_devices = device->fs_devices;
1586 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1587 	list_del_rcu(&device->dev_list);
1588 
1589 	device->fs_devices->num_devices--;
1590 	device->fs_devices->total_devices--;
1591 
1592 	if (device->missing)
1593 		root->fs_info->fs_devices->missing_devices--;
1594 
1595 	next_device = list_entry(root->fs_info->fs_devices->devices.next,
1596 				 struct btrfs_device, dev_list);
1597 	if (device->bdev == root->fs_info->sb->s_bdev)
1598 		root->fs_info->sb->s_bdev = next_device->bdev;
1599 	if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1600 		root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1601 
1602 	if (device->bdev)
1603 		device->fs_devices->open_devices--;
1604 
1605 	call_rcu(&device->rcu, free_device);
1606 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1607 
1608 	num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1609 	btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1610 
1611 	if (cur_devices->open_devices == 0) {
1612 		struct btrfs_fs_devices *fs_devices;
1613 		fs_devices = root->fs_info->fs_devices;
1614 		while (fs_devices) {
1615 			if (fs_devices->seed == cur_devices)
1616 				break;
1617 			fs_devices = fs_devices->seed;
1618 		}
1619 		fs_devices->seed = cur_devices->seed;
1620 		cur_devices->seed = NULL;
1621 		lock_chunks(root);
1622 		__btrfs_close_devices(cur_devices);
1623 		unlock_chunks(root);
1624 		free_fs_devices(cur_devices);
1625 	}
1626 
1627 	root->fs_info->num_tolerated_disk_barrier_failures =
1628 		btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1629 
1630 	/*
1631 	 * at this point, the device is zero sized.  We want to
1632 	 * remove it from the devices list and zero out the old super
1633 	 */
1634 	if (clear_super && disk_super) {
1635 		/* make sure this device isn't detected as part of
1636 		 * the FS anymore
1637 		 */
1638 		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1639 		set_buffer_dirty(bh);
1640 		sync_dirty_buffer(bh);
1641 	}
1642 
1643 	ret = 0;
1644 
1645 	/* Notify udev that device has changed */
1646 	if (bdev)
1647 		btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1648 
1649 error_brelse:
1650 	brelse(bh);
1651 	if (bdev)
1652 		blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1653 out:
1654 	mutex_unlock(&uuid_mutex);
1655 	return ret;
1656 error_undo:
1657 	if (device->writeable) {
1658 		lock_chunks(root);
1659 		list_add(&device->dev_alloc_list,
1660 			 &root->fs_info->fs_devices->alloc_list);
1661 		unlock_chunks(root);
1662 		root->fs_info->fs_devices->rw_devices++;
1663 	}
1664 	goto error_brelse;
1665 }
1666 
1667 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1668 				 struct btrfs_device *srcdev)
1669 {
1670 	WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1671 	list_del_rcu(&srcdev->dev_list);
1672 	list_del_rcu(&srcdev->dev_alloc_list);
1673 	fs_info->fs_devices->num_devices--;
1674 	if (srcdev->missing) {
1675 		fs_info->fs_devices->missing_devices--;
1676 		fs_info->fs_devices->rw_devices++;
1677 	}
1678 	if (srcdev->can_discard)
1679 		fs_info->fs_devices->num_can_discard--;
1680 	if (srcdev->bdev)
1681 		fs_info->fs_devices->open_devices--;
1682 
1683 	call_rcu(&srcdev->rcu, free_device);
1684 }
1685 
1686 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1687 				      struct btrfs_device *tgtdev)
1688 {
1689 	struct btrfs_device *next_device;
1690 
1691 	WARN_ON(!tgtdev);
1692 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
1693 	if (tgtdev->bdev) {
1694 		btrfs_scratch_superblock(tgtdev);
1695 		fs_info->fs_devices->open_devices--;
1696 	}
1697 	fs_info->fs_devices->num_devices--;
1698 	if (tgtdev->can_discard)
1699 		fs_info->fs_devices->num_can_discard++;
1700 
1701 	next_device = list_entry(fs_info->fs_devices->devices.next,
1702 				 struct btrfs_device, dev_list);
1703 	if (tgtdev->bdev == fs_info->sb->s_bdev)
1704 		fs_info->sb->s_bdev = next_device->bdev;
1705 	if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1706 		fs_info->fs_devices->latest_bdev = next_device->bdev;
1707 	list_del_rcu(&tgtdev->dev_list);
1708 
1709 	call_rcu(&tgtdev->rcu, free_device);
1710 
1711 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1712 }
1713 
1714 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1715 				     struct btrfs_device **device)
1716 {
1717 	int ret = 0;
1718 	struct btrfs_super_block *disk_super;
1719 	u64 devid;
1720 	u8 *dev_uuid;
1721 	struct block_device *bdev;
1722 	struct buffer_head *bh;
1723 
1724 	*device = NULL;
1725 	ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1726 				    root->fs_info->bdev_holder, 0, &bdev, &bh);
1727 	if (ret)
1728 		return ret;
1729 	disk_super = (struct btrfs_super_block *)bh->b_data;
1730 	devid = btrfs_stack_device_id(&disk_super->dev_item);
1731 	dev_uuid = disk_super->dev_item.uuid;
1732 	*device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1733 				    disk_super->fsid);
1734 	brelse(bh);
1735 	if (!*device)
1736 		ret = -ENOENT;
1737 	blkdev_put(bdev, FMODE_READ);
1738 	return ret;
1739 }
1740 
1741 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1742 					 char *device_path,
1743 					 struct btrfs_device **device)
1744 {
1745 	*device = NULL;
1746 	if (strcmp(device_path, "missing") == 0) {
1747 		struct list_head *devices;
1748 		struct btrfs_device *tmp;
1749 
1750 		devices = &root->fs_info->fs_devices->devices;
1751 		/*
1752 		 * It is safe to read the devices since the volume_mutex
1753 		 * is held by the caller.
1754 		 */
1755 		list_for_each_entry(tmp, devices, dev_list) {
1756 			if (tmp->in_fs_metadata && !tmp->bdev) {
1757 				*device = tmp;
1758 				break;
1759 			}
1760 		}
1761 
1762 		if (!*device) {
1763 			pr_err("btrfs: no missing device found\n");
1764 			return -ENOENT;
1765 		}
1766 
1767 		return 0;
1768 	} else {
1769 		return btrfs_find_device_by_path(root, device_path, device);
1770 	}
1771 }
1772 
1773 /*
1774  * does all the dirty work required for changing file system's UUID.
1775  */
1776 static int btrfs_prepare_sprout(struct btrfs_root *root)
1777 {
1778 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1779 	struct btrfs_fs_devices *old_devices;
1780 	struct btrfs_fs_devices *seed_devices;
1781 	struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1782 	struct btrfs_device *device;
1783 	u64 super_flags;
1784 
1785 	BUG_ON(!mutex_is_locked(&uuid_mutex));
1786 	if (!fs_devices->seeding)
1787 		return -EINVAL;
1788 
1789 	seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1790 	if (!seed_devices)
1791 		return -ENOMEM;
1792 
1793 	old_devices = clone_fs_devices(fs_devices);
1794 	if (IS_ERR(old_devices)) {
1795 		kfree(seed_devices);
1796 		return PTR_ERR(old_devices);
1797 	}
1798 
1799 	list_add(&old_devices->list, &fs_uuids);
1800 
1801 	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1802 	seed_devices->opened = 1;
1803 	INIT_LIST_HEAD(&seed_devices->devices);
1804 	INIT_LIST_HEAD(&seed_devices->alloc_list);
1805 	mutex_init(&seed_devices->device_list_mutex);
1806 
1807 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1808 	list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1809 			      synchronize_rcu);
1810 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1811 
1812 	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1813 	list_for_each_entry(device, &seed_devices->devices, dev_list) {
1814 		device->fs_devices = seed_devices;
1815 	}
1816 
1817 	fs_devices->seeding = 0;
1818 	fs_devices->num_devices = 0;
1819 	fs_devices->open_devices = 0;
1820 	fs_devices->total_devices = 0;
1821 	fs_devices->seed = seed_devices;
1822 
1823 	generate_random_uuid(fs_devices->fsid);
1824 	memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1825 	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1826 	super_flags = btrfs_super_flags(disk_super) &
1827 		      ~BTRFS_SUPER_FLAG_SEEDING;
1828 	btrfs_set_super_flags(disk_super, super_flags);
1829 
1830 	return 0;
1831 }
1832 
1833 /*
1834  * strore the expected generation for seed devices in device items.
1835  */
1836 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1837 			       struct btrfs_root *root)
1838 {
1839 	struct btrfs_path *path;
1840 	struct extent_buffer *leaf;
1841 	struct btrfs_dev_item *dev_item;
1842 	struct btrfs_device *device;
1843 	struct btrfs_key key;
1844 	u8 fs_uuid[BTRFS_UUID_SIZE];
1845 	u8 dev_uuid[BTRFS_UUID_SIZE];
1846 	u64 devid;
1847 	int ret;
1848 
1849 	path = btrfs_alloc_path();
1850 	if (!path)
1851 		return -ENOMEM;
1852 
1853 	root = root->fs_info->chunk_root;
1854 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1855 	key.offset = 0;
1856 	key.type = BTRFS_DEV_ITEM_KEY;
1857 
1858 	while (1) {
1859 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1860 		if (ret < 0)
1861 			goto error;
1862 
1863 		leaf = path->nodes[0];
1864 next_slot:
1865 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1866 			ret = btrfs_next_leaf(root, path);
1867 			if (ret > 0)
1868 				break;
1869 			if (ret < 0)
1870 				goto error;
1871 			leaf = path->nodes[0];
1872 			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1873 			btrfs_release_path(path);
1874 			continue;
1875 		}
1876 
1877 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1878 		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1879 		    key.type != BTRFS_DEV_ITEM_KEY)
1880 			break;
1881 
1882 		dev_item = btrfs_item_ptr(leaf, path->slots[0],
1883 					  struct btrfs_dev_item);
1884 		devid = btrfs_device_id(leaf, dev_item);
1885 		read_extent_buffer(leaf, dev_uuid,
1886 				   (unsigned long)btrfs_device_uuid(dev_item),
1887 				   BTRFS_UUID_SIZE);
1888 		read_extent_buffer(leaf, fs_uuid,
1889 				   (unsigned long)btrfs_device_fsid(dev_item),
1890 				   BTRFS_UUID_SIZE);
1891 		device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1892 					   fs_uuid);
1893 		BUG_ON(!device); /* Logic error */
1894 
1895 		if (device->fs_devices->seeding) {
1896 			btrfs_set_device_generation(leaf, dev_item,
1897 						    device->generation);
1898 			btrfs_mark_buffer_dirty(leaf);
1899 		}
1900 
1901 		path->slots[0]++;
1902 		goto next_slot;
1903 	}
1904 	ret = 0;
1905 error:
1906 	btrfs_free_path(path);
1907 	return ret;
1908 }
1909 
1910 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1911 {
1912 	struct request_queue *q;
1913 	struct btrfs_trans_handle *trans;
1914 	struct btrfs_device *device;
1915 	struct block_device *bdev;
1916 	struct list_head *devices;
1917 	struct super_block *sb = root->fs_info->sb;
1918 	struct rcu_string *name;
1919 	u64 total_bytes;
1920 	int seeding_dev = 0;
1921 	int ret = 0;
1922 
1923 	if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1924 		return -EROFS;
1925 
1926 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1927 				  root->fs_info->bdev_holder);
1928 	if (IS_ERR(bdev))
1929 		return PTR_ERR(bdev);
1930 
1931 	if (root->fs_info->fs_devices->seeding) {
1932 		seeding_dev = 1;
1933 		down_write(&sb->s_umount);
1934 		mutex_lock(&uuid_mutex);
1935 	}
1936 
1937 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
1938 
1939 	devices = &root->fs_info->fs_devices->devices;
1940 
1941 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1942 	list_for_each_entry(device, devices, dev_list) {
1943 		if (device->bdev == bdev) {
1944 			ret = -EEXIST;
1945 			mutex_unlock(
1946 				&root->fs_info->fs_devices->device_list_mutex);
1947 			goto error;
1948 		}
1949 	}
1950 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1951 
1952 	device = kzalloc(sizeof(*device), GFP_NOFS);
1953 	if (!device) {
1954 		/* we can safely leave the fs_devices entry around */
1955 		ret = -ENOMEM;
1956 		goto error;
1957 	}
1958 
1959 	name = rcu_string_strdup(device_path, GFP_NOFS);
1960 	if (!name) {
1961 		kfree(device);
1962 		ret = -ENOMEM;
1963 		goto error;
1964 	}
1965 	rcu_assign_pointer(device->name, name);
1966 
1967 	ret = find_next_devid(root, &device->devid);
1968 	if (ret) {
1969 		rcu_string_free(device->name);
1970 		kfree(device);
1971 		goto error;
1972 	}
1973 
1974 	trans = btrfs_start_transaction(root, 0);
1975 	if (IS_ERR(trans)) {
1976 		rcu_string_free(device->name);
1977 		kfree(device);
1978 		ret = PTR_ERR(trans);
1979 		goto error;
1980 	}
1981 
1982 	lock_chunks(root);
1983 
1984 	q = bdev_get_queue(bdev);
1985 	if (blk_queue_discard(q))
1986 		device->can_discard = 1;
1987 	device->writeable = 1;
1988 	device->work.func = pending_bios_fn;
1989 	generate_random_uuid(device->uuid);
1990 	spin_lock_init(&device->io_lock);
1991 	device->generation = trans->transid;
1992 	device->io_width = root->sectorsize;
1993 	device->io_align = root->sectorsize;
1994 	device->sector_size = root->sectorsize;
1995 	device->total_bytes = i_size_read(bdev->bd_inode);
1996 	device->disk_total_bytes = device->total_bytes;
1997 	device->dev_root = root->fs_info->dev_root;
1998 	device->bdev = bdev;
1999 	device->in_fs_metadata = 1;
2000 	device->is_tgtdev_for_dev_replace = 0;
2001 	device->mode = FMODE_EXCL;
2002 	set_blocksize(device->bdev, 4096);
2003 
2004 	if (seeding_dev) {
2005 		sb->s_flags &= ~MS_RDONLY;
2006 		ret = btrfs_prepare_sprout(root);
2007 		BUG_ON(ret); /* -ENOMEM */
2008 	}
2009 
2010 	device->fs_devices = root->fs_info->fs_devices;
2011 
2012 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2013 	list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2014 	list_add(&device->dev_alloc_list,
2015 		 &root->fs_info->fs_devices->alloc_list);
2016 	root->fs_info->fs_devices->num_devices++;
2017 	root->fs_info->fs_devices->open_devices++;
2018 	root->fs_info->fs_devices->rw_devices++;
2019 	root->fs_info->fs_devices->total_devices++;
2020 	if (device->can_discard)
2021 		root->fs_info->fs_devices->num_can_discard++;
2022 	root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2023 
2024 	spin_lock(&root->fs_info->free_chunk_lock);
2025 	root->fs_info->free_chunk_space += device->total_bytes;
2026 	spin_unlock(&root->fs_info->free_chunk_lock);
2027 
2028 	if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2029 		root->fs_info->fs_devices->rotating = 1;
2030 
2031 	total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2032 	btrfs_set_super_total_bytes(root->fs_info->super_copy,
2033 				    total_bytes + device->total_bytes);
2034 
2035 	total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2036 	btrfs_set_super_num_devices(root->fs_info->super_copy,
2037 				    total_bytes + 1);
2038 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2039 
2040 	if (seeding_dev) {
2041 		ret = init_first_rw_device(trans, root, device);
2042 		if (ret) {
2043 			btrfs_abort_transaction(trans, root, ret);
2044 			goto error_trans;
2045 		}
2046 		ret = btrfs_finish_sprout(trans, root);
2047 		if (ret) {
2048 			btrfs_abort_transaction(trans, root, ret);
2049 			goto error_trans;
2050 		}
2051 	} else {
2052 		ret = btrfs_add_device(trans, root, device);
2053 		if (ret) {
2054 			btrfs_abort_transaction(trans, root, ret);
2055 			goto error_trans;
2056 		}
2057 	}
2058 
2059 	/*
2060 	 * we've got more storage, clear any full flags on the space
2061 	 * infos
2062 	 */
2063 	btrfs_clear_space_info_full(root->fs_info);
2064 
2065 	unlock_chunks(root);
2066 	root->fs_info->num_tolerated_disk_barrier_failures =
2067 		btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2068 	ret = btrfs_commit_transaction(trans, root);
2069 
2070 	if (seeding_dev) {
2071 		mutex_unlock(&uuid_mutex);
2072 		up_write(&sb->s_umount);
2073 
2074 		if (ret) /* transaction commit */
2075 			return ret;
2076 
2077 		ret = btrfs_relocate_sys_chunks(root);
2078 		if (ret < 0)
2079 			btrfs_error(root->fs_info, ret,
2080 				    "Failed to relocate sys chunks after "
2081 				    "device initialization. This can be fixed "
2082 				    "using the \"btrfs balance\" command.");
2083 		trans = btrfs_attach_transaction(root);
2084 		if (IS_ERR(trans)) {
2085 			if (PTR_ERR(trans) == -ENOENT)
2086 				return 0;
2087 			return PTR_ERR(trans);
2088 		}
2089 		ret = btrfs_commit_transaction(trans, root);
2090 	}
2091 
2092 	return ret;
2093 
2094 error_trans:
2095 	unlock_chunks(root);
2096 	btrfs_end_transaction(trans, root);
2097 	rcu_string_free(device->name);
2098 	kfree(device);
2099 error:
2100 	blkdev_put(bdev, FMODE_EXCL);
2101 	if (seeding_dev) {
2102 		mutex_unlock(&uuid_mutex);
2103 		up_write(&sb->s_umount);
2104 	}
2105 	return ret;
2106 }
2107 
2108 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2109 				  struct btrfs_device **device_out)
2110 {
2111 	struct request_queue *q;
2112 	struct btrfs_device *device;
2113 	struct block_device *bdev;
2114 	struct btrfs_fs_info *fs_info = root->fs_info;
2115 	struct list_head *devices;
2116 	struct rcu_string *name;
2117 	int ret = 0;
2118 
2119 	*device_out = NULL;
2120 	if (fs_info->fs_devices->seeding)
2121 		return -EINVAL;
2122 
2123 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2124 				  fs_info->bdev_holder);
2125 	if (IS_ERR(bdev))
2126 		return PTR_ERR(bdev);
2127 
2128 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
2129 
2130 	devices = &fs_info->fs_devices->devices;
2131 	list_for_each_entry(device, devices, dev_list) {
2132 		if (device->bdev == bdev) {
2133 			ret = -EEXIST;
2134 			goto error;
2135 		}
2136 	}
2137 
2138 	device = kzalloc(sizeof(*device), GFP_NOFS);
2139 	if (!device) {
2140 		ret = -ENOMEM;
2141 		goto error;
2142 	}
2143 
2144 	name = rcu_string_strdup(device_path, GFP_NOFS);
2145 	if (!name) {
2146 		kfree(device);
2147 		ret = -ENOMEM;
2148 		goto error;
2149 	}
2150 	rcu_assign_pointer(device->name, name);
2151 
2152 	q = bdev_get_queue(bdev);
2153 	if (blk_queue_discard(q))
2154 		device->can_discard = 1;
2155 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2156 	device->writeable = 1;
2157 	device->work.func = pending_bios_fn;
2158 	generate_random_uuid(device->uuid);
2159 	device->devid = BTRFS_DEV_REPLACE_DEVID;
2160 	spin_lock_init(&device->io_lock);
2161 	device->generation = 0;
2162 	device->io_width = root->sectorsize;
2163 	device->io_align = root->sectorsize;
2164 	device->sector_size = root->sectorsize;
2165 	device->total_bytes = i_size_read(bdev->bd_inode);
2166 	device->disk_total_bytes = device->total_bytes;
2167 	device->dev_root = fs_info->dev_root;
2168 	device->bdev = bdev;
2169 	device->in_fs_metadata = 1;
2170 	device->is_tgtdev_for_dev_replace = 1;
2171 	device->mode = FMODE_EXCL;
2172 	set_blocksize(device->bdev, 4096);
2173 	device->fs_devices = fs_info->fs_devices;
2174 	list_add(&device->dev_list, &fs_info->fs_devices->devices);
2175 	fs_info->fs_devices->num_devices++;
2176 	fs_info->fs_devices->open_devices++;
2177 	if (device->can_discard)
2178 		fs_info->fs_devices->num_can_discard++;
2179 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2180 
2181 	*device_out = device;
2182 	return ret;
2183 
2184 error:
2185 	blkdev_put(bdev, FMODE_EXCL);
2186 	return ret;
2187 }
2188 
2189 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2190 					      struct btrfs_device *tgtdev)
2191 {
2192 	WARN_ON(fs_info->fs_devices->rw_devices == 0);
2193 	tgtdev->io_width = fs_info->dev_root->sectorsize;
2194 	tgtdev->io_align = fs_info->dev_root->sectorsize;
2195 	tgtdev->sector_size = fs_info->dev_root->sectorsize;
2196 	tgtdev->dev_root = fs_info->dev_root;
2197 	tgtdev->in_fs_metadata = 1;
2198 }
2199 
2200 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2201 					struct btrfs_device *device)
2202 {
2203 	int ret;
2204 	struct btrfs_path *path;
2205 	struct btrfs_root *root;
2206 	struct btrfs_dev_item *dev_item;
2207 	struct extent_buffer *leaf;
2208 	struct btrfs_key key;
2209 
2210 	root = device->dev_root->fs_info->chunk_root;
2211 
2212 	path = btrfs_alloc_path();
2213 	if (!path)
2214 		return -ENOMEM;
2215 
2216 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2217 	key.type = BTRFS_DEV_ITEM_KEY;
2218 	key.offset = device->devid;
2219 
2220 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2221 	if (ret < 0)
2222 		goto out;
2223 
2224 	if (ret > 0) {
2225 		ret = -ENOENT;
2226 		goto out;
2227 	}
2228 
2229 	leaf = path->nodes[0];
2230 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2231 
2232 	btrfs_set_device_id(leaf, dev_item, device->devid);
2233 	btrfs_set_device_type(leaf, dev_item, device->type);
2234 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2235 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2236 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2237 	btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2238 	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2239 	btrfs_mark_buffer_dirty(leaf);
2240 
2241 out:
2242 	btrfs_free_path(path);
2243 	return ret;
2244 }
2245 
2246 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2247 		      struct btrfs_device *device, u64 new_size)
2248 {
2249 	struct btrfs_super_block *super_copy =
2250 		device->dev_root->fs_info->super_copy;
2251 	u64 old_total = btrfs_super_total_bytes(super_copy);
2252 	u64 diff = new_size - device->total_bytes;
2253 
2254 	if (!device->writeable)
2255 		return -EACCES;
2256 	if (new_size <= device->total_bytes ||
2257 	    device->is_tgtdev_for_dev_replace)
2258 		return -EINVAL;
2259 
2260 	btrfs_set_super_total_bytes(super_copy, old_total + diff);
2261 	device->fs_devices->total_rw_bytes += diff;
2262 
2263 	device->total_bytes = new_size;
2264 	device->disk_total_bytes = new_size;
2265 	btrfs_clear_space_info_full(device->dev_root->fs_info);
2266 
2267 	return btrfs_update_device(trans, device);
2268 }
2269 
2270 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2271 		      struct btrfs_device *device, u64 new_size)
2272 {
2273 	int ret;
2274 	lock_chunks(device->dev_root);
2275 	ret = __btrfs_grow_device(trans, device, new_size);
2276 	unlock_chunks(device->dev_root);
2277 	return ret;
2278 }
2279 
2280 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2281 			    struct btrfs_root *root,
2282 			    u64 chunk_tree, u64 chunk_objectid,
2283 			    u64 chunk_offset)
2284 {
2285 	int ret;
2286 	struct btrfs_path *path;
2287 	struct btrfs_key key;
2288 
2289 	root = root->fs_info->chunk_root;
2290 	path = btrfs_alloc_path();
2291 	if (!path)
2292 		return -ENOMEM;
2293 
2294 	key.objectid = chunk_objectid;
2295 	key.offset = chunk_offset;
2296 	key.type = BTRFS_CHUNK_ITEM_KEY;
2297 
2298 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2299 	if (ret < 0)
2300 		goto out;
2301 	else if (ret > 0) { /* Logic error or corruption */
2302 		btrfs_error(root->fs_info, -ENOENT,
2303 			    "Failed lookup while freeing chunk.");
2304 		ret = -ENOENT;
2305 		goto out;
2306 	}
2307 
2308 	ret = btrfs_del_item(trans, root, path);
2309 	if (ret < 0)
2310 		btrfs_error(root->fs_info, ret,
2311 			    "Failed to delete chunk item.");
2312 out:
2313 	btrfs_free_path(path);
2314 	return ret;
2315 }
2316 
2317 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2318 			chunk_offset)
2319 {
2320 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2321 	struct btrfs_disk_key *disk_key;
2322 	struct btrfs_chunk *chunk;
2323 	u8 *ptr;
2324 	int ret = 0;
2325 	u32 num_stripes;
2326 	u32 array_size;
2327 	u32 len = 0;
2328 	u32 cur;
2329 	struct btrfs_key key;
2330 
2331 	array_size = btrfs_super_sys_array_size(super_copy);
2332 
2333 	ptr = super_copy->sys_chunk_array;
2334 	cur = 0;
2335 
2336 	while (cur < array_size) {
2337 		disk_key = (struct btrfs_disk_key *)ptr;
2338 		btrfs_disk_key_to_cpu(&key, disk_key);
2339 
2340 		len = sizeof(*disk_key);
2341 
2342 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2343 			chunk = (struct btrfs_chunk *)(ptr + len);
2344 			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2345 			len += btrfs_chunk_item_size(num_stripes);
2346 		} else {
2347 			ret = -EIO;
2348 			break;
2349 		}
2350 		if (key.objectid == chunk_objectid &&
2351 		    key.offset == chunk_offset) {
2352 			memmove(ptr, ptr + len, array_size - (cur + len));
2353 			array_size -= len;
2354 			btrfs_set_super_sys_array_size(super_copy, array_size);
2355 		} else {
2356 			ptr += len;
2357 			cur += len;
2358 		}
2359 	}
2360 	return ret;
2361 }
2362 
2363 static int btrfs_relocate_chunk(struct btrfs_root *root,
2364 			 u64 chunk_tree, u64 chunk_objectid,
2365 			 u64 chunk_offset)
2366 {
2367 	struct extent_map_tree *em_tree;
2368 	struct btrfs_root *extent_root;
2369 	struct btrfs_trans_handle *trans;
2370 	struct extent_map *em;
2371 	struct map_lookup *map;
2372 	int ret;
2373 	int i;
2374 
2375 	root = root->fs_info->chunk_root;
2376 	extent_root = root->fs_info->extent_root;
2377 	em_tree = &root->fs_info->mapping_tree.map_tree;
2378 
2379 	ret = btrfs_can_relocate(extent_root, chunk_offset);
2380 	if (ret)
2381 		return -ENOSPC;
2382 
2383 	/* step one, relocate all the extents inside this chunk */
2384 	ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2385 	if (ret)
2386 		return ret;
2387 
2388 	trans = btrfs_start_transaction(root, 0);
2389 	if (IS_ERR(trans)) {
2390 		ret = PTR_ERR(trans);
2391 		btrfs_std_error(root->fs_info, ret);
2392 		return ret;
2393 	}
2394 
2395 	lock_chunks(root);
2396 
2397 	/*
2398 	 * step two, delete the device extents and the
2399 	 * chunk tree entries
2400 	 */
2401 	read_lock(&em_tree->lock);
2402 	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2403 	read_unlock(&em_tree->lock);
2404 
2405 	BUG_ON(!em || em->start > chunk_offset ||
2406 	       em->start + em->len < chunk_offset);
2407 	map = (struct map_lookup *)em->bdev;
2408 
2409 	for (i = 0; i < map->num_stripes; i++) {
2410 		ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2411 					    map->stripes[i].physical);
2412 		BUG_ON(ret);
2413 
2414 		if (map->stripes[i].dev) {
2415 			ret = btrfs_update_device(trans, map->stripes[i].dev);
2416 			BUG_ON(ret);
2417 		}
2418 	}
2419 	ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2420 			       chunk_offset);
2421 
2422 	BUG_ON(ret);
2423 
2424 	trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2425 
2426 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2427 		ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2428 		BUG_ON(ret);
2429 	}
2430 
2431 	ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2432 	BUG_ON(ret);
2433 
2434 	write_lock(&em_tree->lock);
2435 	remove_extent_mapping(em_tree, em);
2436 	write_unlock(&em_tree->lock);
2437 
2438 	kfree(map);
2439 	em->bdev = NULL;
2440 
2441 	/* once for the tree */
2442 	free_extent_map(em);
2443 	/* once for us */
2444 	free_extent_map(em);
2445 
2446 	unlock_chunks(root);
2447 	btrfs_end_transaction(trans, root);
2448 	return 0;
2449 }
2450 
2451 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2452 {
2453 	struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2454 	struct btrfs_path *path;
2455 	struct extent_buffer *leaf;
2456 	struct btrfs_chunk *chunk;
2457 	struct btrfs_key key;
2458 	struct btrfs_key found_key;
2459 	u64 chunk_tree = chunk_root->root_key.objectid;
2460 	u64 chunk_type;
2461 	bool retried = false;
2462 	int failed = 0;
2463 	int ret;
2464 
2465 	path = btrfs_alloc_path();
2466 	if (!path)
2467 		return -ENOMEM;
2468 
2469 again:
2470 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2471 	key.offset = (u64)-1;
2472 	key.type = BTRFS_CHUNK_ITEM_KEY;
2473 
2474 	while (1) {
2475 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2476 		if (ret < 0)
2477 			goto error;
2478 		BUG_ON(ret == 0); /* Corruption */
2479 
2480 		ret = btrfs_previous_item(chunk_root, path, key.objectid,
2481 					  key.type);
2482 		if (ret < 0)
2483 			goto error;
2484 		if (ret > 0)
2485 			break;
2486 
2487 		leaf = path->nodes[0];
2488 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2489 
2490 		chunk = btrfs_item_ptr(leaf, path->slots[0],
2491 				       struct btrfs_chunk);
2492 		chunk_type = btrfs_chunk_type(leaf, chunk);
2493 		btrfs_release_path(path);
2494 
2495 		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2496 			ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2497 						   found_key.objectid,
2498 						   found_key.offset);
2499 			if (ret == -ENOSPC)
2500 				failed++;
2501 			else if (ret)
2502 				BUG();
2503 		}
2504 
2505 		if (found_key.offset == 0)
2506 			break;
2507 		key.offset = found_key.offset - 1;
2508 	}
2509 	ret = 0;
2510 	if (failed && !retried) {
2511 		failed = 0;
2512 		retried = true;
2513 		goto again;
2514 	} else if (failed && retried) {
2515 		WARN_ON(1);
2516 		ret = -ENOSPC;
2517 	}
2518 error:
2519 	btrfs_free_path(path);
2520 	return ret;
2521 }
2522 
2523 static int insert_balance_item(struct btrfs_root *root,
2524 			       struct btrfs_balance_control *bctl)
2525 {
2526 	struct btrfs_trans_handle *trans;
2527 	struct btrfs_balance_item *item;
2528 	struct btrfs_disk_balance_args disk_bargs;
2529 	struct btrfs_path *path;
2530 	struct extent_buffer *leaf;
2531 	struct btrfs_key key;
2532 	int ret, err;
2533 
2534 	path = btrfs_alloc_path();
2535 	if (!path)
2536 		return -ENOMEM;
2537 
2538 	trans = btrfs_start_transaction(root, 0);
2539 	if (IS_ERR(trans)) {
2540 		btrfs_free_path(path);
2541 		return PTR_ERR(trans);
2542 	}
2543 
2544 	key.objectid = BTRFS_BALANCE_OBJECTID;
2545 	key.type = BTRFS_BALANCE_ITEM_KEY;
2546 	key.offset = 0;
2547 
2548 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2549 				      sizeof(*item));
2550 	if (ret)
2551 		goto out;
2552 
2553 	leaf = path->nodes[0];
2554 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2555 
2556 	memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2557 
2558 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2559 	btrfs_set_balance_data(leaf, item, &disk_bargs);
2560 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2561 	btrfs_set_balance_meta(leaf, item, &disk_bargs);
2562 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2563 	btrfs_set_balance_sys(leaf, item, &disk_bargs);
2564 
2565 	btrfs_set_balance_flags(leaf, item, bctl->flags);
2566 
2567 	btrfs_mark_buffer_dirty(leaf);
2568 out:
2569 	btrfs_free_path(path);
2570 	err = btrfs_commit_transaction(trans, root);
2571 	if (err && !ret)
2572 		ret = err;
2573 	return ret;
2574 }
2575 
2576 static int del_balance_item(struct btrfs_root *root)
2577 {
2578 	struct btrfs_trans_handle *trans;
2579 	struct btrfs_path *path;
2580 	struct btrfs_key key;
2581 	int ret, err;
2582 
2583 	path = btrfs_alloc_path();
2584 	if (!path)
2585 		return -ENOMEM;
2586 
2587 	trans = btrfs_start_transaction(root, 0);
2588 	if (IS_ERR(trans)) {
2589 		btrfs_free_path(path);
2590 		return PTR_ERR(trans);
2591 	}
2592 
2593 	key.objectid = BTRFS_BALANCE_OBJECTID;
2594 	key.type = BTRFS_BALANCE_ITEM_KEY;
2595 	key.offset = 0;
2596 
2597 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2598 	if (ret < 0)
2599 		goto out;
2600 	if (ret > 0) {
2601 		ret = -ENOENT;
2602 		goto out;
2603 	}
2604 
2605 	ret = btrfs_del_item(trans, root, path);
2606 out:
2607 	btrfs_free_path(path);
2608 	err = btrfs_commit_transaction(trans, root);
2609 	if (err && !ret)
2610 		ret = err;
2611 	return ret;
2612 }
2613 
2614 /*
2615  * This is a heuristic used to reduce the number of chunks balanced on
2616  * resume after balance was interrupted.
2617  */
2618 static void update_balance_args(struct btrfs_balance_control *bctl)
2619 {
2620 	/*
2621 	 * Turn on soft mode for chunk types that were being converted.
2622 	 */
2623 	if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2624 		bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2625 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2626 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2627 	if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2628 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2629 
2630 	/*
2631 	 * Turn on usage filter if is not already used.  The idea is
2632 	 * that chunks that we have already balanced should be
2633 	 * reasonably full.  Don't do it for chunks that are being
2634 	 * converted - that will keep us from relocating unconverted
2635 	 * (albeit full) chunks.
2636 	 */
2637 	if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2638 	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2639 		bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2640 		bctl->data.usage = 90;
2641 	}
2642 	if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2643 	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2644 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2645 		bctl->sys.usage = 90;
2646 	}
2647 	if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2648 	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2649 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2650 		bctl->meta.usage = 90;
2651 	}
2652 }
2653 
2654 /*
2655  * Should be called with both balance and volume mutexes held to
2656  * serialize other volume operations (add_dev/rm_dev/resize) with
2657  * restriper.  Same goes for unset_balance_control.
2658  */
2659 static void set_balance_control(struct btrfs_balance_control *bctl)
2660 {
2661 	struct btrfs_fs_info *fs_info = bctl->fs_info;
2662 
2663 	BUG_ON(fs_info->balance_ctl);
2664 
2665 	spin_lock(&fs_info->balance_lock);
2666 	fs_info->balance_ctl = bctl;
2667 	spin_unlock(&fs_info->balance_lock);
2668 }
2669 
2670 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2671 {
2672 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2673 
2674 	BUG_ON(!fs_info->balance_ctl);
2675 
2676 	spin_lock(&fs_info->balance_lock);
2677 	fs_info->balance_ctl = NULL;
2678 	spin_unlock(&fs_info->balance_lock);
2679 
2680 	kfree(bctl);
2681 }
2682 
2683 /*
2684  * Balance filters.  Return 1 if chunk should be filtered out
2685  * (should not be balanced).
2686  */
2687 static int chunk_profiles_filter(u64 chunk_type,
2688 				 struct btrfs_balance_args *bargs)
2689 {
2690 	chunk_type = chunk_to_extended(chunk_type) &
2691 				BTRFS_EXTENDED_PROFILE_MASK;
2692 
2693 	if (bargs->profiles & chunk_type)
2694 		return 0;
2695 
2696 	return 1;
2697 }
2698 
2699 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2700 			      struct btrfs_balance_args *bargs)
2701 {
2702 	struct btrfs_block_group_cache *cache;
2703 	u64 chunk_used, user_thresh;
2704 	int ret = 1;
2705 
2706 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2707 	chunk_used = btrfs_block_group_used(&cache->item);
2708 
2709 	if (bargs->usage == 0)
2710 		user_thresh = 1;
2711 	else if (bargs->usage > 100)
2712 		user_thresh = cache->key.offset;
2713 	else
2714 		user_thresh = div_factor_fine(cache->key.offset,
2715 					      bargs->usage);
2716 
2717 	if (chunk_used < user_thresh)
2718 		ret = 0;
2719 
2720 	btrfs_put_block_group(cache);
2721 	return ret;
2722 }
2723 
2724 static int chunk_devid_filter(struct extent_buffer *leaf,
2725 			      struct btrfs_chunk *chunk,
2726 			      struct btrfs_balance_args *bargs)
2727 {
2728 	struct btrfs_stripe *stripe;
2729 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2730 	int i;
2731 
2732 	for (i = 0; i < num_stripes; i++) {
2733 		stripe = btrfs_stripe_nr(chunk, i);
2734 		if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2735 			return 0;
2736 	}
2737 
2738 	return 1;
2739 }
2740 
2741 /* [pstart, pend) */
2742 static int chunk_drange_filter(struct extent_buffer *leaf,
2743 			       struct btrfs_chunk *chunk,
2744 			       u64 chunk_offset,
2745 			       struct btrfs_balance_args *bargs)
2746 {
2747 	struct btrfs_stripe *stripe;
2748 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2749 	u64 stripe_offset;
2750 	u64 stripe_length;
2751 	int factor;
2752 	int i;
2753 
2754 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2755 		return 0;
2756 
2757 	if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2758 	     BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2759 		factor = num_stripes / 2;
2760 	} else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2761 		factor = num_stripes - 1;
2762 	} else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2763 		factor = num_stripes - 2;
2764 	} else {
2765 		factor = num_stripes;
2766 	}
2767 
2768 	for (i = 0; i < num_stripes; i++) {
2769 		stripe = btrfs_stripe_nr(chunk, i);
2770 		if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2771 			continue;
2772 
2773 		stripe_offset = btrfs_stripe_offset(leaf, stripe);
2774 		stripe_length = btrfs_chunk_length(leaf, chunk);
2775 		do_div(stripe_length, factor);
2776 
2777 		if (stripe_offset < bargs->pend &&
2778 		    stripe_offset + stripe_length > bargs->pstart)
2779 			return 0;
2780 	}
2781 
2782 	return 1;
2783 }
2784 
2785 /* [vstart, vend) */
2786 static int chunk_vrange_filter(struct extent_buffer *leaf,
2787 			       struct btrfs_chunk *chunk,
2788 			       u64 chunk_offset,
2789 			       struct btrfs_balance_args *bargs)
2790 {
2791 	if (chunk_offset < bargs->vend &&
2792 	    chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2793 		/* at least part of the chunk is inside this vrange */
2794 		return 0;
2795 
2796 	return 1;
2797 }
2798 
2799 static int chunk_soft_convert_filter(u64 chunk_type,
2800 				     struct btrfs_balance_args *bargs)
2801 {
2802 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2803 		return 0;
2804 
2805 	chunk_type = chunk_to_extended(chunk_type) &
2806 				BTRFS_EXTENDED_PROFILE_MASK;
2807 
2808 	if (bargs->target == chunk_type)
2809 		return 1;
2810 
2811 	return 0;
2812 }
2813 
2814 static int should_balance_chunk(struct btrfs_root *root,
2815 				struct extent_buffer *leaf,
2816 				struct btrfs_chunk *chunk, u64 chunk_offset)
2817 {
2818 	struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2819 	struct btrfs_balance_args *bargs = NULL;
2820 	u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2821 
2822 	/* type filter */
2823 	if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2824 	      (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2825 		return 0;
2826 	}
2827 
2828 	if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2829 		bargs = &bctl->data;
2830 	else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2831 		bargs = &bctl->sys;
2832 	else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2833 		bargs = &bctl->meta;
2834 
2835 	/* profiles filter */
2836 	if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2837 	    chunk_profiles_filter(chunk_type, bargs)) {
2838 		return 0;
2839 	}
2840 
2841 	/* usage filter */
2842 	if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2843 	    chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2844 		return 0;
2845 	}
2846 
2847 	/* devid filter */
2848 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2849 	    chunk_devid_filter(leaf, chunk, bargs)) {
2850 		return 0;
2851 	}
2852 
2853 	/* drange filter, makes sense only with devid filter */
2854 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2855 	    chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2856 		return 0;
2857 	}
2858 
2859 	/* vrange filter */
2860 	if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2861 	    chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2862 		return 0;
2863 	}
2864 
2865 	/* soft profile changing mode */
2866 	if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2867 	    chunk_soft_convert_filter(chunk_type, bargs)) {
2868 		return 0;
2869 	}
2870 
2871 	return 1;
2872 }
2873 
2874 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2875 {
2876 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2877 	struct btrfs_root *chunk_root = fs_info->chunk_root;
2878 	struct btrfs_root *dev_root = fs_info->dev_root;
2879 	struct list_head *devices;
2880 	struct btrfs_device *device;
2881 	u64 old_size;
2882 	u64 size_to_free;
2883 	struct btrfs_chunk *chunk;
2884 	struct btrfs_path *path;
2885 	struct btrfs_key key;
2886 	struct btrfs_key found_key;
2887 	struct btrfs_trans_handle *trans;
2888 	struct extent_buffer *leaf;
2889 	int slot;
2890 	int ret;
2891 	int enospc_errors = 0;
2892 	bool counting = true;
2893 
2894 	/* step one make some room on all the devices */
2895 	devices = &fs_info->fs_devices->devices;
2896 	list_for_each_entry(device, devices, dev_list) {
2897 		old_size = device->total_bytes;
2898 		size_to_free = div_factor(old_size, 1);
2899 		size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2900 		if (!device->writeable ||
2901 		    device->total_bytes - device->bytes_used > size_to_free ||
2902 		    device->is_tgtdev_for_dev_replace)
2903 			continue;
2904 
2905 		ret = btrfs_shrink_device(device, old_size - size_to_free);
2906 		if (ret == -ENOSPC)
2907 			break;
2908 		BUG_ON(ret);
2909 
2910 		trans = btrfs_start_transaction(dev_root, 0);
2911 		BUG_ON(IS_ERR(trans));
2912 
2913 		ret = btrfs_grow_device(trans, device, old_size);
2914 		BUG_ON(ret);
2915 
2916 		btrfs_end_transaction(trans, dev_root);
2917 	}
2918 
2919 	/* step two, relocate all the chunks */
2920 	path = btrfs_alloc_path();
2921 	if (!path) {
2922 		ret = -ENOMEM;
2923 		goto error;
2924 	}
2925 
2926 	/* zero out stat counters */
2927 	spin_lock(&fs_info->balance_lock);
2928 	memset(&bctl->stat, 0, sizeof(bctl->stat));
2929 	spin_unlock(&fs_info->balance_lock);
2930 again:
2931 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2932 	key.offset = (u64)-1;
2933 	key.type = BTRFS_CHUNK_ITEM_KEY;
2934 
2935 	while (1) {
2936 		if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2937 		    atomic_read(&fs_info->balance_cancel_req)) {
2938 			ret = -ECANCELED;
2939 			goto error;
2940 		}
2941 
2942 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2943 		if (ret < 0)
2944 			goto error;
2945 
2946 		/*
2947 		 * this shouldn't happen, it means the last relocate
2948 		 * failed
2949 		 */
2950 		if (ret == 0)
2951 			BUG(); /* FIXME break ? */
2952 
2953 		ret = btrfs_previous_item(chunk_root, path, 0,
2954 					  BTRFS_CHUNK_ITEM_KEY);
2955 		if (ret) {
2956 			ret = 0;
2957 			break;
2958 		}
2959 
2960 		leaf = path->nodes[0];
2961 		slot = path->slots[0];
2962 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
2963 
2964 		if (found_key.objectid != key.objectid)
2965 			break;
2966 
2967 		/* chunk zero is special */
2968 		if (found_key.offset == 0)
2969 			break;
2970 
2971 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2972 
2973 		if (!counting) {
2974 			spin_lock(&fs_info->balance_lock);
2975 			bctl->stat.considered++;
2976 			spin_unlock(&fs_info->balance_lock);
2977 		}
2978 
2979 		ret = should_balance_chunk(chunk_root, leaf, chunk,
2980 					   found_key.offset);
2981 		btrfs_release_path(path);
2982 		if (!ret)
2983 			goto loop;
2984 
2985 		if (counting) {
2986 			spin_lock(&fs_info->balance_lock);
2987 			bctl->stat.expected++;
2988 			spin_unlock(&fs_info->balance_lock);
2989 			goto loop;
2990 		}
2991 
2992 		ret = btrfs_relocate_chunk(chunk_root,
2993 					   chunk_root->root_key.objectid,
2994 					   found_key.objectid,
2995 					   found_key.offset);
2996 		if (ret && ret != -ENOSPC)
2997 			goto error;
2998 		if (ret == -ENOSPC) {
2999 			enospc_errors++;
3000 		} else {
3001 			spin_lock(&fs_info->balance_lock);
3002 			bctl->stat.completed++;
3003 			spin_unlock(&fs_info->balance_lock);
3004 		}
3005 loop:
3006 		key.offset = found_key.offset - 1;
3007 	}
3008 
3009 	if (counting) {
3010 		btrfs_release_path(path);
3011 		counting = false;
3012 		goto again;
3013 	}
3014 error:
3015 	btrfs_free_path(path);
3016 	if (enospc_errors) {
3017 		printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3018 		       enospc_errors);
3019 		if (!ret)
3020 			ret = -ENOSPC;
3021 	}
3022 
3023 	return ret;
3024 }
3025 
3026 /**
3027  * alloc_profile_is_valid - see if a given profile is valid and reduced
3028  * @flags: profile to validate
3029  * @extended: if true @flags is treated as an extended profile
3030  */
3031 static int alloc_profile_is_valid(u64 flags, int extended)
3032 {
3033 	u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3034 			       BTRFS_BLOCK_GROUP_PROFILE_MASK);
3035 
3036 	flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3037 
3038 	/* 1) check that all other bits are zeroed */
3039 	if (flags & ~mask)
3040 		return 0;
3041 
3042 	/* 2) see if profile is reduced */
3043 	if (flags == 0)
3044 		return !extended; /* "0" is valid for usual profiles */
3045 
3046 	/* true if exactly one bit set */
3047 	return (flags & (flags - 1)) == 0;
3048 }
3049 
3050 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3051 {
3052 	/* cancel requested || normal exit path */
3053 	return atomic_read(&fs_info->balance_cancel_req) ||
3054 		(atomic_read(&fs_info->balance_pause_req) == 0 &&
3055 		 atomic_read(&fs_info->balance_cancel_req) == 0);
3056 }
3057 
3058 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3059 {
3060 	int ret;
3061 
3062 	unset_balance_control(fs_info);
3063 	ret = del_balance_item(fs_info->tree_root);
3064 	if (ret)
3065 		btrfs_std_error(fs_info, ret);
3066 
3067 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3068 }
3069 
3070 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3071 			       struct btrfs_ioctl_balance_args *bargs);
3072 
3073 /*
3074  * Should be called with both balance and volume mutexes held
3075  */
3076 int btrfs_balance(struct btrfs_balance_control *bctl,
3077 		  struct btrfs_ioctl_balance_args *bargs)
3078 {
3079 	struct btrfs_fs_info *fs_info = bctl->fs_info;
3080 	u64 allowed;
3081 	int mixed = 0;
3082 	int ret;
3083 	u64 num_devices;
3084 	unsigned seq;
3085 
3086 	if (btrfs_fs_closing(fs_info) ||
3087 	    atomic_read(&fs_info->balance_pause_req) ||
3088 	    atomic_read(&fs_info->balance_cancel_req)) {
3089 		ret = -EINVAL;
3090 		goto out;
3091 	}
3092 
3093 	allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3094 	if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3095 		mixed = 1;
3096 
3097 	/*
3098 	 * In case of mixed groups both data and meta should be picked,
3099 	 * and identical options should be given for both of them.
3100 	 */
3101 	allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3102 	if (mixed && (bctl->flags & allowed)) {
3103 		if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3104 		    !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3105 		    memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3106 			printk(KERN_ERR "btrfs: with mixed groups data and "
3107 			       "metadata balance options must be the same\n");
3108 			ret = -EINVAL;
3109 			goto out;
3110 		}
3111 	}
3112 
3113 	num_devices = fs_info->fs_devices->num_devices;
3114 	btrfs_dev_replace_lock(&fs_info->dev_replace);
3115 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3116 		BUG_ON(num_devices < 1);
3117 		num_devices--;
3118 	}
3119 	btrfs_dev_replace_unlock(&fs_info->dev_replace);
3120 	allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3121 	if (num_devices == 1)
3122 		allowed |= BTRFS_BLOCK_GROUP_DUP;
3123 	else if (num_devices > 1)
3124 		allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3125 	if (num_devices > 2)
3126 		allowed |= BTRFS_BLOCK_GROUP_RAID5;
3127 	if (num_devices > 3)
3128 		allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3129 			    BTRFS_BLOCK_GROUP_RAID6);
3130 	if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3131 	    (!alloc_profile_is_valid(bctl->data.target, 1) ||
3132 	     (bctl->data.target & ~allowed))) {
3133 		printk(KERN_ERR "btrfs: unable to start balance with target "
3134 		       "data profile %llu\n",
3135 		       (unsigned long long)bctl->data.target);
3136 		ret = -EINVAL;
3137 		goto out;
3138 	}
3139 	if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3140 	    (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3141 	     (bctl->meta.target & ~allowed))) {
3142 		printk(KERN_ERR "btrfs: unable to start balance with target "
3143 		       "metadata profile %llu\n",
3144 		       (unsigned long long)bctl->meta.target);
3145 		ret = -EINVAL;
3146 		goto out;
3147 	}
3148 	if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3149 	    (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3150 	     (bctl->sys.target & ~allowed))) {
3151 		printk(KERN_ERR "btrfs: unable to start balance with target "
3152 		       "system profile %llu\n",
3153 		       (unsigned long long)bctl->sys.target);
3154 		ret = -EINVAL;
3155 		goto out;
3156 	}
3157 
3158 	/* allow dup'ed data chunks only in mixed mode */
3159 	if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3160 	    (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3161 		printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3162 		ret = -EINVAL;
3163 		goto out;
3164 	}
3165 
3166 	/* allow to reduce meta or sys integrity only if force set */
3167 	allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3168 			BTRFS_BLOCK_GROUP_RAID10 |
3169 			BTRFS_BLOCK_GROUP_RAID5 |
3170 			BTRFS_BLOCK_GROUP_RAID6;
3171 	do {
3172 		seq = read_seqbegin(&fs_info->profiles_lock);
3173 
3174 		if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3175 		     (fs_info->avail_system_alloc_bits & allowed) &&
3176 		     !(bctl->sys.target & allowed)) ||
3177 		    ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3178 		     (fs_info->avail_metadata_alloc_bits & allowed) &&
3179 		     !(bctl->meta.target & allowed))) {
3180 			if (bctl->flags & BTRFS_BALANCE_FORCE) {
3181 				printk(KERN_INFO "btrfs: force reducing metadata "
3182 				       "integrity\n");
3183 			} else {
3184 				printk(KERN_ERR "btrfs: balance will reduce metadata "
3185 				       "integrity, use force if you want this\n");
3186 				ret = -EINVAL;
3187 				goto out;
3188 			}
3189 		}
3190 	} while (read_seqretry(&fs_info->profiles_lock, seq));
3191 
3192 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3193 		int num_tolerated_disk_barrier_failures;
3194 		u64 target = bctl->sys.target;
3195 
3196 		num_tolerated_disk_barrier_failures =
3197 			btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3198 		if (num_tolerated_disk_barrier_failures > 0 &&
3199 		    (target &
3200 		     (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3201 		      BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3202 			num_tolerated_disk_barrier_failures = 0;
3203 		else if (num_tolerated_disk_barrier_failures > 1 &&
3204 			 (target &
3205 			  (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3206 			num_tolerated_disk_barrier_failures = 1;
3207 
3208 		fs_info->num_tolerated_disk_barrier_failures =
3209 			num_tolerated_disk_barrier_failures;
3210 	}
3211 
3212 	ret = insert_balance_item(fs_info->tree_root, bctl);
3213 	if (ret && ret != -EEXIST)
3214 		goto out;
3215 
3216 	if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3217 		BUG_ON(ret == -EEXIST);
3218 		set_balance_control(bctl);
3219 	} else {
3220 		BUG_ON(ret != -EEXIST);
3221 		spin_lock(&fs_info->balance_lock);
3222 		update_balance_args(bctl);
3223 		spin_unlock(&fs_info->balance_lock);
3224 	}
3225 
3226 	atomic_inc(&fs_info->balance_running);
3227 	mutex_unlock(&fs_info->balance_mutex);
3228 
3229 	ret = __btrfs_balance(fs_info);
3230 
3231 	mutex_lock(&fs_info->balance_mutex);
3232 	atomic_dec(&fs_info->balance_running);
3233 
3234 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3235 		fs_info->num_tolerated_disk_barrier_failures =
3236 			btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3237 	}
3238 
3239 	if (bargs) {
3240 		memset(bargs, 0, sizeof(*bargs));
3241 		update_ioctl_balance_args(fs_info, 0, bargs);
3242 	}
3243 
3244 	if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3245 	    balance_need_close(fs_info)) {
3246 		__cancel_balance(fs_info);
3247 	}
3248 
3249 	wake_up(&fs_info->balance_wait_q);
3250 
3251 	return ret;
3252 out:
3253 	if (bctl->flags & BTRFS_BALANCE_RESUME)
3254 		__cancel_balance(fs_info);
3255 	else {
3256 		kfree(bctl);
3257 		atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3258 	}
3259 	return ret;
3260 }
3261 
3262 static int balance_kthread(void *data)
3263 {
3264 	struct btrfs_fs_info *fs_info = data;
3265 	int ret = 0;
3266 
3267 	mutex_lock(&fs_info->volume_mutex);
3268 	mutex_lock(&fs_info->balance_mutex);
3269 
3270 	if (fs_info->balance_ctl) {
3271 		printk(KERN_INFO "btrfs: continuing balance\n");
3272 		ret = btrfs_balance(fs_info->balance_ctl, NULL);
3273 	}
3274 
3275 	mutex_unlock(&fs_info->balance_mutex);
3276 	mutex_unlock(&fs_info->volume_mutex);
3277 
3278 	return ret;
3279 }
3280 
3281 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3282 {
3283 	struct task_struct *tsk;
3284 
3285 	spin_lock(&fs_info->balance_lock);
3286 	if (!fs_info->balance_ctl) {
3287 		spin_unlock(&fs_info->balance_lock);
3288 		return 0;
3289 	}
3290 	spin_unlock(&fs_info->balance_lock);
3291 
3292 	if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3293 		printk(KERN_INFO "btrfs: force skipping balance\n");
3294 		return 0;
3295 	}
3296 
3297 	tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3298 	if (IS_ERR(tsk))
3299 		return PTR_ERR(tsk);
3300 
3301 	return 0;
3302 }
3303 
3304 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3305 {
3306 	struct btrfs_balance_control *bctl;
3307 	struct btrfs_balance_item *item;
3308 	struct btrfs_disk_balance_args disk_bargs;
3309 	struct btrfs_path *path;
3310 	struct extent_buffer *leaf;
3311 	struct btrfs_key key;
3312 	int ret;
3313 
3314 	path = btrfs_alloc_path();
3315 	if (!path)
3316 		return -ENOMEM;
3317 
3318 	key.objectid = BTRFS_BALANCE_OBJECTID;
3319 	key.type = BTRFS_BALANCE_ITEM_KEY;
3320 	key.offset = 0;
3321 
3322 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3323 	if (ret < 0)
3324 		goto out;
3325 	if (ret > 0) { /* ret = -ENOENT; */
3326 		ret = 0;
3327 		goto out;
3328 	}
3329 
3330 	bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3331 	if (!bctl) {
3332 		ret = -ENOMEM;
3333 		goto out;
3334 	}
3335 
3336 	leaf = path->nodes[0];
3337 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3338 
3339 	bctl->fs_info = fs_info;
3340 	bctl->flags = btrfs_balance_flags(leaf, item);
3341 	bctl->flags |= BTRFS_BALANCE_RESUME;
3342 
3343 	btrfs_balance_data(leaf, item, &disk_bargs);
3344 	btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3345 	btrfs_balance_meta(leaf, item, &disk_bargs);
3346 	btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3347 	btrfs_balance_sys(leaf, item, &disk_bargs);
3348 	btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3349 
3350 	WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3351 
3352 	mutex_lock(&fs_info->volume_mutex);
3353 	mutex_lock(&fs_info->balance_mutex);
3354 
3355 	set_balance_control(bctl);
3356 
3357 	mutex_unlock(&fs_info->balance_mutex);
3358 	mutex_unlock(&fs_info->volume_mutex);
3359 out:
3360 	btrfs_free_path(path);
3361 	return ret;
3362 }
3363 
3364 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3365 {
3366 	int ret = 0;
3367 
3368 	mutex_lock(&fs_info->balance_mutex);
3369 	if (!fs_info->balance_ctl) {
3370 		mutex_unlock(&fs_info->balance_mutex);
3371 		return -ENOTCONN;
3372 	}
3373 
3374 	if (atomic_read(&fs_info->balance_running)) {
3375 		atomic_inc(&fs_info->balance_pause_req);
3376 		mutex_unlock(&fs_info->balance_mutex);
3377 
3378 		wait_event(fs_info->balance_wait_q,
3379 			   atomic_read(&fs_info->balance_running) == 0);
3380 
3381 		mutex_lock(&fs_info->balance_mutex);
3382 		/* we are good with balance_ctl ripped off from under us */
3383 		BUG_ON(atomic_read(&fs_info->balance_running));
3384 		atomic_dec(&fs_info->balance_pause_req);
3385 	} else {
3386 		ret = -ENOTCONN;
3387 	}
3388 
3389 	mutex_unlock(&fs_info->balance_mutex);
3390 	return ret;
3391 }
3392 
3393 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3394 {
3395 	mutex_lock(&fs_info->balance_mutex);
3396 	if (!fs_info->balance_ctl) {
3397 		mutex_unlock(&fs_info->balance_mutex);
3398 		return -ENOTCONN;
3399 	}
3400 
3401 	atomic_inc(&fs_info->balance_cancel_req);
3402 	/*
3403 	 * if we are running just wait and return, balance item is
3404 	 * deleted in btrfs_balance in this case
3405 	 */
3406 	if (atomic_read(&fs_info->balance_running)) {
3407 		mutex_unlock(&fs_info->balance_mutex);
3408 		wait_event(fs_info->balance_wait_q,
3409 			   atomic_read(&fs_info->balance_running) == 0);
3410 		mutex_lock(&fs_info->balance_mutex);
3411 	} else {
3412 		/* __cancel_balance needs volume_mutex */
3413 		mutex_unlock(&fs_info->balance_mutex);
3414 		mutex_lock(&fs_info->volume_mutex);
3415 		mutex_lock(&fs_info->balance_mutex);
3416 
3417 		if (fs_info->balance_ctl)
3418 			__cancel_balance(fs_info);
3419 
3420 		mutex_unlock(&fs_info->volume_mutex);
3421 	}
3422 
3423 	BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3424 	atomic_dec(&fs_info->balance_cancel_req);
3425 	mutex_unlock(&fs_info->balance_mutex);
3426 	return 0;
3427 }
3428 
3429 /*
3430  * shrinking a device means finding all of the device extents past
3431  * the new size, and then following the back refs to the chunks.
3432  * The chunk relocation code actually frees the device extent
3433  */
3434 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3435 {
3436 	struct btrfs_trans_handle *trans;
3437 	struct btrfs_root *root = device->dev_root;
3438 	struct btrfs_dev_extent *dev_extent = NULL;
3439 	struct btrfs_path *path;
3440 	u64 length;
3441 	u64 chunk_tree;
3442 	u64 chunk_objectid;
3443 	u64 chunk_offset;
3444 	int ret;
3445 	int slot;
3446 	int failed = 0;
3447 	bool retried = false;
3448 	struct extent_buffer *l;
3449 	struct btrfs_key key;
3450 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3451 	u64 old_total = btrfs_super_total_bytes(super_copy);
3452 	u64 old_size = device->total_bytes;
3453 	u64 diff = device->total_bytes - new_size;
3454 
3455 	if (device->is_tgtdev_for_dev_replace)
3456 		return -EINVAL;
3457 
3458 	path = btrfs_alloc_path();
3459 	if (!path)
3460 		return -ENOMEM;
3461 
3462 	path->reada = 2;
3463 
3464 	lock_chunks(root);
3465 
3466 	device->total_bytes = new_size;
3467 	if (device->writeable) {
3468 		device->fs_devices->total_rw_bytes -= diff;
3469 		spin_lock(&root->fs_info->free_chunk_lock);
3470 		root->fs_info->free_chunk_space -= diff;
3471 		spin_unlock(&root->fs_info->free_chunk_lock);
3472 	}
3473 	unlock_chunks(root);
3474 
3475 again:
3476 	key.objectid = device->devid;
3477 	key.offset = (u64)-1;
3478 	key.type = BTRFS_DEV_EXTENT_KEY;
3479 
3480 	do {
3481 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3482 		if (ret < 0)
3483 			goto done;
3484 
3485 		ret = btrfs_previous_item(root, path, 0, key.type);
3486 		if (ret < 0)
3487 			goto done;
3488 		if (ret) {
3489 			ret = 0;
3490 			btrfs_release_path(path);
3491 			break;
3492 		}
3493 
3494 		l = path->nodes[0];
3495 		slot = path->slots[0];
3496 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3497 
3498 		if (key.objectid != device->devid) {
3499 			btrfs_release_path(path);
3500 			break;
3501 		}
3502 
3503 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3504 		length = btrfs_dev_extent_length(l, dev_extent);
3505 
3506 		if (key.offset + length <= new_size) {
3507 			btrfs_release_path(path);
3508 			break;
3509 		}
3510 
3511 		chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3512 		chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3513 		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3514 		btrfs_release_path(path);
3515 
3516 		ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3517 					   chunk_offset);
3518 		if (ret && ret != -ENOSPC)
3519 			goto done;
3520 		if (ret == -ENOSPC)
3521 			failed++;
3522 	} while (key.offset-- > 0);
3523 
3524 	if (failed && !retried) {
3525 		failed = 0;
3526 		retried = true;
3527 		goto again;
3528 	} else if (failed && retried) {
3529 		ret = -ENOSPC;
3530 		lock_chunks(root);
3531 
3532 		device->total_bytes = old_size;
3533 		if (device->writeable)
3534 			device->fs_devices->total_rw_bytes += diff;
3535 		spin_lock(&root->fs_info->free_chunk_lock);
3536 		root->fs_info->free_chunk_space += diff;
3537 		spin_unlock(&root->fs_info->free_chunk_lock);
3538 		unlock_chunks(root);
3539 		goto done;
3540 	}
3541 
3542 	/* Shrinking succeeded, else we would be at "done". */
3543 	trans = btrfs_start_transaction(root, 0);
3544 	if (IS_ERR(trans)) {
3545 		ret = PTR_ERR(trans);
3546 		goto done;
3547 	}
3548 
3549 	lock_chunks(root);
3550 
3551 	device->disk_total_bytes = new_size;
3552 	/* Now btrfs_update_device() will change the on-disk size. */
3553 	ret = btrfs_update_device(trans, device);
3554 	if (ret) {
3555 		unlock_chunks(root);
3556 		btrfs_end_transaction(trans, root);
3557 		goto done;
3558 	}
3559 	WARN_ON(diff > old_total);
3560 	btrfs_set_super_total_bytes(super_copy, old_total - diff);
3561 	unlock_chunks(root);
3562 	btrfs_end_transaction(trans, root);
3563 done:
3564 	btrfs_free_path(path);
3565 	return ret;
3566 }
3567 
3568 static int btrfs_add_system_chunk(struct btrfs_root *root,
3569 			   struct btrfs_key *key,
3570 			   struct btrfs_chunk *chunk, int item_size)
3571 {
3572 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3573 	struct btrfs_disk_key disk_key;
3574 	u32 array_size;
3575 	u8 *ptr;
3576 
3577 	array_size = btrfs_super_sys_array_size(super_copy);
3578 	if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3579 		return -EFBIG;
3580 
3581 	ptr = super_copy->sys_chunk_array + array_size;
3582 	btrfs_cpu_key_to_disk(&disk_key, key);
3583 	memcpy(ptr, &disk_key, sizeof(disk_key));
3584 	ptr += sizeof(disk_key);
3585 	memcpy(ptr, chunk, item_size);
3586 	item_size += sizeof(disk_key);
3587 	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3588 	return 0;
3589 }
3590 
3591 /*
3592  * sort the devices in descending order by max_avail, total_avail
3593  */
3594 static int btrfs_cmp_device_info(const void *a, const void *b)
3595 {
3596 	const struct btrfs_device_info *di_a = a;
3597 	const struct btrfs_device_info *di_b = b;
3598 
3599 	if (di_a->max_avail > di_b->max_avail)
3600 		return -1;
3601 	if (di_a->max_avail < di_b->max_avail)
3602 		return 1;
3603 	if (di_a->total_avail > di_b->total_avail)
3604 		return -1;
3605 	if (di_a->total_avail < di_b->total_avail)
3606 		return 1;
3607 	return 0;
3608 }
3609 
3610 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3611 	[BTRFS_RAID_RAID10] = {
3612 		.sub_stripes	= 2,
3613 		.dev_stripes	= 1,
3614 		.devs_max	= 0,	/* 0 == as many as possible */
3615 		.devs_min	= 4,
3616 		.devs_increment	= 2,
3617 		.ncopies	= 2,
3618 	},
3619 	[BTRFS_RAID_RAID1] = {
3620 		.sub_stripes	= 1,
3621 		.dev_stripes	= 1,
3622 		.devs_max	= 2,
3623 		.devs_min	= 2,
3624 		.devs_increment	= 2,
3625 		.ncopies	= 2,
3626 	},
3627 	[BTRFS_RAID_DUP] = {
3628 		.sub_stripes	= 1,
3629 		.dev_stripes	= 2,
3630 		.devs_max	= 1,
3631 		.devs_min	= 1,
3632 		.devs_increment	= 1,
3633 		.ncopies	= 2,
3634 	},
3635 	[BTRFS_RAID_RAID0] = {
3636 		.sub_stripes	= 1,
3637 		.dev_stripes	= 1,
3638 		.devs_max	= 0,
3639 		.devs_min	= 2,
3640 		.devs_increment	= 1,
3641 		.ncopies	= 1,
3642 	},
3643 	[BTRFS_RAID_SINGLE] = {
3644 		.sub_stripes	= 1,
3645 		.dev_stripes	= 1,
3646 		.devs_max	= 1,
3647 		.devs_min	= 1,
3648 		.devs_increment	= 1,
3649 		.ncopies	= 1,
3650 	},
3651 	[BTRFS_RAID_RAID5] = {
3652 		.sub_stripes	= 1,
3653 		.dev_stripes	= 1,
3654 		.devs_max	= 0,
3655 		.devs_min	= 2,
3656 		.devs_increment	= 1,
3657 		.ncopies	= 2,
3658 	},
3659 	[BTRFS_RAID_RAID6] = {
3660 		.sub_stripes	= 1,
3661 		.dev_stripes	= 1,
3662 		.devs_max	= 0,
3663 		.devs_min	= 3,
3664 		.devs_increment	= 1,
3665 		.ncopies	= 3,
3666 	},
3667 };
3668 
3669 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3670 {
3671 	/* TODO allow them to set a preferred stripe size */
3672 	return 64 * 1024;
3673 }
3674 
3675 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3676 {
3677 	if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3678 		return;
3679 
3680 	btrfs_set_fs_incompat(info, RAID56);
3681 }
3682 
3683 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3684 			       struct btrfs_root *extent_root,
3685 			       struct map_lookup **map_ret,
3686 			       u64 *num_bytes_out, u64 *stripe_size_out,
3687 			       u64 start, u64 type)
3688 {
3689 	struct btrfs_fs_info *info = extent_root->fs_info;
3690 	struct btrfs_fs_devices *fs_devices = info->fs_devices;
3691 	struct list_head *cur;
3692 	struct map_lookup *map = NULL;
3693 	struct extent_map_tree *em_tree;
3694 	struct extent_map *em;
3695 	struct btrfs_device_info *devices_info = NULL;
3696 	u64 total_avail;
3697 	int num_stripes;	/* total number of stripes to allocate */
3698 	int data_stripes;	/* number of stripes that count for
3699 				   block group size */
3700 	int sub_stripes;	/* sub_stripes info for map */
3701 	int dev_stripes;	/* stripes per dev */
3702 	int devs_max;		/* max devs to use */
3703 	int devs_min;		/* min devs needed */
3704 	int devs_increment;	/* ndevs has to be a multiple of this */
3705 	int ncopies;		/* how many copies to data has */
3706 	int ret;
3707 	u64 max_stripe_size;
3708 	u64 max_chunk_size;
3709 	u64 stripe_size;
3710 	u64 num_bytes;
3711 	u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3712 	int ndevs;
3713 	int i;
3714 	int j;
3715 	int index;
3716 
3717 	BUG_ON(!alloc_profile_is_valid(type, 0));
3718 
3719 	if (list_empty(&fs_devices->alloc_list))
3720 		return -ENOSPC;
3721 
3722 	index = __get_raid_index(type);
3723 
3724 	sub_stripes = btrfs_raid_array[index].sub_stripes;
3725 	dev_stripes = btrfs_raid_array[index].dev_stripes;
3726 	devs_max = btrfs_raid_array[index].devs_max;
3727 	devs_min = btrfs_raid_array[index].devs_min;
3728 	devs_increment = btrfs_raid_array[index].devs_increment;
3729 	ncopies = btrfs_raid_array[index].ncopies;
3730 
3731 	if (type & BTRFS_BLOCK_GROUP_DATA) {
3732 		max_stripe_size = 1024 * 1024 * 1024;
3733 		max_chunk_size = 10 * max_stripe_size;
3734 	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3735 		/* for larger filesystems, use larger metadata chunks */
3736 		if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3737 			max_stripe_size = 1024 * 1024 * 1024;
3738 		else
3739 			max_stripe_size = 256 * 1024 * 1024;
3740 		max_chunk_size = max_stripe_size;
3741 	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3742 		max_stripe_size = 32 * 1024 * 1024;
3743 		max_chunk_size = 2 * max_stripe_size;
3744 	} else {
3745 		printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3746 		       type);
3747 		BUG_ON(1);
3748 	}
3749 
3750 	/* we don't want a chunk larger than 10% of writeable space */
3751 	max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3752 			     max_chunk_size);
3753 
3754 	devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3755 			       GFP_NOFS);
3756 	if (!devices_info)
3757 		return -ENOMEM;
3758 
3759 	cur = fs_devices->alloc_list.next;
3760 
3761 	/*
3762 	 * in the first pass through the devices list, we gather information
3763 	 * about the available holes on each device.
3764 	 */
3765 	ndevs = 0;
3766 	while (cur != &fs_devices->alloc_list) {
3767 		struct btrfs_device *device;
3768 		u64 max_avail;
3769 		u64 dev_offset;
3770 
3771 		device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3772 
3773 		cur = cur->next;
3774 
3775 		if (!device->writeable) {
3776 			WARN(1, KERN_ERR
3777 			       "btrfs: read-only device in alloc_list\n");
3778 			continue;
3779 		}
3780 
3781 		if (!device->in_fs_metadata ||
3782 		    device->is_tgtdev_for_dev_replace)
3783 			continue;
3784 
3785 		if (device->total_bytes > device->bytes_used)
3786 			total_avail = device->total_bytes - device->bytes_used;
3787 		else
3788 			total_avail = 0;
3789 
3790 		/* If there is no space on this device, skip it. */
3791 		if (total_avail == 0)
3792 			continue;
3793 
3794 		ret = find_free_dev_extent(device,
3795 					   max_stripe_size * dev_stripes,
3796 					   &dev_offset, &max_avail);
3797 		if (ret && ret != -ENOSPC)
3798 			goto error;
3799 
3800 		if (ret == 0)
3801 			max_avail = max_stripe_size * dev_stripes;
3802 
3803 		if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3804 			continue;
3805 
3806 		if (ndevs == fs_devices->rw_devices) {
3807 			WARN(1, "%s: found more than %llu devices\n",
3808 			     __func__, fs_devices->rw_devices);
3809 			break;
3810 		}
3811 		devices_info[ndevs].dev_offset = dev_offset;
3812 		devices_info[ndevs].max_avail = max_avail;
3813 		devices_info[ndevs].total_avail = total_avail;
3814 		devices_info[ndevs].dev = device;
3815 		++ndevs;
3816 	}
3817 
3818 	/*
3819 	 * now sort the devices by hole size / available space
3820 	 */
3821 	sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3822 	     btrfs_cmp_device_info, NULL);
3823 
3824 	/* round down to number of usable stripes */
3825 	ndevs -= ndevs % devs_increment;
3826 
3827 	if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3828 		ret = -ENOSPC;
3829 		goto error;
3830 	}
3831 
3832 	if (devs_max && ndevs > devs_max)
3833 		ndevs = devs_max;
3834 	/*
3835 	 * the primary goal is to maximize the number of stripes, so use as many
3836 	 * devices as possible, even if the stripes are not maximum sized.
3837 	 */
3838 	stripe_size = devices_info[ndevs-1].max_avail;
3839 	num_stripes = ndevs * dev_stripes;
3840 
3841 	/*
3842 	 * this will have to be fixed for RAID1 and RAID10 over
3843 	 * more drives
3844 	 */
3845 	data_stripes = num_stripes / ncopies;
3846 
3847 	if (type & BTRFS_BLOCK_GROUP_RAID5) {
3848 		raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
3849 				 btrfs_super_stripesize(info->super_copy));
3850 		data_stripes = num_stripes - 1;
3851 	}
3852 	if (type & BTRFS_BLOCK_GROUP_RAID6) {
3853 		raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
3854 				 btrfs_super_stripesize(info->super_copy));
3855 		data_stripes = num_stripes - 2;
3856 	}
3857 
3858 	/*
3859 	 * Use the number of data stripes to figure out how big this chunk
3860 	 * is really going to be in terms of logical address space,
3861 	 * and compare that answer with the max chunk size
3862 	 */
3863 	if (stripe_size * data_stripes > max_chunk_size) {
3864 		u64 mask = (1ULL << 24) - 1;
3865 		stripe_size = max_chunk_size;
3866 		do_div(stripe_size, data_stripes);
3867 
3868 		/* bump the answer up to a 16MB boundary */
3869 		stripe_size = (stripe_size + mask) & ~mask;
3870 
3871 		/* but don't go higher than the limits we found
3872 		 * while searching for free extents
3873 		 */
3874 		if (stripe_size > devices_info[ndevs-1].max_avail)
3875 			stripe_size = devices_info[ndevs-1].max_avail;
3876 	}
3877 
3878 	do_div(stripe_size, dev_stripes);
3879 
3880 	/* align to BTRFS_STRIPE_LEN */
3881 	do_div(stripe_size, raid_stripe_len);
3882 	stripe_size *= raid_stripe_len;
3883 
3884 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3885 	if (!map) {
3886 		ret = -ENOMEM;
3887 		goto error;
3888 	}
3889 	map->num_stripes = num_stripes;
3890 
3891 	for (i = 0; i < ndevs; ++i) {
3892 		for (j = 0; j < dev_stripes; ++j) {
3893 			int s = i * dev_stripes + j;
3894 			map->stripes[s].dev = devices_info[i].dev;
3895 			map->stripes[s].physical = devices_info[i].dev_offset +
3896 						   j * stripe_size;
3897 		}
3898 	}
3899 	map->sector_size = extent_root->sectorsize;
3900 	map->stripe_len = raid_stripe_len;
3901 	map->io_align = raid_stripe_len;
3902 	map->io_width = raid_stripe_len;
3903 	map->type = type;
3904 	map->sub_stripes = sub_stripes;
3905 
3906 	*map_ret = map;
3907 	num_bytes = stripe_size * data_stripes;
3908 
3909 	*stripe_size_out = stripe_size;
3910 	*num_bytes_out = num_bytes;
3911 
3912 	trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3913 
3914 	em = alloc_extent_map();
3915 	if (!em) {
3916 		ret = -ENOMEM;
3917 		goto error;
3918 	}
3919 	em->bdev = (struct block_device *)map;
3920 	em->start = start;
3921 	em->len = num_bytes;
3922 	em->block_start = 0;
3923 	em->block_len = em->len;
3924 
3925 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3926 	write_lock(&em_tree->lock);
3927 	ret = add_extent_mapping(em_tree, em, 0);
3928 	write_unlock(&em_tree->lock);
3929 	if (ret) {
3930 		free_extent_map(em);
3931 		goto error;
3932 	}
3933 
3934 	for (i = 0; i < map->num_stripes; ++i) {
3935 		struct btrfs_device *device;
3936 		u64 dev_offset;
3937 
3938 		device = map->stripes[i].dev;
3939 		dev_offset = map->stripes[i].physical;
3940 
3941 		ret = btrfs_alloc_dev_extent(trans, device,
3942 				info->chunk_root->root_key.objectid,
3943 				BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3944 				start, dev_offset, stripe_size);
3945 		if (ret)
3946 			goto error_dev_extent;
3947 	}
3948 
3949 	ret = btrfs_make_block_group(trans, extent_root, 0, type,
3950 				     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3951 				     start, num_bytes);
3952 	if (ret) {
3953 		i = map->num_stripes - 1;
3954 		goto error_dev_extent;
3955 	}
3956 
3957 	free_extent_map(em);
3958 	check_raid56_incompat_flag(extent_root->fs_info, type);
3959 
3960 	kfree(devices_info);
3961 	return 0;
3962 
3963 error_dev_extent:
3964 	for (; i >= 0; i--) {
3965 		struct btrfs_device *device;
3966 		int err;
3967 
3968 		device = map->stripes[i].dev;
3969 		err = btrfs_free_dev_extent(trans, device, start);
3970 		if (err) {
3971 			btrfs_abort_transaction(trans, extent_root, err);
3972 			break;
3973 		}
3974 	}
3975 	write_lock(&em_tree->lock);
3976 	remove_extent_mapping(em_tree, em);
3977 	write_unlock(&em_tree->lock);
3978 
3979 	/* One for our allocation */
3980 	free_extent_map(em);
3981 	/* One for the tree reference */
3982 	free_extent_map(em);
3983 error:
3984 	kfree(map);
3985 	kfree(devices_info);
3986 	return ret;
3987 }
3988 
3989 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3990 				struct btrfs_root *extent_root,
3991 				struct map_lookup *map, u64 chunk_offset,
3992 				u64 chunk_size, u64 stripe_size)
3993 {
3994 	u64 dev_offset;
3995 	struct btrfs_key key;
3996 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3997 	struct btrfs_device *device;
3998 	struct btrfs_chunk *chunk;
3999 	struct btrfs_stripe *stripe;
4000 	size_t item_size = btrfs_chunk_item_size(map->num_stripes);
4001 	int index = 0;
4002 	int ret;
4003 
4004 	chunk = kzalloc(item_size, GFP_NOFS);
4005 	if (!chunk)
4006 		return -ENOMEM;
4007 
4008 	index = 0;
4009 	while (index < map->num_stripes) {
4010 		device = map->stripes[index].dev;
4011 		device->bytes_used += stripe_size;
4012 		ret = btrfs_update_device(trans, device);
4013 		if (ret)
4014 			goto out_free;
4015 		index++;
4016 	}
4017 
4018 	spin_lock(&extent_root->fs_info->free_chunk_lock);
4019 	extent_root->fs_info->free_chunk_space -= (stripe_size *
4020 						   map->num_stripes);
4021 	spin_unlock(&extent_root->fs_info->free_chunk_lock);
4022 
4023 	index = 0;
4024 	stripe = &chunk->stripe;
4025 	while (index < map->num_stripes) {
4026 		device = map->stripes[index].dev;
4027 		dev_offset = map->stripes[index].physical;
4028 
4029 		btrfs_set_stack_stripe_devid(stripe, device->devid);
4030 		btrfs_set_stack_stripe_offset(stripe, dev_offset);
4031 		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4032 		stripe++;
4033 		index++;
4034 	}
4035 
4036 	btrfs_set_stack_chunk_length(chunk, chunk_size);
4037 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4038 	btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4039 	btrfs_set_stack_chunk_type(chunk, map->type);
4040 	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4041 	btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4042 	btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4043 	btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4044 	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4045 
4046 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4047 	key.type = BTRFS_CHUNK_ITEM_KEY;
4048 	key.offset = chunk_offset;
4049 
4050 	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4051 
4052 	if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4053 		/*
4054 		 * TODO: Cleanup of inserted chunk root in case of
4055 		 * failure.
4056 		 */
4057 		ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4058 					     item_size);
4059 	}
4060 
4061 out_free:
4062 	kfree(chunk);
4063 	return ret;
4064 }
4065 
4066 /*
4067  * Chunk allocation falls into two parts. The first part does works
4068  * that make the new allocated chunk useable, but not do any operation
4069  * that modifies the chunk tree. The second part does the works that
4070  * require modifying the chunk tree. This division is important for the
4071  * bootstrap process of adding storage to a seed btrfs.
4072  */
4073 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4074 		      struct btrfs_root *extent_root, u64 type)
4075 {
4076 	u64 chunk_offset;
4077 	u64 chunk_size;
4078 	u64 stripe_size;
4079 	struct map_lookup *map;
4080 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4081 	int ret;
4082 
4083 	ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4084 			      &chunk_offset);
4085 	if (ret)
4086 		return ret;
4087 
4088 	ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4089 				  &stripe_size, chunk_offset, type);
4090 	if (ret)
4091 		return ret;
4092 
4093 	ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4094 				   chunk_size, stripe_size);
4095 	if (ret)
4096 		return ret;
4097 	return 0;
4098 }
4099 
4100 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4101 					 struct btrfs_root *root,
4102 					 struct btrfs_device *device)
4103 {
4104 	u64 chunk_offset;
4105 	u64 sys_chunk_offset;
4106 	u64 chunk_size;
4107 	u64 sys_chunk_size;
4108 	u64 stripe_size;
4109 	u64 sys_stripe_size;
4110 	u64 alloc_profile;
4111 	struct map_lookup *map;
4112 	struct map_lookup *sys_map;
4113 	struct btrfs_fs_info *fs_info = root->fs_info;
4114 	struct btrfs_root *extent_root = fs_info->extent_root;
4115 	int ret;
4116 
4117 	ret = find_next_chunk(fs_info->chunk_root,
4118 			      BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
4119 	if (ret)
4120 		return ret;
4121 
4122 	alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4123 	ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4124 				  &stripe_size, chunk_offset, alloc_profile);
4125 	if (ret)
4126 		return ret;
4127 
4128 	sys_chunk_offset = chunk_offset + chunk_size;
4129 
4130 	alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4131 	ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
4132 				  &sys_chunk_size, &sys_stripe_size,
4133 				  sys_chunk_offset, alloc_profile);
4134 	if (ret) {
4135 		btrfs_abort_transaction(trans, root, ret);
4136 		goto out;
4137 	}
4138 
4139 	ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4140 	if (ret) {
4141 		btrfs_abort_transaction(trans, root, ret);
4142 		goto out;
4143 	}
4144 
4145 	/*
4146 	 * Modifying chunk tree needs allocating new blocks from both
4147 	 * system block group and metadata block group. So we only can
4148 	 * do operations require modifying the chunk tree after both
4149 	 * block groups were created.
4150 	 */
4151 	ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4152 				   chunk_size, stripe_size);
4153 	if (ret) {
4154 		btrfs_abort_transaction(trans, root, ret);
4155 		goto out;
4156 	}
4157 
4158 	ret = __finish_chunk_alloc(trans, extent_root, sys_map,
4159 				   sys_chunk_offset, sys_chunk_size,
4160 				   sys_stripe_size);
4161 	if (ret)
4162 		btrfs_abort_transaction(trans, root, ret);
4163 
4164 out:
4165 
4166 	return ret;
4167 }
4168 
4169 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4170 {
4171 	struct extent_map *em;
4172 	struct map_lookup *map;
4173 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4174 	int readonly = 0;
4175 	int i;
4176 
4177 	read_lock(&map_tree->map_tree.lock);
4178 	em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4179 	read_unlock(&map_tree->map_tree.lock);
4180 	if (!em)
4181 		return 1;
4182 
4183 	if (btrfs_test_opt(root, DEGRADED)) {
4184 		free_extent_map(em);
4185 		return 0;
4186 	}
4187 
4188 	map = (struct map_lookup *)em->bdev;
4189 	for (i = 0; i < map->num_stripes; i++) {
4190 		if (!map->stripes[i].dev->writeable) {
4191 			readonly = 1;
4192 			break;
4193 		}
4194 	}
4195 	free_extent_map(em);
4196 	return readonly;
4197 }
4198 
4199 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4200 {
4201 	extent_map_tree_init(&tree->map_tree);
4202 }
4203 
4204 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4205 {
4206 	struct extent_map *em;
4207 
4208 	while (1) {
4209 		write_lock(&tree->map_tree.lock);
4210 		em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4211 		if (em)
4212 			remove_extent_mapping(&tree->map_tree, em);
4213 		write_unlock(&tree->map_tree.lock);
4214 		if (!em)
4215 			break;
4216 		kfree(em->bdev);
4217 		/* once for us */
4218 		free_extent_map(em);
4219 		/* once for the tree */
4220 		free_extent_map(em);
4221 	}
4222 }
4223 
4224 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4225 {
4226 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4227 	struct extent_map *em;
4228 	struct map_lookup *map;
4229 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4230 	int ret;
4231 
4232 	read_lock(&em_tree->lock);
4233 	em = lookup_extent_mapping(em_tree, logical, len);
4234 	read_unlock(&em_tree->lock);
4235 
4236 	/*
4237 	 * We could return errors for these cases, but that could get ugly and
4238 	 * we'd probably do the same thing which is just not do anything else
4239 	 * and exit, so return 1 so the callers don't try to use other copies.
4240 	 */
4241 	if (!em) {
4242 		btrfs_emerg(fs_info, "No mapping for %Lu-%Lu\n", logical,
4243 			    logical+len);
4244 		return 1;
4245 	}
4246 
4247 	if (em->start > logical || em->start + em->len < logical) {
4248 		btrfs_emerg(fs_info, "Invalid mapping for %Lu-%Lu, got "
4249 			    "%Lu-%Lu\n", logical, logical+len, em->start,
4250 			    em->start + em->len);
4251 		return 1;
4252 	}
4253 
4254 	map = (struct map_lookup *)em->bdev;
4255 	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4256 		ret = map->num_stripes;
4257 	else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4258 		ret = map->sub_stripes;
4259 	else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4260 		ret = 2;
4261 	else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4262 		ret = 3;
4263 	else
4264 		ret = 1;
4265 	free_extent_map(em);
4266 
4267 	btrfs_dev_replace_lock(&fs_info->dev_replace);
4268 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4269 		ret++;
4270 	btrfs_dev_replace_unlock(&fs_info->dev_replace);
4271 
4272 	return ret;
4273 }
4274 
4275 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4276 				    struct btrfs_mapping_tree *map_tree,
4277 				    u64 logical)
4278 {
4279 	struct extent_map *em;
4280 	struct map_lookup *map;
4281 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4282 	unsigned long len = root->sectorsize;
4283 
4284 	read_lock(&em_tree->lock);
4285 	em = lookup_extent_mapping(em_tree, logical, len);
4286 	read_unlock(&em_tree->lock);
4287 	BUG_ON(!em);
4288 
4289 	BUG_ON(em->start > logical || em->start + em->len < logical);
4290 	map = (struct map_lookup *)em->bdev;
4291 	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4292 			 BTRFS_BLOCK_GROUP_RAID6)) {
4293 		len = map->stripe_len * nr_data_stripes(map);
4294 	}
4295 	free_extent_map(em);
4296 	return len;
4297 }
4298 
4299 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4300 			   u64 logical, u64 len, int mirror_num)
4301 {
4302 	struct extent_map *em;
4303 	struct map_lookup *map;
4304 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4305 	int ret = 0;
4306 
4307 	read_lock(&em_tree->lock);
4308 	em = lookup_extent_mapping(em_tree, logical, len);
4309 	read_unlock(&em_tree->lock);
4310 	BUG_ON(!em);
4311 
4312 	BUG_ON(em->start > logical || em->start + em->len < logical);
4313 	map = (struct map_lookup *)em->bdev;
4314 	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4315 			 BTRFS_BLOCK_GROUP_RAID6))
4316 		ret = 1;
4317 	free_extent_map(em);
4318 	return ret;
4319 }
4320 
4321 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4322 			    struct map_lookup *map, int first, int num,
4323 			    int optimal, int dev_replace_is_ongoing)
4324 {
4325 	int i;
4326 	int tolerance;
4327 	struct btrfs_device *srcdev;
4328 
4329 	if (dev_replace_is_ongoing &&
4330 	    fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4331 	     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4332 		srcdev = fs_info->dev_replace.srcdev;
4333 	else
4334 		srcdev = NULL;
4335 
4336 	/*
4337 	 * try to avoid the drive that is the source drive for a
4338 	 * dev-replace procedure, only choose it if no other non-missing
4339 	 * mirror is available
4340 	 */
4341 	for (tolerance = 0; tolerance < 2; tolerance++) {
4342 		if (map->stripes[optimal].dev->bdev &&
4343 		    (tolerance || map->stripes[optimal].dev != srcdev))
4344 			return optimal;
4345 		for (i = first; i < first + num; i++) {
4346 			if (map->stripes[i].dev->bdev &&
4347 			    (tolerance || map->stripes[i].dev != srcdev))
4348 				return i;
4349 		}
4350 	}
4351 
4352 	/* we couldn't find one that doesn't fail.  Just return something
4353 	 * and the io error handling code will clean up eventually
4354 	 */
4355 	return optimal;
4356 }
4357 
4358 static inline int parity_smaller(u64 a, u64 b)
4359 {
4360 	return a > b;
4361 }
4362 
4363 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4364 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4365 {
4366 	struct btrfs_bio_stripe s;
4367 	int i;
4368 	u64 l;
4369 	int again = 1;
4370 
4371 	while (again) {
4372 		again = 0;
4373 		for (i = 0; i < bbio->num_stripes - 1; i++) {
4374 			if (parity_smaller(raid_map[i], raid_map[i+1])) {
4375 				s = bbio->stripes[i];
4376 				l = raid_map[i];
4377 				bbio->stripes[i] = bbio->stripes[i+1];
4378 				raid_map[i] = raid_map[i+1];
4379 				bbio->stripes[i+1] = s;
4380 				raid_map[i+1] = l;
4381 				again = 1;
4382 			}
4383 		}
4384 	}
4385 }
4386 
4387 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4388 			     u64 logical, u64 *length,
4389 			     struct btrfs_bio **bbio_ret,
4390 			     int mirror_num, u64 **raid_map_ret)
4391 {
4392 	struct extent_map *em;
4393 	struct map_lookup *map;
4394 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4395 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4396 	u64 offset;
4397 	u64 stripe_offset;
4398 	u64 stripe_end_offset;
4399 	u64 stripe_nr;
4400 	u64 stripe_nr_orig;
4401 	u64 stripe_nr_end;
4402 	u64 stripe_len;
4403 	u64 *raid_map = NULL;
4404 	int stripe_index;
4405 	int i;
4406 	int ret = 0;
4407 	int num_stripes;
4408 	int max_errors = 0;
4409 	struct btrfs_bio *bbio = NULL;
4410 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4411 	int dev_replace_is_ongoing = 0;
4412 	int num_alloc_stripes;
4413 	int patch_the_first_stripe_for_dev_replace = 0;
4414 	u64 physical_to_patch_in_first_stripe = 0;
4415 	u64 raid56_full_stripe_start = (u64)-1;
4416 
4417 	read_lock(&em_tree->lock);
4418 	em = lookup_extent_mapping(em_tree, logical, *length);
4419 	read_unlock(&em_tree->lock);
4420 
4421 	if (!em) {
4422 		btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4423 			(unsigned long long)logical,
4424 			(unsigned long long)*length);
4425 		return -EINVAL;
4426 	}
4427 
4428 	if (em->start > logical || em->start + em->len < logical) {
4429 		btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4430 			   "found %Lu-%Lu\n", logical, em->start,
4431 			   em->start + em->len);
4432 		return -EINVAL;
4433 	}
4434 
4435 	map = (struct map_lookup *)em->bdev;
4436 	offset = logical - em->start;
4437 
4438 	if (mirror_num > map->num_stripes)
4439 		mirror_num = 0;
4440 
4441 	stripe_len = map->stripe_len;
4442 	stripe_nr = offset;
4443 	/*
4444 	 * stripe_nr counts the total number of stripes we have to stride
4445 	 * to get to this block
4446 	 */
4447 	do_div(stripe_nr, stripe_len);
4448 
4449 	stripe_offset = stripe_nr * stripe_len;
4450 	BUG_ON(offset < stripe_offset);
4451 
4452 	/* stripe_offset is the offset of this block in its stripe*/
4453 	stripe_offset = offset - stripe_offset;
4454 
4455 	/* if we're here for raid56, we need to know the stripe aligned start */
4456 	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4457 		unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4458 		raid56_full_stripe_start = offset;
4459 
4460 		/* allow a write of a full stripe, but make sure we don't
4461 		 * allow straddling of stripes
4462 		 */
4463 		do_div(raid56_full_stripe_start, full_stripe_len);
4464 		raid56_full_stripe_start *= full_stripe_len;
4465 	}
4466 
4467 	if (rw & REQ_DISCARD) {
4468 		/* we don't discard raid56 yet */
4469 		if (map->type &
4470 		    (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4471 			ret = -EOPNOTSUPP;
4472 			goto out;
4473 		}
4474 		*length = min_t(u64, em->len - offset, *length);
4475 	} else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4476 		u64 max_len;
4477 		/* For writes to RAID[56], allow a full stripeset across all disks.
4478 		   For other RAID types and for RAID[56] reads, just allow a single
4479 		   stripe (on a single disk). */
4480 		if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4481 		    (rw & REQ_WRITE)) {
4482 			max_len = stripe_len * nr_data_stripes(map) -
4483 				(offset - raid56_full_stripe_start);
4484 		} else {
4485 			/* we limit the length of each bio to what fits in a stripe */
4486 			max_len = stripe_len - stripe_offset;
4487 		}
4488 		*length = min_t(u64, em->len - offset, max_len);
4489 	} else {
4490 		*length = em->len - offset;
4491 	}
4492 
4493 	/* This is for when we're called from btrfs_merge_bio_hook() and all
4494 	   it cares about is the length */
4495 	if (!bbio_ret)
4496 		goto out;
4497 
4498 	btrfs_dev_replace_lock(dev_replace);
4499 	dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4500 	if (!dev_replace_is_ongoing)
4501 		btrfs_dev_replace_unlock(dev_replace);
4502 
4503 	if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4504 	    !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4505 	    dev_replace->tgtdev != NULL) {
4506 		/*
4507 		 * in dev-replace case, for repair case (that's the only
4508 		 * case where the mirror is selected explicitly when
4509 		 * calling btrfs_map_block), blocks left of the left cursor
4510 		 * can also be read from the target drive.
4511 		 * For REQ_GET_READ_MIRRORS, the target drive is added as
4512 		 * the last one to the array of stripes. For READ, it also
4513 		 * needs to be supported using the same mirror number.
4514 		 * If the requested block is not left of the left cursor,
4515 		 * EIO is returned. This can happen because btrfs_num_copies()
4516 		 * returns one more in the dev-replace case.
4517 		 */
4518 		u64 tmp_length = *length;
4519 		struct btrfs_bio *tmp_bbio = NULL;
4520 		int tmp_num_stripes;
4521 		u64 srcdev_devid = dev_replace->srcdev->devid;
4522 		int index_srcdev = 0;
4523 		int found = 0;
4524 		u64 physical_of_found = 0;
4525 
4526 		ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4527 			     logical, &tmp_length, &tmp_bbio, 0, NULL);
4528 		if (ret) {
4529 			WARN_ON(tmp_bbio != NULL);
4530 			goto out;
4531 		}
4532 
4533 		tmp_num_stripes = tmp_bbio->num_stripes;
4534 		if (mirror_num > tmp_num_stripes) {
4535 			/*
4536 			 * REQ_GET_READ_MIRRORS does not contain this
4537 			 * mirror, that means that the requested area
4538 			 * is not left of the left cursor
4539 			 */
4540 			ret = -EIO;
4541 			kfree(tmp_bbio);
4542 			goto out;
4543 		}
4544 
4545 		/*
4546 		 * process the rest of the function using the mirror_num
4547 		 * of the source drive. Therefore look it up first.
4548 		 * At the end, patch the device pointer to the one of the
4549 		 * target drive.
4550 		 */
4551 		for (i = 0; i < tmp_num_stripes; i++) {
4552 			if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4553 				/*
4554 				 * In case of DUP, in order to keep it
4555 				 * simple, only add the mirror with the
4556 				 * lowest physical address
4557 				 */
4558 				if (found &&
4559 				    physical_of_found <=
4560 				     tmp_bbio->stripes[i].physical)
4561 					continue;
4562 				index_srcdev = i;
4563 				found = 1;
4564 				physical_of_found =
4565 					tmp_bbio->stripes[i].physical;
4566 			}
4567 		}
4568 
4569 		if (found) {
4570 			mirror_num = index_srcdev + 1;
4571 			patch_the_first_stripe_for_dev_replace = 1;
4572 			physical_to_patch_in_first_stripe = physical_of_found;
4573 		} else {
4574 			WARN_ON(1);
4575 			ret = -EIO;
4576 			kfree(tmp_bbio);
4577 			goto out;
4578 		}
4579 
4580 		kfree(tmp_bbio);
4581 	} else if (mirror_num > map->num_stripes) {
4582 		mirror_num = 0;
4583 	}
4584 
4585 	num_stripes = 1;
4586 	stripe_index = 0;
4587 	stripe_nr_orig = stripe_nr;
4588 	stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4589 	do_div(stripe_nr_end, map->stripe_len);
4590 	stripe_end_offset = stripe_nr_end * map->stripe_len -
4591 			    (offset + *length);
4592 
4593 	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4594 		if (rw & REQ_DISCARD)
4595 			num_stripes = min_t(u64, map->num_stripes,
4596 					    stripe_nr_end - stripe_nr_orig);
4597 		stripe_index = do_div(stripe_nr, map->num_stripes);
4598 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4599 		if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4600 			num_stripes = map->num_stripes;
4601 		else if (mirror_num)
4602 			stripe_index = mirror_num - 1;
4603 		else {
4604 			stripe_index = find_live_mirror(fs_info, map, 0,
4605 					    map->num_stripes,
4606 					    current->pid % map->num_stripes,
4607 					    dev_replace_is_ongoing);
4608 			mirror_num = stripe_index + 1;
4609 		}
4610 
4611 	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4612 		if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4613 			num_stripes = map->num_stripes;
4614 		} else if (mirror_num) {
4615 			stripe_index = mirror_num - 1;
4616 		} else {
4617 			mirror_num = 1;
4618 		}
4619 
4620 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4621 		int factor = map->num_stripes / map->sub_stripes;
4622 
4623 		stripe_index = do_div(stripe_nr, factor);
4624 		stripe_index *= map->sub_stripes;
4625 
4626 		if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4627 			num_stripes = map->sub_stripes;
4628 		else if (rw & REQ_DISCARD)
4629 			num_stripes = min_t(u64, map->sub_stripes *
4630 					    (stripe_nr_end - stripe_nr_orig),
4631 					    map->num_stripes);
4632 		else if (mirror_num)
4633 			stripe_index += mirror_num - 1;
4634 		else {
4635 			int old_stripe_index = stripe_index;
4636 			stripe_index = find_live_mirror(fs_info, map,
4637 					      stripe_index,
4638 					      map->sub_stripes, stripe_index +
4639 					      current->pid % map->sub_stripes,
4640 					      dev_replace_is_ongoing);
4641 			mirror_num = stripe_index - old_stripe_index + 1;
4642 		}
4643 
4644 	} else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4645 				BTRFS_BLOCK_GROUP_RAID6)) {
4646 		u64 tmp;
4647 
4648 		if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4649 		    && raid_map_ret) {
4650 			int i, rot;
4651 
4652 			/* push stripe_nr back to the start of the full stripe */
4653 			stripe_nr = raid56_full_stripe_start;
4654 			do_div(stripe_nr, stripe_len);
4655 
4656 			stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4657 
4658 			/* RAID[56] write or recovery. Return all stripes */
4659 			num_stripes = map->num_stripes;
4660 			max_errors = nr_parity_stripes(map);
4661 
4662 			raid_map = kmalloc(sizeof(u64) * num_stripes,
4663 					   GFP_NOFS);
4664 			if (!raid_map) {
4665 				ret = -ENOMEM;
4666 				goto out;
4667 			}
4668 
4669 			/* Work out the disk rotation on this stripe-set */
4670 			tmp = stripe_nr;
4671 			rot = do_div(tmp, num_stripes);
4672 
4673 			/* Fill in the logical address of each stripe */
4674 			tmp = stripe_nr * nr_data_stripes(map);
4675 			for (i = 0; i < nr_data_stripes(map); i++)
4676 				raid_map[(i+rot) % num_stripes] =
4677 					em->start + (tmp + i) * map->stripe_len;
4678 
4679 			raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4680 			if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4681 				raid_map[(i+rot+1) % num_stripes] =
4682 					RAID6_Q_STRIPE;
4683 
4684 			*length = map->stripe_len;
4685 			stripe_index = 0;
4686 			stripe_offset = 0;
4687 		} else {
4688 			/*
4689 			 * Mirror #0 or #1 means the original data block.
4690 			 * Mirror #2 is RAID5 parity block.
4691 			 * Mirror #3 is RAID6 Q block.
4692 			 */
4693 			stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4694 			if (mirror_num > 1)
4695 				stripe_index = nr_data_stripes(map) +
4696 						mirror_num - 2;
4697 
4698 			/* We distribute the parity blocks across stripes */
4699 			tmp = stripe_nr + stripe_index;
4700 			stripe_index = do_div(tmp, map->num_stripes);
4701 		}
4702 	} else {
4703 		/*
4704 		 * after this do_div call, stripe_nr is the number of stripes
4705 		 * on this device we have to walk to find the data, and
4706 		 * stripe_index is the number of our device in the stripe array
4707 		 */
4708 		stripe_index = do_div(stripe_nr, map->num_stripes);
4709 		mirror_num = stripe_index + 1;
4710 	}
4711 	BUG_ON(stripe_index >= map->num_stripes);
4712 
4713 	num_alloc_stripes = num_stripes;
4714 	if (dev_replace_is_ongoing) {
4715 		if (rw & (REQ_WRITE | REQ_DISCARD))
4716 			num_alloc_stripes <<= 1;
4717 		if (rw & REQ_GET_READ_MIRRORS)
4718 			num_alloc_stripes++;
4719 	}
4720 	bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4721 	if (!bbio) {
4722 		ret = -ENOMEM;
4723 		goto out;
4724 	}
4725 	atomic_set(&bbio->error, 0);
4726 
4727 	if (rw & REQ_DISCARD) {
4728 		int factor = 0;
4729 		int sub_stripes = 0;
4730 		u64 stripes_per_dev = 0;
4731 		u32 remaining_stripes = 0;
4732 		u32 last_stripe = 0;
4733 
4734 		if (map->type &
4735 		    (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4736 			if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4737 				sub_stripes = 1;
4738 			else
4739 				sub_stripes = map->sub_stripes;
4740 
4741 			factor = map->num_stripes / sub_stripes;
4742 			stripes_per_dev = div_u64_rem(stripe_nr_end -
4743 						      stripe_nr_orig,
4744 						      factor,
4745 						      &remaining_stripes);
4746 			div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4747 			last_stripe *= sub_stripes;
4748 		}
4749 
4750 		for (i = 0; i < num_stripes; i++) {
4751 			bbio->stripes[i].physical =
4752 				map->stripes[stripe_index].physical +
4753 				stripe_offset + stripe_nr * map->stripe_len;
4754 			bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4755 
4756 			if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4757 					 BTRFS_BLOCK_GROUP_RAID10)) {
4758 				bbio->stripes[i].length = stripes_per_dev *
4759 							  map->stripe_len;
4760 
4761 				if (i / sub_stripes < remaining_stripes)
4762 					bbio->stripes[i].length +=
4763 						map->stripe_len;
4764 
4765 				/*
4766 				 * Special for the first stripe and
4767 				 * the last stripe:
4768 				 *
4769 				 * |-------|...|-------|
4770 				 *     |----------|
4771 				 *    off     end_off
4772 				 */
4773 				if (i < sub_stripes)
4774 					bbio->stripes[i].length -=
4775 						stripe_offset;
4776 
4777 				if (stripe_index >= last_stripe &&
4778 				    stripe_index <= (last_stripe +
4779 						     sub_stripes - 1))
4780 					bbio->stripes[i].length -=
4781 						stripe_end_offset;
4782 
4783 				if (i == sub_stripes - 1)
4784 					stripe_offset = 0;
4785 			} else
4786 				bbio->stripes[i].length = *length;
4787 
4788 			stripe_index++;
4789 			if (stripe_index == map->num_stripes) {
4790 				/* This could only happen for RAID0/10 */
4791 				stripe_index = 0;
4792 				stripe_nr++;
4793 			}
4794 		}
4795 	} else {
4796 		for (i = 0; i < num_stripes; i++) {
4797 			bbio->stripes[i].physical =
4798 				map->stripes[stripe_index].physical +
4799 				stripe_offset +
4800 				stripe_nr * map->stripe_len;
4801 			bbio->stripes[i].dev =
4802 				map->stripes[stripe_index].dev;
4803 			stripe_index++;
4804 		}
4805 	}
4806 
4807 	if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4808 		if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4809 				 BTRFS_BLOCK_GROUP_RAID10 |
4810 				 BTRFS_BLOCK_GROUP_RAID5 |
4811 				 BTRFS_BLOCK_GROUP_DUP)) {
4812 			max_errors = 1;
4813 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4814 			max_errors = 2;
4815 		}
4816 	}
4817 
4818 	if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4819 	    dev_replace->tgtdev != NULL) {
4820 		int index_where_to_add;
4821 		u64 srcdev_devid = dev_replace->srcdev->devid;
4822 
4823 		/*
4824 		 * duplicate the write operations while the dev replace
4825 		 * procedure is running. Since the copying of the old disk
4826 		 * to the new disk takes place at run time while the
4827 		 * filesystem is mounted writable, the regular write
4828 		 * operations to the old disk have to be duplicated to go
4829 		 * to the new disk as well.
4830 		 * Note that device->missing is handled by the caller, and
4831 		 * that the write to the old disk is already set up in the
4832 		 * stripes array.
4833 		 */
4834 		index_where_to_add = num_stripes;
4835 		for (i = 0; i < num_stripes; i++) {
4836 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
4837 				/* write to new disk, too */
4838 				struct btrfs_bio_stripe *new =
4839 					bbio->stripes + index_where_to_add;
4840 				struct btrfs_bio_stripe *old =
4841 					bbio->stripes + i;
4842 
4843 				new->physical = old->physical;
4844 				new->length = old->length;
4845 				new->dev = dev_replace->tgtdev;
4846 				index_where_to_add++;
4847 				max_errors++;
4848 			}
4849 		}
4850 		num_stripes = index_where_to_add;
4851 	} else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4852 		   dev_replace->tgtdev != NULL) {
4853 		u64 srcdev_devid = dev_replace->srcdev->devid;
4854 		int index_srcdev = 0;
4855 		int found = 0;
4856 		u64 physical_of_found = 0;
4857 
4858 		/*
4859 		 * During the dev-replace procedure, the target drive can
4860 		 * also be used to read data in case it is needed to repair
4861 		 * a corrupt block elsewhere. This is possible if the
4862 		 * requested area is left of the left cursor. In this area,
4863 		 * the target drive is a full copy of the source drive.
4864 		 */
4865 		for (i = 0; i < num_stripes; i++) {
4866 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
4867 				/*
4868 				 * In case of DUP, in order to keep it
4869 				 * simple, only add the mirror with the
4870 				 * lowest physical address
4871 				 */
4872 				if (found &&
4873 				    physical_of_found <=
4874 				     bbio->stripes[i].physical)
4875 					continue;
4876 				index_srcdev = i;
4877 				found = 1;
4878 				physical_of_found = bbio->stripes[i].physical;
4879 			}
4880 		}
4881 		if (found) {
4882 			u64 length = map->stripe_len;
4883 
4884 			if (physical_of_found + length <=
4885 			    dev_replace->cursor_left) {
4886 				struct btrfs_bio_stripe *tgtdev_stripe =
4887 					bbio->stripes + num_stripes;
4888 
4889 				tgtdev_stripe->physical = physical_of_found;
4890 				tgtdev_stripe->length =
4891 					bbio->stripes[index_srcdev].length;
4892 				tgtdev_stripe->dev = dev_replace->tgtdev;
4893 
4894 				num_stripes++;
4895 			}
4896 		}
4897 	}
4898 
4899 	*bbio_ret = bbio;
4900 	bbio->num_stripes = num_stripes;
4901 	bbio->max_errors = max_errors;
4902 	bbio->mirror_num = mirror_num;
4903 
4904 	/*
4905 	 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4906 	 * mirror_num == num_stripes + 1 && dev_replace target drive is
4907 	 * available as a mirror
4908 	 */
4909 	if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4910 		WARN_ON(num_stripes > 1);
4911 		bbio->stripes[0].dev = dev_replace->tgtdev;
4912 		bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4913 		bbio->mirror_num = map->num_stripes + 1;
4914 	}
4915 	if (raid_map) {
4916 		sort_parity_stripes(bbio, raid_map);
4917 		*raid_map_ret = raid_map;
4918 	}
4919 out:
4920 	if (dev_replace_is_ongoing)
4921 		btrfs_dev_replace_unlock(dev_replace);
4922 	free_extent_map(em);
4923 	return ret;
4924 }
4925 
4926 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4927 		      u64 logical, u64 *length,
4928 		      struct btrfs_bio **bbio_ret, int mirror_num)
4929 {
4930 	return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4931 				 mirror_num, NULL);
4932 }
4933 
4934 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4935 		     u64 chunk_start, u64 physical, u64 devid,
4936 		     u64 **logical, int *naddrs, int *stripe_len)
4937 {
4938 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4939 	struct extent_map *em;
4940 	struct map_lookup *map;
4941 	u64 *buf;
4942 	u64 bytenr;
4943 	u64 length;
4944 	u64 stripe_nr;
4945 	u64 rmap_len;
4946 	int i, j, nr = 0;
4947 
4948 	read_lock(&em_tree->lock);
4949 	em = lookup_extent_mapping(em_tree, chunk_start, 1);
4950 	read_unlock(&em_tree->lock);
4951 
4952 	if (!em) {
4953 		printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n",
4954 		       chunk_start);
4955 		return -EIO;
4956 	}
4957 
4958 	if (em->start != chunk_start) {
4959 		printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n",
4960 		       em->start, chunk_start);
4961 		free_extent_map(em);
4962 		return -EIO;
4963 	}
4964 	map = (struct map_lookup *)em->bdev;
4965 
4966 	length = em->len;
4967 	rmap_len = map->stripe_len;
4968 
4969 	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4970 		do_div(length, map->num_stripes / map->sub_stripes);
4971 	else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4972 		do_div(length, map->num_stripes);
4973 	else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4974 			      BTRFS_BLOCK_GROUP_RAID6)) {
4975 		do_div(length, nr_data_stripes(map));
4976 		rmap_len = map->stripe_len * nr_data_stripes(map);
4977 	}
4978 
4979 	buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4980 	BUG_ON(!buf); /* -ENOMEM */
4981 
4982 	for (i = 0; i < map->num_stripes; i++) {
4983 		if (devid && map->stripes[i].dev->devid != devid)
4984 			continue;
4985 		if (map->stripes[i].physical > physical ||
4986 		    map->stripes[i].physical + length <= physical)
4987 			continue;
4988 
4989 		stripe_nr = physical - map->stripes[i].physical;
4990 		do_div(stripe_nr, map->stripe_len);
4991 
4992 		if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4993 			stripe_nr = stripe_nr * map->num_stripes + i;
4994 			do_div(stripe_nr, map->sub_stripes);
4995 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4996 			stripe_nr = stripe_nr * map->num_stripes + i;
4997 		} /* else if RAID[56], multiply by nr_data_stripes().
4998 		   * Alternatively, just use rmap_len below instead of
4999 		   * map->stripe_len */
5000 
5001 		bytenr = chunk_start + stripe_nr * rmap_len;
5002 		WARN_ON(nr >= map->num_stripes);
5003 		for (j = 0; j < nr; j++) {
5004 			if (buf[j] == bytenr)
5005 				break;
5006 		}
5007 		if (j == nr) {
5008 			WARN_ON(nr >= map->num_stripes);
5009 			buf[nr++] = bytenr;
5010 		}
5011 	}
5012 
5013 	*logical = buf;
5014 	*naddrs = nr;
5015 	*stripe_len = rmap_len;
5016 
5017 	free_extent_map(em);
5018 	return 0;
5019 }
5020 
5021 static void btrfs_end_bio(struct bio *bio, int err)
5022 {
5023 	struct btrfs_bio *bbio = bio->bi_private;
5024 	int is_orig_bio = 0;
5025 
5026 	if (err) {
5027 		atomic_inc(&bbio->error);
5028 		if (err == -EIO || err == -EREMOTEIO) {
5029 			unsigned int stripe_index =
5030 				btrfs_io_bio(bio)->stripe_index;
5031 			struct btrfs_device *dev;
5032 
5033 			BUG_ON(stripe_index >= bbio->num_stripes);
5034 			dev = bbio->stripes[stripe_index].dev;
5035 			if (dev->bdev) {
5036 				if (bio->bi_rw & WRITE)
5037 					btrfs_dev_stat_inc(dev,
5038 						BTRFS_DEV_STAT_WRITE_ERRS);
5039 				else
5040 					btrfs_dev_stat_inc(dev,
5041 						BTRFS_DEV_STAT_READ_ERRS);
5042 				if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5043 					btrfs_dev_stat_inc(dev,
5044 						BTRFS_DEV_STAT_FLUSH_ERRS);
5045 				btrfs_dev_stat_print_on_error(dev);
5046 			}
5047 		}
5048 	}
5049 
5050 	if (bio == bbio->orig_bio)
5051 		is_orig_bio = 1;
5052 
5053 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
5054 		if (!is_orig_bio) {
5055 			bio_put(bio);
5056 			bio = bbio->orig_bio;
5057 		}
5058 		bio->bi_private = bbio->private;
5059 		bio->bi_end_io = bbio->end_io;
5060 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5061 		/* only send an error to the higher layers if it is
5062 		 * beyond the tolerance of the btrfs bio
5063 		 */
5064 		if (atomic_read(&bbio->error) > bbio->max_errors) {
5065 			err = -EIO;
5066 		} else {
5067 			/*
5068 			 * this bio is actually up to date, we didn't
5069 			 * go over the max number of errors
5070 			 */
5071 			set_bit(BIO_UPTODATE, &bio->bi_flags);
5072 			err = 0;
5073 		}
5074 		kfree(bbio);
5075 
5076 		bio_endio(bio, err);
5077 	} else if (!is_orig_bio) {
5078 		bio_put(bio);
5079 	}
5080 }
5081 
5082 struct async_sched {
5083 	struct bio *bio;
5084 	int rw;
5085 	struct btrfs_fs_info *info;
5086 	struct btrfs_work work;
5087 };
5088 
5089 /*
5090  * see run_scheduled_bios for a description of why bios are collected for
5091  * async submit.
5092  *
5093  * This will add one bio to the pending list for a device and make sure
5094  * the work struct is scheduled.
5095  */
5096 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5097 					struct btrfs_device *device,
5098 					int rw, struct bio *bio)
5099 {
5100 	int should_queue = 1;
5101 	struct btrfs_pending_bios *pending_bios;
5102 
5103 	if (device->missing || !device->bdev) {
5104 		bio_endio(bio, -EIO);
5105 		return;
5106 	}
5107 
5108 	/* don't bother with additional async steps for reads, right now */
5109 	if (!(rw & REQ_WRITE)) {
5110 		bio_get(bio);
5111 		btrfsic_submit_bio(rw, bio);
5112 		bio_put(bio);
5113 		return;
5114 	}
5115 
5116 	/*
5117 	 * nr_async_bios allows us to reliably return congestion to the
5118 	 * higher layers.  Otherwise, the async bio makes it appear we have
5119 	 * made progress against dirty pages when we've really just put it
5120 	 * on a queue for later
5121 	 */
5122 	atomic_inc(&root->fs_info->nr_async_bios);
5123 	WARN_ON(bio->bi_next);
5124 	bio->bi_next = NULL;
5125 	bio->bi_rw |= rw;
5126 
5127 	spin_lock(&device->io_lock);
5128 	if (bio->bi_rw & REQ_SYNC)
5129 		pending_bios = &device->pending_sync_bios;
5130 	else
5131 		pending_bios = &device->pending_bios;
5132 
5133 	if (pending_bios->tail)
5134 		pending_bios->tail->bi_next = bio;
5135 
5136 	pending_bios->tail = bio;
5137 	if (!pending_bios->head)
5138 		pending_bios->head = bio;
5139 	if (device->running_pending)
5140 		should_queue = 0;
5141 
5142 	spin_unlock(&device->io_lock);
5143 
5144 	if (should_queue)
5145 		btrfs_queue_worker(&root->fs_info->submit_workers,
5146 				   &device->work);
5147 }
5148 
5149 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5150 		       sector_t sector)
5151 {
5152 	struct bio_vec *prev;
5153 	struct request_queue *q = bdev_get_queue(bdev);
5154 	unsigned short max_sectors = queue_max_sectors(q);
5155 	struct bvec_merge_data bvm = {
5156 		.bi_bdev = bdev,
5157 		.bi_sector = sector,
5158 		.bi_rw = bio->bi_rw,
5159 	};
5160 
5161 	if (bio->bi_vcnt == 0) {
5162 		WARN_ON(1);
5163 		return 1;
5164 	}
5165 
5166 	prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5167 	if (bio_sectors(bio) > max_sectors)
5168 		return 0;
5169 
5170 	if (!q->merge_bvec_fn)
5171 		return 1;
5172 
5173 	bvm.bi_size = bio->bi_size - prev->bv_len;
5174 	if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5175 		return 0;
5176 	return 1;
5177 }
5178 
5179 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5180 			      struct bio *bio, u64 physical, int dev_nr,
5181 			      int rw, int async)
5182 {
5183 	struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5184 
5185 	bio->bi_private = bbio;
5186 	btrfs_io_bio(bio)->stripe_index = dev_nr;
5187 	bio->bi_end_io = btrfs_end_bio;
5188 	bio->bi_sector = physical >> 9;
5189 #ifdef DEBUG
5190 	{
5191 		struct rcu_string *name;
5192 
5193 		rcu_read_lock();
5194 		name = rcu_dereference(dev->name);
5195 		pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5196 			 "(%s id %llu), size=%u\n", rw,
5197 			 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5198 			 name->str, dev->devid, bio->bi_size);
5199 		rcu_read_unlock();
5200 	}
5201 #endif
5202 	bio->bi_bdev = dev->bdev;
5203 	if (async)
5204 		btrfs_schedule_bio(root, dev, rw, bio);
5205 	else
5206 		btrfsic_submit_bio(rw, bio);
5207 }
5208 
5209 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5210 			      struct bio *first_bio, struct btrfs_device *dev,
5211 			      int dev_nr, int rw, int async)
5212 {
5213 	struct bio_vec *bvec = first_bio->bi_io_vec;
5214 	struct bio *bio;
5215 	int nr_vecs = bio_get_nr_vecs(dev->bdev);
5216 	u64 physical = bbio->stripes[dev_nr].physical;
5217 
5218 again:
5219 	bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5220 	if (!bio)
5221 		return -ENOMEM;
5222 
5223 	while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5224 		if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5225 				 bvec->bv_offset) < bvec->bv_len) {
5226 			u64 len = bio->bi_size;
5227 
5228 			atomic_inc(&bbio->stripes_pending);
5229 			submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5230 					  rw, async);
5231 			physical += len;
5232 			goto again;
5233 		}
5234 		bvec++;
5235 	}
5236 
5237 	submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5238 	return 0;
5239 }
5240 
5241 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5242 {
5243 	atomic_inc(&bbio->error);
5244 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
5245 		bio->bi_private = bbio->private;
5246 		bio->bi_end_io = bbio->end_io;
5247 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5248 		bio->bi_sector = logical >> 9;
5249 		kfree(bbio);
5250 		bio_endio(bio, -EIO);
5251 	}
5252 }
5253 
5254 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5255 		  int mirror_num, int async_submit)
5256 {
5257 	struct btrfs_device *dev;
5258 	struct bio *first_bio = bio;
5259 	u64 logical = (u64)bio->bi_sector << 9;
5260 	u64 length = 0;
5261 	u64 map_length;
5262 	u64 *raid_map = NULL;
5263 	int ret;
5264 	int dev_nr = 0;
5265 	int total_devs = 1;
5266 	struct btrfs_bio *bbio = NULL;
5267 
5268 	length = bio->bi_size;
5269 	map_length = length;
5270 
5271 	ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5272 			      mirror_num, &raid_map);
5273 	if (ret) /* -ENOMEM */
5274 		return ret;
5275 
5276 	total_devs = bbio->num_stripes;
5277 	bbio->orig_bio = first_bio;
5278 	bbio->private = first_bio->bi_private;
5279 	bbio->end_io = first_bio->bi_end_io;
5280 	atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5281 
5282 	if (raid_map) {
5283 		/* In this case, map_length has been set to the length of
5284 		   a single stripe; not the whole write */
5285 		if (rw & WRITE) {
5286 			return raid56_parity_write(root, bio, bbio,
5287 						   raid_map, map_length);
5288 		} else {
5289 			return raid56_parity_recover(root, bio, bbio,
5290 						     raid_map, map_length,
5291 						     mirror_num);
5292 		}
5293 	}
5294 
5295 	if (map_length < length) {
5296 		btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5297 			(unsigned long long)logical,
5298 			(unsigned long long)length,
5299 			(unsigned long long)map_length);
5300 		BUG();
5301 	}
5302 
5303 	while (dev_nr < total_devs) {
5304 		dev = bbio->stripes[dev_nr].dev;
5305 		if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5306 			bbio_error(bbio, first_bio, logical);
5307 			dev_nr++;
5308 			continue;
5309 		}
5310 
5311 		/*
5312 		 * Check and see if we're ok with this bio based on it's size
5313 		 * and offset with the given device.
5314 		 */
5315 		if (!bio_size_ok(dev->bdev, first_bio,
5316 				 bbio->stripes[dev_nr].physical >> 9)) {
5317 			ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5318 						 dev_nr, rw, async_submit);
5319 			BUG_ON(ret);
5320 			dev_nr++;
5321 			continue;
5322 		}
5323 
5324 		if (dev_nr < total_devs - 1) {
5325 			bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5326 			BUG_ON(!bio); /* -ENOMEM */
5327 		} else {
5328 			bio = first_bio;
5329 		}
5330 
5331 		submit_stripe_bio(root, bbio, bio,
5332 				  bbio->stripes[dev_nr].physical, dev_nr, rw,
5333 				  async_submit);
5334 		dev_nr++;
5335 	}
5336 	return 0;
5337 }
5338 
5339 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5340 				       u8 *uuid, u8 *fsid)
5341 {
5342 	struct btrfs_device *device;
5343 	struct btrfs_fs_devices *cur_devices;
5344 
5345 	cur_devices = fs_info->fs_devices;
5346 	while (cur_devices) {
5347 		if (!fsid ||
5348 		    !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5349 			device = __find_device(&cur_devices->devices,
5350 					       devid, uuid);
5351 			if (device)
5352 				return device;
5353 		}
5354 		cur_devices = cur_devices->seed;
5355 	}
5356 	return NULL;
5357 }
5358 
5359 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5360 					    u64 devid, u8 *dev_uuid)
5361 {
5362 	struct btrfs_device *device;
5363 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5364 
5365 	device = kzalloc(sizeof(*device), GFP_NOFS);
5366 	if (!device)
5367 		return NULL;
5368 	list_add(&device->dev_list,
5369 		 &fs_devices->devices);
5370 	device->dev_root = root->fs_info->dev_root;
5371 	device->devid = devid;
5372 	device->work.func = pending_bios_fn;
5373 	device->fs_devices = fs_devices;
5374 	device->missing = 1;
5375 	fs_devices->num_devices++;
5376 	fs_devices->missing_devices++;
5377 	spin_lock_init(&device->io_lock);
5378 	INIT_LIST_HEAD(&device->dev_alloc_list);
5379 	memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
5380 	return device;
5381 }
5382 
5383 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5384 			  struct extent_buffer *leaf,
5385 			  struct btrfs_chunk *chunk)
5386 {
5387 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5388 	struct map_lookup *map;
5389 	struct extent_map *em;
5390 	u64 logical;
5391 	u64 length;
5392 	u64 devid;
5393 	u8 uuid[BTRFS_UUID_SIZE];
5394 	int num_stripes;
5395 	int ret;
5396 	int i;
5397 
5398 	logical = key->offset;
5399 	length = btrfs_chunk_length(leaf, chunk);
5400 
5401 	read_lock(&map_tree->map_tree.lock);
5402 	em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5403 	read_unlock(&map_tree->map_tree.lock);
5404 
5405 	/* already mapped? */
5406 	if (em && em->start <= logical && em->start + em->len > logical) {
5407 		free_extent_map(em);
5408 		return 0;
5409 	} else if (em) {
5410 		free_extent_map(em);
5411 	}
5412 
5413 	em = alloc_extent_map();
5414 	if (!em)
5415 		return -ENOMEM;
5416 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5417 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5418 	if (!map) {
5419 		free_extent_map(em);
5420 		return -ENOMEM;
5421 	}
5422 
5423 	em->bdev = (struct block_device *)map;
5424 	em->start = logical;
5425 	em->len = length;
5426 	em->orig_start = 0;
5427 	em->block_start = 0;
5428 	em->block_len = em->len;
5429 
5430 	map->num_stripes = num_stripes;
5431 	map->io_width = btrfs_chunk_io_width(leaf, chunk);
5432 	map->io_align = btrfs_chunk_io_align(leaf, chunk);
5433 	map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5434 	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5435 	map->type = btrfs_chunk_type(leaf, chunk);
5436 	map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5437 	for (i = 0; i < num_stripes; i++) {
5438 		map->stripes[i].physical =
5439 			btrfs_stripe_offset_nr(leaf, chunk, i);
5440 		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5441 		read_extent_buffer(leaf, uuid, (unsigned long)
5442 				   btrfs_stripe_dev_uuid_nr(chunk, i),
5443 				   BTRFS_UUID_SIZE);
5444 		map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5445 							uuid, NULL);
5446 		if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5447 			kfree(map);
5448 			free_extent_map(em);
5449 			return -EIO;
5450 		}
5451 		if (!map->stripes[i].dev) {
5452 			map->stripes[i].dev =
5453 				add_missing_dev(root, devid, uuid);
5454 			if (!map->stripes[i].dev) {
5455 				kfree(map);
5456 				free_extent_map(em);
5457 				return -EIO;
5458 			}
5459 		}
5460 		map->stripes[i].dev->in_fs_metadata = 1;
5461 	}
5462 
5463 	write_lock(&map_tree->map_tree.lock);
5464 	ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5465 	write_unlock(&map_tree->map_tree.lock);
5466 	BUG_ON(ret); /* Tree corruption */
5467 	free_extent_map(em);
5468 
5469 	return 0;
5470 }
5471 
5472 static void fill_device_from_item(struct extent_buffer *leaf,
5473 				 struct btrfs_dev_item *dev_item,
5474 				 struct btrfs_device *device)
5475 {
5476 	unsigned long ptr;
5477 
5478 	device->devid = btrfs_device_id(leaf, dev_item);
5479 	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5480 	device->total_bytes = device->disk_total_bytes;
5481 	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5482 	device->type = btrfs_device_type(leaf, dev_item);
5483 	device->io_align = btrfs_device_io_align(leaf, dev_item);
5484 	device->io_width = btrfs_device_io_width(leaf, dev_item);
5485 	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5486 	WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5487 	device->is_tgtdev_for_dev_replace = 0;
5488 
5489 	ptr = (unsigned long)btrfs_device_uuid(dev_item);
5490 	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5491 }
5492 
5493 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5494 {
5495 	struct btrfs_fs_devices *fs_devices;
5496 	int ret;
5497 
5498 	BUG_ON(!mutex_is_locked(&uuid_mutex));
5499 
5500 	fs_devices = root->fs_info->fs_devices->seed;
5501 	while (fs_devices) {
5502 		if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5503 			ret = 0;
5504 			goto out;
5505 		}
5506 		fs_devices = fs_devices->seed;
5507 	}
5508 
5509 	fs_devices = find_fsid(fsid);
5510 	if (!fs_devices) {
5511 		ret = -ENOENT;
5512 		goto out;
5513 	}
5514 
5515 	fs_devices = clone_fs_devices(fs_devices);
5516 	if (IS_ERR(fs_devices)) {
5517 		ret = PTR_ERR(fs_devices);
5518 		goto out;
5519 	}
5520 
5521 	ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5522 				   root->fs_info->bdev_holder);
5523 	if (ret) {
5524 		free_fs_devices(fs_devices);
5525 		goto out;
5526 	}
5527 
5528 	if (!fs_devices->seeding) {
5529 		__btrfs_close_devices(fs_devices);
5530 		free_fs_devices(fs_devices);
5531 		ret = -EINVAL;
5532 		goto out;
5533 	}
5534 
5535 	fs_devices->seed = root->fs_info->fs_devices->seed;
5536 	root->fs_info->fs_devices->seed = fs_devices;
5537 out:
5538 	return ret;
5539 }
5540 
5541 static int read_one_dev(struct btrfs_root *root,
5542 			struct extent_buffer *leaf,
5543 			struct btrfs_dev_item *dev_item)
5544 {
5545 	struct btrfs_device *device;
5546 	u64 devid;
5547 	int ret;
5548 	u8 fs_uuid[BTRFS_UUID_SIZE];
5549 	u8 dev_uuid[BTRFS_UUID_SIZE];
5550 
5551 	devid = btrfs_device_id(leaf, dev_item);
5552 	read_extent_buffer(leaf, dev_uuid,
5553 			   (unsigned long)btrfs_device_uuid(dev_item),
5554 			   BTRFS_UUID_SIZE);
5555 	read_extent_buffer(leaf, fs_uuid,
5556 			   (unsigned long)btrfs_device_fsid(dev_item),
5557 			   BTRFS_UUID_SIZE);
5558 
5559 	if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5560 		ret = open_seed_devices(root, fs_uuid);
5561 		if (ret && !btrfs_test_opt(root, DEGRADED))
5562 			return ret;
5563 	}
5564 
5565 	device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5566 	if (!device || !device->bdev) {
5567 		if (!btrfs_test_opt(root, DEGRADED))
5568 			return -EIO;
5569 
5570 		if (!device) {
5571 			btrfs_warn(root->fs_info, "devid %llu missing",
5572 				(unsigned long long)devid);
5573 			device = add_missing_dev(root, devid, dev_uuid);
5574 			if (!device)
5575 				return -ENOMEM;
5576 		} else if (!device->missing) {
5577 			/*
5578 			 * this happens when a device that was properly setup
5579 			 * in the device info lists suddenly goes bad.
5580 			 * device->bdev is NULL, and so we have to set
5581 			 * device->missing to one here
5582 			 */
5583 			root->fs_info->fs_devices->missing_devices++;
5584 			device->missing = 1;
5585 		}
5586 	}
5587 
5588 	if (device->fs_devices != root->fs_info->fs_devices) {
5589 		BUG_ON(device->writeable);
5590 		if (device->generation !=
5591 		    btrfs_device_generation(leaf, dev_item))
5592 			return -EINVAL;
5593 	}
5594 
5595 	fill_device_from_item(leaf, dev_item, device);
5596 	device->dev_root = root->fs_info->dev_root;
5597 	device->in_fs_metadata = 1;
5598 	if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5599 		device->fs_devices->total_rw_bytes += device->total_bytes;
5600 		spin_lock(&root->fs_info->free_chunk_lock);
5601 		root->fs_info->free_chunk_space += device->total_bytes -
5602 			device->bytes_used;
5603 		spin_unlock(&root->fs_info->free_chunk_lock);
5604 	}
5605 	ret = 0;
5606 	return ret;
5607 }
5608 
5609 int btrfs_read_sys_array(struct btrfs_root *root)
5610 {
5611 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5612 	struct extent_buffer *sb;
5613 	struct btrfs_disk_key *disk_key;
5614 	struct btrfs_chunk *chunk;
5615 	u8 *ptr;
5616 	unsigned long sb_ptr;
5617 	int ret = 0;
5618 	u32 num_stripes;
5619 	u32 array_size;
5620 	u32 len = 0;
5621 	u32 cur;
5622 	struct btrfs_key key;
5623 
5624 	sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5625 					  BTRFS_SUPER_INFO_SIZE);
5626 	if (!sb)
5627 		return -ENOMEM;
5628 	btrfs_set_buffer_uptodate(sb);
5629 	btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5630 	/*
5631 	 * The sb extent buffer is artifical and just used to read the system array.
5632 	 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5633 	 * pages up-to-date when the page is larger: extent does not cover the
5634 	 * whole page and consequently check_page_uptodate does not find all
5635 	 * the page's extents up-to-date (the hole beyond sb),
5636 	 * write_extent_buffer then triggers a WARN_ON.
5637 	 *
5638 	 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5639 	 * but sb spans only this function. Add an explicit SetPageUptodate call
5640 	 * to silence the warning eg. on PowerPC 64.
5641 	 */
5642 	if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5643 		SetPageUptodate(sb->pages[0]);
5644 
5645 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5646 	array_size = btrfs_super_sys_array_size(super_copy);
5647 
5648 	ptr = super_copy->sys_chunk_array;
5649 	sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5650 	cur = 0;
5651 
5652 	while (cur < array_size) {
5653 		disk_key = (struct btrfs_disk_key *)ptr;
5654 		btrfs_disk_key_to_cpu(&key, disk_key);
5655 
5656 		len = sizeof(*disk_key); ptr += len;
5657 		sb_ptr += len;
5658 		cur += len;
5659 
5660 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5661 			chunk = (struct btrfs_chunk *)sb_ptr;
5662 			ret = read_one_chunk(root, &key, sb, chunk);
5663 			if (ret)
5664 				break;
5665 			num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5666 			len = btrfs_chunk_item_size(num_stripes);
5667 		} else {
5668 			ret = -EIO;
5669 			break;
5670 		}
5671 		ptr += len;
5672 		sb_ptr += len;
5673 		cur += len;
5674 	}
5675 	free_extent_buffer(sb);
5676 	return ret;
5677 }
5678 
5679 int btrfs_read_chunk_tree(struct btrfs_root *root)
5680 {
5681 	struct btrfs_path *path;
5682 	struct extent_buffer *leaf;
5683 	struct btrfs_key key;
5684 	struct btrfs_key found_key;
5685 	int ret;
5686 	int slot;
5687 
5688 	root = root->fs_info->chunk_root;
5689 
5690 	path = btrfs_alloc_path();
5691 	if (!path)
5692 		return -ENOMEM;
5693 
5694 	mutex_lock(&uuid_mutex);
5695 	lock_chunks(root);
5696 
5697 	/* first we search for all of the device items, and then we
5698 	 * read in all of the chunk items.  This way we can create chunk
5699 	 * mappings that reference all of the devices that are afound
5700 	 */
5701 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5702 	key.offset = 0;
5703 	key.type = 0;
5704 again:
5705 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5706 	if (ret < 0)
5707 		goto error;
5708 	while (1) {
5709 		leaf = path->nodes[0];
5710 		slot = path->slots[0];
5711 		if (slot >= btrfs_header_nritems(leaf)) {
5712 			ret = btrfs_next_leaf(root, path);
5713 			if (ret == 0)
5714 				continue;
5715 			if (ret < 0)
5716 				goto error;
5717 			break;
5718 		}
5719 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
5720 		if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5721 			if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5722 				break;
5723 			if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5724 				struct btrfs_dev_item *dev_item;
5725 				dev_item = btrfs_item_ptr(leaf, slot,
5726 						  struct btrfs_dev_item);
5727 				ret = read_one_dev(root, leaf, dev_item);
5728 				if (ret)
5729 					goto error;
5730 			}
5731 		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5732 			struct btrfs_chunk *chunk;
5733 			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5734 			ret = read_one_chunk(root, &found_key, leaf, chunk);
5735 			if (ret)
5736 				goto error;
5737 		}
5738 		path->slots[0]++;
5739 	}
5740 	if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5741 		key.objectid = 0;
5742 		btrfs_release_path(path);
5743 		goto again;
5744 	}
5745 	ret = 0;
5746 error:
5747 	unlock_chunks(root);
5748 	mutex_unlock(&uuid_mutex);
5749 
5750 	btrfs_free_path(path);
5751 	return ret;
5752 }
5753 
5754 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5755 {
5756 	int i;
5757 
5758 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5759 		btrfs_dev_stat_reset(dev, i);
5760 }
5761 
5762 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5763 {
5764 	struct btrfs_key key;
5765 	struct btrfs_key found_key;
5766 	struct btrfs_root *dev_root = fs_info->dev_root;
5767 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5768 	struct extent_buffer *eb;
5769 	int slot;
5770 	int ret = 0;
5771 	struct btrfs_device *device;
5772 	struct btrfs_path *path = NULL;
5773 	int i;
5774 
5775 	path = btrfs_alloc_path();
5776 	if (!path) {
5777 		ret = -ENOMEM;
5778 		goto out;
5779 	}
5780 
5781 	mutex_lock(&fs_devices->device_list_mutex);
5782 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
5783 		int item_size;
5784 		struct btrfs_dev_stats_item *ptr;
5785 
5786 		key.objectid = 0;
5787 		key.type = BTRFS_DEV_STATS_KEY;
5788 		key.offset = device->devid;
5789 		ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5790 		if (ret) {
5791 			__btrfs_reset_dev_stats(device);
5792 			device->dev_stats_valid = 1;
5793 			btrfs_release_path(path);
5794 			continue;
5795 		}
5796 		slot = path->slots[0];
5797 		eb = path->nodes[0];
5798 		btrfs_item_key_to_cpu(eb, &found_key, slot);
5799 		item_size = btrfs_item_size_nr(eb, slot);
5800 
5801 		ptr = btrfs_item_ptr(eb, slot,
5802 				     struct btrfs_dev_stats_item);
5803 
5804 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5805 			if (item_size >= (1 + i) * sizeof(__le64))
5806 				btrfs_dev_stat_set(device, i,
5807 					btrfs_dev_stats_value(eb, ptr, i));
5808 			else
5809 				btrfs_dev_stat_reset(device, i);
5810 		}
5811 
5812 		device->dev_stats_valid = 1;
5813 		btrfs_dev_stat_print_on_load(device);
5814 		btrfs_release_path(path);
5815 	}
5816 	mutex_unlock(&fs_devices->device_list_mutex);
5817 
5818 out:
5819 	btrfs_free_path(path);
5820 	return ret < 0 ? ret : 0;
5821 }
5822 
5823 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5824 				struct btrfs_root *dev_root,
5825 				struct btrfs_device *device)
5826 {
5827 	struct btrfs_path *path;
5828 	struct btrfs_key key;
5829 	struct extent_buffer *eb;
5830 	struct btrfs_dev_stats_item *ptr;
5831 	int ret;
5832 	int i;
5833 
5834 	key.objectid = 0;
5835 	key.type = BTRFS_DEV_STATS_KEY;
5836 	key.offset = device->devid;
5837 
5838 	path = btrfs_alloc_path();
5839 	BUG_ON(!path);
5840 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5841 	if (ret < 0) {
5842 		printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5843 			      ret, rcu_str_deref(device->name));
5844 		goto out;
5845 	}
5846 
5847 	if (ret == 0 &&
5848 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5849 		/* need to delete old one and insert a new one */
5850 		ret = btrfs_del_item(trans, dev_root, path);
5851 		if (ret != 0) {
5852 			printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5853 				      rcu_str_deref(device->name), ret);
5854 			goto out;
5855 		}
5856 		ret = 1;
5857 	}
5858 
5859 	if (ret == 1) {
5860 		/* need to insert a new item */
5861 		btrfs_release_path(path);
5862 		ret = btrfs_insert_empty_item(trans, dev_root, path,
5863 					      &key, sizeof(*ptr));
5864 		if (ret < 0) {
5865 			printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5866 				      rcu_str_deref(device->name), ret);
5867 			goto out;
5868 		}
5869 	}
5870 
5871 	eb = path->nodes[0];
5872 	ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5873 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5874 		btrfs_set_dev_stats_value(eb, ptr, i,
5875 					  btrfs_dev_stat_read(device, i));
5876 	btrfs_mark_buffer_dirty(eb);
5877 
5878 out:
5879 	btrfs_free_path(path);
5880 	return ret;
5881 }
5882 
5883 /*
5884  * called from commit_transaction. Writes all changed device stats to disk.
5885  */
5886 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5887 			struct btrfs_fs_info *fs_info)
5888 {
5889 	struct btrfs_root *dev_root = fs_info->dev_root;
5890 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5891 	struct btrfs_device *device;
5892 	int ret = 0;
5893 
5894 	mutex_lock(&fs_devices->device_list_mutex);
5895 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
5896 		if (!device->dev_stats_valid || !device->dev_stats_dirty)
5897 			continue;
5898 
5899 		ret = update_dev_stat_item(trans, dev_root, device);
5900 		if (!ret)
5901 			device->dev_stats_dirty = 0;
5902 	}
5903 	mutex_unlock(&fs_devices->device_list_mutex);
5904 
5905 	return ret;
5906 }
5907 
5908 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5909 {
5910 	btrfs_dev_stat_inc(dev, index);
5911 	btrfs_dev_stat_print_on_error(dev);
5912 }
5913 
5914 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5915 {
5916 	if (!dev->dev_stats_valid)
5917 		return;
5918 	printk_ratelimited_in_rcu(KERN_ERR
5919 			   "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5920 			   rcu_str_deref(dev->name),
5921 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5922 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5923 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5924 			   btrfs_dev_stat_read(dev,
5925 					       BTRFS_DEV_STAT_CORRUPTION_ERRS),
5926 			   btrfs_dev_stat_read(dev,
5927 					       BTRFS_DEV_STAT_GENERATION_ERRS));
5928 }
5929 
5930 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5931 {
5932 	int i;
5933 
5934 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5935 		if (btrfs_dev_stat_read(dev, i) != 0)
5936 			break;
5937 	if (i == BTRFS_DEV_STAT_VALUES_MAX)
5938 		return; /* all values == 0, suppress message */
5939 
5940 	printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5941 	       rcu_str_deref(dev->name),
5942 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5943 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5944 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5945 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5946 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5947 }
5948 
5949 int btrfs_get_dev_stats(struct btrfs_root *root,
5950 			struct btrfs_ioctl_get_dev_stats *stats)
5951 {
5952 	struct btrfs_device *dev;
5953 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5954 	int i;
5955 
5956 	mutex_lock(&fs_devices->device_list_mutex);
5957 	dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5958 	mutex_unlock(&fs_devices->device_list_mutex);
5959 
5960 	if (!dev) {
5961 		printk(KERN_WARNING
5962 		       "btrfs: get dev_stats failed, device not found\n");
5963 		return -ENODEV;
5964 	} else if (!dev->dev_stats_valid) {
5965 		printk(KERN_WARNING
5966 		       "btrfs: get dev_stats failed, not yet valid\n");
5967 		return -ENODEV;
5968 	} else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5969 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5970 			if (stats->nr_items > i)
5971 				stats->values[i] =
5972 					btrfs_dev_stat_read_and_reset(dev, i);
5973 			else
5974 				btrfs_dev_stat_reset(dev, i);
5975 		}
5976 	} else {
5977 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5978 			if (stats->nr_items > i)
5979 				stats->values[i] = btrfs_dev_stat_read(dev, i);
5980 	}
5981 	if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5982 		stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5983 	return 0;
5984 }
5985 
5986 int btrfs_scratch_superblock(struct btrfs_device *device)
5987 {
5988 	struct buffer_head *bh;
5989 	struct btrfs_super_block *disk_super;
5990 
5991 	bh = btrfs_read_dev_super(device->bdev);
5992 	if (!bh)
5993 		return -EINVAL;
5994 	disk_super = (struct btrfs_super_block *)bh->b_data;
5995 
5996 	memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5997 	set_buffer_dirty(bh);
5998 	sync_dirty_buffer(bh);
5999 	brelse(bh);
6000 
6001 	return 0;
6002 }
6003