xref: /openbmc/linux/fs/btrfs/volumes.c (revision 81d67439)
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 <asm/div64.h>
27 #include "compat.h"
28 #include "ctree.h"
29 #include "extent_map.h"
30 #include "disk-io.h"
31 #include "transaction.h"
32 #include "print-tree.h"
33 #include "volumes.h"
34 #include "async-thread.h"
35 
36 static int init_first_rw_device(struct btrfs_trans_handle *trans,
37 				struct btrfs_root *root,
38 				struct btrfs_device *device);
39 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
40 
41 static DEFINE_MUTEX(uuid_mutex);
42 static LIST_HEAD(fs_uuids);
43 
44 static void lock_chunks(struct btrfs_root *root)
45 {
46 	mutex_lock(&root->fs_info->chunk_mutex);
47 }
48 
49 static void unlock_chunks(struct btrfs_root *root)
50 {
51 	mutex_unlock(&root->fs_info->chunk_mutex);
52 }
53 
54 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
55 {
56 	struct btrfs_device *device;
57 	WARN_ON(fs_devices->opened);
58 	while (!list_empty(&fs_devices->devices)) {
59 		device = list_entry(fs_devices->devices.next,
60 				    struct btrfs_device, dev_list);
61 		list_del(&device->dev_list);
62 		kfree(device->name);
63 		kfree(device);
64 	}
65 	kfree(fs_devices);
66 }
67 
68 int btrfs_cleanup_fs_uuids(void)
69 {
70 	struct btrfs_fs_devices *fs_devices;
71 
72 	while (!list_empty(&fs_uuids)) {
73 		fs_devices = list_entry(fs_uuids.next,
74 					struct btrfs_fs_devices, list);
75 		list_del(&fs_devices->list);
76 		free_fs_devices(fs_devices);
77 	}
78 	return 0;
79 }
80 
81 static noinline struct btrfs_device *__find_device(struct list_head *head,
82 						   u64 devid, u8 *uuid)
83 {
84 	struct btrfs_device *dev;
85 
86 	list_for_each_entry(dev, head, dev_list) {
87 		if (dev->devid == devid &&
88 		    (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
89 			return dev;
90 		}
91 	}
92 	return NULL;
93 }
94 
95 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
96 {
97 	struct btrfs_fs_devices *fs_devices;
98 
99 	list_for_each_entry(fs_devices, &fs_uuids, list) {
100 		if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
101 			return fs_devices;
102 	}
103 	return NULL;
104 }
105 
106 static void requeue_list(struct btrfs_pending_bios *pending_bios,
107 			struct bio *head, struct bio *tail)
108 {
109 
110 	struct bio *old_head;
111 
112 	old_head = pending_bios->head;
113 	pending_bios->head = head;
114 	if (pending_bios->tail)
115 		tail->bi_next = old_head;
116 	else
117 		pending_bios->tail = tail;
118 }
119 
120 /*
121  * we try to collect pending bios for a device so we don't get a large
122  * number of procs sending bios down to the same device.  This greatly
123  * improves the schedulers ability to collect and merge the bios.
124  *
125  * But, it also turns into a long list of bios to process and that is sure
126  * to eventually make the worker thread block.  The solution here is to
127  * make some progress and then put this work struct back at the end of
128  * the list if the block device is congested.  This way, multiple devices
129  * can make progress from a single worker thread.
130  */
131 static noinline int run_scheduled_bios(struct btrfs_device *device)
132 {
133 	struct bio *pending;
134 	struct backing_dev_info *bdi;
135 	struct btrfs_fs_info *fs_info;
136 	struct btrfs_pending_bios *pending_bios;
137 	struct bio *tail;
138 	struct bio *cur;
139 	int again = 0;
140 	unsigned long num_run;
141 	unsigned long batch_run = 0;
142 	unsigned long limit;
143 	unsigned long last_waited = 0;
144 	int force_reg = 0;
145 	struct blk_plug plug;
146 
147 	/*
148 	 * this function runs all the bios we've collected for
149 	 * a particular device.  We don't want to wander off to
150 	 * another device without first sending all of these down.
151 	 * So, setup a plug here and finish it off before we return
152 	 */
153 	blk_start_plug(&plug);
154 
155 	bdi = blk_get_backing_dev_info(device->bdev);
156 	fs_info = device->dev_root->fs_info;
157 	limit = btrfs_async_submit_limit(fs_info);
158 	limit = limit * 2 / 3;
159 
160 loop:
161 	spin_lock(&device->io_lock);
162 
163 loop_lock:
164 	num_run = 0;
165 
166 	/* take all the bios off the list at once and process them
167 	 * later on (without the lock held).  But, remember the
168 	 * tail and other pointers so the bios can be properly reinserted
169 	 * into the list if we hit congestion
170 	 */
171 	if (!force_reg && device->pending_sync_bios.head) {
172 		pending_bios = &device->pending_sync_bios;
173 		force_reg = 1;
174 	} else {
175 		pending_bios = &device->pending_bios;
176 		force_reg = 0;
177 	}
178 
179 	pending = pending_bios->head;
180 	tail = pending_bios->tail;
181 	WARN_ON(pending && !tail);
182 
183 	/*
184 	 * if pending was null this time around, no bios need processing
185 	 * at all and we can stop.  Otherwise it'll loop back up again
186 	 * and do an additional check so no bios are missed.
187 	 *
188 	 * device->running_pending is used to synchronize with the
189 	 * schedule_bio code.
190 	 */
191 	if (device->pending_sync_bios.head == NULL &&
192 	    device->pending_bios.head == NULL) {
193 		again = 0;
194 		device->running_pending = 0;
195 	} else {
196 		again = 1;
197 		device->running_pending = 1;
198 	}
199 
200 	pending_bios->head = NULL;
201 	pending_bios->tail = NULL;
202 
203 	spin_unlock(&device->io_lock);
204 
205 	while (pending) {
206 
207 		rmb();
208 		/* we want to work on both lists, but do more bios on the
209 		 * sync list than the regular list
210 		 */
211 		if ((num_run > 32 &&
212 		    pending_bios != &device->pending_sync_bios &&
213 		    device->pending_sync_bios.head) ||
214 		   (num_run > 64 && pending_bios == &device->pending_sync_bios &&
215 		    device->pending_bios.head)) {
216 			spin_lock(&device->io_lock);
217 			requeue_list(pending_bios, pending, tail);
218 			goto loop_lock;
219 		}
220 
221 		cur = pending;
222 		pending = pending->bi_next;
223 		cur->bi_next = NULL;
224 		atomic_dec(&fs_info->nr_async_bios);
225 
226 		if (atomic_read(&fs_info->nr_async_bios) < limit &&
227 		    waitqueue_active(&fs_info->async_submit_wait))
228 			wake_up(&fs_info->async_submit_wait);
229 
230 		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
231 
232 		submit_bio(cur->bi_rw, cur);
233 		num_run++;
234 		batch_run++;
235 		if (need_resched())
236 			cond_resched();
237 
238 		/*
239 		 * we made progress, there is more work to do and the bdi
240 		 * is now congested.  Back off and let other work structs
241 		 * run instead
242 		 */
243 		if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
244 		    fs_info->fs_devices->open_devices > 1) {
245 			struct io_context *ioc;
246 
247 			ioc = current->io_context;
248 
249 			/*
250 			 * the main goal here is that we don't want to
251 			 * block if we're going to be able to submit
252 			 * more requests without blocking.
253 			 *
254 			 * This code does two great things, it pokes into
255 			 * the elevator code from a filesystem _and_
256 			 * it makes assumptions about how batching works.
257 			 */
258 			if (ioc && ioc->nr_batch_requests > 0 &&
259 			    time_before(jiffies, ioc->last_waited + HZ/50UL) &&
260 			    (last_waited == 0 ||
261 			     ioc->last_waited == last_waited)) {
262 				/*
263 				 * we want to go through our batch of
264 				 * requests and stop.  So, we copy out
265 				 * the ioc->last_waited time and test
266 				 * against it before looping
267 				 */
268 				last_waited = ioc->last_waited;
269 				if (need_resched())
270 					cond_resched();
271 				continue;
272 			}
273 			spin_lock(&device->io_lock);
274 			requeue_list(pending_bios, pending, tail);
275 			device->running_pending = 1;
276 
277 			spin_unlock(&device->io_lock);
278 			btrfs_requeue_work(&device->work);
279 			goto done;
280 		}
281 	}
282 
283 	cond_resched();
284 	if (again)
285 		goto loop;
286 
287 	spin_lock(&device->io_lock);
288 	if (device->pending_bios.head || device->pending_sync_bios.head)
289 		goto loop_lock;
290 	spin_unlock(&device->io_lock);
291 
292 done:
293 	blk_finish_plug(&plug);
294 	return 0;
295 }
296 
297 static void pending_bios_fn(struct btrfs_work *work)
298 {
299 	struct btrfs_device *device;
300 
301 	device = container_of(work, struct btrfs_device, work);
302 	run_scheduled_bios(device);
303 }
304 
305 static noinline int device_list_add(const char *path,
306 			   struct btrfs_super_block *disk_super,
307 			   u64 devid, struct btrfs_fs_devices **fs_devices_ret)
308 {
309 	struct btrfs_device *device;
310 	struct btrfs_fs_devices *fs_devices;
311 	u64 found_transid = btrfs_super_generation(disk_super);
312 	char *name;
313 
314 	fs_devices = find_fsid(disk_super->fsid);
315 	if (!fs_devices) {
316 		fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
317 		if (!fs_devices)
318 			return -ENOMEM;
319 		INIT_LIST_HEAD(&fs_devices->devices);
320 		INIT_LIST_HEAD(&fs_devices->alloc_list);
321 		list_add(&fs_devices->list, &fs_uuids);
322 		memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
323 		fs_devices->latest_devid = devid;
324 		fs_devices->latest_trans = found_transid;
325 		mutex_init(&fs_devices->device_list_mutex);
326 		device = NULL;
327 	} else {
328 		device = __find_device(&fs_devices->devices, devid,
329 				       disk_super->dev_item.uuid);
330 	}
331 	if (!device) {
332 		if (fs_devices->opened)
333 			return -EBUSY;
334 
335 		device = kzalloc(sizeof(*device), GFP_NOFS);
336 		if (!device) {
337 			/* we can safely leave the fs_devices entry around */
338 			return -ENOMEM;
339 		}
340 		device->devid = devid;
341 		device->work.func = pending_bios_fn;
342 		memcpy(device->uuid, disk_super->dev_item.uuid,
343 		       BTRFS_UUID_SIZE);
344 		spin_lock_init(&device->io_lock);
345 		device->name = kstrdup(path, GFP_NOFS);
346 		if (!device->name) {
347 			kfree(device);
348 			return -ENOMEM;
349 		}
350 		INIT_LIST_HEAD(&device->dev_alloc_list);
351 
352 		mutex_lock(&fs_devices->device_list_mutex);
353 		list_add_rcu(&device->dev_list, &fs_devices->devices);
354 		mutex_unlock(&fs_devices->device_list_mutex);
355 
356 		device->fs_devices = fs_devices;
357 		fs_devices->num_devices++;
358 	} else if (!device->name || strcmp(device->name, path)) {
359 		name = kstrdup(path, GFP_NOFS);
360 		if (!name)
361 			return -ENOMEM;
362 		kfree(device->name);
363 		device->name = name;
364 		if (device->missing) {
365 			fs_devices->missing_devices--;
366 			device->missing = 0;
367 		}
368 	}
369 
370 	if (found_transid > fs_devices->latest_trans) {
371 		fs_devices->latest_devid = devid;
372 		fs_devices->latest_trans = found_transid;
373 	}
374 	*fs_devices_ret = fs_devices;
375 	return 0;
376 }
377 
378 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
379 {
380 	struct btrfs_fs_devices *fs_devices;
381 	struct btrfs_device *device;
382 	struct btrfs_device *orig_dev;
383 
384 	fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
385 	if (!fs_devices)
386 		return ERR_PTR(-ENOMEM);
387 
388 	INIT_LIST_HEAD(&fs_devices->devices);
389 	INIT_LIST_HEAD(&fs_devices->alloc_list);
390 	INIT_LIST_HEAD(&fs_devices->list);
391 	mutex_init(&fs_devices->device_list_mutex);
392 	fs_devices->latest_devid = orig->latest_devid;
393 	fs_devices->latest_trans = orig->latest_trans;
394 	memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
395 
396 	/* We have held the volume lock, it is safe to get the devices. */
397 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
398 		device = kzalloc(sizeof(*device), GFP_NOFS);
399 		if (!device)
400 			goto error;
401 
402 		device->name = kstrdup(orig_dev->name, GFP_NOFS);
403 		if (!device->name) {
404 			kfree(device);
405 			goto error;
406 		}
407 
408 		device->devid = orig_dev->devid;
409 		device->work.func = pending_bios_fn;
410 		memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
411 		spin_lock_init(&device->io_lock);
412 		INIT_LIST_HEAD(&device->dev_list);
413 		INIT_LIST_HEAD(&device->dev_alloc_list);
414 
415 		list_add(&device->dev_list, &fs_devices->devices);
416 		device->fs_devices = fs_devices;
417 		fs_devices->num_devices++;
418 	}
419 	return fs_devices;
420 error:
421 	free_fs_devices(fs_devices);
422 	return ERR_PTR(-ENOMEM);
423 }
424 
425 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
426 {
427 	struct btrfs_device *device, *next;
428 
429 	mutex_lock(&uuid_mutex);
430 again:
431 	/* This is the initialized path, it is safe to release the devices. */
432 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
433 		if (device->in_fs_metadata)
434 			continue;
435 
436 		if (device->bdev) {
437 			blkdev_put(device->bdev, device->mode);
438 			device->bdev = NULL;
439 			fs_devices->open_devices--;
440 		}
441 		if (device->writeable) {
442 			list_del_init(&device->dev_alloc_list);
443 			device->writeable = 0;
444 			fs_devices->rw_devices--;
445 		}
446 		list_del_init(&device->dev_list);
447 		fs_devices->num_devices--;
448 		kfree(device->name);
449 		kfree(device);
450 	}
451 
452 	if (fs_devices->seed) {
453 		fs_devices = fs_devices->seed;
454 		goto again;
455 	}
456 
457 	mutex_unlock(&uuid_mutex);
458 	return 0;
459 }
460 
461 static void __free_device(struct work_struct *work)
462 {
463 	struct btrfs_device *device;
464 
465 	device = container_of(work, struct btrfs_device, rcu_work);
466 
467 	if (device->bdev)
468 		blkdev_put(device->bdev, device->mode);
469 
470 	kfree(device->name);
471 	kfree(device);
472 }
473 
474 static void free_device(struct rcu_head *head)
475 {
476 	struct btrfs_device *device;
477 
478 	device = container_of(head, struct btrfs_device, rcu);
479 
480 	INIT_WORK(&device->rcu_work, __free_device);
481 	schedule_work(&device->rcu_work);
482 }
483 
484 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
485 {
486 	struct btrfs_device *device;
487 
488 	if (--fs_devices->opened > 0)
489 		return 0;
490 
491 	mutex_lock(&fs_devices->device_list_mutex);
492 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
493 		struct btrfs_device *new_device;
494 
495 		if (device->bdev)
496 			fs_devices->open_devices--;
497 
498 		if (device->writeable) {
499 			list_del_init(&device->dev_alloc_list);
500 			fs_devices->rw_devices--;
501 		}
502 
503 		new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
504 		BUG_ON(!new_device);
505 		memcpy(new_device, device, sizeof(*new_device));
506 		new_device->name = kstrdup(device->name, GFP_NOFS);
507 		BUG_ON(device->name && !new_device->name);
508 		new_device->bdev = NULL;
509 		new_device->writeable = 0;
510 		new_device->in_fs_metadata = 0;
511 		list_replace_rcu(&device->dev_list, &new_device->dev_list);
512 
513 		call_rcu(&device->rcu, free_device);
514 	}
515 	mutex_unlock(&fs_devices->device_list_mutex);
516 
517 	WARN_ON(fs_devices->open_devices);
518 	WARN_ON(fs_devices->rw_devices);
519 	fs_devices->opened = 0;
520 	fs_devices->seeding = 0;
521 
522 	return 0;
523 }
524 
525 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
526 {
527 	struct btrfs_fs_devices *seed_devices = NULL;
528 	int ret;
529 
530 	mutex_lock(&uuid_mutex);
531 	ret = __btrfs_close_devices(fs_devices);
532 	if (!fs_devices->opened) {
533 		seed_devices = fs_devices->seed;
534 		fs_devices->seed = NULL;
535 	}
536 	mutex_unlock(&uuid_mutex);
537 
538 	while (seed_devices) {
539 		fs_devices = seed_devices;
540 		seed_devices = fs_devices->seed;
541 		__btrfs_close_devices(fs_devices);
542 		free_fs_devices(fs_devices);
543 	}
544 	return ret;
545 }
546 
547 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
548 				fmode_t flags, void *holder)
549 {
550 	struct block_device *bdev;
551 	struct list_head *head = &fs_devices->devices;
552 	struct btrfs_device *device;
553 	struct block_device *latest_bdev = NULL;
554 	struct buffer_head *bh;
555 	struct btrfs_super_block *disk_super;
556 	u64 latest_devid = 0;
557 	u64 latest_transid = 0;
558 	u64 devid;
559 	int seeding = 1;
560 	int ret = 0;
561 
562 	flags |= FMODE_EXCL;
563 
564 	list_for_each_entry(device, head, dev_list) {
565 		if (device->bdev)
566 			continue;
567 		if (!device->name)
568 			continue;
569 
570 		bdev = blkdev_get_by_path(device->name, flags, holder);
571 		if (IS_ERR(bdev)) {
572 			printk(KERN_INFO "open %s failed\n", device->name);
573 			goto error;
574 		}
575 		set_blocksize(bdev, 4096);
576 
577 		bh = btrfs_read_dev_super(bdev);
578 		if (!bh) {
579 			ret = -EINVAL;
580 			goto error_close;
581 		}
582 
583 		disk_super = (struct btrfs_super_block *)bh->b_data;
584 		devid = btrfs_stack_device_id(&disk_super->dev_item);
585 		if (devid != device->devid)
586 			goto error_brelse;
587 
588 		if (memcmp(device->uuid, disk_super->dev_item.uuid,
589 			   BTRFS_UUID_SIZE))
590 			goto error_brelse;
591 
592 		device->generation = btrfs_super_generation(disk_super);
593 		if (!latest_transid || device->generation > latest_transid) {
594 			latest_devid = devid;
595 			latest_transid = device->generation;
596 			latest_bdev = bdev;
597 		}
598 
599 		if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
600 			device->writeable = 0;
601 		} else {
602 			device->writeable = !bdev_read_only(bdev);
603 			seeding = 0;
604 		}
605 
606 		device->bdev = bdev;
607 		device->in_fs_metadata = 0;
608 		device->mode = flags;
609 
610 		if (!blk_queue_nonrot(bdev_get_queue(bdev)))
611 			fs_devices->rotating = 1;
612 
613 		fs_devices->open_devices++;
614 		if (device->writeable) {
615 			fs_devices->rw_devices++;
616 			list_add(&device->dev_alloc_list,
617 				 &fs_devices->alloc_list);
618 		}
619 		brelse(bh);
620 		continue;
621 
622 error_brelse:
623 		brelse(bh);
624 error_close:
625 		blkdev_put(bdev, flags);
626 error:
627 		continue;
628 	}
629 	if (fs_devices->open_devices == 0) {
630 		ret = -EIO;
631 		goto out;
632 	}
633 	fs_devices->seeding = seeding;
634 	fs_devices->opened = 1;
635 	fs_devices->latest_bdev = latest_bdev;
636 	fs_devices->latest_devid = latest_devid;
637 	fs_devices->latest_trans = latest_transid;
638 	fs_devices->total_rw_bytes = 0;
639 out:
640 	return ret;
641 }
642 
643 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
644 		       fmode_t flags, void *holder)
645 {
646 	int ret;
647 
648 	mutex_lock(&uuid_mutex);
649 	if (fs_devices->opened) {
650 		fs_devices->opened++;
651 		ret = 0;
652 	} else {
653 		ret = __btrfs_open_devices(fs_devices, flags, holder);
654 	}
655 	mutex_unlock(&uuid_mutex);
656 	return ret;
657 }
658 
659 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
660 			  struct btrfs_fs_devices **fs_devices_ret)
661 {
662 	struct btrfs_super_block *disk_super;
663 	struct block_device *bdev;
664 	struct buffer_head *bh;
665 	int ret;
666 	u64 devid;
667 	u64 transid;
668 
669 	mutex_lock(&uuid_mutex);
670 
671 	flags |= FMODE_EXCL;
672 	bdev = blkdev_get_by_path(path, flags, holder);
673 
674 	if (IS_ERR(bdev)) {
675 		ret = PTR_ERR(bdev);
676 		goto error;
677 	}
678 
679 	ret = set_blocksize(bdev, 4096);
680 	if (ret)
681 		goto error_close;
682 	bh = btrfs_read_dev_super(bdev);
683 	if (!bh) {
684 		ret = -EINVAL;
685 		goto error_close;
686 	}
687 	disk_super = (struct btrfs_super_block *)bh->b_data;
688 	devid = btrfs_stack_device_id(&disk_super->dev_item);
689 	transid = btrfs_super_generation(disk_super);
690 	if (disk_super->label[0])
691 		printk(KERN_INFO "device label %s ", disk_super->label);
692 	else
693 		printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
694 	printk(KERN_CONT "devid %llu transid %llu %s\n",
695 	       (unsigned long long)devid, (unsigned long long)transid, path);
696 	ret = device_list_add(path, disk_super, devid, fs_devices_ret);
697 
698 	brelse(bh);
699 error_close:
700 	blkdev_put(bdev, flags);
701 error:
702 	mutex_unlock(&uuid_mutex);
703 	return ret;
704 }
705 
706 /* helper to account the used device space in the range */
707 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
708 				   u64 end, u64 *length)
709 {
710 	struct btrfs_key key;
711 	struct btrfs_root *root = device->dev_root;
712 	struct btrfs_dev_extent *dev_extent;
713 	struct btrfs_path *path;
714 	u64 extent_end;
715 	int ret;
716 	int slot;
717 	struct extent_buffer *l;
718 
719 	*length = 0;
720 
721 	if (start >= device->total_bytes)
722 		return 0;
723 
724 	path = btrfs_alloc_path();
725 	if (!path)
726 		return -ENOMEM;
727 	path->reada = 2;
728 
729 	key.objectid = device->devid;
730 	key.offset = start;
731 	key.type = BTRFS_DEV_EXTENT_KEY;
732 
733 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
734 	if (ret < 0)
735 		goto out;
736 	if (ret > 0) {
737 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
738 		if (ret < 0)
739 			goto out;
740 	}
741 
742 	while (1) {
743 		l = path->nodes[0];
744 		slot = path->slots[0];
745 		if (slot >= btrfs_header_nritems(l)) {
746 			ret = btrfs_next_leaf(root, path);
747 			if (ret == 0)
748 				continue;
749 			if (ret < 0)
750 				goto out;
751 
752 			break;
753 		}
754 		btrfs_item_key_to_cpu(l, &key, slot);
755 
756 		if (key.objectid < device->devid)
757 			goto next;
758 
759 		if (key.objectid > device->devid)
760 			break;
761 
762 		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
763 			goto next;
764 
765 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
766 		extent_end = key.offset + btrfs_dev_extent_length(l,
767 								  dev_extent);
768 		if (key.offset <= start && extent_end > end) {
769 			*length = end - start + 1;
770 			break;
771 		} else if (key.offset <= start && extent_end > start)
772 			*length += extent_end - start;
773 		else if (key.offset > start && extent_end <= end)
774 			*length += extent_end - key.offset;
775 		else if (key.offset > start && key.offset <= end) {
776 			*length += end - key.offset + 1;
777 			break;
778 		} else if (key.offset > end)
779 			break;
780 
781 next:
782 		path->slots[0]++;
783 	}
784 	ret = 0;
785 out:
786 	btrfs_free_path(path);
787 	return ret;
788 }
789 
790 /*
791  * find_free_dev_extent - find free space in the specified device
792  * @trans:	transaction handler
793  * @device:	the device which we search the free space in
794  * @num_bytes:	the size of the free space that we need
795  * @start:	store the start of the free space.
796  * @len:	the size of the free space. that we find, or the size of the max
797  * 		free space if we don't find suitable free space
798  *
799  * this uses a pretty simple search, the expectation is that it is
800  * called very infrequently and that a given device has a small number
801  * of extents
802  *
803  * @start is used to store the start of the free space if we find. But if we
804  * don't find suitable free space, it will be used to store the start position
805  * of the max free space.
806  *
807  * @len is used to store the size of the free space that we find.
808  * But if we don't find suitable free space, it is used to store the size of
809  * the max free space.
810  */
811 int find_free_dev_extent(struct btrfs_trans_handle *trans,
812 			 struct btrfs_device *device, u64 num_bytes,
813 			 u64 *start, u64 *len)
814 {
815 	struct btrfs_key key;
816 	struct btrfs_root *root = device->dev_root;
817 	struct btrfs_dev_extent *dev_extent;
818 	struct btrfs_path *path;
819 	u64 hole_size;
820 	u64 max_hole_start;
821 	u64 max_hole_size;
822 	u64 extent_end;
823 	u64 search_start;
824 	u64 search_end = device->total_bytes;
825 	int ret;
826 	int slot;
827 	struct extent_buffer *l;
828 
829 	/* FIXME use last free of some kind */
830 
831 	/* we don't want to overwrite the superblock on the drive,
832 	 * so we make sure to start at an offset of at least 1MB
833 	 */
834 	search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
835 
836 	max_hole_start = search_start;
837 	max_hole_size = 0;
838 
839 	if (search_start >= search_end) {
840 		ret = -ENOSPC;
841 		goto error;
842 	}
843 
844 	path = btrfs_alloc_path();
845 	if (!path) {
846 		ret = -ENOMEM;
847 		goto error;
848 	}
849 	path->reada = 2;
850 
851 	key.objectid = device->devid;
852 	key.offset = search_start;
853 	key.type = BTRFS_DEV_EXTENT_KEY;
854 
855 	ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
856 	if (ret < 0)
857 		goto out;
858 	if (ret > 0) {
859 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
860 		if (ret < 0)
861 			goto out;
862 	}
863 
864 	while (1) {
865 		l = path->nodes[0];
866 		slot = path->slots[0];
867 		if (slot >= btrfs_header_nritems(l)) {
868 			ret = btrfs_next_leaf(root, path);
869 			if (ret == 0)
870 				continue;
871 			if (ret < 0)
872 				goto out;
873 
874 			break;
875 		}
876 		btrfs_item_key_to_cpu(l, &key, slot);
877 
878 		if (key.objectid < device->devid)
879 			goto next;
880 
881 		if (key.objectid > device->devid)
882 			break;
883 
884 		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
885 			goto next;
886 
887 		if (key.offset > search_start) {
888 			hole_size = key.offset - search_start;
889 
890 			if (hole_size > max_hole_size) {
891 				max_hole_start = search_start;
892 				max_hole_size = hole_size;
893 			}
894 
895 			/*
896 			 * If this free space is greater than which we need,
897 			 * it must be the max free space that we have found
898 			 * until now, so max_hole_start must point to the start
899 			 * of this free space and the length of this free space
900 			 * is stored in max_hole_size. Thus, we return
901 			 * max_hole_start and max_hole_size and go back to the
902 			 * caller.
903 			 */
904 			if (hole_size >= num_bytes) {
905 				ret = 0;
906 				goto out;
907 			}
908 		}
909 
910 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
911 		extent_end = key.offset + btrfs_dev_extent_length(l,
912 								  dev_extent);
913 		if (extent_end > search_start)
914 			search_start = extent_end;
915 next:
916 		path->slots[0]++;
917 		cond_resched();
918 	}
919 
920 	hole_size = search_end- search_start;
921 	if (hole_size > max_hole_size) {
922 		max_hole_start = search_start;
923 		max_hole_size = hole_size;
924 	}
925 
926 	/* See above. */
927 	if (hole_size < num_bytes)
928 		ret = -ENOSPC;
929 	else
930 		ret = 0;
931 
932 out:
933 	btrfs_free_path(path);
934 error:
935 	*start = max_hole_start;
936 	if (len)
937 		*len = max_hole_size;
938 	return ret;
939 }
940 
941 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
942 			  struct btrfs_device *device,
943 			  u64 start)
944 {
945 	int ret;
946 	struct btrfs_path *path;
947 	struct btrfs_root *root = device->dev_root;
948 	struct btrfs_key key;
949 	struct btrfs_key found_key;
950 	struct extent_buffer *leaf = NULL;
951 	struct btrfs_dev_extent *extent = NULL;
952 
953 	path = btrfs_alloc_path();
954 	if (!path)
955 		return -ENOMEM;
956 
957 	key.objectid = device->devid;
958 	key.offset = start;
959 	key.type = BTRFS_DEV_EXTENT_KEY;
960 
961 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
962 	if (ret > 0) {
963 		ret = btrfs_previous_item(root, path, key.objectid,
964 					  BTRFS_DEV_EXTENT_KEY);
965 		if (ret)
966 			goto out;
967 		leaf = path->nodes[0];
968 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
969 		extent = btrfs_item_ptr(leaf, path->slots[0],
970 					struct btrfs_dev_extent);
971 		BUG_ON(found_key.offset > start || found_key.offset +
972 		       btrfs_dev_extent_length(leaf, extent) < start);
973 	} else if (ret == 0) {
974 		leaf = path->nodes[0];
975 		extent = btrfs_item_ptr(leaf, path->slots[0],
976 					struct btrfs_dev_extent);
977 	}
978 	BUG_ON(ret);
979 
980 	if (device->bytes_used > 0)
981 		device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
982 	ret = btrfs_del_item(trans, root, path);
983 
984 out:
985 	btrfs_free_path(path);
986 	return ret;
987 }
988 
989 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
990 			   struct btrfs_device *device,
991 			   u64 chunk_tree, u64 chunk_objectid,
992 			   u64 chunk_offset, u64 start, u64 num_bytes)
993 {
994 	int ret;
995 	struct btrfs_path *path;
996 	struct btrfs_root *root = device->dev_root;
997 	struct btrfs_dev_extent *extent;
998 	struct extent_buffer *leaf;
999 	struct btrfs_key key;
1000 
1001 	WARN_ON(!device->in_fs_metadata);
1002 	path = btrfs_alloc_path();
1003 	if (!path)
1004 		return -ENOMEM;
1005 
1006 	key.objectid = device->devid;
1007 	key.offset = start;
1008 	key.type = BTRFS_DEV_EXTENT_KEY;
1009 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1010 				      sizeof(*extent));
1011 	BUG_ON(ret);
1012 
1013 	leaf = path->nodes[0];
1014 	extent = btrfs_item_ptr(leaf, path->slots[0],
1015 				struct btrfs_dev_extent);
1016 	btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1017 	btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1018 	btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1019 
1020 	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1021 		    (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1022 		    BTRFS_UUID_SIZE);
1023 
1024 	btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1025 	btrfs_mark_buffer_dirty(leaf);
1026 	btrfs_free_path(path);
1027 	return ret;
1028 }
1029 
1030 static noinline int find_next_chunk(struct btrfs_root *root,
1031 				    u64 objectid, u64 *offset)
1032 {
1033 	struct btrfs_path *path;
1034 	int ret;
1035 	struct btrfs_key key;
1036 	struct btrfs_chunk *chunk;
1037 	struct btrfs_key found_key;
1038 
1039 	path = btrfs_alloc_path();
1040 	BUG_ON(!path);
1041 
1042 	key.objectid = objectid;
1043 	key.offset = (u64)-1;
1044 	key.type = BTRFS_CHUNK_ITEM_KEY;
1045 
1046 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1047 	if (ret < 0)
1048 		goto error;
1049 
1050 	BUG_ON(ret == 0);
1051 
1052 	ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1053 	if (ret) {
1054 		*offset = 0;
1055 	} else {
1056 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1057 				      path->slots[0]);
1058 		if (found_key.objectid != objectid)
1059 			*offset = 0;
1060 		else {
1061 			chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1062 					       struct btrfs_chunk);
1063 			*offset = found_key.offset +
1064 				btrfs_chunk_length(path->nodes[0], chunk);
1065 		}
1066 	}
1067 	ret = 0;
1068 error:
1069 	btrfs_free_path(path);
1070 	return ret;
1071 }
1072 
1073 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1074 {
1075 	int ret;
1076 	struct btrfs_key key;
1077 	struct btrfs_key found_key;
1078 	struct btrfs_path *path;
1079 
1080 	root = root->fs_info->chunk_root;
1081 
1082 	path = btrfs_alloc_path();
1083 	if (!path)
1084 		return -ENOMEM;
1085 
1086 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1087 	key.type = BTRFS_DEV_ITEM_KEY;
1088 	key.offset = (u64)-1;
1089 
1090 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1091 	if (ret < 0)
1092 		goto error;
1093 
1094 	BUG_ON(ret == 0);
1095 
1096 	ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1097 				  BTRFS_DEV_ITEM_KEY);
1098 	if (ret) {
1099 		*objectid = 1;
1100 	} else {
1101 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1102 				      path->slots[0]);
1103 		*objectid = found_key.offset + 1;
1104 	}
1105 	ret = 0;
1106 error:
1107 	btrfs_free_path(path);
1108 	return ret;
1109 }
1110 
1111 /*
1112  * the device information is stored in the chunk root
1113  * the btrfs_device struct should be fully filled in
1114  */
1115 int btrfs_add_device(struct btrfs_trans_handle *trans,
1116 		     struct btrfs_root *root,
1117 		     struct btrfs_device *device)
1118 {
1119 	int ret;
1120 	struct btrfs_path *path;
1121 	struct btrfs_dev_item *dev_item;
1122 	struct extent_buffer *leaf;
1123 	struct btrfs_key key;
1124 	unsigned long ptr;
1125 
1126 	root = root->fs_info->chunk_root;
1127 
1128 	path = btrfs_alloc_path();
1129 	if (!path)
1130 		return -ENOMEM;
1131 
1132 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1133 	key.type = BTRFS_DEV_ITEM_KEY;
1134 	key.offset = device->devid;
1135 
1136 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1137 				      sizeof(*dev_item));
1138 	if (ret)
1139 		goto out;
1140 
1141 	leaf = path->nodes[0];
1142 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1143 
1144 	btrfs_set_device_id(leaf, dev_item, device->devid);
1145 	btrfs_set_device_generation(leaf, dev_item, 0);
1146 	btrfs_set_device_type(leaf, dev_item, device->type);
1147 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1148 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1149 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1150 	btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1151 	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1152 	btrfs_set_device_group(leaf, dev_item, 0);
1153 	btrfs_set_device_seek_speed(leaf, dev_item, 0);
1154 	btrfs_set_device_bandwidth(leaf, dev_item, 0);
1155 	btrfs_set_device_start_offset(leaf, dev_item, 0);
1156 
1157 	ptr = (unsigned long)btrfs_device_uuid(dev_item);
1158 	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1159 	ptr = (unsigned long)btrfs_device_fsid(dev_item);
1160 	write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1161 	btrfs_mark_buffer_dirty(leaf);
1162 
1163 	ret = 0;
1164 out:
1165 	btrfs_free_path(path);
1166 	return ret;
1167 }
1168 
1169 static int btrfs_rm_dev_item(struct btrfs_root *root,
1170 			     struct btrfs_device *device)
1171 {
1172 	int ret;
1173 	struct btrfs_path *path;
1174 	struct btrfs_key key;
1175 	struct btrfs_trans_handle *trans;
1176 
1177 	root = root->fs_info->chunk_root;
1178 
1179 	path = btrfs_alloc_path();
1180 	if (!path)
1181 		return -ENOMEM;
1182 
1183 	trans = btrfs_start_transaction(root, 0);
1184 	if (IS_ERR(trans)) {
1185 		btrfs_free_path(path);
1186 		return PTR_ERR(trans);
1187 	}
1188 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1189 	key.type = BTRFS_DEV_ITEM_KEY;
1190 	key.offset = device->devid;
1191 	lock_chunks(root);
1192 
1193 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1194 	if (ret < 0)
1195 		goto out;
1196 
1197 	if (ret > 0) {
1198 		ret = -ENOENT;
1199 		goto out;
1200 	}
1201 
1202 	ret = btrfs_del_item(trans, root, path);
1203 	if (ret)
1204 		goto out;
1205 out:
1206 	btrfs_free_path(path);
1207 	unlock_chunks(root);
1208 	btrfs_commit_transaction(trans, root);
1209 	return ret;
1210 }
1211 
1212 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1213 {
1214 	struct btrfs_device *device;
1215 	struct btrfs_device *next_device;
1216 	struct block_device *bdev;
1217 	struct buffer_head *bh = NULL;
1218 	struct btrfs_super_block *disk_super;
1219 	struct btrfs_fs_devices *cur_devices;
1220 	u64 all_avail;
1221 	u64 devid;
1222 	u64 num_devices;
1223 	u8 *dev_uuid;
1224 	int ret = 0;
1225 	bool clear_super = false;
1226 
1227 	mutex_lock(&uuid_mutex);
1228 	mutex_lock(&root->fs_info->volume_mutex);
1229 
1230 	all_avail = root->fs_info->avail_data_alloc_bits |
1231 		root->fs_info->avail_system_alloc_bits |
1232 		root->fs_info->avail_metadata_alloc_bits;
1233 
1234 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1235 	    root->fs_info->fs_devices->num_devices <= 4) {
1236 		printk(KERN_ERR "btrfs: unable to go below four devices "
1237 		       "on raid10\n");
1238 		ret = -EINVAL;
1239 		goto out;
1240 	}
1241 
1242 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1243 	    root->fs_info->fs_devices->num_devices <= 2) {
1244 		printk(KERN_ERR "btrfs: unable to go below two "
1245 		       "devices on raid1\n");
1246 		ret = -EINVAL;
1247 		goto out;
1248 	}
1249 
1250 	if (strcmp(device_path, "missing") == 0) {
1251 		struct list_head *devices;
1252 		struct btrfs_device *tmp;
1253 
1254 		device = NULL;
1255 		devices = &root->fs_info->fs_devices->devices;
1256 		/*
1257 		 * It is safe to read the devices since the volume_mutex
1258 		 * is held.
1259 		 */
1260 		list_for_each_entry(tmp, devices, dev_list) {
1261 			if (tmp->in_fs_metadata && !tmp->bdev) {
1262 				device = tmp;
1263 				break;
1264 			}
1265 		}
1266 		bdev = NULL;
1267 		bh = NULL;
1268 		disk_super = NULL;
1269 		if (!device) {
1270 			printk(KERN_ERR "btrfs: no missing devices found to "
1271 			       "remove\n");
1272 			goto out;
1273 		}
1274 	} else {
1275 		bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1276 					  root->fs_info->bdev_holder);
1277 		if (IS_ERR(bdev)) {
1278 			ret = PTR_ERR(bdev);
1279 			goto out;
1280 		}
1281 
1282 		set_blocksize(bdev, 4096);
1283 		bh = btrfs_read_dev_super(bdev);
1284 		if (!bh) {
1285 			ret = -EINVAL;
1286 			goto error_close;
1287 		}
1288 		disk_super = (struct btrfs_super_block *)bh->b_data;
1289 		devid = btrfs_stack_device_id(&disk_super->dev_item);
1290 		dev_uuid = disk_super->dev_item.uuid;
1291 		device = btrfs_find_device(root, devid, dev_uuid,
1292 					   disk_super->fsid);
1293 		if (!device) {
1294 			ret = -ENOENT;
1295 			goto error_brelse;
1296 		}
1297 	}
1298 
1299 	if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1300 		printk(KERN_ERR "btrfs: unable to remove the only writeable "
1301 		       "device\n");
1302 		ret = -EINVAL;
1303 		goto error_brelse;
1304 	}
1305 
1306 	if (device->writeable) {
1307 		lock_chunks(root);
1308 		list_del_init(&device->dev_alloc_list);
1309 		unlock_chunks(root);
1310 		root->fs_info->fs_devices->rw_devices--;
1311 		clear_super = true;
1312 	}
1313 
1314 	ret = btrfs_shrink_device(device, 0);
1315 	if (ret)
1316 		goto error_undo;
1317 
1318 	ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1319 	if (ret)
1320 		goto error_undo;
1321 
1322 	device->in_fs_metadata = 0;
1323 	btrfs_scrub_cancel_dev(root, device);
1324 
1325 	/*
1326 	 * the device list mutex makes sure that we don't change
1327 	 * the device list while someone else is writing out all
1328 	 * the device supers.
1329 	 */
1330 
1331 	cur_devices = device->fs_devices;
1332 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1333 	list_del_rcu(&device->dev_list);
1334 
1335 	device->fs_devices->num_devices--;
1336 
1337 	if (device->missing)
1338 		root->fs_info->fs_devices->missing_devices--;
1339 
1340 	next_device = list_entry(root->fs_info->fs_devices->devices.next,
1341 				 struct btrfs_device, dev_list);
1342 	if (device->bdev == root->fs_info->sb->s_bdev)
1343 		root->fs_info->sb->s_bdev = next_device->bdev;
1344 	if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1345 		root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1346 
1347 	if (device->bdev)
1348 		device->fs_devices->open_devices--;
1349 
1350 	call_rcu(&device->rcu, free_device);
1351 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1352 
1353 	num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1354 	btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1355 
1356 	if (cur_devices->open_devices == 0) {
1357 		struct btrfs_fs_devices *fs_devices;
1358 		fs_devices = root->fs_info->fs_devices;
1359 		while (fs_devices) {
1360 			if (fs_devices->seed == cur_devices)
1361 				break;
1362 			fs_devices = fs_devices->seed;
1363 		}
1364 		fs_devices->seed = cur_devices->seed;
1365 		cur_devices->seed = NULL;
1366 		lock_chunks(root);
1367 		__btrfs_close_devices(cur_devices);
1368 		unlock_chunks(root);
1369 		free_fs_devices(cur_devices);
1370 	}
1371 
1372 	/*
1373 	 * at this point, the device is zero sized.  We want to
1374 	 * remove it from the devices list and zero out the old super
1375 	 */
1376 	if (clear_super) {
1377 		/* make sure this device isn't detected as part of
1378 		 * the FS anymore
1379 		 */
1380 		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1381 		set_buffer_dirty(bh);
1382 		sync_dirty_buffer(bh);
1383 	}
1384 
1385 	ret = 0;
1386 
1387 error_brelse:
1388 	brelse(bh);
1389 error_close:
1390 	if (bdev)
1391 		blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1392 out:
1393 	mutex_unlock(&root->fs_info->volume_mutex);
1394 	mutex_unlock(&uuid_mutex);
1395 	return ret;
1396 error_undo:
1397 	if (device->writeable) {
1398 		lock_chunks(root);
1399 		list_add(&device->dev_alloc_list,
1400 			 &root->fs_info->fs_devices->alloc_list);
1401 		unlock_chunks(root);
1402 		root->fs_info->fs_devices->rw_devices++;
1403 	}
1404 	goto error_brelse;
1405 }
1406 
1407 /*
1408  * does all the dirty work required for changing file system's UUID.
1409  */
1410 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1411 				struct btrfs_root *root)
1412 {
1413 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1414 	struct btrfs_fs_devices *old_devices;
1415 	struct btrfs_fs_devices *seed_devices;
1416 	struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1417 	struct btrfs_device *device;
1418 	u64 super_flags;
1419 
1420 	BUG_ON(!mutex_is_locked(&uuid_mutex));
1421 	if (!fs_devices->seeding)
1422 		return -EINVAL;
1423 
1424 	seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1425 	if (!seed_devices)
1426 		return -ENOMEM;
1427 
1428 	old_devices = clone_fs_devices(fs_devices);
1429 	if (IS_ERR(old_devices)) {
1430 		kfree(seed_devices);
1431 		return PTR_ERR(old_devices);
1432 	}
1433 
1434 	list_add(&old_devices->list, &fs_uuids);
1435 
1436 	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1437 	seed_devices->opened = 1;
1438 	INIT_LIST_HEAD(&seed_devices->devices);
1439 	INIT_LIST_HEAD(&seed_devices->alloc_list);
1440 	mutex_init(&seed_devices->device_list_mutex);
1441 
1442 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1443 	list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1444 			      synchronize_rcu);
1445 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1446 
1447 	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1448 	list_for_each_entry(device, &seed_devices->devices, dev_list) {
1449 		device->fs_devices = seed_devices;
1450 	}
1451 
1452 	fs_devices->seeding = 0;
1453 	fs_devices->num_devices = 0;
1454 	fs_devices->open_devices = 0;
1455 	fs_devices->seed = seed_devices;
1456 
1457 	generate_random_uuid(fs_devices->fsid);
1458 	memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1459 	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1460 	super_flags = btrfs_super_flags(disk_super) &
1461 		      ~BTRFS_SUPER_FLAG_SEEDING;
1462 	btrfs_set_super_flags(disk_super, super_flags);
1463 
1464 	return 0;
1465 }
1466 
1467 /*
1468  * strore the expected generation for seed devices in device items.
1469  */
1470 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1471 			       struct btrfs_root *root)
1472 {
1473 	struct btrfs_path *path;
1474 	struct extent_buffer *leaf;
1475 	struct btrfs_dev_item *dev_item;
1476 	struct btrfs_device *device;
1477 	struct btrfs_key key;
1478 	u8 fs_uuid[BTRFS_UUID_SIZE];
1479 	u8 dev_uuid[BTRFS_UUID_SIZE];
1480 	u64 devid;
1481 	int ret;
1482 
1483 	path = btrfs_alloc_path();
1484 	if (!path)
1485 		return -ENOMEM;
1486 
1487 	root = root->fs_info->chunk_root;
1488 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1489 	key.offset = 0;
1490 	key.type = BTRFS_DEV_ITEM_KEY;
1491 
1492 	while (1) {
1493 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1494 		if (ret < 0)
1495 			goto error;
1496 
1497 		leaf = path->nodes[0];
1498 next_slot:
1499 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1500 			ret = btrfs_next_leaf(root, path);
1501 			if (ret > 0)
1502 				break;
1503 			if (ret < 0)
1504 				goto error;
1505 			leaf = path->nodes[0];
1506 			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1507 			btrfs_release_path(path);
1508 			continue;
1509 		}
1510 
1511 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1512 		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1513 		    key.type != BTRFS_DEV_ITEM_KEY)
1514 			break;
1515 
1516 		dev_item = btrfs_item_ptr(leaf, path->slots[0],
1517 					  struct btrfs_dev_item);
1518 		devid = btrfs_device_id(leaf, dev_item);
1519 		read_extent_buffer(leaf, dev_uuid,
1520 				   (unsigned long)btrfs_device_uuid(dev_item),
1521 				   BTRFS_UUID_SIZE);
1522 		read_extent_buffer(leaf, fs_uuid,
1523 				   (unsigned long)btrfs_device_fsid(dev_item),
1524 				   BTRFS_UUID_SIZE);
1525 		device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1526 		BUG_ON(!device);
1527 
1528 		if (device->fs_devices->seeding) {
1529 			btrfs_set_device_generation(leaf, dev_item,
1530 						    device->generation);
1531 			btrfs_mark_buffer_dirty(leaf);
1532 		}
1533 
1534 		path->slots[0]++;
1535 		goto next_slot;
1536 	}
1537 	ret = 0;
1538 error:
1539 	btrfs_free_path(path);
1540 	return ret;
1541 }
1542 
1543 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1544 {
1545 	struct btrfs_trans_handle *trans;
1546 	struct btrfs_device *device;
1547 	struct block_device *bdev;
1548 	struct list_head *devices;
1549 	struct super_block *sb = root->fs_info->sb;
1550 	u64 total_bytes;
1551 	int seeding_dev = 0;
1552 	int ret = 0;
1553 
1554 	if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1555 		return -EINVAL;
1556 
1557 	bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1558 				  root->fs_info->bdev_holder);
1559 	if (IS_ERR(bdev))
1560 		return PTR_ERR(bdev);
1561 
1562 	if (root->fs_info->fs_devices->seeding) {
1563 		seeding_dev = 1;
1564 		down_write(&sb->s_umount);
1565 		mutex_lock(&uuid_mutex);
1566 	}
1567 
1568 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
1569 	mutex_lock(&root->fs_info->volume_mutex);
1570 
1571 	devices = &root->fs_info->fs_devices->devices;
1572 	/*
1573 	 * we have the volume lock, so we don't need the extra
1574 	 * device list mutex while reading the list here.
1575 	 */
1576 	list_for_each_entry(device, devices, dev_list) {
1577 		if (device->bdev == bdev) {
1578 			ret = -EEXIST;
1579 			goto error;
1580 		}
1581 	}
1582 
1583 	device = kzalloc(sizeof(*device), GFP_NOFS);
1584 	if (!device) {
1585 		/* we can safely leave the fs_devices entry around */
1586 		ret = -ENOMEM;
1587 		goto error;
1588 	}
1589 
1590 	device->name = kstrdup(device_path, GFP_NOFS);
1591 	if (!device->name) {
1592 		kfree(device);
1593 		ret = -ENOMEM;
1594 		goto error;
1595 	}
1596 
1597 	ret = find_next_devid(root, &device->devid);
1598 	if (ret) {
1599 		kfree(device->name);
1600 		kfree(device);
1601 		goto error;
1602 	}
1603 
1604 	trans = btrfs_start_transaction(root, 0);
1605 	if (IS_ERR(trans)) {
1606 		kfree(device->name);
1607 		kfree(device);
1608 		ret = PTR_ERR(trans);
1609 		goto error;
1610 	}
1611 
1612 	lock_chunks(root);
1613 
1614 	device->writeable = 1;
1615 	device->work.func = pending_bios_fn;
1616 	generate_random_uuid(device->uuid);
1617 	spin_lock_init(&device->io_lock);
1618 	device->generation = trans->transid;
1619 	device->io_width = root->sectorsize;
1620 	device->io_align = root->sectorsize;
1621 	device->sector_size = root->sectorsize;
1622 	device->total_bytes = i_size_read(bdev->bd_inode);
1623 	device->disk_total_bytes = device->total_bytes;
1624 	device->dev_root = root->fs_info->dev_root;
1625 	device->bdev = bdev;
1626 	device->in_fs_metadata = 1;
1627 	device->mode = FMODE_EXCL;
1628 	set_blocksize(device->bdev, 4096);
1629 
1630 	if (seeding_dev) {
1631 		sb->s_flags &= ~MS_RDONLY;
1632 		ret = btrfs_prepare_sprout(trans, root);
1633 		BUG_ON(ret);
1634 	}
1635 
1636 	device->fs_devices = root->fs_info->fs_devices;
1637 
1638 	/*
1639 	 * we don't want write_supers to jump in here with our device
1640 	 * half setup
1641 	 */
1642 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1643 	list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1644 	list_add(&device->dev_alloc_list,
1645 		 &root->fs_info->fs_devices->alloc_list);
1646 	root->fs_info->fs_devices->num_devices++;
1647 	root->fs_info->fs_devices->open_devices++;
1648 	root->fs_info->fs_devices->rw_devices++;
1649 	root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1650 
1651 	if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1652 		root->fs_info->fs_devices->rotating = 1;
1653 
1654 	total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1655 	btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1656 				    total_bytes + device->total_bytes);
1657 
1658 	total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1659 	btrfs_set_super_num_devices(&root->fs_info->super_copy,
1660 				    total_bytes + 1);
1661 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1662 
1663 	if (seeding_dev) {
1664 		ret = init_first_rw_device(trans, root, device);
1665 		BUG_ON(ret);
1666 		ret = btrfs_finish_sprout(trans, root);
1667 		BUG_ON(ret);
1668 	} else {
1669 		ret = btrfs_add_device(trans, root, device);
1670 	}
1671 
1672 	/*
1673 	 * we've got more storage, clear any full flags on the space
1674 	 * infos
1675 	 */
1676 	btrfs_clear_space_info_full(root->fs_info);
1677 
1678 	unlock_chunks(root);
1679 	btrfs_commit_transaction(trans, root);
1680 
1681 	if (seeding_dev) {
1682 		mutex_unlock(&uuid_mutex);
1683 		up_write(&sb->s_umount);
1684 
1685 		ret = btrfs_relocate_sys_chunks(root);
1686 		BUG_ON(ret);
1687 	}
1688 out:
1689 	mutex_unlock(&root->fs_info->volume_mutex);
1690 	return ret;
1691 error:
1692 	blkdev_put(bdev, FMODE_EXCL);
1693 	if (seeding_dev) {
1694 		mutex_unlock(&uuid_mutex);
1695 		up_write(&sb->s_umount);
1696 	}
1697 	goto out;
1698 }
1699 
1700 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1701 					struct btrfs_device *device)
1702 {
1703 	int ret;
1704 	struct btrfs_path *path;
1705 	struct btrfs_root *root;
1706 	struct btrfs_dev_item *dev_item;
1707 	struct extent_buffer *leaf;
1708 	struct btrfs_key key;
1709 
1710 	root = device->dev_root->fs_info->chunk_root;
1711 
1712 	path = btrfs_alloc_path();
1713 	if (!path)
1714 		return -ENOMEM;
1715 
1716 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1717 	key.type = BTRFS_DEV_ITEM_KEY;
1718 	key.offset = device->devid;
1719 
1720 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1721 	if (ret < 0)
1722 		goto out;
1723 
1724 	if (ret > 0) {
1725 		ret = -ENOENT;
1726 		goto out;
1727 	}
1728 
1729 	leaf = path->nodes[0];
1730 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1731 
1732 	btrfs_set_device_id(leaf, dev_item, device->devid);
1733 	btrfs_set_device_type(leaf, dev_item, device->type);
1734 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1735 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1736 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1737 	btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1738 	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1739 	btrfs_mark_buffer_dirty(leaf);
1740 
1741 out:
1742 	btrfs_free_path(path);
1743 	return ret;
1744 }
1745 
1746 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1747 		      struct btrfs_device *device, u64 new_size)
1748 {
1749 	struct btrfs_super_block *super_copy =
1750 		&device->dev_root->fs_info->super_copy;
1751 	u64 old_total = btrfs_super_total_bytes(super_copy);
1752 	u64 diff = new_size - device->total_bytes;
1753 
1754 	if (!device->writeable)
1755 		return -EACCES;
1756 	if (new_size <= device->total_bytes)
1757 		return -EINVAL;
1758 
1759 	btrfs_set_super_total_bytes(super_copy, old_total + diff);
1760 	device->fs_devices->total_rw_bytes += diff;
1761 
1762 	device->total_bytes = new_size;
1763 	device->disk_total_bytes = new_size;
1764 	btrfs_clear_space_info_full(device->dev_root->fs_info);
1765 
1766 	return btrfs_update_device(trans, device);
1767 }
1768 
1769 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1770 		      struct btrfs_device *device, u64 new_size)
1771 {
1772 	int ret;
1773 	lock_chunks(device->dev_root);
1774 	ret = __btrfs_grow_device(trans, device, new_size);
1775 	unlock_chunks(device->dev_root);
1776 	return ret;
1777 }
1778 
1779 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1780 			    struct btrfs_root *root,
1781 			    u64 chunk_tree, u64 chunk_objectid,
1782 			    u64 chunk_offset)
1783 {
1784 	int ret;
1785 	struct btrfs_path *path;
1786 	struct btrfs_key key;
1787 
1788 	root = root->fs_info->chunk_root;
1789 	path = btrfs_alloc_path();
1790 	if (!path)
1791 		return -ENOMEM;
1792 
1793 	key.objectid = chunk_objectid;
1794 	key.offset = chunk_offset;
1795 	key.type = BTRFS_CHUNK_ITEM_KEY;
1796 
1797 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1798 	BUG_ON(ret);
1799 
1800 	ret = btrfs_del_item(trans, root, path);
1801 
1802 	btrfs_free_path(path);
1803 	return ret;
1804 }
1805 
1806 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1807 			chunk_offset)
1808 {
1809 	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1810 	struct btrfs_disk_key *disk_key;
1811 	struct btrfs_chunk *chunk;
1812 	u8 *ptr;
1813 	int ret = 0;
1814 	u32 num_stripes;
1815 	u32 array_size;
1816 	u32 len = 0;
1817 	u32 cur;
1818 	struct btrfs_key key;
1819 
1820 	array_size = btrfs_super_sys_array_size(super_copy);
1821 
1822 	ptr = super_copy->sys_chunk_array;
1823 	cur = 0;
1824 
1825 	while (cur < array_size) {
1826 		disk_key = (struct btrfs_disk_key *)ptr;
1827 		btrfs_disk_key_to_cpu(&key, disk_key);
1828 
1829 		len = sizeof(*disk_key);
1830 
1831 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1832 			chunk = (struct btrfs_chunk *)(ptr + len);
1833 			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1834 			len += btrfs_chunk_item_size(num_stripes);
1835 		} else {
1836 			ret = -EIO;
1837 			break;
1838 		}
1839 		if (key.objectid == chunk_objectid &&
1840 		    key.offset == chunk_offset) {
1841 			memmove(ptr, ptr + len, array_size - (cur + len));
1842 			array_size -= len;
1843 			btrfs_set_super_sys_array_size(super_copy, array_size);
1844 		} else {
1845 			ptr += len;
1846 			cur += len;
1847 		}
1848 	}
1849 	return ret;
1850 }
1851 
1852 static int btrfs_relocate_chunk(struct btrfs_root *root,
1853 			 u64 chunk_tree, u64 chunk_objectid,
1854 			 u64 chunk_offset)
1855 {
1856 	struct extent_map_tree *em_tree;
1857 	struct btrfs_root *extent_root;
1858 	struct btrfs_trans_handle *trans;
1859 	struct extent_map *em;
1860 	struct map_lookup *map;
1861 	int ret;
1862 	int i;
1863 
1864 	root = root->fs_info->chunk_root;
1865 	extent_root = root->fs_info->extent_root;
1866 	em_tree = &root->fs_info->mapping_tree.map_tree;
1867 
1868 	ret = btrfs_can_relocate(extent_root, chunk_offset);
1869 	if (ret)
1870 		return -ENOSPC;
1871 
1872 	/* step one, relocate all the extents inside this chunk */
1873 	ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1874 	if (ret)
1875 		return ret;
1876 
1877 	trans = btrfs_start_transaction(root, 0);
1878 	BUG_ON(IS_ERR(trans));
1879 
1880 	lock_chunks(root);
1881 
1882 	/*
1883 	 * step two, delete the device extents and the
1884 	 * chunk tree entries
1885 	 */
1886 	read_lock(&em_tree->lock);
1887 	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1888 	read_unlock(&em_tree->lock);
1889 
1890 	BUG_ON(em->start > chunk_offset ||
1891 	       em->start + em->len < chunk_offset);
1892 	map = (struct map_lookup *)em->bdev;
1893 
1894 	for (i = 0; i < map->num_stripes; i++) {
1895 		ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1896 					    map->stripes[i].physical);
1897 		BUG_ON(ret);
1898 
1899 		if (map->stripes[i].dev) {
1900 			ret = btrfs_update_device(trans, map->stripes[i].dev);
1901 			BUG_ON(ret);
1902 		}
1903 	}
1904 	ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1905 			       chunk_offset);
1906 
1907 	BUG_ON(ret);
1908 
1909 	trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1910 
1911 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1912 		ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1913 		BUG_ON(ret);
1914 	}
1915 
1916 	ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1917 	BUG_ON(ret);
1918 
1919 	write_lock(&em_tree->lock);
1920 	remove_extent_mapping(em_tree, em);
1921 	write_unlock(&em_tree->lock);
1922 
1923 	kfree(map);
1924 	em->bdev = NULL;
1925 
1926 	/* once for the tree */
1927 	free_extent_map(em);
1928 	/* once for us */
1929 	free_extent_map(em);
1930 
1931 	unlock_chunks(root);
1932 	btrfs_end_transaction(trans, root);
1933 	return 0;
1934 }
1935 
1936 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1937 {
1938 	struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1939 	struct btrfs_path *path;
1940 	struct extent_buffer *leaf;
1941 	struct btrfs_chunk *chunk;
1942 	struct btrfs_key key;
1943 	struct btrfs_key found_key;
1944 	u64 chunk_tree = chunk_root->root_key.objectid;
1945 	u64 chunk_type;
1946 	bool retried = false;
1947 	int failed = 0;
1948 	int ret;
1949 
1950 	path = btrfs_alloc_path();
1951 	if (!path)
1952 		return -ENOMEM;
1953 
1954 again:
1955 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1956 	key.offset = (u64)-1;
1957 	key.type = BTRFS_CHUNK_ITEM_KEY;
1958 
1959 	while (1) {
1960 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1961 		if (ret < 0)
1962 			goto error;
1963 		BUG_ON(ret == 0);
1964 
1965 		ret = btrfs_previous_item(chunk_root, path, key.objectid,
1966 					  key.type);
1967 		if (ret < 0)
1968 			goto error;
1969 		if (ret > 0)
1970 			break;
1971 
1972 		leaf = path->nodes[0];
1973 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1974 
1975 		chunk = btrfs_item_ptr(leaf, path->slots[0],
1976 				       struct btrfs_chunk);
1977 		chunk_type = btrfs_chunk_type(leaf, chunk);
1978 		btrfs_release_path(path);
1979 
1980 		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1981 			ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1982 						   found_key.objectid,
1983 						   found_key.offset);
1984 			if (ret == -ENOSPC)
1985 				failed++;
1986 			else if (ret)
1987 				BUG();
1988 		}
1989 
1990 		if (found_key.offset == 0)
1991 			break;
1992 		key.offset = found_key.offset - 1;
1993 	}
1994 	ret = 0;
1995 	if (failed && !retried) {
1996 		failed = 0;
1997 		retried = true;
1998 		goto again;
1999 	} else if (failed && retried) {
2000 		WARN_ON(1);
2001 		ret = -ENOSPC;
2002 	}
2003 error:
2004 	btrfs_free_path(path);
2005 	return ret;
2006 }
2007 
2008 static u64 div_factor(u64 num, int factor)
2009 {
2010 	if (factor == 10)
2011 		return num;
2012 	num *= factor;
2013 	do_div(num, 10);
2014 	return num;
2015 }
2016 
2017 int btrfs_balance(struct btrfs_root *dev_root)
2018 {
2019 	int ret;
2020 	struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2021 	struct btrfs_device *device;
2022 	u64 old_size;
2023 	u64 size_to_free;
2024 	struct btrfs_path *path;
2025 	struct btrfs_key key;
2026 	struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2027 	struct btrfs_trans_handle *trans;
2028 	struct btrfs_key found_key;
2029 
2030 	if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2031 		return -EROFS;
2032 
2033 	if (!capable(CAP_SYS_ADMIN))
2034 		return -EPERM;
2035 
2036 	mutex_lock(&dev_root->fs_info->volume_mutex);
2037 	dev_root = dev_root->fs_info->dev_root;
2038 
2039 	/* step one make some room on all the devices */
2040 	list_for_each_entry(device, devices, dev_list) {
2041 		old_size = device->total_bytes;
2042 		size_to_free = div_factor(old_size, 1);
2043 		size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2044 		if (!device->writeable ||
2045 		    device->total_bytes - device->bytes_used > size_to_free)
2046 			continue;
2047 
2048 		ret = btrfs_shrink_device(device, old_size - size_to_free);
2049 		if (ret == -ENOSPC)
2050 			break;
2051 		BUG_ON(ret);
2052 
2053 		trans = btrfs_start_transaction(dev_root, 0);
2054 		BUG_ON(IS_ERR(trans));
2055 
2056 		ret = btrfs_grow_device(trans, device, old_size);
2057 		BUG_ON(ret);
2058 
2059 		btrfs_end_transaction(trans, dev_root);
2060 	}
2061 
2062 	/* step two, relocate all the chunks */
2063 	path = btrfs_alloc_path();
2064 	BUG_ON(!path);
2065 
2066 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2067 	key.offset = (u64)-1;
2068 	key.type = BTRFS_CHUNK_ITEM_KEY;
2069 
2070 	while (1) {
2071 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2072 		if (ret < 0)
2073 			goto error;
2074 
2075 		/*
2076 		 * this shouldn't happen, it means the last relocate
2077 		 * failed
2078 		 */
2079 		if (ret == 0)
2080 			break;
2081 
2082 		ret = btrfs_previous_item(chunk_root, path, 0,
2083 					  BTRFS_CHUNK_ITEM_KEY);
2084 		if (ret)
2085 			break;
2086 
2087 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2088 				      path->slots[0]);
2089 		if (found_key.objectid != key.objectid)
2090 			break;
2091 
2092 		/* chunk zero is special */
2093 		if (found_key.offset == 0)
2094 			break;
2095 
2096 		btrfs_release_path(path);
2097 		ret = btrfs_relocate_chunk(chunk_root,
2098 					   chunk_root->root_key.objectid,
2099 					   found_key.objectid,
2100 					   found_key.offset);
2101 		if (ret && ret != -ENOSPC)
2102 			goto error;
2103 		key.offset = found_key.offset - 1;
2104 	}
2105 	ret = 0;
2106 error:
2107 	btrfs_free_path(path);
2108 	mutex_unlock(&dev_root->fs_info->volume_mutex);
2109 	return ret;
2110 }
2111 
2112 /*
2113  * shrinking a device means finding all of the device extents past
2114  * the new size, and then following the back refs to the chunks.
2115  * The chunk relocation code actually frees the device extent
2116  */
2117 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2118 {
2119 	struct btrfs_trans_handle *trans;
2120 	struct btrfs_root *root = device->dev_root;
2121 	struct btrfs_dev_extent *dev_extent = NULL;
2122 	struct btrfs_path *path;
2123 	u64 length;
2124 	u64 chunk_tree;
2125 	u64 chunk_objectid;
2126 	u64 chunk_offset;
2127 	int ret;
2128 	int slot;
2129 	int failed = 0;
2130 	bool retried = false;
2131 	struct extent_buffer *l;
2132 	struct btrfs_key key;
2133 	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2134 	u64 old_total = btrfs_super_total_bytes(super_copy);
2135 	u64 old_size = device->total_bytes;
2136 	u64 diff = device->total_bytes - new_size;
2137 
2138 	if (new_size >= device->total_bytes)
2139 		return -EINVAL;
2140 
2141 	path = btrfs_alloc_path();
2142 	if (!path)
2143 		return -ENOMEM;
2144 
2145 	path->reada = 2;
2146 
2147 	lock_chunks(root);
2148 
2149 	device->total_bytes = new_size;
2150 	if (device->writeable)
2151 		device->fs_devices->total_rw_bytes -= diff;
2152 	unlock_chunks(root);
2153 
2154 again:
2155 	key.objectid = device->devid;
2156 	key.offset = (u64)-1;
2157 	key.type = BTRFS_DEV_EXTENT_KEY;
2158 
2159 	while (1) {
2160 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2161 		if (ret < 0)
2162 			goto done;
2163 
2164 		ret = btrfs_previous_item(root, path, 0, key.type);
2165 		if (ret < 0)
2166 			goto done;
2167 		if (ret) {
2168 			ret = 0;
2169 			btrfs_release_path(path);
2170 			break;
2171 		}
2172 
2173 		l = path->nodes[0];
2174 		slot = path->slots[0];
2175 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2176 
2177 		if (key.objectid != device->devid) {
2178 			btrfs_release_path(path);
2179 			break;
2180 		}
2181 
2182 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2183 		length = btrfs_dev_extent_length(l, dev_extent);
2184 
2185 		if (key.offset + length <= new_size) {
2186 			btrfs_release_path(path);
2187 			break;
2188 		}
2189 
2190 		chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2191 		chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2192 		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2193 		btrfs_release_path(path);
2194 
2195 		ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2196 					   chunk_offset);
2197 		if (ret && ret != -ENOSPC)
2198 			goto done;
2199 		if (ret == -ENOSPC)
2200 			failed++;
2201 		key.offset -= 1;
2202 	}
2203 
2204 	if (failed && !retried) {
2205 		failed = 0;
2206 		retried = true;
2207 		goto again;
2208 	} else if (failed && retried) {
2209 		ret = -ENOSPC;
2210 		lock_chunks(root);
2211 
2212 		device->total_bytes = old_size;
2213 		if (device->writeable)
2214 			device->fs_devices->total_rw_bytes += diff;
2215 		unlock_chunks(root);
2216 		goto done;
2217 	}
2218 
2219 	/* Shrinking succeeded, else we would be at "done". */
2220 	trans = btrfs_start_transaction(root, 0);
2221 	if (IS_ERR(trans)) {
2222 		ret = PTR_ERR(trans);
2223 		goto done;
2224 	}
2225 
2226 	lock_chunks(root);
2227 
2228 	device->disk_total_bytes = new_size;
2229 	/* Now btrfs_update_device() will change the on-disk size. */
2230 	ret = btrfs_update_device(trans, device);
2231 	if (ret) {
2232 		unlock_chunks(root);
2233 		btrfs_end_transaction(trans, root);
2234 		goto done;
2235 	}
2236 	WARN_ON(diff > old_total);
2237 	btrfs_set_super_total_bytes(super_copy, old_total - diff);
2238 	unlock_chunks(root);
2239 	btrfs_end_transaction(trans, root);
2240 done:
2241 	btrfs_free_path(path);
2242 	return ret;
2243 }
2244 
2245 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2246 			   struct btrfs_root *root,
2247 			   struct btrfs_key *key,
2248 			   struct btrfs_chunk *chunk, int item_size)
2249 {
2250 	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2251 	struct btrfs_disk_key disk_key;
2252 	u32 array_size;
2253 	u8 *ptr;
2254 
2255 	array_size = btrfs_super_sys_array_size(super_copy);
2256 	if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2257 		return -EFBIG;
2258 
2259 	ptr = super_copy->sys_chunk_array + array_size;
2260 	btrfs_cpu_key_to_disk(&disk_key, key);
2261 	memcpy(ptr, &disk_key, sizeof(disk_key));
2262 	ptr += sizeof(disk_key);
2263 	memcpy(ptr, chunk, item_size);
2264 	item_size += sizeof(disk_key);
2265 	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2266 	return 0;
2267 }
2268 
2269 /*
2270  * sort the devices in descending order by max_avail, total_avail
2271  */
2272 static int btrfs_cmp_device_info(const void *a, const void *b)
2273 {
2274 	const struct btrfs_device_info *di_a = a;
2275 	const struct btrfs_device_info *di_b = b;
2276 
2277 	if (di_a->max_avail > di_b->max_avail)
2278 		return -1;
2279 	if (di_a->max_avail < di_b->max_avail)
2280 		return 1;
2281 	if (di_a->total_avail > di_b->total_avail)
2282 		return -1;
2283 	if (di_a->total_avail < di_b->total_avail)
2284 		return 1;
2285 	return 0;
2286 }
2287 
2288 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2289 			       struct btrfs_root *extent_root,
2290 			       struct map_lookup **map_ret,
2291 			       u64 *num_bytes_out, u64 *stripe_size_out,
2292 			       u64 start, u64 type)
2293 {
2294 	struct btrfs_fs_info *info = extent_root->fs_info;
2295 	struct btrfs_fs_devices *fs_devices = info->fs_devices;
2296 	struct list_head *cur;
2297 	struct map_lookup *map = NULL;
2298 	struct extent_map_tree *em_tree;
2299 	struct extent_map *em;
2300 	struct btrfs_device_info *devices_info = NULL;
2301 	u64 total_avail;
2302 	int num_stripes;	/* total number of stripes to allocate */
2303 	int sub_stripes;	/* sub_stripes info for map */
2304 	int dev_stripes;	/* stripes per dev */
2305 	int devs_max;		/* max devs to use */
2306 	int devs_min;		/* min devs needed */
2307 	int devs_increment;	/* ndevs has to be a multiple of this */
2308 	int ncopies;		/* how many copies to data has */
2309 	int ret;
2310 	u64 max_stripe_size;
2311 	u64 max_chunk_size;
2312 	u64 stripe_size;
2313 	u64 num_bytes;
2314 	int ndevs;
2315 	int i;
2316 	int j;
2317 
2318 	if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2319 	    (type & BTRFS_BLOCK_GROUP_DUP)) {
2320 		WARN_ON(1);
2321 		type &= ~BTRFS_BLOCK_GROUP_DUP;
2322 	}
2323 
2324 	if (list_empty(&fs_devices->alloc_list))
2325 		return -ENOSPC;
2326 
2327 	sub_stripes = 1;
2328 	dev_stripes = 1;
2329 	devs_increment = 1;
2330 	ncopies = 1;
2331 	devs_max = 0;	/* 0 == as many as possible */
2332 	devs_min = 1;
2333 
2334 	/*
2335 	 * define the properties of each RAID type.
2336 	 * FIXME: move this to a global table and use it in all RAID
2337 	 * calculation code
2338 	 */
2339 	if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2340 		dev_stripes = 2;
2341 		ncopies = 2;
2342 		devs_max = 1;
2343 	} else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2344 		devs_min = 2;
2345 	} else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2346 		devs_increment = 2;
2347 		ncopies = 2;
2348 		devs_max = 2;
2349 		devs_min = 2;
2350 	} else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2351 		sub_stripes = 2;
2352 		devs_increment = 2;
2353 		ncopies = 2;
2354 		devs_min = 4;
2355 	} else {
2356 		devs_max = 1;
2357 	}
2358 
2359 	if (type & BTRFS_BLOCK_GROUP_DATA) {
2360 		max_stripe_size = 1024 * 1024 * 1024;
2361 		max_chunk_size = 10 * max_stripe_size;
2362 	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2363 		max_stripe_size = 256 * 1024 * 1024;
2364 		max_chunk_size = max_stripe_size;
2365 	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2366 		max_stripe_size = 8 * 1024 * 1024;
2367 		max_chunk_size = 2 * max_stripe_size;
2368 	} else {
2369 		printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2370 		       type);
2371 		BUG_ON(1);
2372 	}
2373 
2374 	/* we don't want a chunk larger than 10% of writeable space */
2375 	max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2376 			     max_chunk_size);
2377 
2378 	devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2379 			       GFP_NOFS);
2380 	if (!devices_info)
2381 		return -ENOMEM;
2382 
2383 	cur = fs_devices->alloc_list.next;
2384 
2385 	/*
2386 	 * in the first pass through the devices list, we gather information
2387 	 * about the available holes on each device.
2388 	 */
2389 	ndevs = 0;
2390 	while (cur != &fs_devices->alloc_list) {
2391 		struct btrfs_device *device;
2392 		u64 max_avail;
2393 		u64 dev_offset;
2394 
2395 		device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2396 
2397 		cur = cur->next;
2398 
2399 		if (!device->writeable) {
2400 			printk(KERN_ERR
2401 			       "btrfs: read-only device in alloc_list\n");
2402 			WARN_ON(1);
2403 			continue;
2404 		}
2405 
2406 		if (!device->in_fs_metadata)
2407 			continue;
2408 
2409 		if (device->total_bytes > device->bytes_used)
2410 			total_avail = device->total_bytes - device->bytes_used;
2411 		else
2412 			total_avail = 0;
2413 		/* avail is off by max(alloc_start, 1MB), but that is the same
2414 		 * for all devices, so it doesn't hurt the sorting later on
2415 		 */
2416 
2417 		ret = find_free_dev_extent(trans, device,
2418 					   max_stripe_size * dev_stripes,
2419 					   &dev_offset, &max_avail);
2420 		if (ret && ret != -ENOSPC)
2421 			goto error;
2422 
2423 		if (ret == 0)
2424 			max_avail = max_stripe_size * dev_stripes;
2425 
2426 		if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2427 			continue;
2428 
2429 		devices_info[ndevs].dev_offset = dev_offset;
2430 		devices_info[ndevs].max_avail = max_avail;
2431 		devices_info[ndevs].total_avail = total_avail;
2432 		devices_info[ndevs].dev = device;
2433 		++ndevs;
2434 	}
2435 
2436 	/*
2437 	 * now sort the devices by hole size / available space
2438 	 */
2439 	sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2440 	     btrfs_cmp_device_info, NULL);
2441 
2442 	/* round down to number of usable stripes */
2443 	ndevs -= ndevs % devs_increment;
2444 
2445 	if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2446 		ret = -ENOSPC;
2447 		goto error;
2448 	}
2449 
2450 	if (devs_max && ndevs > devs_max)
2451 		ndevs = devs_max;
2452 	/*
2453 	 * the primary goal is to maximize the number of stripes, so use as many
2454 	 * devices as possible, even if the stripes are not maximum sized.
2455 	 */
2456 	stripe_size = devices_info[ndevs-1].max_avail;
2457 	num_stripes = ndevs * dev_stripes;
2458 
2459 	if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2460 		stripe_size = max_chunk_size * ncopies;
2461 		do_div(stripe_size, num_stripes);
2462 	}
2463 
2464 	do_div(stripe_size, dev_stripes);
2465 	do_div(stripe_size, BTRFS_STRIPE_LEN);
2466 	stripe_size *= BTRFS_STRIPE_LEN;
2467 
2468 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2469 	if (!map) {
2470 		ret = -ENOMEM;
2471 		goto error;
2472 	}
2473 	map->num_stripes = num_stripes;
2474 
2475 	for (i = 0; i < ndevs; ++i) {
2476 		for (j = 0; j < dev_stripes; ++j) {
2477 			int s = i * dev_stripes + j;
2478 			map->stripes[s].dev = devices_info[i].dev;
2479 			map->stripes[s].physical = devices_info[i].dev_offset +
2480 						   j * stripe_size;
2481 		}
2482 	}
2483 	map->sector_size = extent_root->sectorsize;
2484 	map->stripe_len = BTRFS_STRIPE_LEN;
2485 	map->io_align = BTRFS_STRIPE_LEN;
2486 	map->io_width = BTRFS_STRIPE_LEN;
2487 	map->type = type;
2488 	map->sub_stripes = sub_stripes;
2489 
2490 	*map_ret = map;
2491 	num_bytes = stripe_size * (num_stripes / ncopies);
2492 
2493 	*stripe_size_out = stripe_size;
2494 	*num_bytes_out = num_bytes;
2495 
2496 	trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
2497 
2498 	em = alloc_extent_map();
2499 	if (!em) {
2500 		ret = -ENOMEM;
2501 		goto error;
2502 	}
2503 	em->bdev = (struct block_device *)map;
2504 	em->start = start;
2505 	em->len = num_bytes;
2506 	em->block_start = 0;
2507 	em->block_len = em->len;
2508 
2509 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2510 	write_lock(&em_tree->lock);
2511 	ret = add_extent_mapping(em_tree, em);
2512 	write_unlock(&em_tree->lock);
2513 	BUG_ON(ret);
2514 	free_extent_map(em);
2515 
2516 	ret = btrfs_make_block_group(trans, extent_root, 0, type,
2517 				     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2518 				     start, num_bytes);
2519 	BUG_ON(ret);
2520 
2521 	for (i = 0; i < map->num_stripes; ++i) {
2522 		struct btrfs_device *device;
2523 		u64 dev_offset;
2524 
2525 		device = map->stripes[i].dev;
2526 		dev_offset = map->stripes[i].physical;
2527 
2528 		ret = btrfs_alloc_dev_extent(trans, device,
2529 				info->chunk_root->root_key.objectid,
2530 				BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2531 				start, dev_offset, stripe_size);
2532 		BUG_ON(ret);
2533 	}
2534 
2535 	kfree(devices_info);
2536 	return 0;
2537 
2538 error:
2539 	kfree(map);
2540 	kfree(devices_info);
2541 	return ret;
2542 }
2543 
2544 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2545 				struct btrfs_root *extent_root,
2546 				struct map_lookup *map, u64 chunk_offset,
2547 				u64 chunk_size, u64 stripe_size)
2548 {
2549 	u64 dev_offset;
2550 	struct btrfs_key key;
2551 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2552 	struct btrfs_device *device;
2553 	struct btrfs_chunk *chunk;
2554 	struct btrfs_stripe *stripe;
2555 	size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2556 	int index = 0;
2557 	int ret;
2558 
2559 	chunk = kzalloc(item_size, GFP_NOFS);
2560 	if (!chunk)
2561 		return -ENOMEM;
2562 
2563 	index = 0;
2564 	while (index < map->num_stripes) {
2565 		device = map->stripes[index].dev;
2566 		device->bytes_used += stripe_size;
2567 		ret = btrfs_update_device(trans, device);
2568 		BUG_ON(ret);
2569 		index++;
2570 	}
2571 
2572 	index = 0;
2573 	stripe = &chunk->stripe;
2574 	while (index < map->num_stripes) {
2575 		device = map->stripes[index].dev;
2576 		dev_offset = map->stripes[index].physical;
2577 
2578 		btrfs_set_stack_stripe_devid(stripe, device->devid);
2579 		btrfs_set_stack_stripe_offset(stripe, dev_offset);
2580 		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2581 		stripe++;
2582 		index++;
2583 	}
2584 
2585 	btrfs_set_stack_chunk_length(chunk, chunk_size);
2586 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2587 	btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2588 	btrfs_set_stack_chunk_type(chunk, map->type);
2589 	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2590 	btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2591 	btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2592 	btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2593 	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2594 
2595 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2596 	key.type = BTRFS_CHUNK_ITEM_KEY;
2597 	key.offset = chunk_offset;
2598 
2599 	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2600 	BUG_ON(ret);
2601 
2602 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2603 		ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2604 					     item_size);
2605 		BUG_ON(ret);
2606 	}
2607 
2608 	kfree(chunk);
2609 	return 0;
2610 }
2611 
2612 /*
2613  * Chunk allocation falls into two parts. The first part does works
2614  * that make the new allocated chunk useable, but not do any operation
2615  * that modifies the chunk tree. The second part does the works that
2616  * require modifying the chunk tree. This division is important for the
2617  * bootstrap process of adding storage to a seed btrfs.
2618  */
2619 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2620 		      struct btrfs_root *extent_root, u64 type)
2621 {
2622 	u64 chunk_offset;
2623 	u64 chunk_size;
2624 	u64 stripe_size;
2625 	struct map_lookup *map;
2626 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2627 	int ret;
2628 
2629 	ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2630 			      &chunk_offset);
2631 	if (ret)
2632 		return ret;
2633 
2634 	ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2635 				  &stripe_size, chunk_offset, type);
2636 	if (ret)
2637 		return ret;
2638 
2639 	ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2640 				   chunk_size, stripe_size);
2641 	BUG_ON(ret);
2642 	return 0;
2643 }
2644 
2645 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2646 					 struct btrfs_root *root,
2647 					 struct btrfs_device *device)
2648 {
2649 	u64 chunk_offset;
2650 	u64 sys_chunk_offset;
2651 	u64 chunk_size;
2652 	u64 sys_chunk_size;
2653 	u64 stripe_size;
2654 	u64 sys_stripe_size;
2655 	u64 alloc_profile;
2656 	struct map_lookup *map;
2657 	struct map_lookup *sys_map;
2658 	struct btrfs_fs_info *fs_info = root->fs_info;
2659 	struct btrfs_root *extent_root = fs_info->extent_root;
2660 	int ret;
2661 
2662 	ret = find_next_chunk(fs_info->chunk_root,
2663 			      BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2664 	BUG_ON(ret);
2665 
2666 	alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2667 			(fs_info->metadata_alloc_profile &
2668 			 fs_info->avail_metadata_alloc_bits);
2669 	alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2670 
2671 	ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2672 				  &stripe_size, chunk_offset, alloc_profile);
2673 	BUG_ON(ret);
2674 
2675 	sys_chunk_offset = chunk_offset + chunk_size;
2676 
2677 	alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2678 			(fs_info->system_alloc_profile &
2679 			 fs_info->avail_system_alloc_bits);
2680 	alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2681 
2682 	ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2683 				  &sys_chunk_size, &sys_stripe_size,
2684 				  sys_chunk_offset, alloc_profile);
2685 	BUG_ON(ret);
2686 
2687 	ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2688 	BUG_ON(ret);
2689 
2690 	/*
2691 	 * Modifying chunk tree needs allocating new blocks from both
2692 	 * system block group and metadata block group. So we only can
2693 	 * do operations require modifying the chunk tree after both
2694 	 * block groups were created.
2695 	 */
2696 	ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2697 				   chunk_size, stripe_size);
2698 	BUG_ON(ret);
2699 
2700 	ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2701 				   sys_chunk_offset, sys_chunk_size,
2702 				   sys_stripe_size);
2703 	BUG_ON(ret);
2704 	return 0;
2705 }
2706 
2707 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2708 {
2709 	struct extent_map *em;
2710 	struct map_lookup *map;
2711 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2712 	int readonly = 0;
2713 	int i;
2714 
2715 	read_lock(&map_tree->map_tree.lock);
2716 	em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2717 	read_unlock(&map_tree->map_tree.lock);
2718 	if (!em)
2719 		return 1;
2720 
2721 	if (btrfs_test_opt(root, DEGRADED)) {
2722 		free_extent_map(em);
2723 		return 0;
2724 	}
2725 
2726 	map = (struct map_lookup *)em->bdev;
2727 	for (i = 0; i < map->num_stripes; i++) {
2728 		if (!map->stripes[i].dev->writeable) {
2729 			readonly = 1;
2730 			break;
2731 		}
2732 	}
2733 	free_extent_map(em);
2734 	return readonly;
2735 }
2736 
2737 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2738 {
2739 	extent_map_tree_init(&tree->map_tree);
2740 }
2741 
2742 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2743 {
2744 	struct extent_map *em;
2745 
2746 	while (1) {
2747 		write_lock(&tree->map_tree.lock);
2748 		em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2749 		if (em)
2750 			remove_extent_mapping(&tree->map_tree, em);
2751 		write_unlock(&tree->map_tree.lock);
2752 		if (!em)
2753 			break;
2754 		kfree(em->bdev);
2755 		/* once for us */
2756 		free_extent_map(em);
2757 		/* once for the tree */
2758 		free_extent_map(em);
2759 	}
2760 }
2761 
2762 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2763 {
2764 	struct extent_map *em;
2765 	struct map_lookup *map;
2766 	struct extent_map_tree *em_tree = &map_tree->map_tree;
2767 	int ret;
2768 
2769 	read_lock(&em_tree->lock);
2770 	em = lookup_extent_mapping(em_tree, logical, len);
2771 	read_unlock(&em_tree->lock);
2772 	BUG_ON(!em);
2773 
2774 	BUG_ON(em->start > logical || em->start + em->len < logical);
2775 	map = (struct map_lookup *)em->bdev;
2776 	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2777 		ret = map->num_stripes;
2778 	else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2779 		ret = map->sub_stripes;
2780 	else
2781 		ret = 1;
2782 	free_extent_map(em);
2783 	return ret;
2784 }
2785 
2786 static int find_live_mirror(struct map_lookup *map, int first, int num,
2787 			    int optimal)
2788 {
2789 	int i;
2790 	if (map->stripes[optimal].dev->bdev)
2791 		return optimal;
2792 	for (i = first; i < first + num; i++) {
2793 		if (map->stripes[i].dev->bdev)
2794 			return i;
2795 	}
2796 	/* we couldn't find one that doesn't fail.  Just return something
2797 	 * and the io error handling code will clean up eventually
2798 	 */
2799 	return optimal;
2800 }
2801 
2802 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2803 			     u64 logical, u64 *length,
2804 			     struct btrfs_multi_bio **multi_ret,
2805 			     int mirror_num)
2806 {
2807 	struct extent_map *em;
2808 	struct map_lookup *map;
2809 	struct extent_map_tree *em_tree = &map_tree->map_tree;
2810 	u64 offset;
2811 	u64 stripe_offset;
2812 	u64 stripe_end_offset;
2813 	u64 stripe_nr;
2814 	u64 stripe_nr_orig;
2815 	u64 stripe_nr_end;
2816 	int stripes_allocated = 8;
2817 	int stripes_required = 1;
2818 	int stripe_index;
2819 	int i;
2820 	int num_stripes;
2821 	int max_errors = 0;
2822 	struct btrfs_multi_bio *multi = NULL;
2823 
2824 	if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2825 		stripes_allocated = 1;
2826 again:
2827 	if (multi_ret) {
2828 		multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2829 				GFP_NOFS);
2830 		if (!multi)
2831 			return -ENOMEM;
2832 
2833 		atomic_set(&multi->error, 0);
2834 	}
2835 
2836 	read_lock(&em_tree->lock);
2837 	em = lookup_extent_mapping(em_tree, logical, *length);
2838 	read_unlock(&em_tree->lock);
2839 
2840 	if (!em) {
2841 		printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2842 		       (unsigned long long)logical,
2843 		       (unsigned long long)*length);
2844 		BUG();
2845 	}
2846 
2847 	BUG_ON(em->start > logical || em->start + em->len < logical);
2848 	map = (struct map_lookup *)em->bdev;
2849 	offset = logical - em->start;
2850 
2851 	if (mirror_num > map->num_stripes)
2852 		mirror_num = 0;
2853 
2854 	/* if our multi bio struct is too small, back off and try again */
2855 	if (rw & REQ_WRITE) {
2856 		if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2857 				 BTRFS_BLOCK_GROUP_DUP)) {
2858 			stripes_required = map->num_stripes;
2859 			max_errors = 1;
2860 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2861 			stripes_required = map->sub_stripes;
2862 			max_errors = 1;
2863 		}
2864 	}
2865 	if (rw & REQ_DISCARD) {
2866 		if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2867 				 BTRFS_BLOCK_GROUP_RAID1 |
2868 				 BTRFS_BLOCK_GROUP_DUP |
2869 				 BTRFS_BLOCK_GROUP_RAID10)) {
2870 			stripes_required = map->num_stripes;
2871 		}
2872 	}
2873 	if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2874 	    stripes_allocated < stripes_required) {
2875 		stripes_allocated = map->num_stripes;
2876 		free_extent_map(em);
2877 		kfree(multi);
2878 		goto again;
2879 	}
2880 	stripe_nr = offset;
2881 	/*
2882 	 * stripe_nr counts the total number of stripes we have to stride
2883 	 * to get to this block
2884 	 */
2885 	do_div(stripe_nr, map->stripe_len);
2886 
2887 	stripe_offset = stripe_nr * map->stripe_len;
2888 	BUG_ON(offset < stripe_offset);
2889 
2890 	/* stripe_offset is the offset of this block in its stripe*/
2891 	stripe_offset = offset - stripe_offset;
2892 
2893 	if (rw & REQ_DISCARD)
2894 		*length = min_t(u64, em->len - offset, *length);
2895 	else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2896 			      BTRFS_BLOCK_GROUP_RAID1 |
2897 			      BTRFS_BLOCK_GROUP_RAID10 |
2898 			      BTRFS_BLOCK_GROUP_DUP)) {
2899 		/* we limit the length of each bio to what fits in a stripe */
2900 		*length = min_t(u64, em->len - offset,
2901 				map->stripe_len - stripe_offset);
2902 	} else {
2903 		*length = em->len - offset;
2904 	}
2905 
2906 	if (!multi_ret)
2907 		goto out;
2908 
2909 	num_stripes = 1;
2910 	stripe_index = 0;
2911 	stripe_nr_orig = stripe_nr;
2912 	stripe_nr_end = (offset + *length + map->stripe_len - 1) &
2913 			(~(map->stripe_len - 1));
2914 	do_div(stripe_nr_end, map->stripe_len);
2915 	stripe_end_offset = stripe_nr_end * map->stripe_len -
2916 			    (offset + *length);
2917 	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2918 		if (rw & REQ_DISCARD)
2919 			num_stripes = min_t(u64, map->num_stripes,
2920 					    stripe_nr_end - stripe_nr_orig);
2921 		stripe_index = do_div(stripe_nr, map->num_stripes);
2922 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2923 		if (rw & (REQ_WRITE | REQ_DISCARD))
2924 			num_stripes = map->num_stripes;
2925 		else if (mirror_num)
2926 			stripe_index = mirror_num - 1;
2927 		else {
2928 			stripe_index = find_live_mirror(map, 0,
2929 					    map->num_stripes,
2930 					    current->pid % map->num_stripes);
2931 		}
2932 
2933 	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2934 		if (rw & (REQ_WRITE | REQ_DISCARD))
2935 			num_stripes = map->num_stripes;
2936 		else if (mirror_num)
2937 			stripe_index = mirror_num - 1;
2938 
2939 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2940 		int factor = map->num_stripes / map->sub_stripes;
2941 
2942 		stripe_index = do_div(stripe_nr, factor);
2943 		stripe_index *= map->sub_stripes;
2944 
2945 		if (rw & REQ_WRITE)
2946 			num_stripes = map->sub_stripes;
2947 		else if (rw & REQ_DISCARD)
2948 			num_stripes = min_t(u64, map->sub_stripes *
2949 					    (stripe_nr_end - stripe_nr_orig),
2950 					    map->num_stripes);
2951 		else if (mirror_num)
2952 			stripe_index += mirror_num - 1;
2953 		else {
2954 			stripe_index = find_live_mirror(map, stripe_index,
2955 					      map->sub_stripes, stripe_index +
2956 					      current->pid % map->sub_stripes);
2957 		}
2958 	} else {
2959 		/*
2960 		 * after this do_div call, stripe_nr is the number of stripes
2961 		 * on this device we have to walk to find the data, and
2962 		 * stripe_index is the number of our device in the stripe array
2963 		 */
2964 		stripe_index = do_div(stripe_nr, map->num_stripes);
2965 	}
2966 	BUG_ON(stripe_index >= map->num_stripes);
2967 
2968 	if (rw & REQ_DISCARD) {
2969 		for (i = 0; i < num_stripes; i++) {
2970 			multi->stripes[i].physical =
2971 				map->stripes[stripe_index].physical +
2972 				stripe_offset + stripe_nr * map->stripe_len;
2973 			multi->stripes[i].dev = map->stripes[stripe_index].dev;
2974 
2975 			if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2976 				u64 stripes;
2977 				u32 last_stripe = 0;
2978 				int j;
2979 
2980 				div_u64_rem(stripe_nr_end - 1,
2981 					    map->num_stripes,
2982 					    &last_stripe);
2983 
2984 				for (j = 0; j < map->num_stripes; j++) {
2985 					u32 test;
2986 
2987 					div_u64_rem(stripe_nr_end - 1 - j,
2988 						    map->num_stripes, &test);
2989 					if (test == stripe_index)
2990 						break;
2991 				}
2992 				stripes = stripe_nr_end - 1 - j;
2993 				do_div(stripes, map->num_stripes);
2994 				multi->stripes[i].length = map->stripe_len *
2995 					(stripes - stripe_nr + 1);
2996 
2997 				if (i == 0) {
2998 					multi->stripes[i].length -=
2999 						stripe_offset;
3000 					stripe_offset = 0;
3001 				}
3002 				if (stripe_index == last_stripe)
3003 					multi->stripes[i].length -=
3004 						stripe_end_offset;
3005 			} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3006 				u64 stripes;
3007 				int j;
3008 				int factor = map->num_stripes /
3009 					     map->sub_stripes;
3010 				u32 last_stripe = 0;
3011 
3012 				div_u64_rem(stripe_nr_end - 1,
3013 					    factor, &last_stripe);
3014 				last_stripe *= map->sub_stripes;
3015 
3016 				for (j = 0; j < factor; j++) {
3017 					u32 test;
3018 
3019 					div_u64_rem(stripe_nr_end - 1 - j,
3020 						    factor, &test);
3021 
3022 					if (test ==
3023 					    stripe_index / map->sub_stripes)
3024 						break;
3025 				}
3026 				stripes = stripe_nr_end - 1 - j;
3027 				do_div(stripes, factor);
3028 				multi->stripes[i].length = map->stripe_len *
3029 					(stripes - stripe_nr + 1);
3030 
3031 				if (i < map->sub_stripes) {
3032 					multi->stripes[i].length -=
3033 						stripe_offset;
3034 					if (i == map->sub_stripes - 1)
3035 						stripe_offset = 0;
3036 				}
3037 				if (stripe_index >= last_stripe &&
3038 				    stripe_index <= (last_stripe +
3039 						     map->sub_stripes - 1)) {
3040 					multi->stripes[i].length -=
3041 						stripe_end_offset;
3042 				}
3043 			} else
3044 				multi->stripes[i].length = *length;
3045 
3046 			stripe_index++;
3047 			if (stripe_index == map->num_stripes) {
3048 				/* This could only happen for RAID0/10 */
3049 				stripe_index = 0;
3050 				stripe_nr++;
3051 			}
3052 		}
3053 	} else {
3054 		for (i = 0; i < num_stripes; i++) {
3055 			multi->stripes[i].physical =
3056 				map->stripes[stripe_index].physical +
3057 				stripe_offset +
3058 				stripe_nr * map->stripe_len;
3059 			multi->stripes[i].dev =
3060 				map->stripes[stripe_index].dev;
3061 			stripe_index++;
3062 		}
3063 	}
3064 	if (multi_ret) {
3065 		*multi_ret = multi;
3066 		multi->num_stripes = num_stripes;
3067 		multi->max_errors = max_errors;
3068 	}
3069 out:
3070 	free_extent_map(em);
3071 	return 0;
3072 }
3073 
3074 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3075 		      u64 logical, u64 *length,
3076 		      struct btrfs_multi_bio **multi_ret, int mirror_num)
3077 {
3078 	return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3079 				 mirror_num);
3080 }
3081 
3082 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3083 		     u64 chunk_start, u64 physical, u64 devid,
3084 		     u64 **logical, int *naddrs, int *stripe_len)
3085 {
3086 	struct extent_map_tree *em_tree = &map_tree->map_tree;
3087 	struct extent_map *em;
3088 	struct map_lookup *map;
3089 	u64 *buf;
3090 	u64 bytenr;
3091 	u64 length;
3092 	u64 stripe_nr;
3093 	int i, j, nr = 0;
3094 
3095 	read_lock(&em_tree->lock);
3096 	em = lookup_extent_mapping(em_tree, chunk_start, 1);
3097 	read_unlock(&em_tree->lock);
3098 
3099 	BUG_ON(!em || em->start != chunk_start);
3100 	map = (struct map_lookup *)em->bdev;
3101 
3102 	length = em->len;
3103 	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3104 		do_div(length, map->num_stripes / map->sub_stripes);
3105 	else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3106 		do_div(length, map->num_stripes);
3107 
3108 	buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3109 	BUG_ON(!buf);
3110 
3111 	for (i = 0; i < map->num_stripes; i++) {
3112 		if (devid && map->stripes[i].dev->devid != devid)
3113 			continue;
3114 		if (map->stripes[i].physical > physical ||
3115 		    map->stripes[i].physical + length <= physical)
3116 			continue;
3117 
3118 		stripe_nr = physical - map->stripes[i].physical;
3119 		do_div(stripe_nr, map->stripe_len);
3120 
3121 		if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3122 			stripe_nr = stripe_nr * map->num_stripes + i;
3123 			do_div(stripe_nr, map->sub_stripes);
3124 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3125 			stripe_nr = stripe_nr * map->num_stripes + i;
3126 		}
3127 		bytenr = chunk_start + stripe_nr * map->stripe_len;
3128 		WARN_ON(nr >= map->num_stripes);
3129 		for (j = 0; j < nr; j++) {
3130 			if (buf[j] == bytenr)
3131 				break;
3132 		}
3133 		if (j == nr) {
3134 			WARN_ON(nr >= map->num_stripes);
3135 			buf[nr++] = bytenr;
3136 		}
3137 	}
3138 
3139 	*logical = buf;
3140 	*naddrs = nr;
3141 	*stripe_len = map->stripe_len;
3142 
3143 	free_extent_map(em);
3144 	return 0;
3145 }
3146 
3147 static void end_bio_multi_stripe(struct bio *bio, int err)
3148 {
3149 	struct btrfs_multi_bio *multi = bio->bi_private;
3150 	int is_orig_bio = 0;
3151 
3152 	if (err)
3153 		atomic_inc(&multi->error);
3154 
3155 	if (bio == multi->orig_bio)
3156 		is_orig_bio = 1;
3157 
3158 	if (atomic_dec_and_test(&multi->stripes_pending)) {
3159 		if (!is_orig_bio) {
3160 			bio_put(bio);
3161 			bio = multi->orig_bio;
3162 		}
3163 		bio->bi_private = multi->private;
3164 		bio->bi_end_io = multi->end_io;
3165 		/* only send an error to the higher layers if it is
3166 		 * beyond the tolerance of the multi-bio
3167 		 */
3168 		if (atomic_read(&multi->error) > multi->max_errors) {
3169 			err = -EIO;
3170 		} else if (err) {
3171 			/*
3172 			 * this bio is actually up to date, we didn't
3173 			 * go over the max number of errors
3174 			 */
3175 			set_bit(BIO_UPTODATE, &bio->bi_flags);
3176 			err = 0;
3177 		}
3178 		kfree(multi);
3179 
3180 		bio_endio(bio, err);
3181 	} else if (!is_orig_bio) {
3182 		bio_put(bio);
3183 	}
3184 }
3185 
3186 struct async_sched {
3187 	struct bio *bio;
3188 	int rw;
3189 	struct btrfs_fs_info *info;
3190 	struct btrfs_work work;
3191 };
3192 
3193 /*
3194  * see run_scheduled_bios for a description of why bios are collected for
3195  * async submit.
3196  *
3197  * This will add one bio to the pending list for a device and make sure
3198  * the work struct is scheduled.
3199  */
3200 static noinline int schedule_bio(struct btrfs_root *root,
3201 				 struct btrfs_device *device,
3202 				 int rw, struct bio *bio)
3203 {
3204 	int should_queue = 1;
3205 	struct btrfs_pending_bios *pending_bios;
3206 
3207 	/* don't bother with additional async steps for reads, right now */
3208 	if (!(rw & REQ_WRITE)) {
3209 		bio_get(bio);
3210 		submit_bio(rw, bio);
3211 		bio_put(bio);
3212 		return 0;
3213 	}
3214 
3215 	/*
3216 	 * nr_async_bios allows us to reliably return congestion to the
3217 	 * higher layers.  Otherwise, the async bio makes it appear we have
3218 	 * made progress against dirty pages when we've really just put it
3219 	 * on a queue for later
3220 	 */
3221 	atomic_inc(&root->fs_info->nr_async_bios);
3222 	WARN_ON(bio->bi_next);
3223 	bio->bi_next = NULL;
3224 	bio->bi_rw |= rw;
3225 
3226 	spin_lock(&device->io_lock);
3227 	if (bio->bi_rw & REQ_SYNC)
3228 		pending_bios = &device->pending_sync_bios;
3229 	else
3230 		pending_bios = &device->pending_bios;
3231 
3232 	if (pending_bios->tail)
3233 		pending_bios->tail->bi_next = bio;
3234 
3235 	pending_bios->tail = bio;
3236 	if (!pending_bios->head)
3237 		pending_bios->head = bio;
3238 	if (device->running_pending)
3239 		should_queue = 0;
3240 
3241 	spin_unlock(&device->io_lock);
3242 
3243 	if (should_queue)
3244 		btrfs_queue_worker(&root->fs_info->submit_workers,
3245 				   &device->work);
3246 	return 0;
3247 }
3248 
3249 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3250 		  int mirror_num, int async_submit)
3251 {
3252 	struct btrfs_mapping_tree *map_tree;
3253 	struct btrfs_device *dev;
3254 	struct bio *first_bio = bio;
3255 	u64 logical = (u64)bio->bi_sector << 9;
3256 	u64 length = 0;
3257 	u64 map_length;
3258 	struct btrfs_multi_bio *multi = NULL;
3259 	int ret;
3260 	int dev_nr = 0;
3261 	int total_devs = 1;
3262 
3263 	length = bio->bi_size;
3264 	map_tree = &root->fs_info->mapping_tree;
3265 	map_length = length;
3266 
3267 	ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3268 			      mirror_num);
3269 	BUG_ON(ret);
3270 
3271 	total_devs = multi->num_stripes;
3272 	if (map_length < length) {
3273 		printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3274 		       "len %llu\n", (unsigned long long)logical,
3275 		       (unsigned long long)length,
3276 		       (unsigned long long)map_length);
3277 		BUG();
3278 	}
3279 	multi->end_io = first_bio->bi_end_io;
3280 	multi->private = first_bio->bi_private;
3281 	multi->orig_bio = first_bio;
3282 	atomic_set(&multi->stripes_pending, multi->num_stripes);
3283 
3284 	while (dev_nr < total_devs) {
3285 		if (total_devs > 1) {
3286 			if (dev_nr < total_devs - 1) {
3287 				bio = bio_clone(first_bio, GFP_NOFS);
3288 				BUG_ON(!bio);
3289 			} else {
3290 				bio = first_bio;
3291 			}
3292 			bio->bi_private = multi;
3293 			bio->bi_end_io = end_bio_multi_stripe;
3294 		}
3295 		bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3296 		dev = multi->stripes[dev_nr].dev;
3297 		if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3298 			bio->bi_bdev = dev->bdev;
3299 			if (async_submit)
3300 				schedule_bio(root, dev, rw, bio);
3301 			else
3302 				submit_bio(rw, bio);
3303 		} else {
3304 			bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3305 			bio->bi_sector = logical >> 9;
3306 			bio_endio(bio, -EIO);
3307 		}
3308 		dev_nr++;
3309 	}
3310 	if (total_devs == 1)
3311 		kfree(multi);
3312 	return 0;
3313 }
3314 
3315 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3316 				       u8 *uuid, u8 *fsid)
3317 {
3318 	struct btrfs_device *device;
3319 	struct btrfs_fs_devices *cur_devices;
3320 
3321 	cur_devices = root->fs_info->fs_devices;
3322 	while (cur_devices) {
3323 		if (!fsid ||
3324 		    !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3325 			device = __find_device(&cur_devices->devices,
3326 					       devid, uuid);
3327 			if (device)
3328 				return device;
3329 		}
3330 		cur_devices = cur_devices->seed;
3331 	}
3332 	return NULL;
3333 }
3334 
3335 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3336 					    u64 devid, u8 *dev_uuid)
3337 {
3338 	struct btrfs_device *device;
3339 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3340 
3341 	device = kzalloc(sizeof(*device), GFP_NOFS);
3342 	if (!device)
3343 		return NULL;
3344 	list_add(&device->dev_list,
3345 		 &fs_devices->devices);
3346 	device->dev_root = root->fs_info->dev_root;
3347 	device->devid = devid;
3348 	device->work.func = pending_bios_fn;
3349 	device->fs_devices = fs_devices;
3350 	device->missing = 1;
3351 	fs_devices->num_devices++;
3352 	fs_devices->missing_devices++;
3353 	spin_lock_init(&device->io_lock);
3354 	INIT_LIST_HEAD(&device->dev_alloc_list);
3355 	memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3356 	return device;
3357 }
3358 
3359 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3360 			  struct extent_buffer *leaf,
3361 			  struct btrfs_chunk *chunk)
3362 {
3363 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3364 	struct map_lookup *map;
3365 	struct extent_map *em;
3366 	u64 logical;
3367 	u64 length;
3368 	u64 devid;
3369 	u8 uuid[BTRFS_UUID_SIZE];
3370 	int num_stripes;
3371 	int ret;
3372 	int i;
3373 
3374 	logical = key->offset;
3375 	length = btrfs_chunk_length(leaf, chunk);
3376 
3377 	read_lock(&map_tree->map_tree.lock);
3378 	em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3379 	read_unlock(&map_tree->map_tree.lock);
3380 
3381 	/* already mapped? */
3382 	if (em && em->start <= logical && em->start + em->len > logical) {
3383 		free_extent_map(em);
3384 		return 0;
3385 	} else if (em) {
3386 		free_extent_map(em);
3387 	}
3388 
3389 	em = alloc_extent_map();
3390 	if (!em)
3391 		return -ENOMEM;
3392 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3393 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3394 	if (!map) {
3395 		free_extent_map(em);
3396 		return -ENOMEM;
3397 	}
3398 
3399 	em->bdev = (struct block_device *)map;
3400 	em->start = logical;
3401 	em->len = length;
3402 	em->block_start = 0;
3403 	em->block_len = em->len;
3404 
3405 	map->num_stripes = num_stripes;
3406 	map->io_width = btrfs_chunk_io_width(leaf, chunk);
3407 	map->io_align = btrfs_chunk_io_align(leaf, chunk);
3408 	map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3409 	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3410 	map->type = btrfs_chunk_type(leaf, chunk);
3411 	map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3412 	for (i = 0; i < num_stripes; i++) {
3413 		map->stripes[i].physical =
3414 			btrfs_stripe_offset_nr(leaf, chunk, i);
3415 		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3416 		read_extent_buffer(leaf, uuid, (unsigned long)
3417 				   btrfs_stripe_dev_uuid_nr(chunk, i),
3418 				   BTRFS_UUID_SIZE);
3419 		map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3420 							NULL);
3421 		if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3422 			kfree(map);
3423 			free_extent_map(em);
3424 			return -EIO;
3425 		}
3426 		if (!map->stripes[i].dev) {
3427 			map->stripes[i].dev =
3428 				add_missing_dev(root, devid, uuid);
3429 			if (!map->stripes[i].dev) {
3430 				kfree(map);
3431 				free_extent_map(em);
3432 				return -EIO;
3433 			}
3434 		}
3435 		map->stripes[i].dev->in_fs_metadata = 1;
3436 	}
3437 
3438 	write_lock(&map_tree->map_tree.lock);
3439 	ret = add_extent_mapping(&map_tree->map_tree, em);
3440 	write_unlock(&map_tree->map_tree.lock);
3441 	BUG_ON(ret);
3442 	free_extent_map(em);
3443 
3444 	return 0;
3445 }
3446 
3447 static int fill_device_from_item(struct extent_buffer *leaf,
3448 				 struct btrfs_dev_item *dev_item,
3449 				 struct btrfs_device *device)
3450 {
3451 	unsigned long ptr;
3452 
3453 	device->devid = btrfs_device_id(leaf, dev_item);
3454 	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3455 	device->total_bytes = device->disk_total_bytes;
3456 	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3457 	device->type = btrfs_device_type(leaf, dev_item);
3458 	device->io_align = btrfs_device_io_align(leaf, dev_item);
3459 	device->io_width = btrfs_device_io_width(leaf, dev_item);
3460 	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3461 
3462 	ptr = (unsigned long)btrfs_device_uuid(dev_item);
3463 	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3464 
3465 	return 0;
3466 }
3467 
3468 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3469 {
3470 	struct btrfs_fs_devices *fs_devices;
3471 	int ret;
3472 
3473 	mutex_lock(&uuid_mutex);
3474 
3475 	fs_devices = root->fs_info->fs_devices->seed;
3476 	while (fs_devices) {
3477 		if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3478 			ret = 0;
3479 			goto out;
3480 		}
3481 		fs_devices = fs_devices->seed;
3482 	}
3483 
3484 	fs_devices = find_fsid(fsid);
3485 	if (!fs_devices) {
3486 		ret = -ENOENT;
3487 		goto out;
3488 	}
3489 
3490 	fs_devices = clone_fs_devices(fs_devices);
3491 	if (IS_ERR(fs_devices)) {
3492 		ret = PTR_ERR(fs_devices);
3493 		goto out;
3494 	}
3495 
3496 	ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3497 				   root->fs_info->bdev_holder);
3498 	if (ret)
3499 		goto out;
3500 
3501 	if (!fs_devices->seeding) {
3502 		__btrfs_close_devices(fs_devices);
3503 		free_fs_devices(fs_devices);
3504 		ret = -EINVAL;
3505 		goto out;
3506 	}
3507 
3508 	fs_devices->seed = root->fs_info->fs_devices->seed;
3509 	root->fs_info->fs_devices->seed = fs_devices;
3510 out:
3511 	mutex_unlock(&uuid_mutex);
3512 	return ret;
3513 }
3514 
3515 static int read_one_dev(struct btrfs_root *root,
3516 			struct extent_buffer *leaf,
3517 			struct btrfs_dev_item *dev_item)
3518 {
3519 	struct btrfs_device *device;
3520 	u64 devid;
3521 	int ret;
3522 	u8 fs_uuid[BTRFS_UUID_SIZE];
3523 	u8 dev_uuid[BTRFS_UUID_SIZE];
3524 
3525 	devid = btrfs_device_id(leaf, dev_item);
3526 	read_extent_buffer(leaf, dev_uuid,
3527 			   (unsigned long)btrfs_device_uuid(dev_item),
3528 			   BTRFS_UUID_SIZE);
3529 	read_extent_buffer(leaf, fs_uuid,
3530 			   (unsigned long)btrfs_device_fsid(dev_item),
3531 			   BTRFS_UUID_SIZE);
3532 
3533 	if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3534 		ret = open_seed_devices(root, fs_uuid);
3535 		if (ret && !btrfs_test_opt(root, DEGRADED))
3536 			return ret;
3537 	}
3538 
3539 	device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3540 	if (!device || !device->bdev) {
3541 		if (!btrfs_test_opt(root, DEGRADED))
3542 			return -EIO;
3543 
3544 		if (!device) {
3545 			printk(KERN_WARNING "warning devid %llu missing\n",
3546 			       (unsigned long long)devid);
3547 			device = add_missing_dev(root, devid, dev_uuid);
3548 			if (!device)
3549 				return -ENOMEM;
3550 		} else if (!device->missing) {
3551 			/*
3552 			 * this happens when a device that was properly setup
3553 			 * in the device info lists suddenly goes bad.
3554 			 * device->bdev is NULL, and so we have to set
3555 			 * device->missing to one here
3556 			 */
3557 			root->fs_info->fs_devices->missing_devices++;
3558 			device->missing = 1;
3559 		}
3560 	}
3561 
3562 	if (device->fs_devices != root->fs_info->fs_devices) {
3563 		BUG_ON(device->writeable);
3564 		if (device->generation !=
3565 		    btrfs_device_generation(leaf, dev_item))
3566 			return -EINVAL;
3567 	}
3568 
3569 	fill_device_from_item(leaf, dev_item, device);
3570 	device->dev_root = root->fs_info->dev_root;
3571 	device->in_fs_metadata = 1;
3572 	if (device->writeable)
3573 		device->fs_devices->total_rw_bytes += device->total_bytes;
3574 	ret = 0;
3575 	return ret;
3576 }
3577 
3578 int btrfs_read_sys_array(struct btrfs_root *root)
3579 {
3580 	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3581 	struct extent_buffer *sb;
3582 	struct btrfs_disk_key *disk_key;
3583 	struct btrfs_chunk *chunk;
3584 	u8 *ptr;
3585 	unsigned long sb_ptr;
3586 	int ret = 0;
3587 	u32 num_stripes;
3588 	u32 array_size;
3589 	u32 len = 0;
3590 	u32 cur;
3591 	struct btrfs_key key;
3592 
3593 	sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3594 					  BTRFS_SUPER_INFO_SIZE);
3595 	if (!sb)
3596 		return -ENOMEM;
3597 	btrfs_set_buffer_uptodate(sb);
3598 	btrfs_set_buffer_lockdep_class(sb, 0);
3599 
3600 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3601 	array_size = btrfs_super_sys_array_size(super_copy);
3602 
3603 	ptr = super_copy->sys_chunk_array;
3604 	sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3605 	cur = 0;
3606 
3607 	while (cur < array_size) {
3608 		disk_key = (struct btrfs_disk_key *)ptr;
3609 		btrfs_disk_key_to_cpu(&key, disk_key);
3610 
3611 		len = sizeof(*disk_key); ptr += len;
3612 		sb_ptr += len;
3613 		cur += len;
3614 
3615 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3616 			chunk = (struct btrfs_chunk *)sb_ptr;
3617 			ret = read_one_chunk(root, &key, sb, chunk);
3618 			if (ret)
3619 				break;
3620 			num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3621 			len = btrfs_chunk_item_size(num_stripes);
3622 		} else {
3623 			ret = -EIO;
3624 			break;
3625 		}
3626 		ptr += len;
3627 		sb_ptr += len;
3628 		cur += len;
3629 	}
3630 	free_extent_buffer(sb);
3631 	return ret;
3632 }
3633 
3634 int btrfs_read_chunk_tree(struct btrfs_root *root)
3635 {
3636 	struct btrfs_path *path;
3637 	struct extent_buffer *leaf;
3638 	struct btrfs_key key;
3639 	struct btrfs_key found_key;
3640 	int ret;
3641 	int slot;
3642 
3643 	root = root->fs_info->chunk_root;
3644 
3645 	path = btrfs_alloc_path();
3646 	if (!path)
3647 		return -ENOMEM;
3648 
3649 	/* first we search for all of the device items, and then we
3650 	 * read in all of the chunk items.  This way we can create chunk
3651 	 * mappings that reference all of the devices that are afound
3652 	 */
3653 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3654 	key.offset = 0;
3655 	key.type = 0;
3656 again:
3657 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3658 	if (ret < 0)
3659 		goto error;
3660 	while (1) {
3661 		leaf = path->nodes[0];
3662 		slot = path->slots[0];
3663 		if (slot >= btrfs_header_nritems(leaf)) {
3664 			ret = btrfs_next_leaf(root, path);
3665 			if (ret == 0)
3666 				continue;
3667 			if (ret < 0)
3668 				goto error;
3669 			break;
3670 		}
3671 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3672 		if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3673 			if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3674 				break;
3675 			if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3676 				struct btrfs_dev_item *dev_item;
3677 				dev_item = btrfs_item_ptr(leaf, slot,
3678 						  struct btrfs_dev_item);
3679 				ret = read_one_dev(root, leaf, dev_item);
3680 				if (ret)
3681 					goto error;
3682 			}
3683 		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3684 			struct btrfs_chunk *chunk;
3685 			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3686 			ret = read_one_chunk(root, &found_key, leaf, chunk);
3687 			if (ret)
3688 				goto error;
3689 		}
3690 		path->slots[0]++;
3691 	}
3692 	if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3693 		key.objectid = 0;
3694 		btrfs_release_path(path);
3695 		goto again;
3696 	}
3697 	ret = 0;
3698 error:
3699 	btrfs_free_path(path);
3700 	return ret;
3701 }
3702