xref: /openbmc/linux/fs/btrfs/volumes.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
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 struct map_lookup {
37 	u64 type;
38 	int io_align;
39 	int io_width;
40 	int stripe_len;
41 	int sector_size;
42 	int num_stripes;
43 	int sub_stripes;
44 	struct btrfs_bio_stripe stripes[];
45 };
46 
47 static int init_first_rw_device(struct btrfs_trans_handle *trans,
48 				struct btrfs_root *root,
49 				struct btrfs_device *device);
50 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
51 
52 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
53 			    (sizeof(struct btrfs_bio_stripe) * (n)))
54 
55 static DEFINE_MUTEX(uuid_mutex);
56 static LIST_HEAD(fs_uuids);
57 
58 void btrfs_lock_volumes(void)
59 {
60 	mutex_lock(&uuid_mutex);
61 }
62 
63 void btrfs_unlock_volumes(void)
64 {
65 	mutex_unlock(&uuid_mutex);
66 }
67 
68 static void lock_chunks(struct btrfs_root *root)
69 {
70 	mutex_lock(&root->fs_info->chunk_mutex);
71 }
72 
73 static void unlock_chunks(struct btrfs_root *root)
74 {
75 	mutex_unlock(&root->fs_info->chunk_mutex);
76 }
77 
78 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
79 {
80 	struct btrfs_device *device;
81 	WARN_ON(fs_devices->opened);
82 	while (!list_empty(&fs_devices->devices)) {
83 		device = list_entry(fs_devices->devices.next,
84 				    struct btrfs_device, dev_list);
85 		list_del(&device->dev_list);
86 		kfree(device->name);
87 		kfree(device);
88 	}
89 	kfree(fs_devices);
90 }
91 
92 int btrfs_cleanup_fs_uuids(void)
93 {
94 	struct btrfs_fs_devices *fs_devices;
95 
96 	while (!list_empty(&fs_uuids)) {
97 		fs_devices = list_entry(fs_uuids.next,
98 					struct btrfs_fs_devices, list);
99 		list_del(&fs_devices->list);
100 		free_fs_devices(fs_devices);
101 	}
102 	return 0;
103 }
104 
105 static noinline struct btrfs_device *__find_device(struct list_head *head,
106 						   u64 devid, u8 *uuid)
107 {
108 	struct btrfs_device *dev;
109 
110 	list_for_each_entry(dev, head, dev_list) {
111 		if (dev->devid == devid &&
112 		    (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
113 			return dev;
114 		}
115 	}
116 	return NULL;
117 }
118 
119 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
120 {
121 	struct btrfs_fs_devices *fs_devices;
122 
123 	list_for_each_entry(fs_devices, &fs_uuids, list) {
124 		if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
125 			return fs_devices;
126 	}
127 	return NULL;
128 }
129 
130 static void requeue_list(struct btrfs_pending_bios *pending_bios,
131 			struct bio *head, struct bio *tail)
132 {
133 
134 	struct bio *old_head;
135 
136 	old_head = pending_bios->head;
137 	pending_bios->head = head;
138 	if (pending_bios->tail)
139 		tail->bi_next = old_head;
140 	else
141 		pending_bios->tail = tail;
142 }
143 
144 /*
145  * we try to collect pending bios for a device so we don't get a large
146  * number of procs sending bios down to the same device.  This greatly
147  * improves the schedulers ability to collect and merge the bios.
148  *
149  * But, it also turns into a long list of bios to process and that is sure
150  * to eventually make the worker thread block.  The solution here is to
151  * make some progress and then put this work struct back at the end of
152  * the list if the block device is congested.  This way, multiple devices
153  * can make progress from a single worker thread.
154  */
155 static noinline int run_scheduled_bios(struct btrfs_device *device)
156 {
157 	struct bio *pending;
158 	struct backing_dev_info *bdi;
159 	struct btrfs_fs_info *fs_info;
160 	struct btrfs_pending_bios *pending_bios;
161 	struct bio *tail;
162 	struct bio *cur;
163 	int again = 0;
164 	unsigned long num_run;
165 	unsigned long num_sync_run;
166 	unsigned long batch_run = 0;
167 	unsigned long limit;
168 	unsigned long last_waited = 0;
169 	int force_reg = 0;
170 
171 	bdi = blk_get_backing_dev_info(device->bdev);
172 	fs_info = device->dev_root->fs_info;
173 	limit = btrfs_async_submit_limit(fs_info);
174 	limit = limit * 2 / 3;
175 
176 	/* we want to make sure that every time we switch from the sync
177 	 * list to the normal list, we unplug
178 	 */
179 	num_sync_run = 0;
180 
181 loop:
182 	spin_lock(&device->io_lock);
183 
184 loop_lock:
185 	num_run = 0;
186 
187 	/* take all the bios off the list at once and process them
188 	 * later on (without the lock held).  But, remember the
189 	 * tail and other pointers so the bios can be properly reinserted
190 	 * into the list if we hit congestion
191 	 */
192 	if (!force_reg && device->pending_sync_bios.head) {
193 		pending_bios = &device->pending_sync_bios;
194 		force_reg = 1;
195 	} else {
196 		pending_bios = &device->pending_bios;
197 		force_reg = 0;
198 	}
199 
200 	pending = pending_bios->head;
201 	tail = pending_bios->tail;
202 	WARN_ON(pending && !tail);
203 
204 	/*
205 	 * if pending was null this time around, no bios need processing
206 	 * at all and we can stop.  Otherwise it'll loop back up again
207 	 * and do an additional check so no bios are missed.
208 	 *
209 	 * device->running_pending is used to synchronize with the
210 	 * schedule_bio code.
211 	 */
212 	if (device->pending_sync_bios.head == NULL &&
213 	    device->pending_bios.head == NULL) {
214 		again = 0;
215 		device->running_pending = 0;
216 	} else {
217 		again = 1;
218 		device->running_pending = 1;
219 	}
220 
221 	pending_bios->head = NULL;
222 	pending_bios->tail = NULL;
223 
224 	spin_unlock(&device->io_lock);
225 
226 	/*
227 	 * if we're doing the regular priority list, make sure we unplug
228 	 * for any high prio bios we've sent down
229 	 */
230 	if (pending_bios == &device->pending_bios && num_sync_run > 0) {
231 		num_sync_run = 0;
232 		blk_run_backing_dev(bdi, NULL);
233 	}
234 
235 	while (pending) {
236 
237 		rmb();
238 		/* we want to work on both lists, but do more bios on the
239 		 * sync list than the regular list
240 		 */
241 		if ((num_run > 32 &&
242 		    pending_bios != &device->pending_sync_bios &&
243 		    device->pending_sync_bios.head) ||
244 		   (num_run > 64 && pending_bios == &device->pending_sync_bios &&
245 		    device->pending_bios.head)) {
246 			spin_lock(&device->io_lock);
247 			requeue_list(pending_bios, pending, tail);
248 			goto loop_lock;
249 		}
250 
251 		cur = pending;
252 		pending = pending->bi_next;
253 		cur->bi_next = NULL;
254 		atomic_dec(&fs_info->nr_async_bios);
255 
256 		if (atomic_read(&fs_info->nr_async_bios) < limit &&
257 		    waitqueue_active(&fs_info->async_submit_wait))
258 			wake_up(&fs_info->async_submit_wait);
259 
260 		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
261 
262 		if (cur->bi_rw & REQ_SYNC)
263 			num_sync_run++;
264 
265 		submit_bio(cur->bi_rw, cur);
266 		num_run++;
267 		batch_run++;
268 		if (need_resched()) {
269 			if (num_sync_run) {
270 				blk_run_backing_dev(bdi, NULL);
271 				num_sync_run = 0;
272 			}
273 			cond_resched();
274 		}
275 
276 		/*
277 		 * we made progress, there is more work to do and the bdi
278 		 * is now congested.  Back off and let other work structs
279 		 * run instead
280 		 */
281 		if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
282 		    fs_info->fs_devices->open_devices > 1) {
283 			struct io_context *ioc;
284 
285 			ioc = current->io_context;
286 
287 			/*
288 			 * the main goal here is that we don't want to
289 			 * block if we're going to be able to submit
290 			 * more requests without blocking.
291 			 *
292 			 * This code does two great things, it pokes into
293 			 * the elevator code from a filesystem _and_
294 			 * it makes assumptions about how batching works.
295 			 */
296 			if (ioc && ioc->nr_batch_requests > 0 &&
297 			    time_before(jiffies, ioc->last_waited + HZ/50UL) &&
298 			    (last_waited == 0 ||
299 			     ioc->last_waited == last_waited)) {
300 				/*
301 				 * we want to go through our batch of
302 				 * requests and stop.  So, we copy out
303 				 * the ioc->last_waited time and test
304 				 * against it before looping
305 				 */
306 				last_waited = ioc->last_waited;
307 				if (need_resched()) {
308 					if (num_sync_run) {
309 						blk_run_backing_dev(bdi, NULL);
310 						num_sync_run = 0;
311 					}
312 					cond_resched();
313 				}
314 				continue;
315 			}
316 			spin_lock(&device->io_lock);
317 			requeue_list(pending_bios, pending, tail);
318 			device->running_pending = 1;
319 
320 			spin_unlock(&device->io_lock);
321 			btrfs_requeue_work(&device->work);
322 			goto done;
323 		}
324 	}
325 
326 	if (num_sync_run) {
327 		num_sync_run = 0;
328 		blk_run_backing_dev(bdi, NULL);
329 	}
330 	/*
331 	 * IO has already been through a long path to get here.  Checksumming,
332 	 * async helper threads, perhaps compression.  We've done a pretty
333 	 * good job of collecting a batch of IO and should just unplug
334 	 * the device right away.
335 	 *
336 	 * This will help anyone who is waiting on the IO, they might have
337 	 * already unplugged, but managed to do so before the bio they
338 	 * cared about found its way down here.
339 	 */
340 	blk_run_backing_dev(bdi, NULL);
341 
342 	cond_resched();
343 	if (again)
344 		goto loop;
345 
346 	spin_lock(&device->io_lock);
347 	if (device->pending_bios.head || device->pending_sync_bios.head)
348 		goto loop_lock;
349 	spin_unlock(&device->io_lock);
350 
351 done:
352 	return 0;
353 }
354 
355 static void pending_bios_fn(struct btrfs_work *work)
356 {
357 	struct btrfs_device *device;
358 
359 	device = container_of(work, struct btrfs_device, work);
360 	run_scheduled_bios(device);
361 }
362 
363 static noinline int device_list_add(const char *path,
364 			   struct btrfs_super_block *disk_super,
365 			   u64 devid, struct btrfs_fs_devices **fs_devices_ret)
366 {
367 	struct btrfs_device *device;
368 	struct btrfs_fs_devices *fs_devices;
369 	u64 found_transid = btrfs_super_generation(disk_super);
370 	char *name;
371 
372 	fs_devices = find_fsid(disk_super->fsid);
373 	if (!fs_devices) {
374 		fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
375 		if (!fs_devices)
376 			return -ENOMEM;
377 		INIT_LIST_HEAD(&fs_devices->devices);
378 		INIT_LIST_HEAD(&fs_devices->alloc_list);
379 		list_add(&fs_devices->list, &fs_uuids);
380 		memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
381 		fs_devices->latest_devid = devid;
382 		fs_devices->latest_trans = found_transid;
383 		mutex_init(&fs_devices->device_list_mutex);
384 		device = NULL;
385 	} else {
386 		device = __find_device(&fs_devices->devices, devid,
387 				       disk_super->dev_item.uuid);
388 	}
389 	if (!device) {
390 		if (fs_devices->opened)
391 			return -EBUSY;
392 
393 		device = kzalloc(sizeof(*device), GFP_NOFS);
394 		if (!device) {
395 			/* we can safely leave the fs_devices entry around */
396 			return -ENOMEM;
397 		}
398 		device->devid = devid;
399 		device->work.func = pending_bios_fn;
400 		memcpy(device->uuid, disk_super->dev_item.uuid,
401 		       BTRFS_UUID_SIZE);
402 		spin_lock_init(&device->io_lock);
403 		device->name = kstrdup(path, GFP_NOFS);
404 		if (!device->name) {
405 			kfree(device);
406 			return -ENOMEM;
407 		}
408 		INIT_LIST_HEAD(&device->dev_alloc_list);
409 
410 		mutex_lock(&fs_devices->device_list_mutex);
411 		list_add(&device->dev_list, &fs_devices->devices);
412 		mutex_unlock(&fs_devices->device_list_mutex);
413 
414 		device->fs_devices = fs_devices;
415 		fs_devices->num_devices++;
416 	} else if (!device->name || strcmp(device->name, path)) {
417 		name = kstrdup(path, GFP_NOFS);
418 		if (!name)
419 			return -ENOMEM;
420 		kfree(device->name);
421 		device->name = name;
422 		if (device->missing) {
423 			fs_devices->missing_devices--;
424 			device->missing = 0;
425 		}
426 	}
427 
428 	if (found_transid > fs_devices->latest_trans) {
429 		fs_devices->latest_devid = devid;
430 		fs_devices->latest_trans = found_transid;
431 	}
432 	*fs_devices_ret = fs_devices;
433 	return 0;
434 }
435 
436 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
437 {
438 	struct btrfs_fs_devices *fs_devices;
439 	struct btrfs_device *device;
440 	struct btrfs_device *orig_dev;
441 
442 	fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
443 	if (!fs_devices)
444 		return ERR_PTR(-ENOMEM);
445 
446 	INIT_LIST_HEAD(&fs_devices->devices);
447 	INIT_LIST_HEAD(&fs_devices->alloc_list);
448 	INIT_LIST_HEAD(&fs_devices->list);
449 	mutex_init(&fs_devices->device_list_mutex);
450 	fs_devices->latest_devid = orig->latest_devid;
451 	fs_devices->latest_trans = orig->latest_trans;
452 	memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
453 
454 	mutex_lock(&orig->device_list_mutex);
455 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
456 		device = kzalloc(sizeof(*device), GFP_NOFS);
457 		if (!device)
458 			goto error;
459 
460 		device->name = kstrdup(orig_dev->name, GFP_NOFS);
461 		if (!device->name) {
462 			kfree(device);
463 			goto error;
464 		}
465 
466 		device->devid = orig_dev->devid;
467 		device->work.func = pending_bios_fn;
468 		memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
469 		spin_lock_init(&device->io_lock);
470 		INIT_LIST_HEAD(&device->dev_list);
471 		INIT_LIST_HEAD(&device->dev_alloc_list);
472 
473 		list_add(&device->dev_list, &fs_devices->devices);
474 		device->fs_devices = fs_devices;
475 		fs_devices->num_devices++;
476 	}
477 	mutex_unlock(&orig->device_list_mutex);
478 	return fs_devices;
479 error:
480 	mutex_unlock(&orig->device_list_mutex);
481 	free_fs_devices(fs_devices);
482 	return ERR_PTR(-ENOMEM);
483 }
484 
485 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
486 {
487 	struct btrfs_device *device, *next;
488 
489 	mutex_lock(&uuid_mutex);
490 again:
491 	mutex_lock(&fs_devices->device_list_mutex);
492 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
493 		if (device->in_fs_metadata)
494 			continue;
495 
496 		if (device->bdev) {
497 			blkdev_put(device->bdev, device->mode);
498 			device->bdev = NULL;
499 			fs_devices->open_devices--;
500 		}
501 		if (device->writeable) {
502 			list_del_init(&device->dev_alloc_list);
503 			device->writeable = 0;
504 			fs_devices->rw_devices--;
505 		}
506 		list_del_init(&device->dev_list);
507 		fs_devices->num_devices--;
508 		kfree(device->name);
509 		kfree(device);
510 	}
511 	mutex_unlock(&fs_devices->device_list_mutex);
512 
513 	if (fs_devices->seed) {
514 		fs_devices = fs_devices->seed;
515 		goto again;
516 	}
517 
518 	mutex_unlock(&uuid_mutex);
519 	return 0;
520 }
521 
522 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
523 {
524 	struct btrfs_device *device;
525 
526 	if (--fs_devices->opened > 0)
527 		return 0;
528 
529 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
530 		if (device->bdev) {
531 			blkdev_put(device->bdev, device->mode);
532 			fs_devices->open_devices--;
533 		}
534 		if (device->writeable) {
535 			list_del_init(&device->dev_alloc_list);
536 			fs_devices->rw_devices--;
537 		}
538 
539 		device->bdev = NULL;
540 		device->writeable = 0;
541 		device->in_fs_metadata = 0;
542 	}
543 	WARN_ON(fs_devices->open_devices);
544 	WARN_ON(fs_devices->rw_devices);
545 	fs_devices->opened = 0;
546 	fs_devices->seeding = 0;
547 
548 	return 0;
549 }
550 
551 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
552 {
553 	struct btrfs_fs_devices *seed_devices = NULL;
554 	int ret;
555 
556 	mutex_lock(&uuid_mutex);
557 	ret = __btrfs_close_devices(fs_devices);
558 	if (!fs_devices->opened) {
559 		seed_devices = fs_devices->seed;
560 		fs_devices->seed = NULL;
561 	}
562 	mutex_unlock(&uuid_mutex);
563 
564 	while (seed_devices) {
565 		fs_devices = seed_devices;
566 		seed_devices = fs_devices->seed;
567 		__btrfs_close_devices(fs_devices);
568 		free_fs_devices(fs_devices);
569 	}
570 	return ret;
571 }
572 
573 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
574 				fmode_t flags, void *holder)
575 {
576 	struct block_device *bdev;
577 	struct list_head *head = &fs_devices->devices;
578 	struct btrfs_device *device;
579 	struct block_device *latest_bdev = NULL;
580 	struct buffer_head *bh;
581 	struct btrfs_super_block *disk_super;
582 	u64 latest_devid = 0;
583 	u64 latest_transid = 0;
584 	u64 devid;
585 	int seeding = 1;
586 	int ret = 0;
587 
588 	flags |= FMODE_EXCL;
589 
590 	list_for_each_entry(device, head, dev_list) {
591 		if (device->bdev)
592 			continue;
593 		if (!device->name)
594 			continue;
595 
596 		bdev = blkdev_get_by_path(device->name, flags, holder);
597 		if (IS_ERR(bdev)) {
598 			printk(KERN_INFO "open %s failed\n", device->name);
599 			goto error;
600 		}
601 		set_blocksize(bdev, 4096);
602 
603 		bh = btrfs_read_dev_super(bdev);
604 		if (!bh) {
605 			ret = -EINVAL;
606 			goto error_close;
607 		}
608 
609 		disk_super = (struct btrfs_super_block *)bh->b_data;
610 		devid = btrfs_stack_device_id(&disk_super->dev_item);
611 		if (devid != device->devid)
612 			goto error_brelse;
613 
614 		if (memcmp(device->uuid, disk_super->dev_item.uuid,
615 			   BTRFS_UUID_SIZE))
616 			goto error_brelse;
617 
618 		device->generation = btrfs_super_generation(disk_super);
619 		if (!latest_transid || device->generation > latest_transid) {
620 			latest_devid = devid;
621 			latest_transid = device->generation;
622 			latest_bdev = bdev;
623 		}
624 
625 		if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
626 			device->writeable = 0;
627 		} else {
628 			device->writeable = !bdev_read_only(bdev);
629 			seeding = 0;
630 		}
631 
632 		device->bdev = bdev;
633 		device->in_fs_metadata = 0;
634 		device->mode = flags;
635 
636 		if (!blk_queue_nonrot(bdev_get_queue(bdev)))
637 			fs_devices->rotating = 1;
638 
639 		fs_devices->open_devices++;
640 		if (device->writeable) {
641 			fs_devices->rw_devices++;
642 			list_add(&device->dev_alloc_list,
643 				 &fs_devices->alloc_list);
644 		}
645 		continue;
646 
647 error_brelse:
648 		brelse(bh);
649 error_close:
650 		blkdev_put(bdev, flags);
651 error:
652 		continue;
653 	}
654 	if (fs_devices->open_devices == 0) {
655 		ret = -EIO;
656 		goto out;
657 	}
658 	fs_devices->seeding = seeding;
659 	fs_devices->opened = 1;
660 	fs_devices->latest_bdev = latest_bdev;
661 	fs_devices->latest_devid = latest_devid;
662 	fs_devices->latest_trans = latest_transid;
663 	fs_devices->total_rw_bytes = 0;
664 out:
665 	return ret;
666 }
667 
668 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
669 		       fmode_t flags, void *holder)
670 {
671 	int ret;
672 
673 	mutex_lock(&uuid_mutex);
674 	if (fs_devices->opened) {
675 		fs_devices->opened++;
676 		ret = 0;
677 	} else {
678 		ret = __btrfs_open_devices(fs_devices, flags, holder);
679 	}
680 	mutex_unlock(&uuid_mutex);
681 	return ret;
682 }
683 
684 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
685 			  struct btrfs_fs_devices **fs_devices_ret)
686 {
687 	struct btrfs_super_block *disk_super;
688 	struct block_device *bdev;
689 	struct buffer_head *bh;
690 	int ret;
691 	u64 devid;
692 	u64 transid;
693 
694 	mutex_lock(&uuid_mutex);
695 
696 	flags |= FMODE_EXCL;
697 	bdev = blkdev_get_by_path(path, flags, holder);
698 
699 	if (IS_ERR(bdev)) {
700 		ret = PTR_ERR(bdev);
701 		goto error;
702 	}
703 
704 	ret = set_blocksize(bdev, 4096);
705 	if (ret)
706 		goto error_close;
707 	bh = btrfs_read_dev_super(bdev);
708 	if (!bh) {
709 		ret = -EINVAL;
710 		goto error_close;
711 	}
712 	disk_super = (struct btrfs_super_block *)bh->b_data;
713 	devid = btrfs_stack_device_id(&disk_super->dev_item);
714 	transid = btrfs_super_generation(disk_super);
715 	if (disk_super->label[0])
716 		printk(KERN_INFO "device label %s ", disk_super->label);
717 	else {
718 		/* FIXME, make a readl uuid parser */
719 		printk(KERN_INFO "device fsid %llx-%llx ",
720 		       *(unsigned long long *)disk_super->fsid,
721 		       *(unsigned long long *)(disk_super->fsid + 8));
722 	}
723 	printk(KERN_CONT "devid %llu transid %llu %s\n",
724 	       (unsigned long long)devid, (unsigned long long)transid, path);
725 	ret = device_list_add(path, disk_super, devid, fs_devices_ret);
726 
727 	brelse(bh);
728 error_close:
729 	blkdev_put(bdev, flags);
730 error:
731 	mutex_unlock(&uuid_mutex);
732 	return ret;
733 }
734 
735 /* helper to account the used device space in the range */
736 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
737 				   u64 end, u64 *length)
738 {
739 	struct btrfs_key key;
740 	struct btrfs_root *root = device->dev_root;
741 	struct btrfs_dev_extent *dev_extent;
742 	struct btrfs_path *path;
743 	u64 extent_end;
744 	int ret;
745 	int slot;
746 	struct extent_buffer *l;
747 
748 	*length = 0;
749 
750 	if (start >= device->total_bytes)
751 		return 0;
752 
753 	path = btrfs_alloc_path();
754 	if (!path)
755 		return -ENOMEM;
756 	path->reada = 2;
757 
758 	key.objectid = device->devid;
759 	key.offset = start;
760 	key.type = BTRFS_DEV_EXTENT_KEY;
761 
762 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
763 	if (ret < 0)
764 		goto out;
765 	if (ret > 0) {
766 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
767 		if (ret < 0)
768 			goto out;
769 	}
770 
771 	while (1) {
772 		l = path->nodes[0];
773 		slot = path->slots[0];
774 		if (slot >= btrfs_header_nritems(l)) {
775 			ret = btrfs_next_leaf(root, path);
776 			if (ret == 0)
777 				continue;
778 			if (ret < 0)
779 				goto out;
780 
781 			break;
782 		}
783 		btrfs_item_key_to_cpu(l, &key, slot);
784 
785 		if (key.objectid < device->devid)
786 			goto next;
787 
788 		if (key.objectid > device->devid)
789 			break;
790 
791 		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
792 			goto next;
793 
794 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
795 		extent_end = key.offset + btrfs_dev_extent_length(l,
796 								  dev_extent);
797 		if (key.offset <= start && extent_end > end) {
798 			*length = end - start + 1;
799 			break;
800 		} else if (key.offset <= start && extent_end > start)
801 			*length += extent_end - start;
802 		else if (key.offset > start && extent_end <= end)
803 			*length += extent_end - key.offset;
804 		else if (key.offset > start && key.offset <= end) {
805 			*length += end - key.offset + 1;
806 			break;
807 		} else if (key.offset > end)
808 			break;
809 
810 next:
811 		path->slots[0]++;
812 	}
813 	ret = 0;
814 out:
815 	btrfs_free_path(path);
816 	return ret;
817 }
818 
819 /*
820  * find_free_dev_extent - find free space in the specified device
821  * @trans:	transaction handler
822  * @device:	the device which we search the free space in
823  * @num_bytes:	the size of the free space that we need
824  * @start:	store the start of the free space.
825  * @len:	the size of the free space. that we find, or the size of the max
826  * 		free space if we don't find suitable free space
827  *
828  * this uses a pretty simple search, the expectation is that it is
829  * called very infrequently and that a given device has a small number
830  * of extents
831  *
832  * @start is used to store the start of the free space if we find. But if we
833  * don't find suitable free space, it will be used to store the start position
834  * of the max free space.
835  *
836  * @len is used to store the size of the free space that we find.
837  * But if we don't find suitable free space, it is used to store the size of
838  * the max free space.
839  */
840 int find_free_dev_extent(struct btrfs_trans_handle *trans,
841 			 struct btrfs_device *device, u64 num_bytes,
842 			 u64 *start, u64 *len)
843 {
844 	struct btrfs_key key;
845 	struct btrfs_root *root = device->dev_root;
846 	struct btrfs_dev_extent *dev_extent;
847 	struct btrfs_path *path;
848 	u64 hole_size;
849 	u64 max_hole_start;
850 	u64 max_hole_size;
851 	u64 extent_end;
852 	u64 search_start;
853 	u64 search_end = device->total_bytes;
854 	int ret;
855 	int slot;
856 	struct extent_buffer *l;
857 
858 	/* FIXME use last free of some kind */
859 
860 	/* we don't want to overwrite the superblock on the drive,
861 	 * so we make sure to start at an offset of at least 1MB
862 	 */
863 	search_start = 1024 * 1024;
864 
865 	if (root->fs_info->alloc_start + num_bytes <= search_end)
866 		search_start = max(root->fs_info->alloc_start, search_start);
867 
868 	max_hole_start = search_start;
869 	max_hole_size = 0;
870 
871 	if (search_start >= search_end) {
872 		ret = -ENOSPC;
873 		goto error;
874 	}
875 
876 	path = btrfs_alloc_path();
877 	if (!path) {
878 		ret = -ENOMEM;
879 		goto error;
880 	}
881 	path->reada = 2;
882 
883 	key.objectid = device->devid;
884 	key.offset = search_start;
885 	key.type = BTRFS_DEV_EXTENT_KEY;
886 
887 	ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
888 	if (ret < 0)
889 		goto out;
890 	if (ret > 0) {
891 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
892 		if (ret < 0)
893 			goto out;
894 	}
895 
896 	while (1) {
897 		l = path->nodes[0];
898 		slot = path->slots[0];
899 		if (slot >= btrfs_header_nritems(l)) {
900 			ret = btrfs_next_leaf(root, path);
901 			if (ret == 0)
902 				continue;
903 			if (ret < 0)
904 				goto out;
905 
906 			break;
907 		}
908 		btrfs_item_key_to_cpu(l, &key, slot);
909 
910 		if (key.objectid < device->devid)
911 			goto next;
912 
913 		if (key.objectid > device->devid)
914 			break;
915 
916 		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
917 			goto next;
918 
919 		if (key.offset > search_start) {
920 			hole_size = key.offset - search_start;
921 
922 			if (hole_size > max_hole_size) {
923 				max_hole_start = search_start;
924 				max_hole_size = hole_size;
925 			}
926 
927 			/*
928 			 * If this free space is greater than which we need,
929 			 * it must be the max free space that we have found
930 			 * until now, so max_hole_start must point to the start
931 			 * of this free space and the length of this free space
932 			 * is stored in max_hole_size. Thus, we return
933 			 * max_hole_start and max_hole_size and go back to the
934 			 * caller.
935 			 */
936 			if (hole_size >= num_bytes) {
937 				ret = 0;
938 				goto out;
939 			}
940 		}
941 
942 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
943 		extent_end = key.offset + btrfs_dev_extent_length(l,
944 								  dev_extent);
945 		if (extent_end > search_start)
946 			search_start = extent_end;
947 next:
948 		path->slots[0]++;
949 		cond_resched();
950 	}
951 
952 	hole_size = search_end- search_start;
953 	if (hole_size > max_hole_size) {
954 		max_hole_start = search_start;
955 		max_hole_size = hole_size;
956 	}
957 
958 	/* See above. */
959 	if (hole_size < num_bytes)
960 		ret = -ENOSPC;
961 	else
962 		ret = 0;
963 
964 out:
965 	btrfs_free_path(path);
966 error:
967 	*start = max_hole_start;
968 	if (len)
969 		*len = max_hole_size;
970 	return ret;
971 }
972 
973 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
974 			  struct btrfs_device *device,
975 			  u64 start)
976 {
977 	int ret;
978 	struct btrfs_path *path;
979 	struct btrfs_root *root = device->dev_root;
980 	struct btrfs_key key;
981 	struct btrfs_key found_key;
982 	struct extent_buffer *leaf = NULL;
983 	struct btrfs_dev_extent *extent = NULL;
984 
985 	path = btrfs_alloc_path();
986 	if (!path)
987 		return -ENOMEM;
988 
989 	key.objectid = device->devid;
990 	key.offset = start;
991 	key.type = BTRFS_DEV_EXTENT_KEY;
992 
993 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
994 	if (ret > 0) {
995 		ret = btrfs_previous_item(root, path, key.objectid,
996 					  BTRFS_DEV_EXTENT_KEY);
997 		BUG_ON(ret);
998 		leaf = path->nodes[0];
999 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1000 		extent = btrfs_item_ptr(leaf, path->slots[0],
1001 					struct btrfs_dev_extent);
1002 		BUG_ON(found_key.offset > start || found_key.offset +
1003 		       btrfs_dev_extent_length(leaf, extent) < start);
1004 		ret = 0;
1005 	} else if (ret == 0) {
1006 		leaf = path->nodes[0];
1007 		extent = btrfs_item_ptr(leaf, path->slots[0],
1008 					struct btrfs_dev_extent);
1009 	}
1010 	BUG_ON(ret);
1011 
1012 	if (device->bytes_used > 0)
1013 		device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
1014 	ret = btrfs_del_item(trans, root, path);
1015 	BUG_ON(ret);
1016 
1017 	btrfs_free_path(path);
1018 	return ret;
1019 }
1020 
1021 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1022 			   struct btrfs_device *device,
1023 			   u64 chunk_tree, u64 chunk_objectid,
1024 			   u64 chunk_offset, u64 start, u64 num_bytes)
1025 {
1026 	int ret;
1027 	struct btrfs_path *path;
1028 	struct btrfs_root *root = device->dev_root;
1029 	struct btrfs_dev_extent *extent;
1030 	struct extent_buffer *leaf;
1031 	struct btrfs_key key;
1032 
1033 	WARN_ON(!device->in_fs_metadata);
1034 	path = btrfs_alloc_path();
1035 	if (!path)
1036 		return -ENOMEM;
1037 
1038 	key.objectid = device->devid;
1039 	key.offset = start;
1040 	key.type = BTRFS_DEV_EXTENT_KEY;
1041 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1042 				      sizeof(*extent));
1043 	BUG_ON(ret);
1044 
1045 	leaf = path->nodes[0];
1046 	extent = btrfs_item_ptr(leaf, path->slots[0],
1047 				struct btrfs_dev_extent);
1048 	btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1049 	btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1050 	btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1051 
1052 	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1053 		    (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1054 		    BTRFS_UUID_SIZE);
1055 
1056 	btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1057 	btrfs_mark_buffer_dirty(leaf);
1058 	btrfs_free_path(path);
1059 	return ret;
1060 }
1061 
1062 static noinline int find_next_chunk(struct btrfs_root *root,
1063 				    u64 objectid, u64 *offset)
1064 {
1065 	struct btrfs_path *path;
1066 	int ret;
1067 	struct btrfs_key key;
1068 	struct btrfs_chunk *chunk;
1069 	struct btrfs_key found_key;
1070 
1071 	path = btrfs_alloc_path();
1072 	BUG_ON(!path);
1073 
1074 	key.objectid = objectid;
1075 	key.offset = (u64)-1;
1076 	key.type = BTRFS_CHUNK_ITEM_KEY;
1077 
1078 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1079 	if (ret < 0)
1080 		goto error;
1081 
1082 	BUG_ON(ret == 0);
1083 
1084 	ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1085 	if (ret) {
1086 		*offset = 0;
1087 	} else {
1088 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1089 				      path->slots[0]);
1090 		if (found_key.objectid != objectid)
1091 			*offset = 0;
1092 		else {
1093 			chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1094 					       struct btrfs_chunk);
1095 			*offset = found_key.offset +
1096 				btrfs_chunk_length(path->nodes[0], chunk);
1097 		}
1098 	}
1099 	ret = 0;
1100 error:
1101 	btrfs_free_path(path);
1102 	return ret;
1103 }
1104 
1105 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1106 {
1107 	int ret;
1108 	struct btrfs_key key;
1109 	struct btrfs_key found_key;
1110 	struct btrfs_path *path;
1111 
1112 	root = root->fs_info->chunk_root;
1113 
1114 	path = btrfs_alloc_path();
1115 	if (!path)
1116 		return -ENOMEM;
1117 
1118 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1119 	key.type = BTRFS_DEV_ITEM_KEY;
1120 	key.offset = (u64)-1;
1121 
1122 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1123 	if (ret < 0)
1124 		goto error;
1125 
1126 	BUG_ON(ret == 0);
1127 
1128 	ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1129 				  BTRFS_DEV_ITEM_KEY);
1130 	if (ret) {
1131 		*objectid = 1;
1132 	} else {
1133 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1134 				      path->slots[0]);
1135 		*objectid = found_key.offset + 1;
1136 	}
1137 	ret = 0;
1138 error:
1139 	btrfs_free_path(path);
1140 	return ret;
1141 }
1142 
1143 /*
1144  * the device information is stored in the chunk root
1145  * the btrfs_device struct should be fully filled in
1146  */
1147 int btrfs_add_device(struct btrfs_trans_handle *trans,
1148 		     struct btrfs_root *root,
1149 		     struct btrfs_device *device)
1150 {
1151 	int ret;
1152 	struct btrfs_path *path;
1153 	struct btrfs_dev_item *dev_item;
1154 	struct extent_buffer *leaf;
1155 	struct btrfs_key key;
1156 	unsigned long ptr;
1157 
1158 	root = root->fs_info->chunk_root;
1159 
1160 	path = btrfs_alloc_path();
1161 	if (!path)
1162 		return -ENOMEM;
1163 
1164 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1165 	key.type = BTRFS_DEV_ITEM_KEY;
1166 	key.offset = device->devid;
1167 
1168 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1169 				      sizeof(*dev_item));
1170 	if (ret)
1171 		goto out;
1172 
1173 	leaf = path->nodes[0];
1174 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1175 
1176 	btrfs_set_device_id(leaf, dev_item, device->devid);
1177 	btrfs_set_device_generation(leaf, dev_item, 0);
1178 	btrfs_set_device_type(leaf, dev_item, device->type);
1179 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1180 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1181 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1182 	btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1183 	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1184 	btrfs_set_device_group(leaf, dev_item, 0);
1185 	btrfs_set_device_seek_speed(leaf, dev_item, 0);
1186 	btrfs_set_device_bandwidth(leaf, dev_item, 0);
1187 	btrfs_set_device_start_offset(leaf, dev_item, 0);
1188 
1189 	ptr = (unsigned long)btrfs_device_uuid(dev_item);
1190 	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1191 	ptr = (unsigned long)btrfs_device_fsid(dev_item);
1192 	write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1193 	btrfs_mark_buffer_dirty(leaf);
1194 
1195 	ret = 0;
1196 out:
1197 	btrfs_free_path(path);
1198 	return ret;
1199 }
1200 
1201 static int btrfs_rm_dev_item(struct btrfs_root *root,
1202 			     struct btrfs_device *device)
1203 {
1204 	int ret;
1205 	struct btrfs_path *path;
1206 	struct btrfs_key key;
1207 	struct btrfs_trans_handle *trans;
1208 
1209 	root = root->fs_info->chunk_root;
1210 
1211 	path = btrfs_alloc_path();
1212 	if (!path)
1213 		return -ENOMEM;
1214 
1215 	trans = btrfs_start_transaction(root, 0);
1216 	if (IS_ERR(trans)) {
1217 		btrfs_free_path(path);
1218 		return PTR_ERR(trans);
1219 	}
1220 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1221 	key.type = BTRFS_DEV_ITEM_KEY;
1222 	key.offset = device->devid;
1223 	lock_chunks(root);
1224 
1225 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1226 	if (ret < 0)
1227 		goto out;
1228 
1229 	if (ret > 0) {
1230 		ret = -ENOENT;
1231 		goto out;
1232 	}
1233 
1234 	ret = btrfs_del_item(trans, root, path);
1235 	if (ret)
1236 		goto out;
1237 out:
1238 	btrfs_free_path(path);
1239 	unlock_chunks(root);
1240 	btrfs_commit_transaction(trans, root);
1241 	return ret;
1242 }
1243 
1244 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1245 {
1246 	struct btrfs_device *device;
1247 	struct btrfs_device *next_device;
1248 	struct block_device *bdev;
1249 	struct buffer_head *bh = NULL;
1250 	struct btrfs_super_block *disk_super;
1251 	u64 all_avail;
1252 	u64 devid;
1253 	u64 num_devices;
1254 	u8 *dev_uuid;
1255 	int ret = 0;
1256 
1257 	mutex_lock(&uuid_mutex);
1258 	mutex_lock(&root->fs_info->volume_mutex);
1259 
1260 	all_avail = root->fs_info->avail_data_alloc_bits |
1261 		root->fs_info->avail_system_alloc_bits |
1262 		root->fs_info->avail_metadata_alloc_bits;
1263 
1264 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1265 	    root->fs_info->fs_devices->num_devices <= 4) {
1266 		printk(KERN_ERR "btrfs: unable to go below four devices "
1267 		       "on raid10\n");
1268 		ret = -EINVAL;
1269 		goto out;
1270 	}
1271 
1272 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1273 	    root->fs_info->fs_devices->num_devices <= 2) {
1274 		printk(KERN_ERR "btrfs: unable to go below two "
1275 		       "devices on raid1\n");
1276 		ret = -EINVAL;
1277 		goto out;
1278 	}
1279 
1280 	if (strcmp(device_path, "missing") == 0) {
1281 		struct list_head *devices;
1282 		struct btrfs_device *tmp;
1283 
1284 		device = NULL;
1285 		devices = &root->fs_info->fs_devices->devices;
1286 		mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1287 		list_for_each_entry(tmp, devices, dev_list) {
1288 			if (tmp->in_fs_metadata && !tmp->bdev) {
1289 				device = tmp;
1290 				break;
1291 			}
1292 		}
1293 		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1294 		bdev = NULL;
1295 		bh = NULL;
1296 		disk_super = NULL;
1297 		if (!device) {
1298 			printk(KERN_ERR "btrfs: no missing devices found to "
1299 			       "remove\n");
1300 			goto out;
1301 		}
1302 	} else {
1303 		bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1304 					  root->fs_info->bdev_holder);
1305 		if (IS_ERR(bdev)) {
1306 			ret = PTR_ERR(bdev);
1307 			goto out;
1308 		}
1309 
1310 		set_blocksize(bdev, 4096);
1311 		bh = btrfs_read_dev_super(bdev);
1312 		if (!bh) {
1313 			ret = -EINVAL;
1314 			goto error_close;
1315 		}
1316 		disk_super = (struct btrfs_super_block *)bh->b_data;
1317 		devid = btrfs_stack_device_id(&disk_super->dev_item);
1318 		dev_uuid = disk_super->dev_item.uuid;
1319 		device = btrfs_find_device(root, devid, dev_uuid,
1320 					   disk_super->fsid);
1321 		if (!device) {
1322 			ret = -ENOENT;
1323 			goto error_brelse;
1324 		}
1325 	}
1326 
1327 	if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1328 		printk(KERN_ERR "btrfs: unable to remove the only writeable "
1329 		       "device\n");
1330 		ret = -EINVAL;
1331 		goto error_brelse;
1332 	}
1333 
1334 	if (device->writeable) {
1335 		list_del_init(&device->dev_alloc_list);
1336 		root->fs_info->fs_devices->rw_devices--;
1337 	}
1338 
1339 	ret = btrfs_shrink_device(device, 0);
1340 	if (ret)
1341 		goto error_brelse;
1342 
1343 	ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1344 	if (ret)
1345 		goto error_brelse;
1346 
1347 	device->in_fs_metadata = 0;
1348 
1349 	/*
1350 	 * the device list mutex makes sure that we don't change
1351 	 * the device list while someone else is writing out all
1352 	 * the device supers.
1353 	 */
1354 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1355 	list_del_init(&device->dev_list);
1356 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1357 
1358 	device->fs_devices->num_devices--;
1359 
1360 	if (device->missing)
1361 		root->fs_info->fs_devices->missing_devices--;
1362 
1363 	next_device = list_entry(root->fs_info->fs_devices->devices.next,
1364 				 struct btrfs_device, dev_list);
1365 	if (device->bdev == root->fs_info->sb->s_bdev)
1366 		root->fs_info->sb->s_bdev = next_device->bdev;
1367 	if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1368 		root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1369 
1370 	if (device->bdev) {
1371 		blkdev_put(device->bdev, device->mode);
1372 		device->bdev = NULL;
1373 		device->fs_devices->open_devices--;
1374 	}
1375 
1376 	num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1377 	btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1378 
1379 	if (device->fs_devices->open_devices == 0) {
1380 		struct btrfs_fs_devices *fs_devices;
1381 		fs_devices = root->fs_info->fs_devices;
1382 		while (fs_devices) {
1383 			if (fs_devices->seed == device->fs_devices)
1384 				break;
1385 			fs_devices = fs_devices->seed;
1386 		}
1387 		fs_devices->seed = device->fs_devices->seed;
1388 		device->fs_devices->seed = NULL;
1389 		__btrfs_close_devices(device->fs_devices);
1390 		free_fs_devices(device->fs_devices);
1391 	}
1392 
1393 	/*
1394 	 * at this point, the device is zero sized.  We want to
1395 	 * remove it from the devices list and zero out the old super
1396 	 */
1397 	if (device->writeable) {
1398 		/* make sure this device isn't detected as part of
1399 		 * the FS anymore
1400 		 */
1401 		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1402 		set_buffer_dirty(bh);
1403 		sync_dirty_buffer(bh);
1404 	}
1405 
1406 	kfree(device->name);
1407 	kfree(device);
1408 	ret = 0;
1409 
1410 error_brelse:
1411 	brelse(bh);
1412 error_close:
1413 	if (bdev)
1414 		blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1415 out:
1416 	mutex_unlock(&root->fs_info->volume_mutex);
1417 	mutex_unlock(&uuid_mutex);
1418 	return ret;
1419 }
1420 
1421 /*
1422  * does all the dirty work required for changing file system's UUID.
1423  */
1424 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1425 				struct btrfs_root *root)
1426 {
1427 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1428 	struct btrfs_fs_devices *old_devices;
1429 	struct btrfs_fs_devices *seed_devices;
1430 	struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1431 	struct btrfs_device *device;
1432 	u64 super_flags;
1433 
1434 	BUG_ON(!mutex_is_locked(&uuid_mutex));
1435 	if (!fs_devices->seeding)
1436 		return -EINVAL;
1437 
1438 	seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1439 	if (!seed_devices)
1440 		return -ENOMEM;
1441 
1442 	old_devices = clone_fs_devices(fs_devices);
1443 	if (IS_ERR(old_devices)) {
1444 		kfree(seed_devices);
1445 		return PTR_ERR(old_devices);
1446 	}
1447 
1448 	list_add(&old_devices->list, &fs_uuids);
1449 
1450 	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1451 	seed_devices->opened = 1;
1452 	INIT_LIST_HEAD(&seed_devices->devices);
1453 	INIT_LIST_HEAD(&seed_devices->alloc_list);
1454 	mutex_init(&seed_devices->device_list_mutex);
1455 	list_splice_init(&fs_devices->devices, &seed_devices->devices);
1456 	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1457 	list_for_each_entry(device, &seed_devices->devices, dev_list) {
1458 		device->fs_devices = seed_devices;
1459 	}
1460 
1461 	fs_devices->seeding = 0;
1462 	fs_devices->num_devices = 0;
1463 	fs_devices->open_devices = 0;
1464 	fs_devices->seed = seed_devices;
1465 
1466 	generate_random_uuid(fs_devices->fsid);
1467 	memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1468 	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1469 	super_flags = btrfs_super_flags(disk_super) &
1470 		      ~BTRFS_SUPER_FLAG_SEEDING;
1471 	btrfs_set_super_flags(disk_super, super_flags);
1472 
1473 	return 0;
1474 }
1475 
1476 /*
1477  * strore the expected generation for seed devices in device items.
1478  */
1479 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1480 			       struct btrfs_root *root)
1481 {
1482 	struct btrfs_path *path;
1483 	struct extent_buffer *leaf;
1484 	struct btrfs_dev_item *dev_item;
1485 	struct btrfs_device *device;
1486 	struct btrfs_key key;
1487 	u8 fs_uuid[BTRFS_UUID_SIZE];
1488 	u8 dev_uuid[BTRFS_UUID_SIZE];
1489 	u64 devid;
1490 	int ret;
1491 
1492 	path = btrfs_alloc_path();
1493 	if (!path)
1494 		return -ENOMEM;
1495 
1496 	root = root->fs_info->chunk_root;
1497 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1498 	key.offset = 0;
1499 	key.type = BTRFS_DEV_ITEM_KEY;
1500 
1501 	while (1) {
1502 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1503 		if (ret < 0)
1504 			goto error;
1505 
1506 		leaf = path->nodes[0];
1507 next_slot:
1508 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1509 			ret = btrfs_next_leaf(root, path);
1510 			if (ret > 0)
1511 				break;
1512 			if (ret < 0)
1513 				goto error;
1514 			leaf = path->nodes[0];
1515 			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1516 			btrfs_release_path(root, path);
1517 			continue;
1518 		}
1519 
1520 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1521 		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1522 		    key.type != BTRFS_DEV_ITEM_KEY)
1523 			break;
1524 
1525 		dev_item = btrfs_item_ptr(leaf, path->slots[0],
1526 					  struct btrfs_dev_item);
1527 		devid = btrfs_device_id(leaf, dev_item);
1528 		read_extent_buffer(leaf, dev_uuid,
1529 				   (unsigned long)btrfs_device_uuid(dev_item),
1530 				   BTRFS_UUID_SIZE);
1531 		read_extent_buffer(leaf, fs_uuid,
1532 				   (unsigned long)btrfs_device_fsid(dev_item),
1533 				   BTRFS_UUID_SIZE);
1534 		device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1535 		BUG_ON(!device);
1536 
1537 		if (device->fs_devices->seeding) {
1538 			btrfs_set_device_generation(leaf, dev_item,
1539 						    device->generation);
1540 			btrfs_mark_buffer_dirty(leaf);
1541 		}
1542 
1543 		path->slots[0]++;
1544 		goto next_slot;
1545 	}
1546 	ret = 0;
1547 error:
1548 	btrfs_free_path(path);
1549 	return ret;
1550 }
1551 
1552 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1553 {
1554 	struct btrfs_trans_handle *trans;
1555 	struct btrfs_device *device;
1556 	struct block_device *bdev;
1557 	struct list_head *devices;
1558 	struct super_block *sb = root->fs_info->sb;
1559 	u64 total_bytes;
1560 	int seeding_dev = 0;
1561 	int ret = 0;
1562 
1563 	if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1564 		return -EINVAL;
1565 
1566 	bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1567 				  root->fs_info->bdev_holder);
1568 	if (IS_ERR(bdev))
1569 		return PTR_ERR(bdev);
1570 
1571 	if (root->fs_info->fs_devices->seeding) {
1572 		seeding_dev = 1;
1573 		down_write(&sb->s_umount);
1574 		mutex_lock(&uuid_mutex);
1575 	}
1576 
1577 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
1578 	mutex_lock(&root->fs_info->volume_mutex);
1579 
1580 	devices = &root->fs_info->fs_devices->devices;
1581 	/*
1582 	 * we have the volume lock, so we don't need the extra
1583 	 * device list mutex while reading the list here.
1584 	 */
1585 	list_for_each_entry(device, devices, dev_list) {
1586 		if (device->bdev == bdev) {
1587 			ret = -EEXIST;
1588 			goto error;
1589 		}
1590 	}
1591 
1592 	device = kzalloc(sizeof(*device), GFP_NOFS);
1593 	if (!device) {
1594 		/* we can safely leave the fs_devices entry around */
1595 		ret = -ENOMEM;
1596 		goto error;
1597 	}
1598 
1599 	device->name = kstrdup(device_path, GFP_NOFS);
1600 	if (!device->name) {
1601 		kfree(device);
1602 		ret = -ENOMEM;
1603 		goto error;
1604 	}
1605 
1606 	ret = find_next_devid(root, &device->devid);
1607 	if (ret) {
1608 		kfree(device->name);
1609 		kfree(device);
1610 		goto error;
1611 	}
1612 
1613 	trans = btrfs_start_transaction(root, 0);
1614 	if (IS_ERR(trans)) {
1615 		kfree(device->name);
1616 		kfree(device);
1617 		ret = PTR_ERR(trans);
1618 		goto error;
1619 	}
1620 
1621 	lock_chunks(root);
1622 
1623 	device->writeable = 1;
1624 	device->work.func = pending_bios_fn;
1625 	generate_random_uuid(device->uuid);
1626 	spin_lock_init(&device->io_lock);
1627 	device->generation = trans->transid;
1628 	device->io_width = root->sectorsize;
1629 	device->io_align = root->sectorsize;
1630 	device->sector_size = root->sectorsize;
1631 	device->total_bytes = i_size_read(bdev->bd_inode);
1632 	device->disk_total_bytes = device->total_bytes;
1633 	device->dev_root = root->fs_info->dev_root;
1634 	device->bdev = bdev;
1635 	device->in_fs_metadata = 1;
1636 	device->mode = 0;
1637 	set_blocksize(device->bdev, 4096);
1638 
1639 	if (seeding_dev) {
1640 		sb->s_flags &= ~MS_RDONLY;
1641 		ret = btrfs_prepare_sprout(trans, root);
1642 		BUG_ON(ret);
1643 	}
1644 
1645 	device->fs_devices = root->fs_info->fs_devices;
1646 
1647 	/*
1648 	 * we don't want write_supers to jump in here with our device
1649 	 * half setup
1650 	 */
1651 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1652 	list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1653 	list_add(&device->dev_alloc_list,
1654 		 &root->fs_info->fs_devices->alloc_list);
1655 	root->fs_info->fs_devices->num_devices++;
1656 	root->fs_info->fs_devices->open_devices++;
1657 	root->fs_info->fs_devices->rw_devices++;
1658 	root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1659 
1660 	if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1661 		root->fs_info->fs_devices->rotating = 1;
1662 
1663 	total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1664 	btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1665 				    total_bytes + device->total_bytes);
1666 
1667 	total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1668 	btrfs_set_super_num_devices(&root->fs_info->super_copy,
1669 				    total_bytes + 1);
1670 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1671 
1672 	if (seeding_dev) {
1673 		ret = init_first_rw_device(trans, root, device);
1674 		BUG_ON(ret);
1675 		ret = btrfs_finish_sprout(trans, root);
1676 		BUG_ON(ret);
1677 	} else {
1678 		ret = btrfs_add_device(trans, root, device);
1679 	}
1680 
1681 	/*
1682 	 * we've got more storage, clear any full flags on the space
1683 	 * infos
1684 	 */
1685 	btrfs_clear_space_info_full(root->fs_info);
1686 
1687 	unlock_chunks(root);
1688 	btrfs_commit_transaction(trans, root);
1689 
1690 	if (seeding_dev) {
1691 		mutex_unlock(&uuid_mutex);
1692 		up_write(&sb->s_umount);
1693 
1694 		ret = btrfs_relocate_sys_chunks(root);
1695 		BUG_ON(ret);
1696 	}
1697 out:
1698 	mutex_unlock(&root->fs_info->volume_mutex);
1699 	return ret;
1700 error:
1701 	blkdev_put(bdev, FMODE_EXCL);
1702 	if (seeding_dev) {
1703 		mutex_unlock(&uuid_mutex);
1704 		up_write(&sb->s_umount);
1705 	}
1706 	goto out;
1707 }
1708 
1709 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1710 					struct btrfs_device *device)
1711 {
1712 	int ret;
1713 	struct btrfs_path *path;
1714 	struct btrfs_root *root;
1715 	struct btrfs_dev_item *dev_item;
1716 	struct extent_buffer *leaf;
1717 	struct btrfs_key key;
1718 
1719 	root = device->dev_root->fs_info->chunk_root;
1720 
1721 	path = btrfs_alloc_path();
1722 	if (!path)
1723 		return -ENOMEM;
1724 
1725 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1726 	key.type = BTRFS_DEV_ITEM_KEY;
1727 	key.offset = device->devid;
1728 
1729 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1730 	if (ret < 0)
1731 		goto out;
1732 
1733 	if (ret > 0) {
1734 		ret = -ENOENT;
1735 		goto out;
1736 	}
1737 
1738 	leaf = path->nodes[0];
1739 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1740 
1741 	btrfs_set_device_id(leaf, dev_item, device->devid);
1742 	btrfs_set_device_type(leaf, dev_item, device->type);
1743 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1744 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1745 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1746 	btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1747 	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1748 	btrfs_mark_buffer_dirty(leaf);
1749 
1750 out:
1751 	btrfs_free_path(path);
1752 	return ret;
1753 }
1754 
1755 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1756 		      struct btrfs_device *device, u64 new_size)
1757 {
1758 	struct btrfs_super_block *super_copy =
1759 		&device->dev_root->fs_info->super_copy;
1760 	u64 old_total = btrfs_super_total_bytes(super_copy);
1761 	u64 diff = new_size - device->total_bytes;
1762 
1763 	if (!device->writeable)
1764 		return -EACCES;
1765 	if (new_size <= device->total_bytes)
1766 		return -EINVAL;
1767 
1768 	btrfs_set_super_total_bytes(super_copy, old_total + diff);
1769 	device->fs_devices->total_rw_bytes += diff;
1770 
1771 	device->total_bytes = new_size;
1772 	device->disk_total_bytes = new_size;
1773 	btrfs_clear_space_info_full(device->dev_root->fs_info);
1774 
1775 	return btrfs_update_device(trans, device);
1776 }
1777 
1778 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1779 		      struct btrfs_device *device, u64 new_size)
1780 {
1781 	int ret;
1782 	lock_chunks(device->dev_root);
1783 	ret = __btrfs_grow_device(trans, device, new_size);
1784 	unlock_chunks(device->dev_root);
1785 	return ret;
1786 }
1787 
1788 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1789 			    struct btrfs_root *root,
1790 			    u64 chunk_tree, u64 chunk_objectid,
1791 			    u64 chunk_offset)
1792 {
1793 	int ret;
1794 	struct btrfs_path *path;
1795 	struct btrfs_key key;
1796 
1797 	root = root->fs_info->chunk_root;
1798 	path = btrfs_alloc_path();
1799 	if (!path)
1800 		return -ENOMEM;
1801 
1802 	key.objectid = chunk_objectid;
1803 	key.offset = chunk_offset;
1804 	key.type = BTRFS_CHUNK_ITEM_KEY;
1805 
1806 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1807 	BUG_ON(ret);
1808 
1809 	ret = btrfs_del_item(trans, root, path);
1810 	BUG_ON(ret);
1811 
1812 	btrfs_free_path(path);
1813 	return 0;
1814 }
1815 
1816 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1817 			chunk_offset)
1818 {
1819 	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1820 	struct btrfs_disk_key *disk_key;
1821 	struct btrfs_chunk *chunk;
1822 	u8 *ptr;
1823 	int ret = 0;
1824 	u32 num_stripes;
1825 	u32 array_size;
1826 	u32 len = 0;
1827 	u32 cur;
1828 	struct btrfs_key key;
1829 
1830 	array_size = btrfs_super_sys_array_size(super_copy);
1831 
1832 	ptr = super_copy->sys_chunk_array;
1833 	cur = 0;
1834 
1835 	while (cur < array_size) {
1836 		disk_key = (struct btrfs_disk_key *)ptr;
1837 		btrfs_disk_key_to_cpu(&key, disk_key);
1838 
1839 		len = sizeof(*disk_key);
1840 
1841 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1842 			chunk = (struct btrfs_chunk *)(ptr + len);
1843 			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1844 			len += btrfs_chunk_item_size(num_stripes);
1845 		} else {
1846 			ret = -EIO;
1847 			break;
1848 		}
1849 		if (key.objectid == chunk_objectid &&
1850 		    key.offset == chunk_offset) {
1851 			memmove(ptr, ptr + len, array_size - (cur + len));
1852 			array_size -= len;
1853 			btrfs_set_super_sys_array_size(super_copy, array_size);
1854 		} else {
1855 			ptr += len;
1856 			cur += len;
1857 		}
1858 	}
1859 	return ret;
1860 }
1861 
1862 static int btrfs_relocate_chunk(struct btrfs_root *root,
1863 			 u64 chunk_tree, u64 chunk_objectid,
1864 			 u64 chunk_offset)
1865 {
1866 	struct extent_map_tree *em_tree;
1867 	struct btrfs_root *extent_root;
1868 	struct btrfs_trans_handle *trans;
1869 	struct extent_map *em;
1870 	struct map_lookup *map;
1871 	int ret;
1872 	int i;
1873 
1874 	root = root->fs_info->chunk_root;
1875 	extent_root = root->fs_info->extent_root;
1876 	em_tree = &root->fs_info->mapping_tree.map_tree;
1877 
1878 	ret = btrfs_can_relocate(extent_root, chunk_offset);
1879 	if (ret)
1880 		return -ENOSPC;
1881 
1882 	/* step one, relocate all the extents inside this chunk */
1883 	ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1884 	if (ret)
1885 		return ret;
1886 
1887 	trans = btrfs_start_transaction(root, 0);
1888 	BUG_ON(IS_ERR(trans));
1889 
1890 	lock_chunks(root);
1891 
1892 	/*
1893 	 * step two, delete the device extents and the
1894 	 * chunk tree entries
1895 	 */
1896 	read_lock(&em_tree->lock);
1897 	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1898 	read_unlock(&em_tree->lock);
1899 
1900 	BUG_ON(em->start > chunk_offset ||
1901 	       em->start + em->len < chunk_offset);
1902 	map = (struct map_lookup *)em->bdev;
1903 
1904 	for (i = 0; i < map->num_stripes; i++) {
1905 		ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1906 					    map->stripes[i].physical);
1907 		BUG_ON(ret);
1908 
1909 		if (map->stripes[i].dev) {
1910 			ret = btrfs_update_device(trans, map->stripes[i].dev);
1911 			BUG_ON(ret);
1912 		}
1913 	}
1914 	ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1915 			       chunk_offset);
1916 
1917 	BUG_ON(ret);
1918 
1919 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1920 		ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1921 		BUG_ON(ret);
1922 	}
1923 
1924 	ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1925 	BUG_ON(ret);
1926 
1927 	write_lock(&em_tree->lock);
1928 	remove_extent_mapping(em_tree, em);
1929 	write_unlock(&em_tree->lock);
1930 
1931 	kfree(map);
1932 	em->bdev = NULL;
1933 
1934 	/* once for the tree */
1935 	free_extent_map(em);
1936 	/* once for us */
1937 	free_extent_map(em);
1938 
1939 	unlock_chunks(root);
1940 	btrfs_end_transaction(trans, root);
1941 	return 0;
1942 }
1943 
1944 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1945 {
1946 	struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1947 	struct btrfs_path *path;
1948 	struct extent_buffer *leaf;
1949 	struct btrfs_chunk *chunk;
1950 	struct btrfs_key key;
1951 	struct btrfs_key found_key;
1952 	u64 chunk_tree = chunk_root->root_key.objectid;
1953 	u64 chunk_type;
1954 	bool retried = false;
1955 	int failed = 0;
1956 	int ret;
1957 
1958 	path = btrfs_alloc_path();
1959 	if (!path)
1960 		return -ENOMEM;
1961 
1962 again:
1963 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1964 	key.offset = (u64)-1;
1965 	key.type = BTRFS_CHUNK_ITEM_KEY;
1966 
1967 	while (1) {
1968 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1969 		if (ret < 0)
1970 			goto error;
1971 		BUG_ON(ret == 0);
1972 
1973 		ret = btrfs_previous_item(chunk_root, path, key.objectid,
1974 					  key.type);
1975 		if (ret < 0)
1976 			goto error;
1977 		if (ret > 0)
1978 			break;
1979 
1980 		leaf = path->nodes[0];
1981 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1982 
1983 		chunk = btrfs_item_ptr(leaf, path->slots[0],
1984 				       struct btrfs_chunk);
1985 		chunk_type = btrfs_chunk_type(leaf, chunk);
1986 		btrfs_release_path(chunk_root, path);
1987 
1988 		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1989 			ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1990 						   found_key.objectid,
1991 						   found_key.offset);
1992 			if (ret == -ENOSPC)
1993 				failed++;
1994 			else if (ret)
1995 				BUG();
1996 		}
1997 
1998 		if (found_key.offset == 0)
1999 			break;
2000 		key.offset = found_key.offset - 1;
2001 	}
2002 	ret = 0;
2003 	if (failed && !retried) {
2004 		failed = 0;
2005 		retried = true;
2006 		goto again;
2007 	} else if (failed && retried) {
2008 		WARN_ON(1);
2009 		ret = -ENOSPC;
2010 	}
2011 error:
2012 	btrfs_free_path(path);
2013 	return ret;
2014 }
2015 
2016 static u64 div_factor(u64 num, int factor)
2017 {
2018 	if (factor == 10)
2019 		return num;
2020 	num *= factor;
2021 	do_div(num, 10);
2022 	return num;
2023 }
2024 
2025 int btrfs_balance(struct btrfs_root *dev_root)
2026 {
2027 	int ret;
2028 	struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2029 	struct btrfs_device *device;
2030 	u64 old_size;
2031 	u64 size_to_free;
2032 	struct btrfs_path *path;
2033 	struct btrfs_key key;
2034 	struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2035 	struct btrfs_trans_handle *trans;
2036 	struct btrfs_key found_key;
2037 
2038 	if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2039 		return -EROFS;
2040 
2041 	if (!capable(CAP_SYS_ADMIN))
2042 		return -EPERM;
2043 
2044 	mutex_lock(&dev_root->fs_info->volume_mutex);
2045 	dev_root = dev_root->fs_info->dev_root;
2046 
2047 	/* step one make some room on all the devices */
2048 	list_for_each_entry(device, devices, dev_list) {
2049 		old_size = device->total_bytes;
2050 		size_to_free = div_factor(old_size, 1);
2051 		size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2052 		if (!device->writeable ||
2053 		    device->total_bytes - device->bytes_used > size_to_free)
2054 			continue;
2055 
2056 		ret = btrfs_shrink_device(device, old_size - size_to_free);
2057 		if (ret == -ENOSPC)
2058 			break;
2059 		BUG_ON(ret);
2060 
2061 		trans = btrfs_start_transaction(dev_root, 0);
2062 		BUG_ON(IS_ERR(trans));
2063 
2064 		ret = btrfs_grow_device(trans, device, old_size);
2065 		BUG_ON(ret);
2066 
2067 		btrfs_end_transaction(trans, dev_root);
2068 	}
2069 
2070 	/* step two, relocate all the chunks */
2071 	path = btrfs_alloc_path();
2072 	BUG_ON(!path);
2073 
2074 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2075 	key.offset = (u64)-1;
2076 	key.type = BTRFS_CHUNK_ITEM_KEY;
2077 
2078 	while (1) {
2079 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2080 		if (ret < 0)
2081 			goto error;
2082 
2083 		/*
2084 		 * this shouldn't happen, it means the last relocate
2085 		 * failed
2086 		 */
2087 		if (ret == 0)
2088 			break;
2089 
2090 		ret = btrfs_previous_item(chunk_root, path, 0,
2091 					  BTRFS_CHUNK_ITEM_KEY);
2092 		if (ret)
2093 			break;
2094 
2095 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2096 				      path->slots[0]);
2097 		if (found_key.objectid != key.objectid)
2098 			break;
2099 
2100 		/* chunk zero is special */
2101 		if (found_key.offset == 0)
2102 			break;
2103 
2104 		btrfs_release_path(chunk_root, path);
2105 		ret = btrfs_relocate_chunk(chunk_root,
2106 					   chunk_root->root_key.objectid,
2107 					   found_key.objectid,
2108 					   found_key.offset);
2109 		BUG_ON(ret && ret != -ENOSPC);
2110 		key.offset = found_key.offset - 1;
2111 	}
2112 	ret = 0;
2113 error:
2114 	btrfs_free_path(path);
2115 	mutex_unlock(&dev_root->fs_info->volume_mutex);
2116 	return ret;
2117 }
2118 
2119 /*
2120  * shrinking a device means finding all of the device extents past
2121  * the new size, and then following the back refs to the chunks.
2122  * The chunk relocation code actually frees the device extent
2123  */
2124 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2125 {
2126 	struct btrfs_trans_handle *trans;
2127 	struct btrfs_root *root = device->dev_root;
2128 	struct btrfs_dev_extent *dev_extent = NULL;
2129 	struct btrfs_path *path;
2130 	u64 length;
2131 	u64 chunk_tree;
2132 	u64 chunk_objectid;
2133 	u64 chunk_offset;
2134 	int ret;
2135 	int slot;
2136 	int failed = 0;
2137 	bool retried = false;
2138 	struct extent_buffer *l;
2139 	struct btrfs_key key;
2140 	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2141 	u64 old_total = btrfs_super_total_bytes(super_copy);
2142 	u64 old_size = device->total_bytes;
2143 	u64 diff = device->total_bytes - new_size;
2144 
2145 	if (new_size >= device->total_bytes)
2146 		return -EINVAL;
2147 
2148 	path = btrfs_alloc_path();
2149 	if (!path)
2150 		return -ENOMEM;
2151 
2152 	path->reada = 2;
2153 
2154 	lock_chunks(root);
2155 
2156 	device->total_bytes = new_size;
2157 	if (device->writeable)
2158 		device->fs_devices->total_rw_bytes -= diff;
2159 	unlock_chunks(root);
2160 
2161 again:
2162 	key.objectid = device->devid;
2163 	key.offset = (u64)-1;
2164 	key.type = BTRFS_DEV_EXTENT_KEY;
2165 
2166 	while (1) {
2167 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2168 		if (ret < 0)
2169 			goto done;
2170 
2171 		ret = btrfs_previous_item(root, path, 0, key.type);
2172 		if (ret < 0)
2173 			goto done;
2174 		if (ret) {
2175 			ret = 0;
2176 			btrfs_release_path(root, path);
2177 			break;
2178 		}
2179 
2180 		l = path->nodes[0];
2181 		slot = path->slots[0];
2182 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2183 
2184 		if (key.objectid != device->devid) {
2185 			btrfs_release_path(root, path);
2186 			break;
2187 		}
2188 
2189 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2190 		length = btrfs_dev_extent_length(l, dev_extent);
2191 
2192 		if (key.offset + length <= new_size) {
2193 			btrfs_release_path(root, path);
2194 			break;
2195 		}
2196 
2197 		chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2198 		chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2199 		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2200 		btrfs_release_path(root, path);
2201 
2202 		ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2203 					   chunk_offset);
2204 		if (ret && ret != -ENOSPC)
2205 			goto done;
2206 		if (ret == -ENOSPC)
2207 			failed++;
2208 		key.offset -= 1;
2209 	}
2210 
2211 	if (failed && !retried) {
2212 		failed = 0;
2213 		retried = true;
2214 		goto again;
2215 	} else if (failed && retried) {
2216 		ret = -ENOSPC;
2217 		lock_chunks(root);
2218 
2219 		device->total_bytes = old_size;
2220 		if (device->writeable)
2221 			device->fs_devices->total_rw_bytes += diff;
2222 		unlock_chunks(root);
2223 		goto done;
2224 	}
2225 
2226 	/* Shrinking succeeded, else we would be at "done". */
2227 	trans = btrfs_start_transaction(root, 0);
2228 	if (IS_ERR(trans)) {
2229 		ret = PTR_ERR(trans);
2230 		goto done;
2231 	}
2232 
2233 	lock_chunks(root);
2234 
2235 	device->disk_total_bytes = new_size;
2236 	/* Now btrfs_update_device() will change the on-disk size. */
2237 	ret = btrfs_update_device(trans, device);
2238 	if (ret) {
2239 		unlock_chunks(root);
2240 		btrfs_end_transaction(trans, root);
2241 		goto done;
2242 	}
2243 	WARN_ON(diff > old_total);
2244 	btrfs_set_super_total_bytes(super_copy, old_total - diff);
2245 	unlock_chunks(root);
2246 	btrfs_end_transaction(trans, root);
2247 done:
2248 	btrfs_free_path(path);
2249 	return ret;
2250 }
2251 
2252 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2253 			   struct btrfs_root *root,
2254 			   struct btrfs_key *key,
2255 			   struct btrfs_chunk *chunk, int item_size)
2256 {
2257 	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2258 	struct btrfs_disk_key disk_key;
2259 	u32 array_size;
2260 	u8 *ptr;
2261 
2262 	array_size = btrfs_super_sys_array_size(super_copy);
2263 	if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2264 		return -EFBIG;
2265 
2266 	ptr = super_copy->sys_chunk_array + array_size;
2267 	btrfs_cpu_key_to_disk(&disk_key, key);
2268 	memcpy(ptr, &disk_key, sizeof(disk_key));
2269 	ptr += sizeof(disk_key);
2270 	memcpy(ptr, chunk, item_size);
2271 	item_size += sizeof(disk_key);
2272 	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2273 	return 0;
2274 }
2275 
2276 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2277 					int num_stripes, int sub_stripes)
2278 {
2279 	if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2280 		return calc_size;
2281 	else if (type & BTRFS_BLOCK_GROUP_RAID10)
2282 		return calc_size * (num_stripes / sub_stripes);
2283 	else
2284 		return calc_size * num_stripes;
2285 }
2286 
2287 /* Used to sort the devices by max_avail(descending sort) */
2288 int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2)
2289 {
2290 	if (((struct btrfs_device_info *)dev_info1)->max_avail >
2291 	    ((struct btrfs_device_info *)dev_info2)->max_avail)
2292 		return -1;
2293 	else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2294 		 ((struct btrfs_device_info *)dev_info2)->max_avail)
2295 		return 1;
2296 	else
2297 		return 0;
2298 }
2299 
2300 static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2301 				 int *num_stripes, int *min_stripes,
2302 				 int *sub_stripes)
2303 {
2304 	*num_stripes = 1;
2305 	*min_stripes = 1;
2306 	*sub_stripes = 0;
2307 
2308 	if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2309 		*num_stripes = fs_devices->rw_devices;
2310 		*min_stripes = 2;
2311 	}
2312 	if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2313 		*num_stripes = 2;
2314 		*min_stripes = 2;
2315 	}
2316 	if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2317 		if (fs_devices->rw_devices < 2)
2318 			return -ENOSPC;
2319 		*num_stripes = 2;
2320 		*min_stripes = 2;
2321 	}
2322 	if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2323 		*num_stripes = fs_devices->rw_devices;
2324 		if (*num_stripes < 4)
2325 			return -ENOSPC;
2326 		*num_stripes &= ~(u32)1;
2327 		*sub_stripes = 2;
2328 		*min_stripes = 4;
2329 	}
2330 
2331 	return 0;
2332 }
2333 
2334 static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2335 				    u64 proposed_size, u64 type,
2336 				    int num_stripes, int small_stripe)
2337 {
2338 	int min_stripe_size = 1 * 1024 * 1024;
2339 	u64 calc_size = proposed_size;
2340 	u64 max_chunk_size = calc_size;
2341 	int ncopies = 1;
2342 
2343 	if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2344 		    BTRFS_BLOCK_GROUP_DUP |
2345 		    BTRFS_BLOCK_GROUP_RAID10))
2346 		ncopies = 2;
2347 
2348 	if (type & BTRFS_BLOCK_GROUP_DATA) {
2349 		max_chunk_size = 10 * calc_size;
2350 		min_stripe_size = 64 * 1024 * 1024;
2351 	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2352 		max_chunk_size = 256 * 1024 * 1024;
2353 		min_stripe_size = 32 * 1024 * 1024;
2354 	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2355 		calc_size = 8 * 1024 * 1024;
2356 		max_chunk_size = calc_size * 2;
2357 		min_stripe_size = 1 * 1024 * 1024;
2358 	}
2359 
2360 	/* we don't want a chunk larger than 10% of writeable space */
2361 	max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2362 			     max_chunk_size);
2363 
2364 	if (calc_size * num_stripes > max_chunk_size * ncopies) {
2365 		calc_size = max_chunk_size * ncopies;
2366 		do_div(calc_size, num_stripes);
2367 		do_div(calc_size, BTRFS_STRIPE_LEN);
2368 		calc_size *= BTRFS_STRIPE_LEN;
2369 	}
2370 
2371 	/* we don't want tiny stripes */
2372 	if (!small_stripe)
2373 		calc_size = max_t(u64, min_stripe_size, calc_size);
2374 
2375 	/*
2376 	 * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2377 	 * we end up with something bigger than a stripe
2378 	 */
2379 	calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2380 
2381 	do_div(calc_size, BTRFS_STRIPE_LEN);
2382 	calc_size *= BTRFS_STRIPE_LEN;
2383 
2384 	return calc_size;
2385 }
2386 
2387 static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2388 						      int num_stripes)
2389 {
2390 	struct map_lookup *new;
2391 	size_t len = map_lookup_size(num_stripes);
2392 
2393 	BUG_ON(map->num_stripes < num_stripes);
2394 
2395 	if (map->num_stripes == num_stripes)
2396 		return map;
2397 
2398 	new = kmalloc(len, GFP_NOFS);
2399 	if (!new) {
2400 		/* just change map->num_stripes */
2401 		map->num_stripes = num_stripes;
2402 		return map;
2403 	}
2404 
2405 	memcpy(new, map, len);
2406 	new->num_stripes = num_stripes;
2407 	kfree(map);
2408 	return new;
2409 }
2410 
2411 /*
2412  * helper to allocate device space from btrfs_device_info, in which we stored
2413  * max free space information of every device. It is used when we can not
2414  * allocate chunks by default size.
2415  *
2416  * By this helper, we can allocate a new chunk as larger as possible.
2417  */
2418 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2419 				    struct btrfs_fs_devices *fs_devices,
2420 				    struct btrfs_device_info *devices,
2421 				    int nr_device, u64 type,
2422 				    struct map_lookup **map_lookup,
2423 				    int min_stripes, u64 *stripe_size)
2424 {
2425 	int i, index, sort_again = 0;
2426 	int min_devices = min_stripes;
2427 	u64 max_avail, min_free;
2428 	struct map_lookup *map = *map_lookup;
2429 	int ret;
2430 
2431 	if (nr_device < min_stripes)
2432 		return -ENOSPC;
2433 
2434 	btrfs_descending_sort_devices(devices, nr_device);
2435 
2436 	max_avail = devices[0].max_avail;
2437 	if (!max_avail)
2438 		return -ENOSPC;
2439 
2440 	for (i = 0; i < nr_device; i++) {
2441 		/*
2442 		 * if dev_offset = 0, it means the free space of this device
2443 		 * is less than what we need, and we didn't search max avail
2444 		 * extent on this device, so do it now.
2445 		 */
2446 		if (!devices[i].dev_offset) {
2447 			ret = find_free_dev_extent(trans, devices[i].dev,
2448 						   max_avail,
2449 						   &devices[i].dev_offset,
2450 						   &devices[i].max_avail);
2451 			if (ret != 0 && ret != -ENOSPC)
2452 				return ret;
2453 			sort_again = 1;
2454 		}
2455 	}
2456 
2457 	/* we update the max avail free extent of each devices, sort again */
2458 	if (sort_again)
2459 		btrfs_descending_sort_devices(devices, nr_device);
2460 
2461 	if (type & BTRFS_BLOCK_GROUP_DUP)
2462 		min_devices = 1;
2463 
2464 	if (!devices[min_devices - 1].max_avail)
2465 		return -ENOSPC;
2466 
2467 	max_avail = devices[min_devices - 1].max_avail;
2468 	if (type & BTRFS_BLOCK_GROUP_DUP)
2469 		do_div(max_avail, 2);
2470 
2471 	max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2472 					     min_stripes, 1);
2473 	if (type & BTRFS_BLOCK_GROUP_DUP)
2474 		min_free = max_avail * 2;
2475 	else
2476 		min_free = max_avail;
2477 
2478 	if (min_free > devices[min_devices - 1].max_avail)
2479 		return -ENOSPC;
2480 
2481 	map = __shrink_map_lookup_stripes(map, min_stripes);
2482 	*stripe_size = max_avail;
2483 
2484 	index = 0;
2485 	for (i = 0; i < min_stripes; i++) {
2486 		map->stripes[i].dev = devices[index].dev;
2487 		map->stripes[i].physical = devices[index].dev_offset;
2488 		if (type & BTRFS_BLOCK_GROUP_DUP) {
2489 			i++;
2490 			map->stripes[i].dev = devices[index].dev;
2491 			map->stripes[i].physical = devices[index].dev_offset +
2492 						   max_avail;
2493 		}
2494 		index++;
2495 	}
2496 	*map_lookup = map;
2497 
2498 	return 0;
2499 }
2500 
2501 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2502 			       struct btrfs_root *extent_root,
2503 			       struct map_lookup **map_ret,
2504 			       u64 *num_bytes, u64 *stripe_size,
2505 			       u64 start, u64 type)
2506 {
2507 	struct btrfs_fs_info *info = extent_root->fs_info;
2508 	struct btrfs_device *device = NULL;
2509 	struct btrfs_fs_devices *fs_devices = info->fs_devices;
2510 	struct list_head *cur;
2511 	struct map_lookup *map;
2512 	struct extent_map_tree *em_tree;
2513 	struct extent_map *em;
2514 	struct btrfs_device_info *devices_info;
2515 	struct list_head private_devs;
2516 	u64 calc_size = 1024 * 1024 * 1024;
2517 	u64 min_free;
2518 	u64 avail;
2519 	u64 dev_offset;
2520 	int num_stripes;
2521 	int min_stripes;
2522 	int sub_stripes;
2523 	int min_devices;	/* the min number of devices we need */
2524 	int i;
2525 	int ret;
2526 	int index;
2527 
2528 	if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2529 	    (type & BTRFS_BLOCK_GROUP_DUP)) {
2530 		WARN_ON(1);
2531 		type &= ~BTRFS_BLOCK_GROUP_DUP;
2532 	}
2533 	if (list_empty(&fs_devices->alloc_list))
2534 		return -ENOSPC;
2535 
2536 	ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2537 				    &min_stripes, &sub_stripes);
2538 	if (ret)
2539 		return ret;
2540 
2541 	devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2542 			       GFP_NOFS);
2543 	if (!devices_info)
2544 		return -ENOMEM;
2545 
2546 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2547 	if (!map) {
2548 		ret = -ENOMEM;
2549 		goto error;
2550 	}
2551 	map->num_stripes = num_stripes;
2552 
2553 	cur = fs_devices->alloc_list.next;
2554 	index = 0;
2555 	i = 0;
2556 
2557 	calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2558 					     num_stripes, 0);
2559 
2560 	if (type & BTRFS_BLOCK_GROUP_DUP) {
2561 		min_free = calc_size * 2;
2562 		min_devices = 1;
2563 	} else {
2564 		min_free = calc_size;
2565 		min_devices = min_stripes;
2566 	}
2567 
2568 	INIT_LIST_HEAD(&private_devs);
2569 	while (index < num_stripes) {
2570 		device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2571 		BUG_ON(!device->writeable);
2572 		if (device->total_bytes > device->bytes_used)
2573 			avail = device->total_bytes - device->bytes_used;
2574 		else
2575 			avail = 0;
2576 		cur = cur->next;
2577 
2578 		if (device->in_fs_metadata && avail >= min_free) {
2579 			ret = find_free_dev_extent(trans, device, min_free,
2580 						   &devices_info[i].dev_offset,
2581 						   &devices_info[i].max_avail);
2582 			if (ret == 0) {
2583 				list_move_tail(&device->dev_alloc_list,
2584 					       &private_devs);
2585 				map->stripes[index].dev = device;
2586 				map->stripes[index].physical =
2587 						devices_info[i].dev_offset;
2588 				index++;
2589 				if (type & BTRFS_BLOCK_GROUP_DUP) {
2590 					map->stripes[index].dev = device;
2591 					map->stripes[index].physical =
2592 						devices_info[i].dev_offset +
2593 						calc_size;
2594 					index++;
2595 				}
2596 			} else if (ret != -ENOSPC)
2597 				goto error;
2598 
2599 			devices_info[i].dev = device;
2600 			i++;
2601 		} else if (device->in_fs_metadata &&
2602 			   avail >= BTRFS_STRIPE_LEN) {
2603 			devices_info[i].dev = device;
2604 			devices_info[i].max_avail = avail;
2605 			i++;
2606 		}
2607 
2608 		if (cur == &fs_devices->alloc_list)
2609 			break;
2610 	}
2611 
2612 	list_splice(&private_devs, &fs_devices->alloc_list);
2613 	if (index < num_stripes) {
2614 		if (index >= min_stripes) {
2615 			num_stripes = index;
2616 			if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2617 				num_stripes /= sub_stripes;
2618 				num_stripes *= sub_stripes;
2619 			}
2620 
2621 			map = __shrink_map_lookup_stripes(map, num_stripes);
2622 		} else if (i >= min_devices) {
2623 			ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2624 						       devices_info, i, type,
2625 						       &map, min_stripes,
2626 						       &calc_size);
2627 			if (ret)
2628 				goto error;
2629 		} else {
2630 			ret = -ENOSPC;
2631 			goto error;
2632 		}
2633 	}
2634 	map->sector_size = extent_root->sectorsize;
2635 	map->stripe_len = BTRFS_STRIPE_LEN;
2636 	map->io_align = BTRFS_STRIPE_LEN;
2637 	map->io_width = BTRFS_STRIPE_LEN;
2638 	map->type = type;
2639 	map->sub_stripes = sub_stripes;
2640 
2641 	*map_ret = map;
2642 	*stripe_size = calc_size;
2643 	*num_bytes = chunk_bytes_by_type(type, calc_size,
2644 					 map->num_stripes, sub_stripes);
2645 
2646 	em = alloc_extent_map(GFP_NOFS);
2647 	if (!em) {
2648 		ret = -ENOMEM;
2649 		goto error;
2650 	}
2651 	em->bdev = (struct block_device *)map;
2652 	em->start = start;
2653 	em->len = *num_bytes;
2654 	em->block_start = 0;
2655 	em->block_len = em->len;
2656 
2657 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2658 	write_lock(&em_tree->lock);
2659 	ret = add_extent_mapping(em_tree, em);
2660 	write_unlock(&em_tree->lock);
2661 	BUG_ON(ret);
2662 	free_extent_map(em);
2663 
2664 	ret = btrfs_make_block_group(trans, extent_root, 0, type,
2665 				     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2666 				     start, *num_bytes);
2667 	BUG_ON(ret);
2668 
2669 	index = 0;
2670 	while (index < map->num_stripes) {
2671 		device = map->stripes[index].dev;
2672 		dev_offset = map->stripes[index].physical;
2673 
2674 		ret = btrfs_alloc_dev_extent(trans, device,
2675 				info->chunk_root->root_key.objectid,
2676 				BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2677 				start, dev_offset, calc_size);
2678 		BUG_ON(ret);
2679 		index++;
2680 	}
2681 
2682 	kfree(devices_info);
2683 	return 0;
2684 
2685 error:
2686 	kfree(map);
2687 	kfree(devices_info);
2688 	return ret;
2689 }
2690 
2691 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2692 				struct btrfs_root *extent_root,
2693 				struct map_lookup *map, u64 chunk_offset,
2694 				u64 chunk_size, u64 stripe_size)
2695 {
2696 	u64 dev_offset;
2697 	struct btrfs_key key;
2698 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2699 	struct btrfs_device *device;
2700 	struct btrfs_chunk *chunk;
2701 	struct btrfs_stripe *stripe;
2702 	size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2703 	int index = 0;
2704 	int ret;
2705 
2706 	chunk = kzalloc(item_size, GFP_NOFS);
2707 	if (!chunk)
2708 		return -ENOMEM;
2709 
2710 	index = 0;
2711 	while (index < map->num_stripes) {
2712 		device = map->stripes[index].dev;
2713 		device->bytes_used += stripe_size;
2714 		ret = btrfs_update_device(trans, device);
2715 		BUG_ON(ret);
2716 		index++;
2717 	}
2718 
2719 	index = 0;
2720 	stripe = &chunk->stripe;
2721 	while (index < map->num_stripes) {
2722 		device = map->stripes[index].dev;
2723 		dev_offset = map->stripes[index].physical;
2724 
2725 		btrfs_set_stack_stripe_devid(stripe, device->devid);
2726 		btrfs_set_stack_stripe_offset(stripe, dev_offset);
2727 		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2728 		stripe++;
2729 		index++;
2730 	}
2731 
2732 	btrfs_set_stack_chunk_length(chunk, chunk_size);
2733 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2734 	btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2735 	btrfs_set_stack_chunk_type(chunk, map->type);
2736 	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2737 	btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2738 	btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2739 	btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2740 	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2741 
2742 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2743 	key.type = BTRFS_CHUNK_ITEM_KEY;
2744 	key.offset = chunk_offset;
2745 
2746 	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2747 	BUG_ON(ret);
2748 
2749 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2750 		ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2751 					     item_size);
2752 		BUG_ON(ret);
2753 	}
2754 	kfree(chunk);
2755 	return 0;
2756 }
2757 
2758 /*
2759  * Chunk allocation falls into two parts. The first part does works
2760  * that make the new allocated chunk useable, but not do any operation
2761  * that modifies the chunk tree. The second part does the works that
2762  * require modifying the chunk tree. This division is important for the
2763  * bootstrap process of adding storage to a seed btrfs.
2764  */
2765 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2766 		      struct btrfs_root *extent_root, u64 type)
2767 {
2768 	u64 chunk_offset;
2769 	u64 chunk_size;
2770 	u64 stripe_size;
2771 	struct map_lookup *map;
2772 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2773 	int ret;
2774 
2775 	ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2776 			      &chunk_offset);
2777 	if (ret)
2778 		return ret;
2779 
2780 	ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2781 				  &stripe_size, chunk_offset, type);
2782 	if (ret)
2783 		return ret;
2784 
2785 	ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2786 				   chunk_size, stripe_size);
2787 	BUG_ON(ret);
2788 	return 0;
2789 }
2790 
2791 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2792 					 struct btrfs_root *root,
2793 					 struct btrfs_device *device)
2794 {
2795 	u64 chunk_offset;
2796 	u64 sys_chunk_offset;
2797 	u64 chunk_size;
2798 	u64 sys_chunk_size;
2799 	u64 stripe_size;
2800 	u64 sys_stripe_size;
2801 	u64 alloc_profile;
2802 	struct map_lookup *map;
2803 	struct map_lookup *sys_map;
2804 	struct btrfs_fs_info *fs_info = root->fs_info;
2805 	struct btrfs_root *extent_root = fs_info->extent_root;
2806 	int ret;
2807 
2808 	ret = find_next_chunk(fs_info->chunk_root,
2809 			      BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2810 	BUG_ON(ret);
2811 
2812 	alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2813 			(fs_info->metadata_alloc_profile &
2814 			 fs_info->avail_metadata_alloc_bits);
2815 	alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2816 
2817 	ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2818 				  &stripe_size, chunk_offset, alloc_profile);
2819 	BUG_ON(ret);
2820 
2821 	sys_chunk_offset = chunk_offset + chunk_size;
2822 
2823 	alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2824 			(fs_info->system_alloc_profile &
2825 			 fs_info->avail_system_alloc_bits);
2826 	alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2827 
2828 	ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2829 				  &sys_chunk_size, &sys_stripe_size,
2830 				  sys_chunk_offset, alloc_profile);
2831 	BUG_ON(ret);
2832 
2833 	ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2834 	BUG_ON(ret);
2835 
2836 	/*
2837 	 * Modifying chunk tree needs allocating new blocks from both
2838 	 * system block group and metadata block group. So we only can
2839 	 * do operations require modifying the chunk tree after both
2840 	 * block groups were created.
2841 	 */
2842 	ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2843 				   chunk_size, stripe_size);
2844 	BUG_ON(ret);
2845 
2846 	ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2847 				   sys_chunk_offset, sys_chunk_size,
2848 				   sys_stripe_size);
2849 	BUG_ON(ret);
2850 	return 0;
2851 }
2852 
2853 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2854 {
2855 	struct extent_map *em;
2856 	struct map_lookup *map;
2857 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2858 	int readonly = 0;
2859 	int i;
2860 
2861 	read_lock(&map_tree->map_tree.lock);
2862 	em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2863 	read_unlock(&map_tree->map_tree.lock);
2864 	if (!em)
2865 		return 1;
2866 
2867 	if (btrfs_test_opt(root, DEGRADED)) {
2868 		free_extent_map(em);
2869 		return 0;
2870 	}
2871 
2872 	map = (struct map_lookup *)em->bdev;
2873 	for (i = 0; i < map->num_stripes; i++) {
2874 		if (!map->stripes[i].dev->writeable) {
2875 			readonly = 1;
2876 			break;
2877 		}
2878 	}
2879 	free_extent_map(em);
2880 	return readonly;
2881 }
2882 
2883 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2884 {
2885 	extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2886 }
2887 
2888 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2889 {
2890 	struct extent_map *em;
2891 
2892 	while (1) {
2893 		write_lock(&tree->map_tree.lock);
2894 		em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2895 		if (em)
2896 			remove_extent_mapping(&tree->map_tree, em);
2897 		write_unlock(&tree->map_tree.lock);
2898 		if (!em)
2899 			break;
2900 		kfree(em->bdev);
2901 		/* once for us */
2902 		free_extent_map(em);
2903 		/* once for the tree */
2904 		free_extent_map(em);
2905 	}
2906 }
2907 
2908 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2909 {
2910 	struct extent_map *em;
2911 	struct map_lookup *map;
2912 	struct extent_map_tree *em_tree = &map_tree->map_tree;
2913 	int ret;
2914 
2915 	read_lock(&em_tree->lock);
2916 	em = lookup_extent_mapping(em_tree, logical, len);
2917 	read_unlock(&em_tree->lock);
2918 	BUG_ON(!em);
2919 
2920 	BUG_ON(em->start > logical || em->start + em->len < logical);
2921 	map = (struct map_lookup *)em->bdev;
2922 	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2923 		ret = map->num_stripes;
2924 	else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2925 		ret = map->sub_stripes;
2926 	else
2927 		ret = 1;
2928 	free_extent_map(em);
2929 	return ret;
2930 }
2931 
2932 static int find_live_mirror(struct map_lookup *map, int first, int num,
2933 			    int optimal)
2934 {
2935 	int i;
2936 	if (map->stripes[optimal].dev->bdev)
2937 		return optimal;
2938 	for (i = first; i < first + num; i++) {
2939 		if (map->stripes[i].dev->bdev)
2940 			return i;
2941 	}
2942 	/* we couldn't find one that doesn't fail.  Just return something
2943 	 * and the io error handling code will clean up eventually
2944 	 */
2945 	return optimal;
2946 }
2947 
2948 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2949 			     u64 logical, u64 *length,
2950 			     struct btrfs_multi_bio **multi_ret,
2951 			     int mirror_num, struct page *unplug_page)
2952 {
2953 	struct extent_map *em;
2954 	struct map_lookup *map;
2955 	struct extent_map_tree *em_tree = &map_tree->map_tree;
2956 	u64 offset;
2957 	u64 stripe_offset;
2958 	u64 stripe_nr;
2959 	int stripes_allocated = 8;
2960 	int stripes_required = 1;
2961 	int stripe_index;
2962 	int i;
2963 	int num_stripes;
2964 	int max_errors = 0;
2965 	struct btrfs_multi_bio *multi = NULL;
2966 
2967 	if (multi_ret && !(rw & REQ_WRITE))
2968 		stripes_allocated = 1;
2969 again:
2970 	if (multi_ret) {
2971 		multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2972 				GFP_NOFS);
2973 		if (!multi)
2974 			return -ENOMEM;
2975 
2976 		atomic_set(&multi->error, 0);
2977 	}
2978 
2979 	read_lock(&em_tree->lock);
2980 	em = lookup_extent_mapping(em_tree, logical, *length);
2981 	read_unlock(&em_tree->lock);
2982 
2983 	if (!em && unplug_page) {
2984 		kfree(multi);
2985 		return 0;
2986 	}
2987 
2988 	if (!em) {
2989 		printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2990 		       (unsigned long long)logical,
2991 		       (unsigned long long)*length);
2992 		BUG();
2993 	}
2994 
2995 	BUG_ON(em->start > logical || em->start + em->len < logical);
2996 	map = (struct map_lookup *)em->bdev;
2997 	offset = logical - em->start;
2998 
2999 	if (mirror_num > map->num_stripes)
3000 		mirror_num = 0;
3001 
3002 	/* if our multi bio struct is too small, back off and try again */
3003 	if (rw & REQ_WRITE) {
3004 		if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3005 				 BTRFS_BLOCK_GROUP_DUP)) {
3006 			stripes_required = map->num_stripes;
3007 			max_errors = 1;
3008 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3009 			stripes_required = map->sub_stripes;
3010 			max_errors = 1;
3011 		}
3012 	}
3013 	if (multi_ret && (rw & REQ_WRITE) &&
3014 	    stripes_allocated < stripes_required) {
3015 		stripes_allocated = map->num_stripes;
3016 		free_extent_map(em);
3017 		kfree(multi);
3018 		goto again;
3019 	}
3020 	stripe_nr = offset;
3021 	/*
3022 	 * stripe_nr counts the total number of stripes we have to stride
3023 	 * to get to this block
3024 	 */
3025 	do_div(stripe_nr, map->stripe_len);
3026 
3027 	stripe_offset = stripe_nr * map->stripe_len;
3028 	BUG_ON(offset < stripe_offset);
3029 
3030 	/* stripe_offset is the offset of this block in its stripe*/
3031 	stripe_offset = offset - stripe_offset;
3032 
3033 	if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3034 			 BTRFS_BLOCK_GROUP_RAID10 |
3035 			 BTRFS_BLOCK_GROUP_DUP)) {
3036 		/* we limit the length of each bio to what fits in a stripe */
3037 		*length = min_t(u64, em->len - offset,
3038 			      map->stripe_len - stripe_offset);
3039 	} else {
3040 		*length = em->len - offset;
3041 	}
3042 
3043 	if (!multi_ret && !unplug_page)
3044 		goto out;
3045 
3046 	num_stripes = 1;
3047 	stripe_index = 0;
3048 	if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3049 		if (unplug_page || (rw & REQ_WRITE))
3050 			num_stripes = map->num_stripes;
3051 		else if (mirror_num)
3052 			stripe_index = mirror_num - 1;
3053 		else {
3054 			stripe_index = find_live_mirror(map, 0,
3055 					    map->num_stripes,
3056 					    current->pid % map->num_stripes);
3057 		}
3058 
3059 	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3060 		if (rw & REQ_WRITE)
3061 			num_stripes = map->num_stripes;
3062 		else if (mirror_num)
3063 			stripe_index = mirror_num - 1;
3064 
3065 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3066 		int factor = map->num_stripes / map->sub_stripes;
3067 
3068 		stripe_index = do_div(stripe_nr, factor);
3069 		stripe_index *= map->sub_stripes;
3070 
3071 		if (unplug_page || (rw & REQ_WRITE))
3072 			num_stripes = map->sub_stripes;
3073 		else if (mirror_num)
3074 			stripe_index += mirror_num - 1;
3075 		else {
3076 			stripe_index = find_live_mirror(map, stripe_index,
3077 					      map->sub_stripes, stripe_index +
3078 					      current->pid % map->sub_stripes);
3079 		}
3080 	} else {
3081 		/*
3082 		 * after this do_div call, stripe_nr is the number of stripes
3083 		 * on this device we have to walk to find the data, and
3084 		 * stripe_index is the number of our device in the stripe array
3085 		 */
3086 		stripe_index = do_div(stripe_nr, map->num_stripes);
3087 	}
3088 	BUG_ON(stripe_index >= map->num_stripes);
3089 
3090 	for (i = 0; i < num_stripes; i++) {
3091 		if (unplug_page) {
3092 			struct btrfs_device *device;
3093 			struct backing_dev_info *bdi;
3094 
3095 			device = map->stripes[stripe_index].dev;
3096 			if (device->bdev) {
3097 				bdi = blk_get_backing_dev_info(device->bdev);
3098 				if (bdi->unplug_io_fn)
3099 					bdi->unplug_io_fn(bdi, unplug_page);
3100 			}
3101 		} else {
3102 			multi->stripes[i].physical =
3103 				map->stripes[stripe_index].physical +
3104 				stripe_offset + stripe_nr * map->stripe_len;
3105 			multi->stripes[i].dev = map->stripes[stripe_index].dev;
3106 		}
3107 		stripe_index++;
3108 	}
3109 	if (multi_ret) {
3110 		*multi_ret = multi;
3111 		multi->num_stripes = num_stripes;
3112 		multi->max_errors = max_errors;
3113 	}
3114 out:
3115 	free_extent_map(em);
3116 	return 0;
3117 }
3118 
3119 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3120 		      u64 logical, u64 *length,
3121 		      struct btrfs_multi_bio **multi_ret, int mirror_num)
3122 {
3123 	return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3124 				 mirror_num, NULL);
3125 }
3126 
3127 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3128 		     u64 chunk_start, u64 physical, u64 devid,
3129 		     u64 **logical, int *naddrs, int *stripe_len)
3130 {
3131 	struct extent_map_tree *em_tree = &map_tree->map_tree;
3132 	struct extent_map *em;
3133 	struct map_lookup *map;
3134 	u64 *buf;
3135 	u64 bytenr;
3136 	u64 length;
3137 	u64 stripe_nr;
3138 	int i, j, nr = 0;
3139 
3140 	read_lock(&em_tree->lock);
3141 	em = lookup_extent_mapping(em_tree, chunk_start, 1);
3142 	read_unlock(&em_tree->lock);
3143 
3144 	BUG_ON(!em || em->start != chunk_start);
3145 	map = (struct map_lookup *)em->bdev;
3146 
3147 	length = em->len;
3148 	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3149 		do_div(length, map->num_stripes / map->sub_stripes);
3150 	else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3151 		do_div(length, map->num_stripes);
3152 
3153 	buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3154 	BUG_ON(!buf);
3155 
3156 	for (i = 0; i < map->num_stripes; i++) {
3157 		if (devid && map->stripes[i].dev->devid != devid)
3158 			continue;
3159 		if (map->stripes[i].physical > physical ||
3160 		    map->stripes[i].physical + length <= physical)
3161 			continue;
3162 
3163 		stripe_nr = physical - map->stripes[i].physical;
3164 		do_div(stripe_nr, map->stripe_len);
3165 
3166 		if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3167 			stripe_nr = stripe_nr * map->num_stripes + i;
3168 			do_div(stripe_nr, map->sub_stripes);
3169 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3170 			stripe_nr = stripe_nr * map->num_stripes + i;
3171 		}
3172 		bytenr = chunk_start + stripe_nr * map->stripe_len;
3173 		WARN_ON(nr >= map->num_stripes);
3174 		for (j = 0; j < nr; j++) {
3175 			if (buf[j] == bytenr)
3176 				break;
3177 		}
3178 		if (j == nr) {
3179 			WARN_ON(nr >= map->num_stripes);
3180 			buf[nr++] = bytenr;
3181 		}
3182 	}
3183 
3184 	*logical = buf;
3185 	*naddrs = nr;
3186 	*stripe_len = map->stripe_len;
3187 
3188 	free_extent_map(em);
3189 	return 0;
3190 }
3191 
3192 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
3193 		      u64 logical, struct page *page)
3194 {
3195 	u64 length = PAGE_CACHE_SIZE;
3196 	return __btrfs_map_block(map_tree, READ, logical, &length,
3197 				 NULL, 0, page);
3198 }
3199 
3200 static void end_bio_multi_stripe(struct bio *bio, int err)
3201 {
3202 	struct btrfs_multi_bio *multi = bio->bi_private;
3203 	int is_orig_bio = 0;
3204 
3205 	if (err)
3206 		atomic_inc(&multi->error);
3207 
3208 	if (bio == multi->orig_bio)
3209 		is_orig_bio = 1;
3210 
3211 	if (atomic_dec_and_test(&multi->stripes_pending)) {
3212 		if (!is_orig_bio) {
3213 			bio_put(bio);
3214 			bio = multi->orig_bio;
3215 		}
3216 		bio->bi_private = multi->private;
3217 		bio->bi_end_io = multi->end_io;
3218 		/* only send an error to the higher layers if it is
3219 		 * beyond the tolerance of the multi-bio
3220 		 */
3221 		if (atomic_read(&multi->error) > multi->max_errors) {
3222 			err = -EIO;
3223 		} else if (err) {
3224 			/*
3225 			 * this bio is actually up to date, we didn't
3226 			 * go over the max number of errors
3227 			 */
3228 			set_bit(BIO_UPTODATE, &bio->bi_flags);
3229 			err = 0;
3230 		}
3231 		kfree(multi);
3232 
3233 		bio_endio(bio, err);
3234 	} else if (!is_orig_bio) {
3235 		bio_put(bio);
3236 	}
3237 }
3238 
3239 struct async_sched {
3240 	struct bio *bio;
3241 	int rw;
3242 	struct btrfs_fs_info *info;
3243 	struct btrfs_work work;
3244 };
3245 
3246 /*
3247  * see run_scheduled_bios for a description of why bios are collected for
3248  * async submit.
3249  *
3250  * This will add one bio to the pending list for a device and make sure
3251  * the work struct is scheduled.
3252  */
3253 static noinline int schedule_bio(struct btrfs_root *root,
3254 				 struct btrfs_device *device,
3255 				 int rw, struct bio *bio)
3256 {
3257 	int should_queue = 1;
3258 	struct btrfs_pending_bios *pending_bios;
3259 
3260 	/* don't bother with additional async steps for reads, right now */
3261 	if (!(rw & REQ_WRITE)) {
3262 		bio_get(bio);
3263 		submit_bio(rw, bio);
3264 		bio_put(bio);
3265 		return 0;
3266 	}
3267 
3268 	/*
3269 	 * nr_async_bios allows us to reliably return congestion to the
3270 	 * higher layers.  Otherwise, the async bio makes it appear we have
3271 	 * made progress against dirty pages when we've really just put it
3272 	 * on a queue for later
3273 	 */
3274 	atomic_inc(&root->fs_info->nr_async_bios);
3275 	WARN_ON(bio->bi_next);
3276 	bio->bi_next = NULL;
3277 	bio->bi_rw |= rw;
3278 
3279 	spin_lock(&device->io_lock);
3280 	if (bio->bi_rw & REQ_SYNC)
3281 		pending_bios = &device->pending_sync_bios;
3282 	else
3283 		pending_bios = &device->pending_bios;
3284 
3285 	if (pending_bios->tail)
3286 		pending_bios->tail->bi_next = bio;
3287 
3288 	pending_bios->tail = bio;
3289 	if (!pending_bios->head)
3290 		pending_bios->head = bio;
3291 	if (device->running_pending)
3292 		should_queue = 0;
3293 
3294 	spin_unlock(&device->io_lock);
3295 
3296 	if (should_queue)
3297 		btrfs_queue_worker(&root->fs_info->submit_workers,
3298 				   &device->work);
3299 	return 0;
3300 }
3301 
3302 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3303 		  int mirror_num, int async_submit)
3304 {
3305 	struct btrfs_mapping_tree *map_tree;
3306 	struct btrfs_device *dev;
3307 	struct bio *first_bio = bio;
3308 	u64 logical = (u64)bio->bi_sector << 9;
3309 	u64 length = 0;
3310 	u64 map_length;
3311 	struct btrfs_multi_bio *multi = NULL;
3312 	int ret;
3313 	int dev_nr = 0;
3314 	int total_devs = 1;
3315 
3316 	length = bio->bi_size;
3317 	map_tree = &root->fs_info->mapping_tree;
3318 	map_length = length;
3319 
3320 	ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3321 			      mirror_num);
3322 	BUG_ON(ret);
3323 
3324 	total_devs = multi->num_stripes;
3325 	if (map_length < length) {
3326 		printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3327 		       "len %llu\n", (unsigned long long)logical,
3328 		       (unsigned long long)length,
3329 		       (unsigned long long)map_length);
3330 		BUG();
3331 	}
3332 	multi->end_io = first_bio->bi_end_io;
3333 	multi->private = first_bio->bi_private;
3334 	multi->orig_bio = first_bio;
3335 	atomic_set(&multi->stripes_pending, multi->num_stripes);
3336 
3337 	while (dev_nr < total_devs) {
3338 		if (total_devs > 1) {
3339 			if (dev_nr < total_devs - 1) {
3340 				bio = bio_clone(first_bio, GFP_NOFS);
3341 				BUG_ON(!bio);
3342 			} else {
3343 				bio = first_bio;
3344 			}
3345 			bio->bi_private = multi;
3346 			bio->bi_end_io = end_bio_multi_stripe;
3347 		}
3348 		bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3349 		dev = multi->stripes[dev_nr].dev;
3350 		if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3351 			bio->bi_bdev = dev->bdev;
3352 			if (async_submit)
3353 				schedule_bio(root, dev, rw, bio);
3354 			else
3355 				submit_bio(rw, bio);
3356 		} else {
3357 			bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3358 			bio->bi_sector = logical >> 9;
3359 			bio_endio(bio, -EIO);
3360 		}
3361 		dev_nr++;
3362 	}
3363 	if (total_devs == 1)
3364 		kfree(multi);
3365 	return 0;
3366 }
3367 
3368 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3369 				       u8 *uuid, u8 *fsid)
3370 {
3371 	struct btrfs_device *device;
3372 	struct btrfs_fs_devices *cur_devices;
3373 
3374 	cur_devices = root->fs_info->fs_devices;
3375 	while (cur_devices) {
3376 		if (!fsid ||
3377 		    !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3378 			device = __find_device(&cur_devices->devices,
3379 					       devid, uuid);
3380 			if (device)
3381 				return device;
3382 		}
3383 		cur_devices = cur_devices->seed;
3384 	}
3385 	return NULL;
3386 }
3387 
3388 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3389 					    u64 devid, u8 *dev_uuid)
3390 {
3391 	struct btrfs_device *device;
3392 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3393 
3394 	device = kzalloc(sizeof(*device), GFP_NOFS);
3395 	if (!device)
3396 		return NULL;
3397 	list_add(&device->dev_list,
3398 		 &fs_devices->devices);
3399 	device->dev_root = root->fs_info->dev_root;
3400 	device->devid = devid;
3401 	device->work.func = pending_bios_fn;
3402 	device->fs_devices = fs_devices;
3403 	device->missing = 1;
3404 	fs_devices->num_devices++;
3405 	fs_devices->missing_devices++;
3406 	spin_lock_init(&device->io_lock);
3407 	INIT_LIST_HEAD(&device->dev_alloc_list);
3408 	memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3409 	return device;
3410 }
3411 
3412 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3413 			  struct extent_buffer *leaf,
3414 			  struct btrfs_chunk *chunk)
3415 {
3416 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3417 	struct map_lookup *map;
3418 	struct extent_map *em;
3419 	u64 logical;
3420 	u64 length;
3421 	u64 devid;
3422 	u8 uuid[BTRFS_UUID_SIZE];
3423 	int num_stripes;
3424 	int ret;
3425 	int i;
3426 
3427 	logical = key->offset;
3428 	length = btrfs_chunk_length(leaf, chunk);
3429 
3430 	read_lock(&map_tree->map_tree.lock);
3431 	em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3432 	read_unlock(&map_tree->map_tree.lock);
3433 
3434 	/* already mapped? */
3435 	if (em && em->start <= logical && em->start + em->len > logical) {
3436 		free_extent_map(em);
3437 		return 0;
3438 	} else if (em) {
3439 		free_extent_map(em);
3440 	}
3441 
3442 	em = alloc_extent_map(GFP_NOFS);
3443 	if (!em)
3444 		return -ENOMEM;
3445 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3446 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3447 	if (!map) {
3448 		free_extent_map(em);
3449 		return -ENOMEM;
3450 	}
3451 
3452 	em->bdev = (struct block_device *)map;
3453 	em->start = logical;
3454 	em->len = length;
3455 	em->block_start = 0;
3456 	em->block_len = em->len;
3457 
3458 	map->num_stripes = num_stripes;
3459 	map->io_width = btrfs_chunk_io_width(leaf, chunk);
3460 	map->io_align = btrfs_chunk_io_align(leaf, chunk);
3461 	map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3462 	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3463 	map->type = btrfs_chunk_type(leaf, chunk);
3464 	map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3465 	for (i = 0; i < num_stripes; i++) {
3466 		map->stripes[i].physical =
3467 			btrfs_stripe_offset_nr(leaf, chunk, i);
3468 		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3469 		read_extent_buffer(leaf, uuid, (unsigned long)
3470 				   btrfs_stripe_dev_uuid_nr(chunk, i),
3471 				   BTRFS_UUID_SIZE);
3472 		map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3473 							NULL);
3474 		if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3475 			kfree(map);
3476 			free_extent_map(em);
3477 			return -EIO;
3478 		}
3479 		if (!map->stripes[i].dev) {
3480 			map->stripes[i].dev =
3481 				add_missing_dev(root, devid, uuid);
3482 			if (!map->stripes[i].dev) {
3483 				kfree(map);
3484 				free_extent_map(em);
3485 				return -EIO;
3486 			}
3487 		}
3488 		map->stripes[i].dev->in_fs_metadata = 1;
3489 	}
3490 
3491 	write_lock(&map_tree->map_tree.lock);
3492 	ret = add_extent_mapping(&map_tree->map_tree, em);
3493 	write_unlock(&map_tree->map_tree.lock);
3494 	BUG_ON(ret);
3495 	free_extent_map(em);
3496 
3497 	return 0;
3498 }
3499 
3500 static int fill_device_from_item(struct extent_buffer *leaf,
3501 				 struct btrfs_dev_item *dev_item,
3502 				 struct btrfs_device *device)
3503 {
3504 	unsigned long ptr;
3505 
3506 	device->devid = btrfs_device_id(leaf, dev_item);
3507 	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3508 	device->total_bytes = device->disk_total_bytes;
3509 	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3510 	device->type = btrfs_device_type(leaf, dev_item);
3511 	device->io_align = btrfs_device_io_align(leaf, dev_item);
3512 	device->io_width = btrfs_device_io_width(leaf, dev_item);
3513 	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3514 
3515 	ptr = (unsigned long)btrfs_device_uuid(dev_item);
3516 	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3517 
3518 	return 0;
3519 }
3520 
3521 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3522 {
3523 	struct btrfs_fs_devices *fs_devices;
3524 	int ret;
3525 
3526 	mutex_lock(&uuid_mutex);
3527 
3528 	fs_devices = root->fs_info->fs_devices->seed;
3529 	while (fs_devices) {
3530 		if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3531 			ret = 0;
3532 			goto out;
3533 		}
3534 		fs_devices = fs_devices->seed;
3535 	}
3536 
3537 	fs_devices = find_fsid(fsid);
3538 	if (!fs_devices) {
3539 		ret = -ENOENT;
3540 		goto out;
3541 	}
3542 
3543 	fs_devices = clone_fs_devices(fs_devices);
3544 	if (IS_ERR(fs_devices)) {
3545 		ret = PTR_ERR(fs_devices);
3546 		goto out;
3547 	}
3548 
3549 	ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3550 				   root->fs_info->bdev_holder);
3551 	if (ret)
3552 		goto out;
3553 
3554 	if (!fs_devices->seeding) {
3555 		__btrfs_close_devices(fs_devices);
3556 		free_fs_devices(fs_devices);
3557 		ret = -EINVAL;
3558 		goto out;
3559 	}
3560 
3561 	fs_devices->seed = root->fs_info->fs_devices->seed;
3562 	root->fs_info->fs_devices->seed = fs_devices;
3563 out:
3564 	mutex_unlock(&uuid_mutex);
3565 	return ret;
3566 }
3567 
3568 static int read_one_dev(struct btrfs_root *root,
3569 			struct extent_buffer *leaf,
3570 			struct btrfs_dev_item *dev_item)
3571 {
3572 	struct btrfs_device *device;
3573 	u64 devid;
3574 	int ret;
3575 	u8 fs_uuid[BTRFS_UUID_SIZE];
3576 	u8 dev_uuid[BTRFS_UUID_SIZE];
3577 
3578 	devid = btrfs_device_id(leaf, dev_item);
3579 	read_extent_buffer(leaf, dev_uuid,
3580 			   (unsigned long)btrfs_device_uuid(dev_item),
3581 			   BTRFS_UUID_SIZE);
3582 	read_extent_buffer(leaf, fs_uuid,
3583 			   (unsigned long)btrfs_device_fsid(dev_item),
3584 			   BTRFS_UUID_SIZE);
3585 
3586 	if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3587 		ret = open_seed_devices(root, fs_uuid);
3588 		if (ret && !btrfs_test_opt(root, DEGRADED))
3589 			return ret;
3590 	}
3591 
3592 	device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3593 	if (!device || !device->bdev) {
3594 		if (!btrfs_test_opt(root, DEGRADED))
3595 			return -EIO;
3596 
3597 		if (!device) {
3598 			printk(KERN_WARNING "warning devid %llu missing\n",
3599 			       (unsigned long long)devid);
3600 			device = add_missing_dev(root, devid, dev_uuid);
3601 			if (!device)
3602 				return -ENOMEM;
3603 		} else if (!device->missing) {
3604 			/*
3605 			 * this happens when a device that was properly setup
3606 			 * in the device info lists suddenly goes bad.
3607 			 * device->bdev is NULL, and so we have to set
3608 			 * device->missing to one here
3609 			 */
3610 			root->fs_info->fs_devices->missing_devices++;
3611 			device->missing = 1;
3612 		}
3613 	}
3614 
3615 	if (device->fs_devices != root->fs_info->fs_devices) {
3616 		BUG_ON(device->writeable);
3617 		if (device->generation !=
3618 		    btrfs_device_generation(leaf, dev_item))
3619 			return -EINVAL;
3620 	}
3621 
3622 	fill_device_from_item(leaf, dev_item, device);
3623 	device->dev_root = root->fs_info->dev_root;
3624 	device->in_fs_metadata = 1;
3625 	if (device->writeable)
3626 		device->fs_devices->total_rw_bytes += device->total_bytes;
3627 	ret = 0;
3628 	return ret;
3629 }
3630 
3631 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3632 {
3633 	struct btrfs_dev_item *dev_item;
3634 
3635 	dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3636 						     dev_item);
3637 	return read_one_dev(root, buf, dev_item);
3638 }
3639 
3640 int btrfs_read_sys_array(struct btrfs_root *root)
3641 {
3642 	struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3643 	struct extent_buffer *sb;
3644 	struct btrfs_disk_key *disk_key;
3645 	struct btrfs_chunk *chunk;
3646 	u8 *ptr;
3647 	unsigned long sb_ptr;
3648 	int ret = 0;
3649 	u32 num_stripes;
3650 	u32 array_size;
3651 	u32 len = 0;
3652 	u32 cur;
3653 	struct btrfs_key key;
3654 
3655 	sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3656 					  BTRFS_SUPER_INFO_SIZE);
3657 	if (!sb)
3658 		return -ENOMEM;
3659 	btrfs_set_buffer_uptodate(sb);
3660 	btrfs_set_buffer_lockdep_class(sb, 0);
3661 
3662 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3663 	array_size = btrfs_super_sys_array_size(super_copy);
3664 
3665 	ptr = super_copy->sys_chunk_array;
3666 	sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3667 	cur = 0;
3668 
3669 	while (cur < array_size) {
3670 		disk_key = (struct btrfs_disk_key *)ptr;
3671 		btrfs_disk_key_to_cpu(&key, disk_key);
3672 
3673 		len = sizeof(*disk_key); ptr += len;
3674 		sb_ptr += len;
3675 		cur += len;
3676 
3677 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3678 			chunk = (struct btrfs_chunk *)sb_ptr;
3679 			ret = read_one_chunk(root, &key, sb, chunk);
3680 			if (ret)
3681 				break;
3682 			num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3683 			len = btrfs_chunk_item_size(num_stripes);
3684 		} else {
3685 			ret = -EIO;
3686 			break;
3687 		}
3688 		ptr += len;
3689 		sb_ptr += len;
3690 		cur += len;
3691 	}
3692 	free_extent_buffer(sb);
3693 	return ret;
3694 }
3695 
3696 int btrfs_read_chunk_tree(struct btrfs_root *root)
3697 {
3698 	struct btrfs_path *path;
3699 	struct extent_buffer *leaf;
3700 	struct btrfs_key key;
3701 	struct btrfs_key found_key;
3702 	int ret;
3703 	int slot;
3704 
3705 	root = root->fs_info->chunk_root;
3706 
3707 	path = btrfs_alloc_path();
3708 	if (!path)
3709 		return -ENOMEM;
3710 
3711 	/* first we search for all of the device items, and then we
3712 	 * read in all of the chunk items.  This way we can create chunk
3713 	 * mappings that reference all of the devices that are afound
3714 	 */
3715 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3716 	key.offset = 0;
3717 	key.type = 0;
3718 again:
3719 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3720 	if (ret < 0)
3721 		goto error;
3722 	while (1) {
3723 		leaf = path->nodes[0];
3724 		slot = path->slots[0];
3725 		if (slot >= btrfs_header_nritems(leaf)) {
3726 			ret = btrfs_next_leaf(root, path);
3727 			if (ret == 0)
3728 				continue;
3729 			if (ret < 0)
3730 				goto error;
3731 			break;
3732 		}
3733 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3734 		if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3735 			if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3736 				break;
3737 			if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3738 				struct btrfs_dev_item *dev_item;
3739 				dev_item = btrfs_item_ptr(leaf, slot,
3740 						  struct btrfs_dev_item);
3741 				ret = read_one_dev(root, leaf, dev_item);
3742 				if (ret)
3743 					goto error;
3744 			}
3745 		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3746 			struct btrfs_chunk *chunk;
3747 			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3748 			ret = read_one_chunk(root, &found_key, leaf, chunk);
3749 			if (ret)
3750 				goto error;
3751 		}
3752 		path->slots[0]++;
3753 	}
3754 	if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3755 		key.objectid = 0;
3756 		btrfs_release_path(root, path);
3757 		goto again;
3758 	}
3759 	ret = 0;
3760 error:
3761 	btrfs_free_path(path);
3762 	return ret;
3763 }
3764