xref: /openbmc/linux/fs/btrfs/volumes.c (revision 5104d265)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <asm/div64.h>
30 #include "compat.h"
31 #include "ctree.h"
32 #include "extent_map.h"
33 #include "disk-io.h"
34 #include "transaction.h"
35 #include "print-tree.h"
36 #include "volumes.h"
37 #include "raid56.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
41 #include "math.h"
42 #include "dev-replace.h"
43 
44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
45 				struct btrfs_root *root,
46 				struct btrfs_device *device);
47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
48 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
49 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
50 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
51 
52 static DEFINE_MUTEX(uuid_mutex);
53 static LIST_HEAD(fs_uuids);
54 
55 static void lock_chunks(struct btrfs_root *root)
56 {
57 	mutex_lock(&root->fs_info->chunk_mutex);
58 }
59 
60 static void unlock_chunks(struct btrfs_root *root)
61 {
62 	mutex_unlock(&root->fs_info->chunk_mutex);
63 }
64 
65 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
66 {
67 	struct btrfs_device *device;
68 	WARN_ON(fs_devices->opened);
69 	while (!list_empty(&fs_devices->devices)) {
70 		device = list_entry(fs_devices->devices.next,
71 				    struct btrfs_device, dev_list);
72 		list_del(&device->dev_list);
73 		rcu_string_free(device->name);
74 		kfree(device);
75 	}
76 	kfree(fs_devices);
77 }
78 
79 static void btrfs_kobject_uevent(struct block_device *bdev,
80 				 enum kobject_action action)
81 {
82 	int ret;
83 
84 	ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
85 	if (ret)
86 		pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
87 			action,
88 			kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
89 			&disk_to_dev(bdev->bd_disk)->kobj);
90 }
91 
92 void btrfs_cleanup_fs_uuids(void)
93 {
94 	struct btrfs_fs_devices *fs_devices;
95 
96 	while (!list_empty(&fs_uuids)) {
97 		fs_devices = list_entry(fs_uuids.next,
98 					struct btrfs_fs_devices, list);
99 		list_del(&fs_devices->list);
100 		free_fs_devices(fs_devices);
101 	}
102 }
103 
104 static noinline struct btrfs_device *__find_device(struct list_head *head,
105 						   u64 devid, u8 *uuid)
106 {
107 	struct btrfs_device *dev;
108 
109 	list_for_each_entry(dev, head, dev_list) {
110 		if (dev->devid == devid &&
111 		    (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
112 			return dev;
113 		}
114 	}
115 	return NULL;
116 }
117 
118 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
119 {
120 	struct btrfs_fs_devices *fs_devices;
121 
122 	list_for_each_entry(fs_devices, &fs_uuids, list) {
123 		if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
124 			return fs_devices;
125 	}
126 	return NULL;
127 }
128 
129 static int
130 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
131 		      int flush, struct block_device **bdev,
132 		      struct buffer_head **bh)
133 {
134 	int ret;
135 
136 	*bdev = blkdev_get_by_path(device_path, flags, holder);
137 
138 	if (IS_ERR(*bdev)) {
139 		ret = PTR_ERR(*bdev);
140 		printk(KERN_INFO "btrfs: open %s failed\n", device_path);
141 		goto error;
142 	}
143 
144 	if (flush)
145 		filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
146 	ret = set_blocksize(*bdev, 4096);
147 	if (ret) {
148 		blkdev_put(*bdev, flags);
149 		goto error;
150 	}
151 	invalidate_bdev(*bdev);
152 	*bh = btrfs_read_dev_super(*bdev);
153 	if (!*bh) {
154 		ret = -EINVAL;
155 		blkdev_put(*bdev, flags);
156 		goto error;
157 	}
158 
159 	return 0;
160 
161 error:
162 	*bdev = NULL;
163 	*bh = NULL;
164 	return ret;
165 }
166 
167 static void requeue_list(struct btrfs_pending_bios *pending_bios,
168 			struct bio *head, struct bio *tail)
169 {
170 
171 	struct bio *old_head;
172 
173 	old_head = pending_bios->head;
174 	pending_bios->head = head;
175 	if (pending_bios->tail)
176 		tail->bi_next = old_head;
177 	else
178 		pending_bios->tail = tail;
179 }
180 
181 /*
182  * we try to collect pending bios for a device so we don't get a large
183  * number of procs sending bios down to the same device.  This greatly
184  * improves the schedulers ability to collect and merge the bios.
185  *
186  * But, it also turns into a long list of bios to process and that is sure
187  * to eventually make the worker thread block.  The solution here is to
188  * make some progress and then put this work struct back at the end of
189  * the list if the block device is congested.  This way, multiple devices
190  * can make progress from a single worker thread.
191  */
192 static noinline void run_scheduled_bios(struct btrfs_device *device)
193 {
194 	struct bio *pending;
195 	struct backing_dev_info *bdi;
196 	struct btrfs_fs_info *fs_info;
197 	struct btrfs_pending_bios *pending_bios;
198 	struct bio *tail;
199 	struct bio *cur;
200 	int again = 0;
201 	unsigned long num_run;
202 	unsigned long batch_run = 0;
203 	unsigned long limit;
204 	unsigned long last_waited = 0;
205 	int force_reg = 0;
206 	int sync_pending = 0;
207 	struct blk_plug plug;
208 
209 	/*
210 	 * this function runs all the bios we've collected for
211 	 * a particular device.  We don't want to wander off to
212 	 * another device without first sending all of these down.
213 	 * So, setup a plug here and finish it off before we return
214 	 */
215 	blk_start_plug(&plug);
216 
217 	bdi = blk_get_backing_dev_info(device->bdev);
218 	fs_info = device->dev_root->fs_info;
219 	limit = btrfs_async_submit_limit(fs_info);
220 	limit = limit * 2 / 3;
221 
222 loop:
223 	spin_lock(&device->io_lock);
224 
225 loop_lock:
226 	num_run = 0;
227 
228 	/* take all the bios off the list at once and process them
229 	 * later on (without the lock held).  But, remember the
230 	 * tail and other pointers so the bios can be properly reinserted
231 	 * into the list if we hit congestion
232 	 */
233 	if (!force_reg && device->pending_sync_bios.head) {
234 		pending_bios = &device->pending_sync_bios;
235 		force_reg = 1;
236 	} else {
237 		pending_bios = &device->pending_bios;
238 		force_reg = 0;
239 	}
240 
241 	pending = pending_bios->head;
242 	tail = pending_bios->tail;
243 	WARN_ON(pending && !tail);
244 
245 	/*
246 	 * if pending was null this time around, no bios need processing
247 	 * at all and we can stop.  Otherwise it'll loop back up again
248 	 * and do an additional check so no bios are missed.
249 	 *
250 	 * device->running_pending is used to synchronize with the
251 	 * schedule_bio code.
252 	 */
253 	if (device->pending_sync_bios.head == NULL &&
254 	    device->pending_bios.head == NULL) {
255 		again = 0;
256 		device->running_pending = 0;
257 	} else {
258 		again = 1;
259 		device->running_pending = 1;
260 	}
261 
262 	pending_bios->head = NULL;
263 	pending_bios->tail = NULL;
264 
265 	spin_unlock(&device->io_lock);
266 
267 	while (pending) {
268 
269 		rmb();
270 		/* we want to work on both lists, but do more bios on the
271 		 * sync list than the regular list
272 		 */
273 		if ((num_run > 32 &&
274 		    pending_bios != &device->pending_sync_bios &&
275 		    device->pending_sync_bios.head) ||
276 		   (num_run > 64 && pending_bios == &device->pending_sync_bios &&
277 		    device->pending_bios.head)) {
278 			spin_lock(&device->io_lock);
279 			requeue_list(pending_bios, pending, tail);
280 			goto loop_lock;
281 		}
282 
283 		cur = pending;
284 		pending = pending->bi_next;
285 		cur->bi_next = NULL;
286 
287 		if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
288 		    waitqueue_active(&fs_info->async_submit_wait))
289 			wake_up(&fs_info->async_submit_wait);
290 
291 		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
292 
293 		/*
294 		 * if we're doing the sync list, record that our
295 		 * plug has some sync requests on it
296 		 *
297 		 * If we're doing the regular list and there are
298 		 * sync requests sitting around, unplug before
299 		 * we add more
300 		 */
301 		if (pending_bios == &device->pending_sync_bios) {
302 			sync_pending = 1;
303 		} else if (sync_pending) {
304 			blk_finish_plug(&plug);
305 			blk_start_plug(&plug);
306 			sync_pending = 0;
307 		}
308 
309 		btrfsic_submit_bio(cur->bi_rw, cur);
310 		num_run++;
311 		batch_run++;
312 		if (need_resched())
313 			cond_resched();
314 
315 		/*
316 		 * we made progress, there is more work to do and the bdi
317 		 * is now congested.  Back off and let other work structs
318 		 * run instead
319 		 */
320 		if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
321 		    fs_info->fs_devices->open_devices > 1) {
322 			struct io_context *ioc;
323 
324 			ioc = current->io_context;
325 
326 			/*
327 			 * the main goal here is that we don't want to
328 			 * block if we're going to be able to submit
329 			 * more requests without blocking.
330 			 *
331 			 * This code does two great things, it pokes into
332 			 * the elevator code from a filesystem _and_
333 			 * it makes assumptions about how batching works.
334 			 */
335 			if (ioc && ioc->nr_batch_requests > 0 &&
336 			    time_before(jiffies, ioc->last_waited + HZ/50UL) &&
337 			    (last_waited == 0 ||
338 			     ioc->last_waited == last_waited)) {
339 				/*
340 				 * we want to go through our batch of
341 				 * requests and stop.  So, we copy out
342 				 * the ioc->last_waited time and test
343 				 * against it before looping
344 				 */
345 				last_waited = ioc->last_waited;
346 				if (need_resched())
347 					cond_resched();
348 				continue;
349 			}
350 			spin_lock(&device->io_lock);
351 			requeue_list(pending_bios, pending, tail);
352 			device->running_pending = 1;
353 
354 			spin_unlock(&device->io_lock);
355 			btrfs_requeue_work(&device->work);
356 			goto done;
357 		}
358 		/* unplug every 64 requests just for good measure */
359 		if (batch_run % 64 == 0) {
360 			blk_finish_plug(&plug);
361 			blk_start_plug(&plug);
362 			sync_pending = 0;
363 		}
364 	}
365 
366 	cond_resched();
367 	if (again)
368 		goto loop;
369 
370 	spin_lock(&device->io_lock);
371 	if (device->pending_bios.head || device->pending_sync_bios.head)
372 		goto loop_lock;
373 	spin_unlock(&device->io_lock);
374 
375 done:
376 	blk_finish_plug(&plug);
377 }
378 
379 static void pending_bios_fn(struct btrfs_work *work)
380 {
381 	struct btrfs_device *device;
382 
383 	device = container_of(work, struct btrfs_device, work);
384 	run_scheduled_bios(device);
385 }
386 
387 static noinline int device_list_add(const char *path,
388 			   struct btrfs_super_block *disk_super,
389 			   u64 devid, struct btrfs_fs_devices **fs_devices_ret)
390 {
391 	struct btrfs_device *device;
392 	struct btrfs_fs_devices *fs_devices;
393 	struct rcu_string *name;
394 	u64 found_transid = btrfs_super_generation(disk_super);
395 
396 	fs_devices = find_fsid(disk_super->fsid);
397 	if (!fs_devices) {
398 		fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
399 		if (!fs_devices)
400 			return -ENOMEM;
401 		INIT_LIST_HEAD(&fs_devices->devices);
402 		INIT_LIST_HEAD(&fs_devices->alloc_list);
403 		list_add(&fs_devices->list, &fs_uuids);
404 		memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
405 		fs_devices->latest_devid = devid;
406 		fs_devices->latest_trans = found_transid;
407 		mutex_init(&fs_devices->device_list_mutex);
408 		device = NULL;
409 	} else {
410 		device = __find_device(&fs_devices->devices, devid,
411 				       disk_super->dev_item.uuid);
412 	}
413 	if (!device) {
414 		if (fs_devices->opened)
415 			return -EBUSY;
416 
417 		device = kzalloc(sizeof(*device), GFP_NOFS);
418 		if (!device) {
419 			/* we can safely leave the fs_devices entry around */
420 			return -ENOMEM;
421 		}
422 		device->devid = devid;
423 		device->dev_stats_valid = 0;
424 		device->work.func = pending_bios_fn;
425 		memcpy(device->uuid, disk_super->dev_item.uuid,
426 		       BTRFS_UUID_SIZE);
427 		spin_lock_init(&device->io_lock);
428 
429 		name = rcu_string_strdup(path, GFP_NOFS);
430 		if (!name) {
431 			kfree(device);
432 			return -ENOMEM;
433 		}
434 		rcu_assign_pointer(device->name, name);
435 		INIT_LIST_HEAD(&device->dev_alloc_list);
436 
437 		/* init readahead state */
438 		spin_lock_init(&device->reada_lock);
439 		device->reada_curr_zone = NULL;
440 		atomic_set(&device->reada_in_flight, 0);
441 		device->reada_next = 0;
442 		INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
443 		INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
444 
445 		mutex_lock(&fs_devices->device_list_mutex);
446 		list_add_rcu(&device->dev_list, &fs_devices->devices);
447 		mutex_unlock(&fs_devices->device_list_mutex);
448 
449 		device->fs_devices = fs_devices;
450 		fs_devices->num_devices++;
451 	} else if (!device->name || strcmp(device->name->str, path)) {
452 		name = rcu_string_strdup(path, GFP_NOFS);
453 		if (!name)
454 			return -ENOMEM;
455 		rcu_string_free(device->name);
456 		rcu_assign_pointer(device->name, name);
457 		if (device->missing) {
458 			fs_devices->missing_devices--;
459 			device->missing = 0;
460 		}
461 	}
462 
463 	if (found_transid > fs_devices->latest_trans) {
464 		fs_devices->latest_devid = devid;
465 		fs_devices->latest_trans = found_transid;
466 	}
467 	*fs_devices_ret = fs_devices;
468 	return 0;
469 }
470 
471 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
472 {
473 	struct btrfs_fs_devices *fs_devices;
474 	struct btrfs_device *device;
475 	struct btrfs_device *orig_dev;
476 
477 	fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
478 	if (!fs_devices)
479 		return ERR_PTR(-ENOMEM);
480 
481 	INIT_LIST_HEAD(&fs_devices->devices);
482 	INIT_LIST_HEAD(&fs_devices->alloc_list);
483 	INIT_LIST_HEAD(&fs_devices->list);
484 	mutex_init(&fs_devices->device_list_mutex);
485 	fs_devices->latest_devid = orig->latest_devid;
486 	fs_devices->latest_trans = orig->latest_trans;
487 	fs_devices->total_devices = orig->total_devices;
488 	memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
489 
490 	/* We have held the volume lock, it is safe to get the devices. */
491 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
492 		struct rcu_string *name;
493 
494 		device = kzalloc(sizeof(*device), GFP_NOFS);
495 		if (!device)
496 			goto error;
497 
498 		/*
499 		 * This is ok to do without rcu read locked because we hold the
500 		 * uuid mutex so nothing we touch in here is going to disappear.
501 		 */
502 		name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
503 		if (!name) {
504 			kfree(device);
505 			goto error;
506 		}
507 		rcu_assign_pointer(device->name, name);
508 
509 		device->devid = orig_dev->devid;
510 		device->work.func = pending_bios_fn;
511 		memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
512 		spin_lock_init(&device->io_lock);
513 		INIT_LIST_HEAD(&device->dev_list);
514 		INIT_LIST_HEAD(&device->dev_alloc_list);
515 
516 		list_add(&device->dev_list, &fs_devices->devices);
517 		device->fs_devices = fs_devices;
518 		fs_devices->num_devices++;
519 	}
520 	return fs_devices;
521 error:
522 	free_fs_devices(fs_devices);
523 	return ERR_PTR(-ENOMEM);
524 }
525 
526 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
527 			       struct btrfs_fs_devices *fs_devices, int step)
528 {
529 	struct btrfs_device *device, *next;
530 
531 	struct block_device *latest_bdev = NULL;
532 	u64 latest_devid = 0;
533 	u64 latest_transid = 0;
534 
535 	mutex_lock(&uuid_mutex);
536 again:
537 	/* This is the initialized path, it is safe to release the devices. */
538 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
539 		if (device->in_fs_metadata) {
540 			if (!device->is_tgtdev_for_dev_replace &&
541 			    (!latest_transid ||
542 			     device->generation > latest_transid)) {
543 				latest_devid = device->devid;
544 				latest_transid = device->generation;
545 				latest_bdev = device->bdev;
546 			}
547 			continue;
548 		}
549 
550 		if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
551 			/*
552 			 * In the first step, keep the device which has
553 			 * the correct fsid and the devid that is used
554 			 * for the dev_replace procedure.
555 			 * In the second step, the dev_replace state is
556 			 * read from the device tree and it is known
557 			 * whether the procedure is really active or
558 			 * not, which means whether this device is
559 			 * used or whether it should be removed.
560 			 */
561 			if (step == 0 || device->is_tgtdev_for_dev_replace) {
562 				continue;
563 			}
564 		}
565 		if (device->bdev) {
566 			blkdev_put(device->bdev, device->mode);
567 			device->bdev = NULL;
568 			fs_devices->open_devices--;
569 		}
570 		if (device->writeable) {
571 			list_del_init(&device->dev_alloc_list);
572 			device->writeable = 0;
573 			if (!device->is_tgtdev_for_dev_replace)
574 				fs_devices->rw_devices--;
575 		}
576 		list_del_init(&device->dev_list);
577 		fs_devices->num_devices--;
578 		rcu_string_free(device->name);
579 		kfree(device);
580 	}
581 
582 	if (fs_devices->seed) {
583 		fs_devices = fs_devices->seed;
584 		goto again;
585 	}
586 
587 	fs_devices->latest_bdev = latest_bdev;
588 	fs_devices->latest_devid = latest_devid;
589 	fs_devices->latest_trans = latest_transid;
590 
591 	mutex_unlock(&uuid_mutex);
592 }
593 
594 static void __free_device(struct work_struct *work)
595 {
596 	struct btrfs_device *device;
597 
598 	device = container_of(work, struct btrfs_device, rcu_work);
599 
600 	if (device->bdev)
601 		blkdev_put(device->bdev, device->mode);
602 
603 	rcu_string_free(device->name);
604 	kfree(device);
605 }
606 
607 static void free_device(struct rcu_head *head)
608 {
609 	struct btrfs_device *device;
610 
611 	device = container_of(head, struct btrfs_device, rcu);
612 
613 	INIT_WORK(&device->rcu_work, __free_device);
614 	schedule_work(&device->rcu_work);
615 }
616 
617 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
618 {
619 	struct btrfs_device *device;
620 
621 	if (--fs_devices->opened > 0)
622 		return 0;
623 
624 	mutex_lock(&fs_devices->device_list_mutex);
625 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
626 		struct btrfs_device *new_device;
627 		struct rcu_string *name;
628 
629 		if (device->bdev)
630 			fs_devices->open_devices--;
631 
632 		if (device->writeable && !device->is_tgtdev_for_dev_replace) {
633 			list_del_init(&device->dev_alloc_list);
634 			fs_devices->rw_devices--;
635 		}
636 
637 		if (device->can_discard)
638 			fs_devices->num_can_discard--;
639 
640 		new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
641 		BUG_ON(!new_device); /* -ENOMEM */
642 		memcpy(new_device, device, sizeof(*new_device));
643 
644 		/* Safe because we are under uuid_mutex */
645 		if (device->name) {
646 			name = rcu_string_strdup(device->name->str, GFP_NOFS);
647 			BUG_ON(device->name && !name); /* -ENOMEM */
648 			rcu_assign_pointer(new_device->name, name);
649 		}
650 		new_device->bdev = NULL;
651 		new_device->writeable = 0;
652 		new_device->in_fs_metadata = 0;
653 		new_device->can_discard = 0;
654 		spin_lock_init(&new_device->io_lock);
655 		list_replace_rcu(&device->dev_list, &new_device->dev_list);
656 
657 		call_rcu(&device->rcu, free_device);
658 	}
659 	mutex_unlock(&fs_devices->device_list_mutex);
660 
661 	WARN_ON(fs_devices->open_devices);
662 	WARN_ON(fs_devices->rw_devices);
663 	fs_devices->opened = 0;
664 	fs_devices->seeding = 0;
665 
666 	return 0;
667 }
668 
669 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
670 {
671 	struct btrfs_fs_devices *seed_devices = NULL;
672 	int ret;
673 
674 	mutex_lock(&uuid_mutex);
675 	ret = __btrfs_close_devices(fs_devices);
676 	if (!fs_devices->opened) {
677 		seed_devices = fs_devices->seed;
678 		fs_devices->seed = NULL;
679 	}
680 	mutex_unlock(&uuid_mutex);
681 
682 	while (seed_devices) {
683 		fs_devices = seed_devices;
684 		seed_devices = fs_devices->seed;
685 		__btrfs_close_devices(fs_devices);
686 		free_fs_devices(fs_devices);
687 	}
688 	/*
689 	 * Wait for rcu kworkers under __btrfs_close_devices
690 	 * to finish all blkdev_puts so device is really
691 	 * free when umount is done.
692 	 */
693 	rcu_barrier();
694 	return ret;
695 }
696 
697 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
698 				fmode_t flags, void *holder)
699 {
700 	struct request_queue *q;
701 	struct block_device *bdev;
702 	struct list_head *head = &fs_devices->devices;
703 	struct btrfs_device *device;
704 	struct block_device *latest_bdev = NULL;
705 	struct buffer_head *bh;
706 	struct btrfs_super_block *disk_super;
707 	u64 latest_devid = 0;
708 	u64 latest_transid = 0;
709 	u64 devid;
710 	int seeding = 1;
711 	int ret = 0;
712 
713 	flags |= FMODE_EXCL;
714 
715 	list_for_each_entry(device, head, dev_list) {
716 		if (device->bdev)
717 			continue;
718 		if (!device->name)
719 			continue;
720 
721 		/* Just open everything we can; ignore failures here */
722 		if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
723 					    &bdev, &bh))
724 			continue;
725 
726 		disk_super = (struct btrfs_super_block *)bh->b_data;
727 		devid = btrfs_stack_device_id(&disk_super->dev_item);
728 		if (devid != device->devid)
729 			goto error_brelse;
730 
731 		if (memcmp(device->uuid, disk_super->dev_item.uuid,
732 			   BTRFS_UUID_SIZE))
733 			goto error_brelse;
734 
735 		device->generation = btrfs_super_generation(disk_super);
736 		if (!latest_transid || device->generation > latest_transid) {
737 			latest_devid = devid;
738 			latest_transid = device->generation;
739 			latest_bdev = bdev;
740 		}
741 
742 		if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
743 			device->writeable = 0;
744 		} else {
745 			device->writeable = !bdev_read_only(bdev);
746 			seeding = 0;
747 		}
748 
749 		q = bdev_get_queue(bdev);
750 		if (blk_queue_discard(q)) {
751 			device->can_discard = 1;
752 			fs_devices->num_can_discard++;
753 		}
754 
755 		device->bdev = bdev;
756 		device->in_fs_metadata = 0;
757 		device->mode = flags;
758 
759 		if (!blk_queue_nonrot(bdev_get_queue(bdev)))
760 			fs_devices->rotating = 1;
761 
762 		fs_devices->open_devices++;
763 		if (device->writeable && !device->is_tgtdev_for_dev_replace) {
764 			fs_devices->rw_devices++;
765 			list_add(&device->dev_alloc_list,
766 				 &fs_devices->alloc_list);
767 		}
768 		brelse(bh);
769 		continue;
770 
771 error_brelse:
772 		brelse(bh);
773 		blkdev_put(bdev, flags);
774 		continue;
775 	}
776 	if (fs_devices->open_devices == 0) {
777 		ret = -EINVAL;
778 		goto out;
779 	}
780 	fs_devices->seeding = seeding;
781 	fs_devices->opened = 1;
782 	fs_devices->latest_bdev = latest_bdev;
783 	fs_devices->latest_devid = latest_devid;
784 	fs_devices->latest_trans = latest_transid;
785 	fs_devices->total_rw_bytes = 0;
786 out:
787 	return ret;
788 }
789 
790 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
791 		       fmode_t flags, void *holder)
792 {
793 	int ret;
794 
795 	mutex_lock(&uuid_mutex);
796 	if (fs_devices->opened) {
797 		fs_devices->opened++;
798 		ret = 0;
799 	} else {
800 		ret = __btrfs_open_devices(fs_devices, flags, holder);
801 	}
802 	mutex_unlock(&uuid_mutex);
803 	return ret;
804 }
805 
806 /*
807  * Look for a btrfs signature on a device. This may be called out of the mount path
808  * and we are not allowed to call set_blocksize during the scan. The superblock
809  * is read via pagecache
810  */
811 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
812 			  struct btrfs_fs_devices **fs_devices_ret)
813 {
814 	struct btrfs_super_block *disk_super;
815 	struct block_device *bdev;
816 	struct page *page;
817 	void *p;
818 	int ret = -EINVAL;
819 	u64 devid;
820 	u64 transid;
821 	u64 total_devices;
822 	u64 bytenr;
823 	pgoff_t index;
824 
825 	/*
826 	 * we would like to check all the supers, but that would make
827 	 * a btrfs mount succeed after a mkfs from a different FS.
828 	 * So, we need to add a special mount option to scan for
829 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
830 	 */
831 	bytenr = btrfs_sb_offset(0);
832 	flags |= FMODE_EXCL;
833 	mutex_lock(&uuid_mutex);
834 
835 	bdev = blkdev_get_by_path(path, flags, holder);
836 
837 	if (IS_ERR(bdev)) {
838 		ret = PTR_ERR(bdev);
839 		goto error;
840 	}
841 
842 	/* make sure our super fits in the device */
843 	if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
844 		goto error_bdev_put;
845 
846 	/* make sure our super fits in the page */
847 	if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
848 		goto error_bdev_put;
849 
850 	/* make sure our super doesn't straddle pages on disk */
851 	index = bytenr >> PAGE_CACHE_SHIFT;
852 	if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
853 		goto error_bdev_put;
854 
855 	/* pull in the page with our super */
856 	page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
857 				   index, GFP_NOFS);
858 
859 	if (IS_ERR_OR_NULL(page))
860 		goto error_bdev_put;
861 
862 	p = kmap(page);
863 
864 	/* align our pointer to the offset of the super block */
865 	disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
866 
867 	if (btrfs_super_bytenr(disk_super) != bytenr ||
868 	    disk_super->magic != cpu_to_le64(BTRFS_MAGIC))
869 		goto error_unmap;
870 
871 	devid = btrfs_stack_device_id(&disk_super->dev_item);
872 	transid = btrfs_super_generation(disk_super);
873 	total_devices = btrfs_super_num_devices(disk_super);
874 
875 	if (disk_super->label[0]) {
876 		if (disk_super->label[BTRFS_LABEL_SIZE - 1])
877 			disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
878 		printk(KERN_INFO "device label %s ", disk_super->label);
879 	} else {
880 		printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
881 	}
882 
883 	printk(KERN_CONT "devid %llu transid %llu %s\n",
884 	       (unsigned long long)devid, (unsigned long long)transid, path);
885 
886 	ret = device_list_add(path, disk_super, devid, fs_devices_ret);
887 	if (!ret && fs_devices_ret)
888 		(*fs_devices_ret)->total_devices = total_devices;
889 
890 error_unmap:
891 	kunmap(page);
892 	page_cache_release(page);
893 
894 error_bdev_put:
895 	blkdev_put(bdev, flags);
896 error:
897 	mutex_unlock(&uuid_mutex);
898 	return ret;
899 }
900 
901 /* helper to account the used device space in the range */
902 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
903 				   u64 end, u64 *length)
904 {
905 	struct btrfs_key key;
906 	struct btrfs_root *root = device->dev_root;
907 	struct btrfs_dev_extent *dev_extent;
908 	struct btrfs_path *path;
909 	u64 extent_end;
910 	int ret;
911 	int slot;
912 	struct extent_buffer *l;
913 
914 	*length = 0;
915 
916 	if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
917 		return 0;
918 
919 	path = btrfs_alloc_path();
920 	if (!path)
921 		return -ENOMEM;
922 	path->reada = 2;
923 
924 	key.objectid = device->devid;
925 	key.offset = start;
926 	key.type = BTRFS_DEV_EXTENT_KEY;
927 
928 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
929 	if (ret < 0)
930 		goto out;
931 	if (ret > 0) {
932 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
933 		if (ret < 0)
934 			goto out;
935 	}
936 
937 	while (1) {
938 		l = path->nodes[0];
939 		slot = path->slots[0];
940 		if (slot >= btrfs_header_nritems(l)) {
941 			ret = btrfs_next_leaf(root, path);
942 			if (ret == 0)
943 				continue;
944 			if (ret < 0)
945 				goto out;
946 
947 			break;
948 		}
949 		btrfs_item_key_to_cpu(l, &key, slot);
950 
951 		if (key.objectid < device->devid)
952 			goto next;
953 
954 		if (key.objectid > device->devid)
955 			break;
956 
957 		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
958 			goto next;
959 
960 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
961 		extent_end = key.offset + btrfs_dev_extent_length(l,
962 								  dev_extent);
963 		if (key.offset <= start && extent_end > end) {
964 			*length = end - start + 1;
965 			break;
966 		} else if (key.offset <= start && extent_end > start)
967 			*length += extent_end - start;
968 		else if (key.offset > start && extent_end <= end)
969 			*length += extent_end - key.offset;
970 		else if (key.offset > start && key.offset <= end) {
971 			*length += end - key.offset + 1;
972 			break;
973 		} else if (key.offset > end)
974 			break;
975 
976 next:
977 		path->slots[0]++;
978 	}
979 	ret = 0;
980 out:
981 	btrfs_free_path(path);
982 	return ret;
983 }
984 
985 static int contains_pending_extent(struct btrfs_trans_handle *trans,
986 				   struct btrfs_device *device,
987 				   u64 *start, u64 len)
988 {
989 	struct extent_map *em;
990 	int ret = 0;
991 
992 	list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
993 		struct map_lookup *map;
994 		int i;
995 
996 		map = (struct map_lookup *)em->bdev;
997 		for (i = 0; i < map->num_stripes; i++) {
998 			if (map->stripes[i].dev != device)
999 				continue;
1000 			if (map->stripes[i].physical >= *start + len ||
1001 			    map->stripes[i].physical + em->orig_block_len <=
1002 			    *start)
1003 				continue;
1004 			*start = map->stripes[i].physical +
1005 				em->orig_block_len;
1006 			ret = 1;
1007 		}
1008 	}
1009 
1010 	return ret;
1011 }
1012 
1013 
1014 /*
1015  * find_free_dev_extent - find free space in the specified device
1016  * @device:	the device which we search the free space in
1017  * @num_bytes:	the size of the free space that we need
1018  * @start:	store the start of the free space.
1019  * @len:	the size of the free space. that we find, or the size of the max
1020  * 		free space if we don't find suitable free space
1021  *
1022  * this uses a pretty simple search, the expectation is that it is
1023  * called very infrequently and that a given device has a small number
1024  * of extents
1025  *
1026  * @start is used to store the start of the free space if we find. But if we
1027  * don't find suitable free space, it will be used to store the start position
1028  * of the max free space.
1029  *
1030  * @len is used to store the size of the free space that we find.
1031  * But if we don't find suitable free space, it is used to store the size of
1032  * the max free space.
1033  */
1034 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1035 			 struct btrfs_device *device, u64 num_bytes,
1036 			 u64 *start, u64 *len)
1037 {
1038 	struct btrfs_key key;
1039 	struct btrfs_root *root = device->dev_root;
1040 	struct btrfs_dev_extent *dev_extent;
1041 	struct btrfs_path *path;
1042 	u64 hole_size;
1043 	u64 max_hole_start;
1044 	u64 max_hole_size;
1045 	u64 extent_end;
1046 	u64 search_start;
1047 	u64 search_end = device->total_bytes;
1048 	int ret;
1049 	int slot;
1050 	struct extent_buffer *l;
1051 
1052 	/* FIXME use last free of some kind */
1053 
1054 	/* we don't want to overwrite the superblock on the drive,
1055 	 * so we make sure to start at an offset of at least 1MB
1056 	 */
1057 	search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1058 
1059 	path = btrfs_alloc_path();
1060 	if (!path)
1061 		return -ENOMEM;
1062 again:
1063 	max_hole_start = search_start;
1064 	max_hole_size = 0;
1065 	hole_size = 0;
1066 
1067 	if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1068 		ret = -ENOSPC;
1069 		goto out;
1070 	}
1071 
1072 	path->reada = 2;
1073 	path->search_commit_root = 1;
1074 	path->skip_locking = 1;
1075 
1076 	key.objectid = device->devid;
1077 	key.offset = search_start;
1078 	key.type = BTRFS_DEV_EXTENT_KEY;
1079 
1080 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1081 	if (ret < 0)
1082 		goto out;
1083 	if (ret > 0) {
1084 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
1085 		if (ret < 0)
1086 			goto out;
1087 	}
1088 
1089 	while (1) {
1090 		l = path->nodes[0];
1091 		slot = path->slots[0];
1092 		if (slot >= btrfs_header_nritems(l)) {
1093 			ret = btrfs_next_leaf(root, path);
1094 			if (ret == 0)
1095 				continue;
1096 			if (ret < 0)
1097 				goto out;
1098 
1099 			break;
1100 		}
1101 		btrfs_item_key_to_cpu(l, &key, slot);
1102 
1103 		if (key.objectid < device->devid)
1104 			goto next;
1105 
1106 		if (key.objectid > device->devid)
1107 			break;
1108 
1109 		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1110 			goto next;
1111 
1112 		if (key.offset > search_start) {
1113 			hole_size = key.offset - search_start;
1114 
1115 			/*
1116 			 * Have to check before we set max_hole_start, otherwise
1117 			 * we could end up sending back this offset anyway.
1118 			 */
1119 			if (contains_pending_extent(trans, device,
1120 						    &search_start,
1121 						    hole_size))
1122 				hole_size = 0;
1123 
1124 			if (hole_size > max_hole_size) {
1125 				max_hole_start = search_start;
1126 				max_hole_size = hole_size;
1127 			}
1128 
1129 			/*
1130 			 * If this free space is greater than which we need,
1131 			 * it must be the max free space that we have found
1132 			 * until now, so max_hole_start must point to the start
1133 			 * of this free space and the length of this free space
1134 			 * is stored in max_hole_size. Thus, we return
1135 			 * max_hole_start and max_hole_size and go back to the
1136 			 * caller.
1137 			 */
1138 			if (hole_size >= num_bytes) {
1139 				ret = 0;
1140 				goto out;
1141 			}
1142 		}
1143 
1144 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1145 		extent_end = key.offset + btrfs_dev_extent_length(l,
1146 								  dev_extent);
1147 		if (extent_end > search_start)
1148 			search_start = extent_end;
1149 next:
1150 		path->slots[0]++;
1151 		cond_resched();
1152 	}
1153 
1154 	/*
1155 	 * At this point, search_start should be the end of
1156 	 * allocated dev extents, and when shrinking the device,
1157 	 * search_end may be smaller than search_start.
1158 	 */
1159 	if (search_end > search_start)
1160 		hole_size = search_end - search_start;
1161 
1162 	if (hole_size > max_hole_size) {
1163 		max_hole_start = search_start;
1164 		max_hole_size = hole_size;
1165 	}
1166 
1167 	if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1168 		btrfs_release_path(path);
1169 		goto again;
1170 	}
1171 
1172 	/* See above. */
1173 	if (hole_size < num_bytes)
1174 		ret = -ENOSPC;
1175 	else
1176 		ret = 0;
1177 
1178 out:
1179 	btrfs_free_path(path);
1180 	*start = max_hole_start;
1181 	if (len)
1182 		*len = max_hole_size;
1183 	return ret;
1184 }
1185 
1186 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1187 			  struct btrfs_device *device,
1188 			  u64 start)
1189 {
1190 	int ret;
1191 	struct btrfs_path *path;
1192 	struct btrfs_root *root = device->dev_root;
1193 	struct btrfs_key key;
1194 	struct btrfs_key found_key;
1195 	struct extent_buffer *leaf = NULL;
1196 	struct btrfs_dev_extent *extent = NULL;
1197 
1198 	path = btrfs_alloc_path();
1199 	if (!path)
1200 		return -ENOMEM;
1201 
1202 	key.objectid = device->devid;
1203 	key.offset = start;
1204 	key.type = BTRFS_DEV_EXTENT_KEY;
1205 again:
1206 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1207 	if (ret > 0) {
1208 		ret = btrfs_previous_item(root, path, key.objectid,
1209 					  BTRFS_DEV_EXTENT_KEY);
1210 		if (ret)
1211 			goto out;
1212 		leaf = path->nodes[0];
1213 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1214 		extent = btrfs_item_ptr(leaf, path->slots[0],
1215 					struct btrfs_dev_extent);
1216 		BUG_ON(found_key.offset > start || found_key.offset +
1217 		       btrfs_dev_extent_length(leaf, extent) < start);
1218 		key = found_key;
1219 		btrfs_release_path(path);
1220 		goto again;
1221 	} else if (ret == 0) {
1222 		leaf = path->nodes[0];
1223 		extent = btrfs_item_ptr(leaf, path->slots[0],
1224 					struct btrfs_dev_extent);
1225 	} else {
1226 		btrfs_error(root->fs_info, ret, "Slot search failed");
1227 		goto out;
1228 	}
1229 
1230 	if (device->bytes_used > 0) {
1231 		u64 len = btrfs_dev_extent_length(leaf, extent);
1232 		device->bytes_used -= len;
1233 		spin_lock(&root->fs_info->free_chunk_lock);
1234 		root->fs_info->free_chunk_space += len;
1235 		spin_unlock(&root->fs_info->free_chunk_lock);
1236 	}
1237 	ret = btrfs_del_item(trans, root, path);
1238 	if (ret) {
1239 		btrfs_error(root->fs_info, ret,
1240 			    "Failed to remove dev extent item");
1241 	}
1242 out:
1243 	btrfs_free_path(path);
1244 	return ret;
1245 }
1246 
1247 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1248 				  struct btrfs_device *device,
1249 				  u64 chunk_tree, u64 chunk_objectid,
1250 				  u64 chunk_offset, u64 start, u64 num_bytes)
1251 {
1252 	int ret;
1253 	struct btrfs_path *path;
1254 	struct btrfs_root *root = device->dev_root;
1255 	struct btrfs_dev_extent *extent;
1256 	struct extent_buffer *leaf;
1257 	struct btrfs_key key;
1258 
1259 	WARN_ON(!device->in_fs_metadata);
1260 	WARN_ON(device->is_tgtdev_for_dev_replace);
1261 	path = btrfs_alloc_path();
1262 	if (!path)
1263 		return -ENOMEM;
1264 
1265 	key.objectid = device->devid;
1266 	key.offset = start;
1267 	key.type = BTRFS_DEV_EXTENT_KEY;
1268 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1269 				      sizeof(*extent));
1270 	if (ret)
1271 		goto out;
1272 
1273 	leaf = path->nodes[0];
1274 	extent = btrfs_item_ptr(leaf, path->slots[0],
1275 				struct btrfs_dev_extent);
1276 	btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1277 	btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1278 	btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1279 
1280 	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1281 		    (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1282 		    BTRFS_UUID_SIZE);
1283 
1284 	btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1285 	btrfs_mark_buffer_dirty(leaf);
1286 out:
1287 	btrfs_free_path(path);
1288 	return ret;
1289 }
1290 
1291 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1292 {
1293 	struct extent_map_tree *em_tree;
1294 	struct extent_map *em;
1295 	struct rb_node *n;
1296 	u64 ret = 0;
1297 
1298 	em_tree = &fs_info->mapping_tree.map_tree;
1299 	read_lock(&em_tree->lock);
1300 	n = rb_last(&em_tree->map);
1301 	if (n) {
1302 		em = rb_entry(n, struct extent_map, rb_node);
1303 		ret = em->start + em->len;
1304 	}
1305 	read_unlock(&em_tree->lock);
1306 
1307 	return ret;
1308 }
1309 
1310 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1311 {
1312 	int ret;
1313 	struct btrfs_key key;
1314 	struct btrfs_key found_key;
1315 	struct btrfs_path *path;
1316 
1317 	root = root->fs_info->chunk_root;
1318 
1319 	path = btrfs_alloc_path();
1320 	if (!path)
1321 		return -ENOMEM;
1322 
1323 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1324 	key.type = BTRFS_DEV_ITEM_KEY;
1325 	key.offset = (u64)-1;
1326 
1327 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1328 	if (ret < 0)
1329 		goto error;
1330 
1331 	BUG_ON(ret == 0); /* Corruption */
1332 
1333 	ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1334 				  BTRFS_DEV_ITEM_KEY);
1335 	if (ret) {
1336 		*objectid = 1;
1337 	} else {
1338 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1339 				      path->slots[0]);
1340 		*objectid = found_key.offset + 1;
1341 	}
1342 	ret = 0;
1343 error:
1344 	btrfs_free_path(path);
1345 	return ret;
1346 }
1347 
1348 /*
1349  * the device information is stored in the chunk root
1350  * the btrfs_device struct should be fully filled in
1351  */
1352 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1353 			    struct btrfs_root *root,
1354 			    struct btrfs_device *device)
1355 {
1356 	int ret;
1357 	struct btrfs_path *path;
1358 	struct btrfs_dev_item *dev_item;
1359 	struct extent_buffer *leaf;
1360 	struct btrfs_key key;
1361 	unsigned long ptr;
1362 
1363 	root = root->fs_info->chunk_root;
1364 
1365 	path = btrfs_alloc_path();
1366 	if (!path)
1367 		return -ENOMEM;
1368 
1369 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1370 	key.type = BTRFS_DEV_ITEM_KEY;
1371 	key.offset = device->devid;
1372 
1373 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1374 				      sizeof(*dev_item));
1375 	if (ret)
1376 		goto out;
1377 
1378 	leaf = path->nodes[0];
1379 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1380 
1381 	btrfs_set_device_id(leaf, dev_item, device->devid);
1382 	btrfs_set_device_generation(leaf, dev_item, 0);
1383 	btrfs_set_device_type(leaf, dev_item, device->type);
1384 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1385 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1386 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1387 	btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1388 	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1389 	btrfs_set_device_group(leaf, dev_item, 0);
1390 	btrfs_set_device_seek_speed(leaf, dev_item, 0);
1391 	btrfs_set_device_bandwidth(leaf, dev_item, 0);
1392 	btrfs_set_device_start_offset(leaf, dev_item, 0);
1393 
1394 	ptr = (unsigned long)btrfs_device_uuid(dev_item);
1395 	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1396 	ptr = (unsigned long)btrfs_device_fsid(dev_item);
1397 	write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1398 	btrfs_mark_buffer_dirty(leaf);
1399 
1400 	ret = 0;
1401 out:
1402 	btrfs_free_path(path);
1403 	return ret;
1404 }
1405 
1406 static int btrfs_rm_dev_item(struct btrfs_root *root,
1407 			     struct btrfs_device *device)
1408 {
1409 	int ret;
1410 	struct btrfs_path *path;
1411 	struct btrfs_key key;
1412 	struct btrfs_trans_handle *trans;
1413 
1414 	root = root->fs_info->chunk_root;
1415 
1416 	path = btrfs_alloc_path();
1417 	if (!path)
1418 		return -ENOMEM;
1419 
1420 	trans = btrfs_start_transaction(root, 0);
1421 	if (IS_ERR(trans)) {
1422 		btrfs_free_path(path);
1423 		return PTR_ERR(trans);
1424 	}
1425 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1426 	key.type = BTRFS_DEV_ITEM_KEY;
1427 	key.offset = device->devid;
1428 	lock_chunks(root);
1429 
1430 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1431 	if (ret < 0)
1432 		goto out;
1433 
1434 	if (ret > 0) {
1435 		ret = -ENOENT;
1436 		goto out;
1437 	}
1438 
1439 	ret = btrfs_del_item(trans, root, path);
1440 	if (ret)
1441 		goto out;
1442 out:
1443 	btrfs_free_path(path);
1444 	unlock_chunks(root);
1445 	btrfs_commit_transaction(trans, root);
1446 	return ret;
1447 }
1448 
1449 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1450 {
1451 	struct btrfs_device *device;
1452 	struct btrfs_device *next_device;
1453 	struct block_device *bdev;
1454 	struct buffer_head *bh = NULL;
1455 	struct btrfs_super_block *disk_super;
1456 	struct btrfs_fs_devices *cur_devices;
1457 	u64 all_avail;
1458 	u64 devid;
1459 	u64 num_devices;
1460 	u8 *dev_uuid;
1461 	unsigned seq;
1462 	int ret = 0;
1463 	bool clear_super = false;
1464 
1465 	mutex_lock(&uuid_mutex);
1466 
1467 	do {
1468 		seq = read_seqbegin(&root->fs_info->profiles_lock);
1469 
1470 		all_avail = root->fs_info->avail_data_alloc_bits |
1471 			    root->fs_info->avail_system_alloc_bits |
1472 			    root->fs_info->avail_metadata_alloc_bits;
1473 	} while (read_seqretry(&root->fs_info->profiles_lock, seq));
1474 
1475 	num_devices = root->fs_info->fs_devices->num_devices;
1476 	btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1477 	if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1478 		WARN_ON(num_devices < 1);
1479 		num_devices--;
1480 	}
1481 	btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1482 
1483 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1484 		ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1485 		goto out;
1486 	}
1487 
1488 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1489 		ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1490 		goto out;
1491 	}
1492 
1493 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1494 	    root->fs_info->fs_devices->rw_devices <= 2) {
1495 		ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1496 		goto out;
1497 	}
1498 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1499 	    root->fs_info->fs_devices->rw_devices <= 3) {
1500 		ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1501 		goto out;
1502 	}
1503 
1504 	if (strcmp(device_path, "missing") == 0) {
1505 		struct list_head *devices;
1506 		struct btrfs_device *tmp;
1507 
1508 		device = NULL;
1509 		devices = &root->fs_info->fs_devices->devices;
1510 		/*
1511 		 * It is safe to read the devices since the volume_mutex
1512 		 * is held.
1513 		 */
1514 		list_for_each_entry(tmp, devices, dev_list) {
1515 			if (tmp->in_fs_metadata &&
1516 			    !tmp->is_tgtdev_for_dev_replace &&
1517 			    !tmp->bdev) {
1518 				device = tmp;
1519 				break;
1520 			}
1521 		}
1522 		bdev = NULL;
1523 		bh = NULL;
1524 		disk_super = NULL;
1525 		if (!device) {
1526 			ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1527 			goto out;
1528 		}
1529 	} else {
1530 		ret = btrfs_get_bdev_and_sb(device_path,
1531 					    FMODE_WRITE | FMODE_EXCL,
1532 					    root->fs_info->bdev_holder, 0,
1533 					    &bdev, &bh);
1534 		if (ret)
1535 			goto out;
1536 		disk_super = (struct btrfs_super_block *)bh->b_data;
1537 		devid = btrfs_stack_device_id(&disk_super->dev_item);
1538 		dev_uuid = disk_super->dev_item.uuid;
1539 		device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1540 					   disk_super->fsid);
1541 		if (!device) {
1542 			ret = -ENOENT;
1543 			goto error_brelse;
1544 		}
1545 	}
1546 
1547 	if (device->is_tgtdev_for_dev_replace) {
1548 		ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1549 		goto error_brelse;
1550 	}
1551 
1552 	if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1553 		ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1554 		goto error_brelse;
1555 	}
1556 
1557 	if (device->writeable) {
1558 		lock_chunks(root);
1559 		list_del_init(&device->dev_alloc_list);
1560 		unlock_chunks(root);
1561 		root->fs_info->fs_devices->rw_devices--;
1562 		clear_super = true;
1563 	}
1564 
1565 	ret = btrfs_shrink_device(device, 0);
1566 	if (ret)
1567 		goto error_undo;
1568 
1569 	/*
1570 	 * TODO: the superblock still includes this device in its num_devices
1571 	 * counter although write_all_supers() is not locked out. This
1572 	 * could give a filesystem state which requires a degraded mount.
1573 	 */
1574 	ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1575 	if (ret)
1576 		goto error_undo;
1577 
1578 	spin_lock(&root->fs_info->free_chunk_lock);
1579 	root->fs_info->free_chunk_space = device->total_bytes -
1580 		device->bytes_used;
1581 	spin_unlock(&root->fs_info->free_chunk_lock);
1582 
1583 	device->in_fs_metadata = 0;
1584 	btrfs_scrub_cancel_dev(root->fs_info, device);
1585 
1586 	/*
1587 	 * the device list mutex makes sure that we don't change
1588 	 * the device list while someone else is writing out all
1589 	 * the device supers.
1590 	 */
1591 
1592 	cur_devices = device->fs_devices;
1593 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1594 	list_del_rcu(&device->dev_list);
1595 
1596 	device->fs_devices->num_devices--;
1597 	device->fs_devices->total_devices--;
1598 
1599 	if (device->missing)
1600 		root->fs_info->fs_devices->missing_devices--;
1601 
1602 	next_device = list_entry(root->fs_info->fs_devices->devices.next,
1603 				 struct btrfs_device, dev_list);
1604 	if (device->bdev == root->fs_info->sb->s_bdev)
1605 		root->fs_info->sb->s_bdev = next_device->bdev;
1606 	if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1607 		root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1608 
1609 	if (device->bdev)
1610 		device->fs_devices->open_devices--;
1611 
1612 	call_rcu(&device->rcu, free_device);
1613 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1614 
1615 	num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1616 	btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1617 
1618 	if (cur_devices->open_devices == 0) {
1619 		struct btrfs_fs_devices *fs_devices;
1620 		fs_devices = root->fs_info->fs_devices;
1621 		while (fs_devices) {
1622 			if (fs_devices->seed == cur_devices)
1623 				break;
1624 			fs_devices = fs_devices->seed;
1625 		}
1626 		fs_devices->seed = cur_devices->seed;
1627 		cur_devices->seed = NULL;
1628 		lock_chunks(root);
1629 		__btrfs_close_devices(cur_devices);
1630 		unlock_chunks(root);
1631 		free_fs_devices(cur_devices);
1632 	}
1633 
1634 	root->fs_info->num_tolerated_disk_barrier_failures =
1635 		btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1636 
1637 	/*
1638 	 * at this point, the device is zero sized.  We want to
1639 	 * remove it from the devices list and zero out the old super
1640 	 */
1641 	if (clear_super && disk_super) {
1642 		/* make sure this device isn't detected as part of
1643 		 * the FS anymore
1644 		 */
1645 		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1646 		set_buffer_dirty(bh);
1647 		sync_dirty_buffer(bh);
1648 	}
1649 
1650 	ret = 0;
1651 
1652 	/* Notify udev that device has changed */
1653 	if (bdev)
1654 		btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1655 
1656 error_brelse:
1657 	brelse(bh);
1658 	if (bdev)
1659 		blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1660 out:
1661 	mutex_unlock(&uuid_mutex);
1662 	return ret;
1663 error_undo:
1664 	if (device->writeable) {
1665 		lock_chunks(root);
1666 		list_add(&device->dev_alloc_list,
1667 			 &root->fs_info->fs_devices->alloc_list);
1668 		unlock_chunks(root);
1669 		root->fs_info->fs_devices->rw_devices++;
1670 	}
1671 	goto error_brelse;
1672 }
1673 
1674 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1675 				 struct btrfs_device *srcdev)
1676 {
1677 	WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1678 	list_del_rcu(&srcdev->dev_list);
1679 	list_del_rcu(&srcdev->dev_alloc_list);
1680 	fs_info->fs_devices->num_devices--;
1681 	if (srcdev->missing) {
1682 		fs_info->fs_devices->missing_devices--;
1683 		fs_info->fs_devices->rw_devices++;
1684 	}
1685 	if (srcdev->can_discard)
1686 		fs_info->fs_devices->num_can_discard--;
1687 	if (srcdev->bdev)
1688 		fs_info->fs_devices->open_devices--;
1689 
1690 	call_rcu(&srcdev->rcu, free_device);
1691 }
1692 
1693 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1694 				      struct btrfs_device *tgtdev)
1695 {
1696 	struct btrfs_device *next_device;
1697 
1698 	WARN_ON(!tgtdev);
1699 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
1700 	if (tgtdev->bdev) {
1701 		btrfs_scratch_superblock(tgtdev);
1702 		fs_info->fs_devices->open_devices--;
1703 	}
1704 	fs_info->fs_devices->num_devices--;
1705 	if (tgtdev->can_discard)
1706 		fs_info->fs_devices->num_can_discard++;
1707 
1708 	next_device = list_entry(fs_info->fs_devices->devices.next,
1709 				 struct btrfs_device, dev_list);
1710 	if (tgtdev->bdev == fs_info->sb->s_bdev)
1711 		fs_info->sb->s_bdev = next_device->bdev;
1712 	if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1713 		fs_info->fs_devices->latest_bdev = next_device->bdev;
1714 	list_del_rcu(&tgtdev->dev_list);
1715 
1716 	call_rcu(&tgtdev->rcu, free_device);
1717 
1718 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1719 }
1720 
1721 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1722 				     struct btrfs_device **device)
1723 {
1724 	int ret = 0;
1725 	struct btrfs_super_block *disk_super;
1726 	u64 devid;
1727 	u8 *dev_uuid;
1728 	struct block_device *bdev;
1729 	struct buffer_head *bh;
1730 
1731 	*device = NULL;
1732 	ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1733 				    root->fs_info->bdev_holder, 0, &bdev, &bh);
1734 	if (ret)
1735 		return ret;
1736 	disk_super = (struct btrfs_super_block *)bh->b_data;
1737 	devid = btrfs_stack_device_id(&disk_super->dev_item);
1738 	dev_uuid = disk_super->dev_item.uuid;
1739 	*device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1740 				    disk_super->fsid);
1741 	brelse(bh);
1742 	if (!*device)
1743 		ret = -ENOENT;
1744 	blkdev_put(bdev, FMODE_READ);
1745 	return ret;
1746 }
1747 
1748 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1749 					 char *device_path,
1750 					 struct btrfs_device **device)
1751 {
1752 	*device = NULL;
1753 	if (strcmp(device_path, "missing") == 0) {
1754 		struct list_head *devices;
1755 		struct btrfs_device *tmp;
1756 
1757 		devices = &root->fs_info->fs_devices->devices;
1758 		/*
1759 		 * It is safe to read the devices since the volume_mutex
1760 		 * is held by the caller.
1761 		 */
1762 		list_for_each_entry(tmp, devices, dev_list) {
1763 			if (tmp->in_fs_metadata && !tmp->bdev) {
1764 				*device = tmp;
1765 				break;
1766 			}
1767 		}
1768 
1769 		if (!*device) {
1770 			pr_err("btrfs: no missing device found\n");
1771 			return -ENOENT;
1772 		}
1773 
1774 		return 0;
1775 	} else {
1776 		return btrfs_find_device_by_path(root, device_path, device);
1777 	}
1778 }
1779 
1780 /*
1781  * does all the dirty work required for changing file system's UUID.
1782  */
1783 static int btrfs_prepare_sprout(struct btrfs_root *root)
1784 {
1785 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1786 	struct btrfs_fs_devices *old_devices;
1787 	struct btrfs_fs_devices *seed_devices;
1788 	struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1789 	struct btrfs_device *device;
1790 	u64 super_flags;
1791 
1792 	BUG_ON(!mutex_is_locked(&uuid_mutex));
1793 	if (!fs_devices->seeding)
1794 		return -EINVAL;
1795 
1796 	seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1797 	if (!seed_devices)
1798 		return -ENOMEM;
1799 
1800 	old_devices = clone_fs_devices(fs_devices);
1801 	if (IS_ERR(old_devices)) {
1802 		kfree(seed_devices);
1803 		return PTR_ERR(old_devices);
1804 	}
1805 
1806 	list_add(&old_devices->list, &fs_uuids);
1807 
1808 	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1809 	seed_devices->opened = 1;
1810 	INIT_LIST_HEAD(&seed_devices->devices);
1811 	INIT_LIST_HEAD(&seed_devices->alloc_list);
1812 	mutex_init(&seed_devices->device_list_mutex);
1813 
1814 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1815 	list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1816 			      synchronize_rcu);
1817 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1818 
1819 	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1820 	list_for_each_entry(device, &seed_devices->devices, dev_list) {
1821 		device->fs_devices = seed_devices;
1822 	}
1823 
1824 	fs_devices->seeding = 0;
1825 	fs_devices->num_devices = 0;
1826 	fs_devices->open_devices = 0;
1827 	fs_devices->total_devices = 0;
1828 	fs_devices->seed = seed_devices;
1829 
1830 	generate_random_uuid(fs_devices->fsid);
1831 	memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1832 	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1833 	super_flags = btrfs_super_flags(disk_super) &
1834 		      ~BTRFS_SUPER_FLAG_SEEDING;
1835 	btrfs_set_super_flags(disk_super, super_flags);
1836 
1837 	return 0;
1838 }
1839 
1840 /*
1841  * strore the expected generation for seed devices in device items.
1842  */
1843 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1844 			       struct btrfs_root *root)
1845 {
1846 	struct btrfs_path *path;
1847 	struct extent_buffer *leaf;
1848 	struct btrfs_dev_item *dev_item;
1849 	struct btrfs_device *device;
1850 	struct btrfs_key key;
1851 	u8 fs_uuid[BTRFS_UUID_SIZE];
1852 	u8 dev_uuid[BTRFS_UUID_SIZE];
1853 	u64 devid;
1854 	int ret;
1855 
1856 	path = btrfs_alloc_path();
1857 	if (!path)
1858 		return -ENOMEM;
1859 
1860 	root = root->fs_info->chunk_root;
1861 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1862 	key.offset = 0;
1863 	key.type = BTRFS_DEV_ITEM_KEY;
1864 
1865 	while (1) {
1866 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1867 		if (ret < 0)
1868 			goto error;
1869 
1870 		leaf = path->nodes[0];
1871 next_slot:
1872 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1873 			ret = btrfs_next_leaf(root, path);
1874 			if (ret > 0)
1875 				break;
1876 			if (ret < 0)
1877 				goto error;
1878 			leaf = path->nodes[0];
1879 			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1880 			btrfs_release_path(path);
1881 			continue;
1882 		}
1883 
1884 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1885 		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1886 		    key.type != BTRFS_DEV_ITEM_KEY)
1887 			break;
1888 
1889 		dev_item = btrfs_item_ptr(leaf, path->slots[0],
1890 					  struct btrfs_dev_item);
1891 		devid = btrfs_device_id(leaf, dev_item);
1892 		read_extent_buffer(leaf, dev_uuid,
1893 				   (unsigned long)btrfs_device_uuid(dev_item),
1894 				   BTRFS_UUID_SIZE);
1895 		read_extent_buffer(leaf, fs_uuid,
1896 				   (unsigned long)btrfs_device_fsid(dev_item),
1897 				   BTRFS_UUID_SIZE);
1898 		device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1899 					   fs_uuid);
1900 		BUG_ON(!device); /* Logic error */
1901 
1902 		if (device->fs_devices->seeding) {
1903 			btrfs_set_device_generation(leaf, dev_item,
1904 						    device->generation);
1905 			btrfs_mark_buffer_dirty(leaf);
1906 		}
1907 
1908 		path->slots[0]++;
1909 		goto next_slot;
1910 	}
1911 	ret = 0;
1912 error:
1913 	btrfs_free_path(path);
1914 	return ret;
1915 }
1916 
1917 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1918 {
1919 	struct request_queue *q;
1920 	struct btrfs_trans_handle *trans;
1921 	struct btrfs_device *device;
1922 	struct block_device *bdev;
1923 	struct list_head *devices;
1924 	struct super_block *sb = root->fs_info->sb;
1925 	struct rcu_string *name;
1926 	u64 total_bytes;
1927 	int seeding_dev = 0;
1928 	int ret = 0;
1929 
1930 	if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1931 		return -EROFS;
1932 
1933 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1934 				  root->fs_info->bdev_holder);
1935 	if (IS_ERR(bdev))
1936 		return PTR_ERR(bdev);
1937 
1938 	if (root->fs_info->fs_devices->seeding) {
1939 		seeding_dev = 1;
1940 		down_write(&sb->s_umount);
1941 		mutex_lock(&uuid_mutex);
1942 	}
1943 
1944 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
1945 
1946 	devices = &root->fs_info->fs_devices->devices;
1947 
1948 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1949 	list_for_each_entry(device, devices, dev_list) {
1950 		if (device->bdev == bdev) {
1951 			ret = -EEXIST;
1952 			mutex_unlock(
1953 				&root->fs_info->fs_devices->device_list_mutex);
1954 			goto error;
1955 		}
1956 	}
1957 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1958 
1959 	device = kzalloc(sizeof(*device), GFP_NOFS);
1960 	if (!device) {
1961 		/* we can safely leave the fs_devices entry around */
1962 		ret = -ENOMEM;
1963 		goto error;
1964 	}
1965 
1966 	name = rcu_string_strdup(device_path, GFP_NOFS);
1967 	if (!name) {
1968 		kfree(device);
1969 		ret = -ENOMEM;
1970 		goto error;
1971 	}
1972 	rcu_assign_pointer(device->name, name);
1973 
1974 	ret = find_next_devid(root, &device->devid);
1975 	if (ret) {
1976 		rcu_string_free(device->name);
1977 		kfree(device);
1978 		goto error;
1979 	}
1980 
1981 	trans = btrfs_start_transaction(root, 0);
1982 	if (IS_ERR(trans)) {
1983 		rcu_string_free(device->name);
1984 		kfree(device);
1985 		ret = PTR_ERR(trans);
1986 		goto error;
1987 	}
1988 
1989 	lock_chunks(root);
1990 
1991 	q = bdev_get_queue(bdev);
1992 	if (blk_queue_discard(q))
1993 		device->can_discard = 1;
1994 	device->writeable = 1;
1995 	device->work.func = pending_bios_fn;
1996 	generate_random_uuid(device->uuid);
1997 	spin_lock_init(&device->io_lock);
1998 	device->generation = trans->transid;
1999 	device->io_width = root->sectorsize;
2000 	device->io_align = root->sectorsize;
2001 	device->sector_size = root->sectorsize;
2002 	device->total_bytes = i_size_read(bdev->bd_inode);
2003 	device->disk_total_bytes = device->total_bytes;
2004 	device->dev_root = root->fs_info->dev_root;
2005 	device->bdev = bdev;
2006 	device->in_fs_metadata = 1;
2007 	device->is_tgtdev_for_dev_replace = 0;
2008 	device->mode = FMODE_EXCL;
2009 	set_blocksize(device->bdev, 4096);
2010 
2011 	if (seeding_dev) {
2012 		sb->s_flags &= ~MS_RDONLY;
2013 		ret = btrfs_prepare_sprout(root);
2014 		BUG_ON(ret); /* -ENOMEM */
2015 	}
2016 
2017 	device->fs_devices = root->fs_info->fs_devices;
2018 
2019 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2020 	list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2021 	list_add(&device->dev_alloc_list,
2022 		 &root->fs_info->fs_devices->alloc_list);
2023 	root->fs_info->fs_devices->num_devices++;
2024 	root->fs_info->fs_devices->open_devices++;
2025 	root->fs_info->fs_devices->rw_devices++;
2026 	root->fs_info->fs_devices->total_devices++;
2027 	if (device->can_discard)
2028 		root->fs_info->fs_devices->num_can_discard++;
2029 	root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2030 
2031 	spin_lock(&root->fs_info->free_chunk_lock);
2032 	root->fs_info->free_chunk_space += device->total_bytes;
2033 	spin_unlock(&root->fs_info->free_chunk_lock);
2034 
2035 	if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2036 		root->fs_info->fs_devices->rotating = 1;
2037 
2038 	total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2039 	btrfs_set_super_total_bytes(root->fs_info->super_copy,
2040 				    total_bytes + device->total_bytes);
2041 
2042 	total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2043 	btrfs_set_super_num_devices(root->fs_info->super_copy,
2044 				    total_bytes + 1);
2045 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2046 
2047 	if (seeding_dev) {
2048 		ret = init_first_rw_device(trans, root, device);
2049 		if (ret) {
2050 			btrfs_abort_transaction(trans, root, ret);
2051 			goto error_trans;
2052 		}
2053 		ret = btrfs_finish_sprout(trans, root);
2054 		if (ret) {
2055 			btrfs_abort_transaction(trans, root, ret);
2056 			goto error_trans;
2057 		}
2058 	} else {
2059 		ret = btrfs_add_device(trans, root, device);
2060 		if (ret) {
2061 			btrfs_abort_transaction(trans, root, ret);
2062 			goto error_trans;
2063 		}
2064 	}
2065 
2066 	/*
2067 	 * we've got more storage, clear any full flags on the space
2068 	 * infos
2069 	 */
2070 	btrfs_clear_space_info_full(root->fs_info);
2071 
2072 	unlock_chunks(root);
2073 	root->fs_info->num_tolerated_disk_barrier_failures =
2074 		btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2075 	ret = btrfs_commit_transaction(trans, root);
2076 
2077 	if (seeding_dev) {
2078 		mutex_unlock(&uuid_mutex);
2079 		up_write(&sb->s_umount);
2080 
2081 		if (ret) /* transaction commit */
2082 			return ret;
2083 
2084 		ret = btrfs_relocate_sys_chunks(root);
2085 		if (ret < 0)
2086 			btrfs_error(root->fs_info, ret,
2087 				    "Failed to relocate sys chunks after "
2088 				    "device initialization. This can be fixed "
2089 				    "using the \"btrfs balance\" command.");
2090 		trans = btrfs_attach_transaction(root);
2091 		if (IS_ERR(trans)) {
2092 			if (PTR_ERR(trans) == -ENOENT)
2093 				return 0;
2094 			return PTR_ERR(trans);
2095 		}
2096 		ret = btrfs_commit_transaction(trans, root);
2097 	}
2098 
2099 	return ret;
2100 
2101 error_trans:
2102 	unlock_chunks(root);
2103 	btrfs_end_transaction(trans, root);
2104 	rcu_string_free(device->name);
2105 	kfree(device);
2106 error:
2107 	blkdev_put(bdev, FMODE_EXCL);
2108 	if (seeding_dev) {
2109 		mutex_unlock(&uuid_mutex);
2110 		up_write(&sb->s_umount);
2111 	}
2112 	return ret;
2113 }
2114 
2115 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2116 				  struct btrfs_device **device_out)
2117 {
2118 	struct request_queue *q;
2119 	struct btrfs_device *device;
2120 	struct block_device *bdev;
2121 	struct btrfs_fs_info *fs_info = root->fs_info;
2122 	struct list_head *devices;
2123 	struct rcu_string *name;
2124 	int ret = 0;
2125 
2126 	*device_out = NULL;
2127 	if (fs_info->fs_devices->seeding)
2128 		return -EINVAL;
2129 
2130 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2131 				  fs_info->bdev_holder);
2132 	if (IS_ERR(bdev))
2133 		return PTR_ERR(bdev);
2134 
2135 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
2136 
2137 	devices = &fs_info->fs_devices->devices;
2138 	list_for_each_entry(device, devices, dev_list) {
2139 		if (device->bdev == bdev) {
2140 			ret = -EEXIST;
2141 			goto error;
2142 		}
2143 	}
2144 
2145 	device = kzalloc(sizeof(*device), GFP_NOFS);
2146 	if (!device) {
2147 		ret = -ENOMEM;
2148 		goto error;
2149 	}
2150 
2151 	name = rcu_string_strdup(device_path, GFP_NOFS);
2152 	if (!name) {
2153 		kfree(device);
2154 		ret = -ENOMEM;
2155 		goto error;
2156 	}
2157 	rcu_assign_pointer(device->name, name);
2158 
2159 	q = bdev_get_queue(bdev);
2160 	if (blk_queue_discard(q))
2161 		device->can_discard = 1;
2162 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2163 	device->writeable = 1;
2164 	device->work.func = pending_bios_fn;
2165 	generate_random_uuid(device->uuid);
2166 	device->devid = BTRFS_DEV_REPLACE_DEVID;
2167 	spin_lock_init(&device->io_lock);
2168 	device->generation = 0;
2169 	device->io_width = root->sectorsize;
2170 	device->io_align = root->sectorsize;
2171 	device->sector_size = root->sectorsize;
2172 	device->total_bytes = i_size_read(bdev->bd_inode);
2173 	device->disk_total_bytes = device->total_bytes;
2174 	device->dev_root = fs_info->dev_root;
2175 	device->bdev = bdev;
2176 	device->in_fs_metadata = 1;
2177 	device->is_tgtdev_for_dev_replace = 1;
2178 	device->mode = FMODE_EXCL;
2179 	set_blocksize(device->bdev, 4096);
2180 	device->fs_devices = fs_info->fs_devices;
2181 	list_add(&device->dev_list, &fs_info->fs_devices->devices);
2182 	fs_info->fs_devices->num_devices++;
2183 	fs_info->fs_devices->open_devices++;
2184 	if (device->can_discard)
2185 		fs_info->fs_devices->num_can_discard++;
2186 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2187 
2188 	*device_out = device;
2189 	return ret;
2190 
2191 error:
2192 	blkdev_put(bdev, FMODE_EXCL);
2193 	return ret;
2194 }
2195 
2196 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2197 					      struct btrfs_device *tgtdev)
2198 {
2199 	WARN_ON(fs_info->fs_devices->rw_devices == 0);
2200 	tgtdev->io_width = fs_info->dev_root->sectorsize;
2201 	tgtdev->io_align = fs_info->dev_root->sectorsize;
2202 	tgtdev->sector_size = fs_info->dev_root->sectorsize;
2203 	tgtdev->dev_root = fs_info->dev_root;
2204 	tgtdev->in_fs_metadata = 1;
2205 }
2206 
2207 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2208 					struct btrfs_device *device)
2209 {
2210 	int ret;
2211 	struct btrfs_path *path;
2212 	struct btrfs_root *root;
2213 	struct btrfs_dev_item *dev_item;
2214 	struct extent_buffer *leaf;
2215 	struct btrfs_key key;
2216 
2217 	root = device->dev_root->fs_info->chunk_root;
2218 
2219 	path = btrfs_alloc_path();
2220 	if (!path)
2221 		return -ENOMEM;
2222 
2223 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2224 	key.type = BTRFS_DEV_ITEM_KEY;
2225 	key.offset = device->devid;
2226 
2227 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2228 	if (ret < 0)
2229 		goto out;
2230 
2231 	if (ret > 0) {
2232 		ret = -ENOENT;
2233 		goto out;
2234 	}
2235 
2236 	leaf = path->nodes[0];
2237 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2238 
2239 	btrfs_set_device_id(leaf, dev_item, device->devid);
2240 	btrfs_set_device_type(leaf, dev_item, device->type);
2241 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2242 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2243 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2244 	btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2245 	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2246 	btrfs_mark_buffer_dirty(leaf);
2247 
2248 out:
2249 	btrfs_free_path(path);
2250 	return ret;
2251 }
2252 
2253 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2254 		      struct btrfs_device *device, u64 new_size)
2255 {
2256 	struct btrfs_super_block *super_copy =
2257 		device->dev_root->fs_info->super_copy;
2258 	u64 old_total = btrfs_super_total_bytes(super_copy);
2259 	u64 diff = new_size - device->total_bytes;
2260 
2261 	if (!device->writeable)
2262 		return -EACCES;
2263 	if (new_size <= device->total_bytes ||
2264 	    device->is_tgtdev_for_dev_replace)
2265 		return -EINVAL;
2266 
2267 	btrfs_set_super_total_bytes(super_copy, old_total + diff);
2268 	device->fs_devices->total_rw_bytes += diff;
2269 
2270 	device->total_bytes = new_size;
2271 	device->disk_total_bytes = new_size;
2272 	btrfs_clear_space_info_full(device->dev_root->fs_info);
2273 
2274 	return btrfs_update_device(trans, device);
2275 }
2276 
2277 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2278 		      struct btrfs_device *device, u64 new_size)
2279 {
2280 	int ret;
2281 	lock_chunks(device->dev_root);
2282 	ret = __btrfs_grow_device(trans, device, new_size);
2283 	unlock_chunks(device->dev_root);
2284 	return ret;
2285 }
2286 
2287 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2288 			    struct btrfs_root *root,
2289 			    u64 chunk_tree, u64 chunk_objectid,
2290 			    u64 chunk_offset)
2291 {
2292 	int ret;
2293 	struct btrfs_path *path;
2294 	struct btrfs_key key;
2295 
2296 	root = root->fs_info->chunk_root;
2297 	path = btrfs_alloc_path();
2298 	if (!path)
2299 		return -ENOMEM;
2300 
2301 	key.objectid = chunk_objectid;
2302 	key.offset = chunk_offset;
2303 	key.type = BTRFS_CHUNK_ITEM_KEY;
2304 
2305 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2306 	if (ret < 0)
2307 		goto out;
2308 	else if (ret > 0) { /* Logic error or corruption */
2309 		btrfs_error(root->fs_info, -ENOENT,
2310 			    "Failed lookup while freeing chunk.");
2311 		ret = -ENOENT;
2312 		goto out;
2313 	}
2314 
2315 	ret = btrfs_del_item(trans, root, path);
2316 	if (ret < 0)
2317 		btrfs_error(root->fs_info, ret,
2318 			    "Failed to delete chunk item.");
2319 out:
2320 	btrfs_free_path(path);
2321 	return ret;
2322 }
2323 
2324 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2325 			chunk_offset)
2326 {
2327 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2328 	struct btrfs_disk_key *disk_key;
2329 	struct btrfs_chunk *chunk;
2330 	u8 *ptr;
2331 	int ret = 0;
2332 	u32 num_stripes;
2333 	u32 array_size;
2334 	u32 len = 0;
2335 	u32 cur;
2336 	struct btrfs_key key;
2337 
2338 	array_size = btrfs_super_sys_array_size(super_copy);
2339 
2340 	ptr = super_copy->sys_chunk_array;
2341 	cur = 0;
2342 
2343 	while (cur < array_size) {
2344 		disk_key = (struct btrfs_disk_key *)ptr;
2345 		btrfs_disk_key_to_cpu(&key, disk_key);
2346 
2347 		len = sizeof(*disk_key);
2348 
2349 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2350 			chunk = (struct btrfs_chunk *)(ptr + len);
2351 			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2352 			len += btrfs_chunk_item_size(num_stripes);
2353 		} else {
2354 			ret = -EIO;
2355 			break;
2356 		}
2357 		if (key.objectid == chunk_objectid &&
2358 		    key.offset == chunk_offset) {
2359 			memmove(ptr, ptr + len, array_size - (cur + len));
2360 			array_size -= len;
2361 			btrfs_set_super_sys_array_size(super_copy, array_size);
2362 		} else {
2363 			ptr += len;
2364 			cur += len;
2365 		}
2366 	}
2367 	return ret;
2368 }
2369 
2370 static int btrfs_relocate_chunk(struct btrfs_root *root,
2371 			 u64 chunk_tree, u64 chunk_objectid,
2372 			 u64 chunk_offset)
2373 {
2374 	struct extent_map_tree *em_tree;
2375 	struct btrfs_root *extent_root;
2376 	struct btrfs_trans_handle *trans;
2377 	struct extent_map *em;
2378 	struct map_lookup *map;
2379 	int ret;
2380 	int i;
2381 
2382 	root = root->fs_info->chunk_root;
2383 	extent_root = root->fs_info->extent_root;
2384 	em_tree = &root->fs_info->mapping_tree.map_tree;
2385 
2386 	ret = btrfs_can_relocate(extent_root, chunk_offset);
2387 	if (ret)
2388 		return -ENOSPC;
2389 
2390 	/* step one, relocate all the extents inside this chunk */
2391 	ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2392 	if (ret)
2393 		return ret;
2394 
2395 	trans = btrfs_start_transaction(root, 0);
2396 	if (IS_ERR(trans)) {
2397 		ret = PTR_ERR(trans);
2398 		btrfs_std_error(root->fs_info, ret);
2399 		return ret;
2400 	}
2401 
2402 	lock_chunks(root);
2403 
2404 	/*
2405 	 * step two, delete the device extents and the
2406 	 * chunk tree entries
2407 	 */
2408 	read_lock(&em_tree->lock);
2409 	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2410 	read_unlock(&em_tree->lock);
2411 
2412 	BUG_ON(!em || em->start > chunk_offset ||
2413 	       em->start + em->len < chunk_offset);
2414 	map = (struct map_lookup *)em->bdev;
2415 
2416 	for (i = 0; i < map->num_stripes; i++) {
2417 		ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2418 					    map->stripes[i].physical);
2419 		BUG_ON(ret);
2420 
2421 		if (map->stripes[i].dev) {
2422 			ret = btrfs_update_device(trans, map->stripes[i].dev);
2423 			BUG_ON(ret);
2424 		}
2425 	}
2426 	ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2427 			       chunk_offset);
2428 
2429 	BUG_ON(ret);
2430 
2431 	trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2432 
2433 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2434 		ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2435 		BUG_ON(ret);
2436 	}
2437 
2438 	ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2439 	BUG_ON(ret);
2440 
2441 	write_lock(&em_tree->lock);
2442 	remove_extent_mapping(em_tree, em);
2443 	write_unlock(&em_tree->lock);
2444 
2445 	kfree(map);
2446 	em->bdev = NULL;
2447 
2448 	/* once for the tree */
2449 	free_extent_map(em);
2450 	/* once for us */
2451 	free_extent_map(em);
2452 
2453 	unlock_chunks(root);
2454 	btrfs_end_transaction(trans, root);
2455 	return 0;
2456 }
2457 
2458 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2459 {
2460 	struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2461 	struct btrfs_path *path;
2462 	struct extent_buffer *leaf;
2463 	struct btrfs_chunk *chunk;
2464 	struct btrfs_key key;
2465 	struct btrfs_key found_key;
2466 	u64 chunk_tree = chunk_root->root_key.objectid;
2467 	u64 chunk_type;
2468 	bool retried = false;
2469 	int failed = 0;
2470 	int ret;
2471 
2472 	path = btrfs_alloc_path();
2473 	if (!path)
2474 		return -ENOMEM;
2475 
2476 again:
2477 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2478 	key.offset = (u64)-1;
2479 	key.type = BTRFS_CHUNK_ITEM_KEY;
2480 
2481 	while (1) {
2482 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2483 		if (ret < 0)
2484 			goto error;
2485 		BUG_ON(ret == 0); /* Corruption */
2486 
2487 		ret = btrfs_previous_item(chunk_root, path, key.objectid,
2488 					  key.type);
2489 		if (ret < 0)
2490 			goto error;
2491 		if (ret > 0)
2492 			break;
2493 
2494 		leaf = path->nodes[0];
2495 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2496 
2497 		chunk = btrfs_item_ptr(leaf, path->slots[0],
2498 				       struct btrfs_chunk);
2499 		chunk_type = btrfs_chunk_type(leaf, chunk);
2500 		btrfs_release_path(path);
2501 
2502 		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2503 			ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2504 						   found_key.objectid,
2505 						   found_key.offset);
2506 			if (ret == -ENOSPC)
2507 				failed++;
2508 			else if (ret)
2509 				BUG();
2510 		}
2511 
2512 		if (found_key.offset == 0)
2513 			break;
2514 		key.offset = found_key.offset - 1;
2515 	}
2516 	ret = 0;
2517 	if (failed && !retried) {
2518 		failed = 0;
2519 		retried = true;
2520 		goto again;
2521 	} else if (failed && retried) {
2522 		WARN_ON(1);
2523 		ret = -ENOSPC;
2524 	}
2525 error:
2526 	btrfs_free_path(path);
2527 	return ret;
2528 }
2529 
2530 static int insert_balance_item(struct btrfs_root *root,
2531 			       struct btrfs_balance_control *bctl)
2532 {
2533 	struct btrfs_trans_handle *trans;
2534 	struct btrfs_balance_item *item;
2535 	struct btrfs_disk_balance_args disk_bargs;
2536 	struct btrfs_path *path;
2537 	struct extent_buffer *leaf;
2538 	struct btrfs_key key;
2539 	int ret, err;
2540 
2541 	path = btrfs_alloc_path();
2542 	if (!path)
2543 		return -ENOMEM;
2544 
2545 	trans = btrfs_start_transaction(root, 0);
2546 	if (IS_ERR(trans)) {
2547 		btrfs_free_path(path);
2548 		return PTR_ERR(trans);
2549 	}
2550 
2551 	key.objectid = BTRFS_BALANCE_OBJECTID;
2552 	key.type = BTRFS_BALANCE_ITEM_KEY;
2553 	key.offset = 0;
2554 
2555 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2556 				      sizeof(*item));
2557 	if (ret)
2558 		goto out;
2559 
2560 	leaf = path->nodes[0];
2561 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2562 
2563 	memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2564 
2565 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2566 	btrfs_set_balance_data(leaf, item, &disk_bargs);
2567 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2568 	btrfs_set_balance_meta(leaf, item, &disk_bargs);
2569 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2570 	btrfs_set_balance_sys(leaf, item, &disk_bargs);
2571 
2572 	btrfs_set_balance_flags(leaf, item, bctl->flags);
2573 
2574 	btrfs_mark_buffer_dirty(leaf);
2575 out:
2576 	btrfs_free_path(path);
2577 	err = btrfs_commit_transaction(trans, root);
2578 	if (err && !ret)
2579 		ret = err;
2580 	return ret;
2581 }
2582 
2583 static int del_balance_item(struct btrfs_root *root)
2584 {
2585 	struct btrfs_trans_handle *trans;
2586 	struct btrfs_path *path;
2587 	struct btrfs_key key;
2588 	int ret, err;
2589 
2590 	path = btrfs_alloc_path();
2591 	if (!path)
2592 		return -ENOMEM;
2593 
2594 	trans = btrfs_start_transaction(root, 0);
2595 	if (IS_ERR(trans)) {
2596 		btrfs_free_path(path);
2597 		return PTR_ERR(trans);
2598 	}
2599 
2600 	key.objectid = BTRFS_BALANCE_OBJECTID;
2601 	key.type = BTRFS_BALANCE_ITEM_KEY;
2602 	key.offset = 0;
2603 
2604 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2605 	if (ret < 0)
2606 		goto out;
2607 	if (ret > 0) {
2608 		ret = -ENOENT;
2609 		goto out;
2610 	}
2611 
2612 	ret = btrfs_del_item(trans, root, path);
2613 out:
2614 	btrfs_free_path(path);
2615 	err = btrfs_commit_transaction(trans, root);
2616 	if (err && !ret)
2617 		ret = err;
2618 	return ret;
2619 }
2620 
2621 /*
2622  * This is a heuristic used to reduce the number of chunks balanced on
2623  * resume after balance was interrupted.
2624  */
2625 static void update_balance_args(struct btrfs_balance_control *bctl)
2626 {
2627 	/*
2628 	 * Turn on soft mode for chunk types that were being converted.
2629 	 */
2630 	if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2631 		bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2632 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2633 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2634 	if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2635 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2636 
2637 	/*
2638 	 * Turn on usage filter if is not already used.  The idea is
2639 	 * that chunks that we have already balanced should be
2640 	 * reasonably full.  Don't do it for chunks that are being
2641 	 * converted - that will keep us from relocating unconverted
2642 	 * (albeit full) chunks.
2643 	 */
2644 	if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2645 	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2646 		bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2647 		bctl->data.usage = 90;
2648 	}
2649 	if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2650 	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2651 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2652 		bctl->sys.usage = 90;
2653 	}
2654 	if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2655 	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2656 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2657 		bctl->meta.usage = 90;
2658 	}
2659 }
2660 
2661 /*
2662  * Should be called with both balance and volume mutexes held to
2663  * serialize other volume operations (add_dev/rm_dev/resize) with
2664  * restriper.  Same goes for unset_balance_control.
2665  */
2666 static void set_balance_control(struct btrfs_balance_control *bctl)
2667 {
2668 	struct btrfs_fs_info *fs_info = bctl->fs_info;
2669 
2670 	BUG_ON(fs_info->balance_ctl);
2671 
2672 	spin_lock(&fs_info->balance_lock);
2673 	fs_info->balance_ctl = bctl;
2674 	spin_unlock(&fs_info->balance_lock);
2675 }
2676 
2677 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2678 {
2679 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2680 
2681 	BUG_ON(!fs_info->balance_ctl);
2682 
2683 	spin_lock(&fs_info->balance_lock);
2684 	fs_info->balance_ctl = NULL;
2685 	spin_unlock(&fs_info->balance_lock);
2686 
2687 	kfree(bctl);
2688 }
2689 
2690 /*
2691  * Balance filters.  Return 1 if chunk should be filtered out
2692  * (should not be balanced).
2693  */
2694 static int chunk_profiles_filter(u64 chunk_type,
2695 				 struct btrfs_balance_args *bargs)
2696 {
2697 	chunk_type = chunk_to_extended(chunk_type) &
2698 				BTRFS_EXTENDED_PROFILE_MASK;
2699 
2700 	if (bargs->profiles & chunk_type)
2701 		return 0;
2702 
2703 	return 1;
2704 }
2705 
2706 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2707 			      struct btrfs_balance_args *bargs)
2708 {
2709 	struct btrfs_block_group_cache *cache;
2710 	u64 chunk_used, user_thresh;
2711 	int ret = 1;
2712 
2713 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2714 	chunk_used = btrfs_block_group_used(&cache->item);
2715 
2716 	if (bargs->usage == 0)
2717 		user_thresh = 1;
2718 	else if (bargs->usage > 100)
2719 		user_thresh = cache->key.offset;
2720 	else
2721 		user_thresh = div_factor_fine(cache->key.offset,
2722 					      bargs->usage);
2723 
2724 	if (chunk_used < user_thresh)
2725 		ret = 0;
2726 
2727 	btrfs_put_block_group(cache);
2728 	return ret;
2729 }
2730 
2731 static int chunk_devid_filter(struct extent_buffer *leaf,
2732 			      struct btrfs_chunk *chunk,
2733 			      struct btrfs_balance_args *bargs)
2734 {
2735 	struct btrfs_stripe *stripe;
2736 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2737 	int i;
2738 
2739 	for (i = 0; i < num_stripes; i++) {
2740 		stripe = btrfs_stripe_nr(chunk, i);
2741 		if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2742 			return 0;
2743 	}
2744 
2745 	return 1;
2746 }
2747 
2748 /* [pstart, pend) */
2749 static int chunk_drange_filter(struct extent_buffer *leaf,
2750 			       struct btrfs_chunk *chunk,
2751 			       u64 chunk_offset,
2752 			       struct btrfs_balance_args *bargs)
2753 {
2754 	struct btrfs_stripe *stripe;
2755 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2756 	u64 stripe_offset;
2757 	u64 stripe_length;
2758 	int factor;
2759 	int i;
2760 
2761 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2762 		return 0;
2763 
2764 	if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2765 	     BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2766 		factor = num_stripes / 2;
2767 	} else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2768 		factor = num_stripes - 1;
2769 	} else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2770 		factor = num_stripes - 2;
2771 	} else {
2772 		factor = num_stripes;
2773 	}
2774 
2775 	for (i = 0; i < num_stripes; i++) {
2776 		stripe = btrfs_stripe_nr(chunk, i);
2777 		if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2778 			continue;
2779 
2780 		stripe_offset = btrfs_stripe_offset(leaf, stripe);
2781 		stripe_length = btrfs_chunk_length(leaf, chunk);
2782 		do_div(stripe_length, factor);
2783 
2784 		if (stripe_offset < bargs->pend &&
2785 		    stripe_offset + stripe_length > bargs->pstart)
2786 			return 0;
2787 	}
2788 
2789 	return 1;
2790 }
2791 
2792 /* [vstart, vend) */
2793 static int chunk_vrange_filter(struct extent_buffer *leaf,
2794 			       struct btrfs_chunk *chunk,
2795 			       u64 chunk_offset,
2796 			       struct btrfs_balance_args *bargs)
2797 {
2798 	if (chunk_offset < bargs->vend &&
2799 	    chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2800 		/* at least part of the chunk is inside this vrange */
2801 		return 0;
2802 
2803 	return 1;
2804 }
2805 
2806 static int chunk_soft_convert_filter(u64 chunk_type,
2807 				     struct btrfs_balance_args *bargs)
2808 {
2809 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2810 		return 0;
2811 
2812 	chunk_type = chunk_to_extended(chunk_type) &
2813 				BTRFS_EXTENDED_PROFILE_MASK;
2814 
2815 	if (bargs->target == chunk_type)
2816 		return 1;
2817 
2818 	return 0;
2819 }
2820 
2821 static int should_balance_chunk(struct btrfs_root *root,
2822 				struct extent_buffer *leaf,
2823 				struct btrfs_chunk *chunk, u64 chunk_offset)
2824 {
2825 	struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2826 	struct btrfs_balance_args *bargs = NULL;
2827 	u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2828 
2829 	/* type filter */
2830 	if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2831 	      (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2832 		return 0;
2833 	}
2834 
2835 	if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2836 		bargs = &bctl->data;
2837 	else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2838 		bargs = &bctl->sys;
2839 	else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2840 		bargs = &bctl->meta;
2841 
2842 	/* profiles filter */
2843 	if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2844 	    chunk_profiles_filter(chunk_type, bargs)) {
2845 		return 0;
2846 	}
2847 
2848 	/* usage filter */
2849 	if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2850 	    chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2851 		return 0;
2852 	}
2853 
2854 	/* devid filter */
2855 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2856 	    chunk_devid_filter(leaf, chunk, bargs)) {
2857 		return 0;
2858 	}
2859 
2860 	/* drange filter, makes sense only with devid filter */
2861 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2862 	    chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2863 		return 0;
2864 	}
2865 
2866 	/* vrange filter */
2867 	if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2868 	    chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2869 		return 0;
2870 	}
2871 
2872 	/* soft profile changing mode */
2873 	if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2874 	    chunk_soft_convert_filter(chunk_type, bargs)) {
2875 		return 0;
2876 	}
2877 
2878 	return 1;
2879 }
2880 
2881 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2882 {
2883 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2884 	struct btrfs_root *chunk_root = fs_info->chunk_root;
2885 	struct btrfs_root *dev_root = fs_info->dev_root;
2886 	struct list_head *devices;
2887 	struct btrfs_device *device;
2888 	u64 old_size;
2889 	u64 size_to_free;
2890 	struct btrfs_chunk *chunk;
2891 	struct btrfs_path *path;
2892 	struct btrfs_key key;
2893 	struct btrfs_key found_key;
2894 	struct btrfs_trans_handle *trans;
2895 	struct extent_buffer *leaf;
2896 	int slot;
2897 	int ret;
2898 	int enospc_errors = 0;
2899 	bool counting = true;
2900 
2901 	/* step one make some room on all the devices */
2902 	devices = &fs_info->fs_devices->devices;
2903 	list_for_each_entry(device, devices, dev_list) {
2904 		old_size = device->total_bytes;
2905 		size_to_free = div_factor(old_size, 1);
2906 		size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2907 		if (!device->writeable ||
2908 		    device->total_bytes - device->bytes_used > size_to_free ||
2909 		    device->is_tgtdev_for_dev_replace)
2910 			continue;
2911 
2912 		ret = btrfs_shrink_device(device, old_size - size_to_free);
2913 		if (ret == -ENOSPC)
2914 			break;
2915 		BUG_ON(ret);
2916 
2917 		trans = btrfs_start_transaction(dev_root, 0);
2918 		BUG_ON(IS_ERR(trans));
2919 
2920 		ret = btrfs_grow_device(trans, device, old_size);
2921 		BUG_ON(ret);
2922 
2923 		btrfs_end_transaction(trans, dev_root);
2924 	}
2925 
2926 	/* step two, relocate all the chunks */
2927 	path = btrfs_alloc_path();
2928 	if (!path) {
2929 		ret = -ENOMEM;
2930 		goto error;
2931 	}
2932 
2933 	/* zero out stat counters */
2934 	spin_lock(&fs_info->balance_lock);
2935 	memset(&bctl->stat, 0, sizeof(bctl->stat));
2936 	spin_unlock(&fs_info->balance_lock);
2937 again:
2938 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2939 	key.offset = (u64)-1;
2940 	key.type = BTRFS_CHUNK_ITEM_KEY;
2941 
2942 	while (1) {
2943 		if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2944 		    atomic_read(&fs_info->balance_cancel_req)) {
2945 			ret = -ECANCELED;
2946 			goto error;
2947 		}
2948 
2949 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2950 		if (ret < 0)
2951 			goto error;
2952 
2953 		/*
2954 		 * this shouldn't happen, it means the last relocate
2955 		 * failed
2956 		 */
2957 		if (ret == 0)
2958 			BUG(); /* FIXME break ? */
2959 
2960 		ret = btrfs_previous_item(chunk_root, path, 0,
2961 					  BTRFS_CHUNK_ITEM_KEY);
2962 		if (ret) {
2963 			ret = 0;
2964 			break;
2965 		}
2966 
2967 		leaf = path->nodes[0];
2968 		slot = path->slots[0];
2969 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
2970 
2971 		if (found_key.objectid != key.objectid)
2972 			break;
2973 
2974 		/* chunk zero is special */
2975 		if (found_key.offset == 0)
2976 			break;
2977 
2978 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2979 
2980 		if (!counting) {
2981 			spin_lock(&fs_info->balance_lock);
2982 			bctl->stat.considered++;
2983 			spin_unlock(&fs_info->balance_lock);
2984 		}
2985 
2986 		ret = should_balance_chunk(chunk_root, leaf, chunk,
2987 					   found_key.offset);
2988 		btrfs_release_path(path);
2989 		if (!ret)
2990 			goto loop;
2991 
2992 		if (counting) {
2993 			spin_lock(&fs_info->balance_lock);
2994 			bctl->stat.expected++;
2995 			spin_unlock(&fs_info->balance_lock);
2996 			goto loop;
2997 		}
2998 
2999 		ret = btrfs_relocate_chunk(chunk_root,
3000 					   chunk_root->root_key.objectid,
3001 					   found_key.objectid,
3002 					   found_key.offset);
3003 		if (ret && ret != -ENOSPC)
3004 			goto error;
3005 		if (ret == -ENOSPC) {
3006 			enospc_errors++;
3007 		} else {
3008 			spin_lock(&fs_info->balance_lock);
3009 			bctl->stat.completed++;
3010 			spin_unlock(&fs_info->balance_lock);
3011 		}
3012 loop:
3013 		key.offset = found_key.offset - 1;
3014 	}
3015 
3016 	if (counting) {
3017 		btrfs_release_path(path);
3018 		counting = false;
3019 		goto again;
3020 	}
3021 error:
3022 	btrfs_free_path(path);
3023 	if (enospc_errors) {
3024 		printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3025 		       enospc_errors);
3026 		if (!ret)
3027 			ret = -ENOSPC;
3028 	}
3029 
3030 	return ret;
3031 }
3032 
3033 /**
3034  * alloc_profile_is_valid - see if a given profile is valid and reduced
3035  * @flags: profile to validate
3036  * @extended: if true @flags is treated as an extended profile
3037  */
3038 static int alloc_profile_is_valid(u64 flags, int extended)
3039 {
3040 	u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3041 			       BTRFS_BLOCK_GROUP_PROFILE_MASK);
3042 
3043 	flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3044 
3045 	/* 1) check that all other bits are zeroed */
3046 	if (flags & ~mask)
3047 		return 0;
3048 
3049 	/* 2) see if profile is reduced */
3050 	if (flags == 0)
3051 		return !extended; /* "0" is valid for usual profiles */
3052 
3053 	/* true if exactly one bit set */
3054 	return (flags & (flags - 1)) == 0;
3055 }
3056 
3057 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3058 {
3059 	/* cancel requested || normal exit path */
3060 	return atomic_read(&fs_info->balance_cancel_req) ||
3061 		(atomic_read(&fs_info->balance_pause_req) == 0 &&
3062 		 atomic_read(&fs_info->balance_cancel_req) == 0);
3063 }
3064 
3065 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3066 {
3067 	int ret;
3068 
3069 	unset_balance_control(fs_info);
3070 	ret = del_balance_item(fs_info->tree_root);
3071 	if (ret)
3072 		btrfs_std_error(fs_info, ret);
3073 
3074 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3075 }
3076 
3077 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3078 			       struct btrfs_ioctl_balance_args *bargs);
3079 
3080 /*
3081  * Should be called with both balance and volume mutexes held
3082  */
3083 int btrfs_balance(struct btrfs_balance_control *bctl,
3084 		  struct btrfs_ioctl_balance_args *bargs)
3085 {
3086 	struct btrfs_fs_info *fs_info = bctl->fs_info;
3087 	u64 allowed;
3088 	int mixed = 0;
3089 	int ret;
3090 	u64 num_devices;
3091 	unsigned seq;
3092 
3093 	if (btrfs_fs_closing(fs_info) ||
3094 	    atomic_read(&fs_info->balance_pause_req) ||
3095 	    atomic_read(&fs_info->balance_cancel_req)) {
3096 		ret = -EINVAL;
3097 		goto out;
3098 	}
3099 
3100 	allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3101 	if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3102 		mixed = 1;
3103 
3104 	/*
3105 	 * In case of mixed groups both data and meta should be picked,
3106 	 * and identical options should be given for both of them.
3107 	 */
3108 	allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3109 	if (mixed && (bctl->flags & allowed)) {
3110 		if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3111 		    !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3112 		    memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3113 			printk(KERN_ERR "btrfs: with mixed groups data and "
3114 			       "metadata balance options must be the same\n");
3115 			ret = -EINVAL;
3116 			goto out;
3117 		}
3118 	}
3119 
3120 	num_devices = fs_info->fs_devices->num_devices;
3121 	btrfs_dev_replace_lock(&fs_info->dev_replace);
3122 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3123 		BUG_ON(num_devices < 1);
3124 		num_devices--;
3125 	}
3126 	btrfs_dev_replace_unlock(&fs_info->dev_replace);
3127 	allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3128 	if (num_devices == 1)
3129 		allowed |= BTRFS_BLOCK_GROUP_DUP;
3130 	else if (num_devices > 1)
3131 		allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3132 	if (num_devices > 2)
3133 		allowed |= BTRFS_BLOCK_GROUP_RAID5;
3134 	if (num_devices > 3)
3135 		allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3136 			    BTRFS_BLOCK_GROUP_RAID6);
3137 	if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3138 	    (!alloc_profile_is_valid(bctl->data.target, 1) ||
3139 	     (bctl->data.target & ~allowed))) {
3140 		printk(KERN_ERR "btrfs: unable to start balance with target "
3141 		       "data profile %llu\n",
3142 		       (unsigned long long)bctl->data.target);
3143 		ret = -EINVAL;
3144 		goto out;
3145 	}
3146 	if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3147 	    (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3148 	     (bctl->meta.target & ~allowed))) {
3149 		printk(KERN_ERR "btrfs: unable to start balance with target "
3150 		       "metadata profile %llu\n",
3151 		       (unsigned long long)bctl->meta.target);
3152 		ret = -EINVAL;
3153 		goto out;
3154 	}
3155 	if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3156 	    (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3157 	     (bctl->sys.target & ~allowed))) {
3158 		printk(KERN_ERR "btrfs: unable to start balance with target "
3159 		       "system profile %llu\n",
3160 		       (unsigned long long)bctl->sys.target);
3161 		ret = -EINVAL;
3162 		goto out;
3163 	}
3164 
3165 	/* allow dup'ed data chunks only in mixed mode */
3166 	if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3167 	    (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3168 		printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3169 		ret = -EINVAL;
3170 		goto out;
3171 	}
3172 
3173 	/* allow to reduce meta or sys integrity only if force set */
3174 	allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3175 			BTRFS_BLOCK_GROUP_RAID10 |
3176 			BTRFS_BLOCK_GROUP_RAID5 |
3177 			BTRFS_BLOCK_GROUP_RAID6;
3178 	do {
3179 		seq = read_seqbegin(&fs_info->profiles_lock);
3180 
3181 		if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3182 		     (fs_info->avail_system_alloc_bits & allowed) &&
3183 		     !(bctl->sys.target & allowed)) ||
3184 		    ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3185 		     (fs_info->avail_metadata_alloc_bits & allowed) &&
3186 		     !(bctl->meta.target & allowed))) {
3187 			if (bctl->flags & BTRFS_BALANCE_FORCE) {
3188 				printk(KERN_INFO "btrfs: force reducing metadata "
3189 				       "integrity\n");
3190 			} else {
3191 				printk(KERN_ERR "btrfs: balance will reduce metadata "
3192 				       "integrity, use force if you want this\n");
3193 				ret = -EINVAL;
3194 				goto out;
3195 			}
3196 		}
3197 	} while (read_seqretry(&fs_info->profiles_lock, seq));
3198 
3199 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3200 		int num_tolerated_disk_barrier_failures;
3201 		u64 target = bctl->sys.target;
3202 
3203 		num_tolerated_disk_barrier_failures =
3204 			btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3205 		if (num_tolerated_disk_barrier_failures > 0 &&
3206 		    (target &
3207 		     (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3208 		      BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3209 			num_tolerated_disk_barrier_failures = 0;
3210 		else if (num_tolerated_disk_barrier_failures > 1 &&
3211 			 (target &
3212 			  (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3213 			num_tolerated_disk_barrier_failures = 1;
3214 
3215 		fs_info->num_tolerated_disk_barrier_failures =
3216 			num_tolerated_disk_barrier_failures;
3217 	}
3218 
3219 	ret = insert_balance_item(fs_info->tree_root, bctl);
3220 	if (ret && ret != -EEXIST)
3221 		goto out;
3222 
3223 	if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3224 		BUG_ON(ret == -EEXIST);
3225 		set_balance_control(bctl);
3226 	} else {
3227 		BUG_ON(ret != -EEXIST);
3228 		spin_lock(&fs_info->balance_lock);
3229 		update_balance_args(bctl);
3230 		spin_unlock(&fs_info->balance_lock);
3231 	}
3232 
3233 	atomic_inc(&fs_info->balance_running);
3234 	mutex_unlock(&fs_info->balance_mutex);
3235 
3236 	ret = __btrfs_balance(fs_info);
3237 
3238 	mutex_lock(&fs_info->balance_mutex);
3239 	atomic_dec(&fs_info->balance_running);
3240 
3241 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3242 		fs_info->num_tolerated_disk_barrier_failures =
3243 			btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3244 	}
3245 
3246 	if (bargs) {
3247 		memset(bargs, 0, sizeof(*bargs));
3248 		update_ioctl_balance_args(fs_info, 0, bargs);
3249 	}
3250 
3251 	if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3252 	    balance_need_close(fs_info)) {
3253 		__cancel_balance(fs_info);
3254 	}
3255 
3256 	wake_up(&fs_info->balance_wait_q);
3257 
3258 	return ret;
3259 out:
3260 	if (bctl->flags & BTRFS_BALANCE_RESUME)
3261 		__cancel_balance(fs_info);
3262 	else {
3263 		kfree(bctl);
3264 		atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3265 	}
3266 	return ret;
3267 }
3268 
3269 static int balance_kthread(void *data)
3270 {
3271 	struct btrfs_fs_info *fs_info = data;
3272 	int ret = 0;
3273 
3274 	mutex_lock(&fs_info->volume_mutex);
3275 	mutex_lock(&fs_info->balance_mutex);
3276 
3277 	if (fs_info->balance_ctl) {
3278 		printk(KERN_INFO "btrfs: continuing balance\n");
3279 		ret = btrfs_balance(fs_info->balance_ctl, NULL);
3280 	}
3281 
3282 	mutex_unlock(&fs_info->balance_mutex);
3283 	mutex_unlock(&fs_info->volume_mutex);
3284 
3285 	return ret;
3286 }
3287 
3288 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3289 {
3290 	struct task_struct *tsk;
3291 
3292 	spin_lock(&fs_info->balance_lock);
3293 	if (!fs_info->balance_ctl) {
3294 		spin_unlock(&fs_info->balance_lock);
3295 		return 0;
3296 	}
3297 	spin_unlock(&fs_info->balance_lock);
3298 
3299 	if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3300 		printk(KERN_INFO "btrfs: force skipping balance\n");
3301 		return 0;
3302 	}
3303 
3304 	tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3305 	return PTR_RET(tsk);
3306 }
3307 
3308 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3309 {
3310 	struct btrfs_balance_control *bctl;
3311 	struct btrfs_balance_item *item;
3312 	struct btrfs_disk_balance_args disk_bargs;
3313 	struct btrfs_path *path;
3314 	struct extent_buffer *leaf;
3315 	struct btrfs_key key;
3316 	int ret;
3317 
3318 	path = btrfs_alloc_path();
3319 	if (!path)
3320 		return -ENOMEM;
3321 
3322 	key.objectid = BTRFS_BALANCE_OBJECTID;
3323 	key.type = BTRFS_BALANCE_ITEM_KEY;
3324 	key.offset = 0;
3325 
3326 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3327 	if (ret < 0)
3328 		goto out;
3329 	if (ret > 0) { /* ret = -ENOENT; */
3330 		ret = 0;
3331 		goto out;
3332 	}
3333 
3334 	bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3335 	if (!bctl) {
3336 		ret = -ENOMEM;
3337 		goto out;
3338 	}
3339 
3340 	leaf = path->nodes[0];
3341 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3342 
3343 	bctl->fs_info = fs_info;
3344 	bctl->flags = btrfs_balance_flags(leaf, item);
3345 	bctl->flags |= BTRFS_BALANCE_RESUME;
3346 
3347 	btrfs_balance_data(leaf, item, &disk_bargs);
3348 	btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3349 	btrfs_balance_meta(leaf, item, &disk_bargs);
3350 	btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3351 	btrfs_balance_sys(leaf, item, &disk_bargs);
3352 	btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3353 
3354 	WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3355 
3356 	mutex_lock(&fs_info->volume_mutex);
3357 	mutex_lock(&fs_info->balance_mutex);
3358 
3359 	set_balance_control(bctl);
3360 
3361 	mutex_unlock(&fs_info->balance_mutex);
3362 	mutex_unlock(&fs_info->volume_mutex);
3363 out:
3364 	btrfs_free_path(path);
3365 	return ret;
3366 }
3367 
3368 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3369 {
3370 	int ret = 0;
3371 
3372 	mutex_lock(&fs_info->balance_mutex);
3373 	if (!fs_info->balance_ctl) {
3374 		mutex_unlock(&fs_info->balance_mutex);
3375 		return -ENOTCONN;
3376 	}
3377 
3378 	if (atomic_read(&fs_info->balance_running)) {
3379 		atomic_inc(&fs_info->balance_pause_req);
3380 		mutex_unlock(&fs_info->balance_mutex);
3381 
3382 		wait_event(fs_info->balance_wait_q,
3383 			   atomic_read(&fs_info->balance_running) == 0);
3384 
3385 		mutex_lock(&fs_info->balance_mutex);
3386 		/* we are good with balance_ctl ripped off from under us */
3387 		BUG_ON(atomic_read(&fs_info->balance_running));
3388 		atomic_dec(&fs_info->balance_pause_req);
3389 	} else {
3390 		ret = -ENOTCONN;
3391 	}
3392 
3393 	mutex_unlock(&fs_info->balance_mutex);
3394 	return ret;
3395 }
3396 
3397 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3398 {
3399 	mutex_lock(&fs_info->balance_mutex);
3400 	if (!fs_info->balance_ctl) {
3401 		mutex_unlock(&fs_info->balance_mutex);
3402 		return -ENOTCONN;
3403 	}
3404 
3405 	atomic_inc(&fs_info->balance_cancel_req);
3406 	/*
3407 	 * if we are running just wait and return, balance item is
3408 	 * deleted in btrfs_balance in this case
3409 	 */
3410 	if (atomic_read(&fs_info->balance_running)) {
3411 		mutex_unlock(&fs_info->balance_mutex);
3412 		wait_event(fs_info->balance_wait_q,
3413 			   atomic_read(&fs_info->balance_running) == 0);
3414 		mutex_lock(&fs_info->balance_mutex);
3415 	} else {
3416 		/* __cancel_balance needs volume_mutex */
3417 		mutex_unlock(&fs_info->balance_mutex);
3418 		mutex_lock(&fs_info->volume_mutex);
3419 		mutex_lock(&fs_info->balance_mutex);
3420 
3421 		if (fs_info->balance_ctl)
3422 			__cancel_balance(fs_info);
3423 
3424 		mutex_unlock(&fs_info->volume_mutex);
3425 	}
3426 
3427 	BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3428 	atomic_dec(&fs_info->balance_cancel_req);
3429 	mutex_unlock(&fs_info->balance_mutex);
3430 	return 0;
3431 }
3432 
3433 /*
3434  * shrinking a device means finding all of the device extents past
3435  * the new size, and then following the back refs to the chunks.
3436  * The chunk relocation code actually frees the device extent
3437  */
3438 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3439 {
3440 	struct btrfs_trans_handle *trans;
3441 	struct btrfs_root *root = device->dev_root;
3442 	struct btrfs_dev_extent *dev_extent = NULL;
3443 	struct btrfs_path *path;
3444 	u64 length;
3445 	u64 chunk_tree;
3446 	u64 chunk_objectid;
3447 	u64 chunk_offset;
3448 	int ret;
3449 	int slot;
3450 	int failed = 0;
3451 	bool retried = false;
3452 	struct extent_buffer *l;
3453 	struct btrfs_key key;
3454 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3455 	u64 old_total = btrfs_super_total_bytes(super_copy);
3456 	u64 old_size = device->total_bytes;
3457 	u64 diff = device->total_bytes - new_size;
3458 
3459 	if (device->is_tgtdev_for_dev_replace)
3460 		return -EINVAL;
3461 
3462 	path = btrfs_alloc_path();
3463 	if (!path)
3464 		return -ENOMEM;
3465 
3466 	path->reada = 2;
3467 
3468 	lock_chunks(root);
3469 
3470 	device->total_bytes = new_size;
3471 	if (device->writeable) {
3472 		device->fs_devices->total_rw_bytes -= diff;
3473 		spin_lock(&root->fs_info->free_chunk_lock);
3474 		root->fs_info->free_chunk_space -= diff;
3475 		spin_unlock(&root->fs_info->free_chunk_lock);
3476 	}
3477 	unlock_chunks(root);
3478 
3479 again:
3480 	key.objectid = device->devid;
3481 	key.offset = (u64)-1;
3482 	key.type = BTRFS_DEV_EXTENT_KEY;
3483 
3484 	do {
3485 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3486 		if (ret < 0)
3487 			goto done;
3488 
3489 		ret = btrfs_previous_item(root, path, 0, key.type);
3490 		if (ret < 0)
3491 			goto done;
3492 		if (ret) {
3493 			ret = 0;
3494 			btrfs_release_path(path);
3495 			break;
3496 		}
3497 
3498 		l = path->nodes[0];
3499 		slot = path->slots[0];
3500 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3501 
3502 		if (key.objectid != device->devid) {
3503 			btrfs_release_path(path);
3504 			break;
3505 		}
3506 
3507 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3508 		length = btrfs_dev_extent_length(l, dev_extent);
3509 
3510 		if (key.offset + length <= new_size) {
3511 			btrfs_release_path(path);
3512 			break;
3513 		}
3514 
3515 		chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3516 		chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3517 		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3518 		btrfs_release_path(path);
3519 
3520 		ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3521 					   chunk_offset);
3522 		if (ret && ret != -ENOSPC)
3523 			goto done;
3524 		if (ret == -ENOSPC)
3525 			failed++;
3526 	} while (key.offset-- > 0);
3527 
3528 	if (failed && !retried) {
3529 		failed = 0;
3530 		retried = true;
3531 		goto again;
3532 	} else if (failed && retried) {
3533 		ret = -ENOSPC;
3534 		lock_chunks(root);
3535 
3536 		device->total_bytes = old_size;
3537 		if (device->writeable)
3538 			device->fs_devices->total_rw_bytes += diff;
3539 		spin_lock(&root->fs_info->free_chunk_lock);
3540 		root->fs_info->free_chunk_space += diff;
3541 		spin_unlock(&root->fs_info->free_chunk_lock);
3542 		unlock_chunks(root);
3543 		goto done;
3544 	}
3545 
3546 	/* Shrinking succeeded, else we would be at "done". */
3547 	trans = btrfs_start_transaction(root, 0);
3548 	if (IS_ERR(trans)) {
3549 		ret = PTR_ERR(trans);
3550 		goto done;
3551 	}
3552 
3553 	lock_chunks(root);
3554 
3555 	device->disk_total_bytes = new_size;
3556 	/* Now btrfs_update_device() will change the on-disk size. */
3557 	ret = btrfs_update_device(trans, device);
3558 	if (ret) {
3559 		unlock_chunks(root);
3560 		btrfs_end_transaction(trans, root);
3561 		goto done;
3562 	}
3563 	WARN_ON(diff > old_total);
3564 	btrfs_set_super_total_bytes(super_copy, old_total - diff);
3565 	unlock_chunks(root);
3566 	btrfs_end_transaction(trans, root);
3567 done:
3568 	btrfs_free_path(path);
3569 	return ret;
3570 }
3571 
3572 static int btrfs_add_system_chunk(struct btrfs_root *root,
3573 			   struct btrfs_key *key,
3574 			   struct btrfs_chunk *chunk, int item_size)
3575 {
3576 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3577 	struct btrfs_disk_key disk_key;
3578 	u32 array_size;
3579 	u8 *ptr;
3580 
3581 	array_size = btrfs_super_sys_array_size(super_copy);
3582 	if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3583 		return -EFBIG;
3584 
3585 	ptr = super_copy->sys_chunk_array + array_size;
3586 	btrfs_cpu_key_to_disk(&disk_key, key);
3587 	memcpy(ptr, &disk_key, sizeof(disk_key));
3588 	ptr += sizeof(disk_key);
3589 	memcpy(ptr, chunk, item_size);
3590 	item_size += sizeof(disk_key);
3591 	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3592 	return 0;
3593 }
3594 
3595 /*
3596  * sort the devices in descending order by max_avail, total_avail
3597  */
3598 static int btrfs_cmp_device_info(const void *a, const void *b)
3599 {
3600 	const struct btrfs_device_info *di_a = a;
3601 	const struct btrfs_device_info *di_b = b;
3602 
3603 	if (di_a->max_avail > di_b->max_avail)
3604 		return -1;
3605 	if (di_a->max_avail < di_b->max_avail)
3606 		return 1;
3607 	if (di_a->total_avail > di_b->total_avail)
3608 		return -1;
3609 	if (di_a->total_avail < di_b->total_avail)
3610 		return 1;
3611 	return 0;
3612 }
3613 
3614 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3615 	[BTRFS_RAID_RAID10] = {
3616 		.sub_stripes	= 2,
3617 		.dev_stripes	= 1,
3618 		.devs_max	= 0,	/* 0 == as many as possible */
3619 		.devs_min	= 4,
3620 		.devs_increment	= 2,
3621 		.ncopies	= 2,
3622 	},
3623 	[BTRFS_RAID_RAID1] = {
3624 		.sub_stripes	= 1,
3625 		.dev_stripes	= 1,
3626 		.devs_max	= 2,
3627 		.devs_min	= 2,
3628 		.devs_increment	= 2,
3629 		.ncopies	= 2,
3630 	},
3631 	[BTRFS_RAID_DUP] = {
3632 		.sub_stripes	= 1,
3633 		.dev_stripes	= 2,
3634 		.devs_max	= 1,
3635 		.devs_min	= 1,
3636 		.devs_increment	= 1,
3637 		.ncopies	= 2,
3638 	},
3639 	[BTRFS_RAID_RAID0] = {
3640 		.sub_stripes	= 1,
3641 		.dev_stripes	= 1,
3642 		.devs_max	= 0,
3643 		.devs_min	= 2,
3644 		.devs_increment	= 1,
3645 		.ncopies	= 1,
3646 	},
3647 	[BTRFS_RAID_SINGLE] = {
3648 		.sub_stripes	= 1,
3649 		.dev_stripes	= 1,
3650 		.devs_max	= 1,
3651 		.devs_min	= 1,
3652 		.devs_increment	= 1,
3653 		.ncopies	= 1,
3654 	},
3655 	[BTRFS_RAID_RAID5] = {
3656 		.sub_stripes	= 1,
3657 		.dev_stripes	= 1,
3658 		.devs_max	= 0,
3659 		.devs_min	= 2,
3660 		.devs_increment	= 1,
3661 		.ncopies	= 2,
3662 	},
3663 	[BTRFS_RAID_RAID6] = {
3664 		.sub_stripes	= 1,
3665 		.dev_stripes	= 1,
3666 		.devs_max	= 0,
3667 		.devs_min	= 3,
3668 		.devs_increment	= 1,
3669 		.ncopies	= 3,
3670 	},
3671 };
3672 
3673 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3674 {
3675 	/* TODO allow them to set a preferred stripe size */
3676 	return 64 * 1024;
3677 }
3678 
3679 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3680 {
3681 	if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3682 		return;
3683 
3684 	btrfs_set_fs_incompat(info, RAID56);
3685 }
3686 
3687 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3688 			       struct btrfs_root *extent_root, u64 start,
3689 			       u64 type)
3690 {
3691 	struct btrfs_fs_info *info = extent_root->fs_info;
3692 	struct btrfs_fs_devices *fs_devices = info->fs_devices;
3693 	struct list_head *cur;
3694 	struct map_lookup *map = NULL;
3695 	struct extent_map_tree *em_tree;
3696 	struct extent_map *em;
3697 	struct btrfs_device_info *devices_info = NULL;
3698 	u64 total_avail;
3699 	int num_stripes;	/* total number of stripes to allocate */
3700 	int data_stripes;	/* number of stripes that count for
3701 				   block group size */
3702 	int sub_stripes;	/* sub_stripes info for map */
3703 	int dev_stripes;	/* stripes per dev */
3704 	int devs_max;		/* max devs to use */
3705 	int devs_min;		/* min devs needed */
3706 	int devs_increment;	/* ndevs has to be a multiple of this */
3707 	int ncopies;		/* how many copies to data has */
3708 	int ret;
3709 	u64 max_stripe_size;
3710 	u64 max_chunk_size;
3711 	u64 stripe_size;
3712 	u64 num_bytes;
3713 	u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3714 	int ndevs;
3715 	int i;
3716 	int j;
3717 	int index;
3718 
3719 	BUG_ON(!alloc_profile_is_valid(type, 0));
3720 
3721 	if (list_empty(&fs_devices->alloc_list))
3722 		return -ENOSPC;
3723 
3724 	index = __get_raid_index(type);
3725 
3726 	sub_stripes = btrfs_raid_array[index].sub_stripes;
3727 	dev_stripes = btrfs_raid_array[index].dev_stripes;
3728 	devs_max = btrfs_raid_array[index].devs_max;
3729 	devs_min = btrfs_raid_array[index].devs_min;
3730 	devs_increment = btrfs_raid_array[index].devs_increment;
3731 	ncopies = btrfs_raid_array[index].ncopies;
3732 
3733 	if (type & BTRFS_BLOCK_GROUP_DATA) {
3734 		max_stripe_size = 1024 * 1024 * 1024;
3735 		max_chunk_size = 10 * max_stripe_size;
3736 	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3737 		/* for larger filesystems, use larger metadata chunks */
3738 		if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3739 			max_stripe_size = 1024 * 1024 * 1024;
3740 		else
3741 			max_stripe_size = 256 * 1024 * 1024;
3742 		max_chunk_size = max_stripe_size;
3743 	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3744 		max_stripe_size = 32 * 1024 * 1024;
3745 		max_chunk_size = 2 * max_stripe_size;
3746 	} else {
3747 		printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3748 		       type);
3749 		BUG_ON(1);
3750 	}
3751 
3752 	/* we don't want a chunk larger than 10% of writeable space */
3753 	max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3754 			     max_chunk_size);
3755 
3756 	devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3757 			       GFP_NOFS);
3758 	if (!devices_info)
3759 		return -ENOMEM;
3760 
3761 	cur = fs_devices->alloc_list.next;
3762 
3763 	/*
3764 	 * in the first pass through the devices list, we gather information
3765 	 * about the available holes on each device.
3766 	 */
3767 	ndevs = 0;
3768 	while (cur != &fs_devices->alloc_list) {
3769 		struct btrfs_device *device;
3770 		u64 max_avail;
3771 		u64 dev_offset;
3772 
3773 		device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3774 
3775 		cur = cur->next;
3776 
3777 		if (!device->writeable) {
3778 			WARN(1, KERN_ERR
3779 			       "btrfs: read-only device in alloc_list\n");
3780 			continue;
3781 		}
3782 
3783 		if (!device->in_fs_metadata ||
3784 		    device->is_tgtdev_for_dev_replace)
3785 			continue;
3786 
3787 		if (device->total_bytes > device->bytes_used)
3788 			total_avail = device->total_bytes - device->bytes_used;
3789 		else
3790 			total_avail = 0;
3791 
3792 		/* If there is no space on this device, skip it. */
3793 		if (total_avail == 0)
3794 			continue;
3795 
3796 		ret = find_free_dev_extent(trans, device,
3797 					   max_stripe_size * dev_stripes,
3798 					   &dev_offset, &max_avail);
3799 		if (ret && ret != -ENOSPC)
3800 			goto error;
3801 
3802 		if (ret == 0)
3803 			max_avail = max_stripe_size * dev_stripes;
3804 
3805 		if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3806 			continue;
3807 
3808 		if (ndevs == fs_devices->rw_devices) {
3809 			WARN(1, "%s: found more than %llu devices\n",
3810 			     __func__, fs_devices->rw_devices);
3811 			break;
3812 		}
3813 		devices_info[ndevs].dev_offset = dev_offset;
3814 		devices_info[ndevs].max_avail = max_avail;
3815 		devices_info[ndevs].total_avail = total_avail;
3816 		devices_info[ndevs].dev = device;
3817 		++ndevs;
3818 	}
3819 
3820 	/*
3821 	 * now sort the devices by hole size / available space
3822 	 */
3823 	sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3824 	     btrfs_cmp_device_info, NULL);
3825 
3826 	/* round down to number of usable stripes */
3827 	ndevs -= ndevs % devs_increment;
3828 
3829 	if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3830 		ret = -ENOSPC;
3831 		goto error;
3832 	}
3833 
3834 	if (devs_max && ndevs > devs_max)
3835 		ndevs = devs_max;
3836 	/*
3837 	 * the primary goal is to maximize the number of stripes, so use as many
3838 	 * devices as possible, even if the stripes are not maximum sized.
3839 	 */
3840 	stripe_size = devices_info[ndevs-1].max_avail;
3841 	num_stripes = ndevs * dev_stripes;
3842 
3843 	/*
3844 	 * this will have to be fixed for RAID1 and RAID10 over
3845 	 * more drives
3846 	 */
3847 	data_stripes = num_stripes / ncopies;
3848 
3849 	if (type & BTRFS_BLOCK_GROUP_RAID5) {
3850 		raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
3851 				 btrfs_super_stripesize(info->super_copy));
3852 		data_stripes = num_stripes - 1;
3853 	}
3854 	if (type & BTRFS_BLOCK_GROUP_RAID6) {
3855 		raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
3856 				 btrfs_super_stripesize(info->super_copy));
3857 		data_stripes = num_stripes - 2;
3858 	}
3859 
3860 	/*
3861 	 * Use the number of data stripes to figure out how big this chunk
3862 	 * is really going to be in terms of logical address space,
3863 	 * and compare that answer with the max chunk size
3864 	 */
3865 	if (stripe_size * data_stripes > max_chunk_size) {
3866 		u64 mask = (1ULL << 24) - 1;
3867 		stripe_size = max_chunk_size;
3868 		do_div(stripe_size, data_stripes);
3869 
3870 		/* bump the answer up to a 16MB boundary */
3871 		stripe_size = (stripe_size + mask) & ~mask;
3872 
3873 		/* but don't go higher than the limits we found
3874 		 * while searching for free extents
3875 		 */
3876 		if (stripe_size > devices_info[ndevs-1].max_avail)
3877 			stripe_size = devices_info[ndevs-1].max_avail;
3878 	}
3879 
3880 	do_div(stripe_size, dev_stripes);
3881 
3882 	/* align to BTRFS_STRIPE_LEN */
3883 	do_div(stripe_size, raid_stripe_len);
3884 	stripe_size *= raid_stripe_len;
3885 
3886 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3887 	if (!map) {
3888 		ret = -ENOMEM;
3889 		goto error;
3890 	}
3891 	map->num_stripes = num_stripes;
3892 
3893 	for (i = 0; i < ndevs; ++i) {
3894 		for (j = 0; j < dev_stripes; ++j) {
3895 			int s = i * dev_stripes + j;
3896 			map->stripes[s].dev = devices_info[i].dev;
3897 			map->stripes[s].physical = devices_info[i].dev_offset +
3898 						   j * stripe_size;
3899 		}
3900 	}
3901 	map->sector_size = extent_root->sectorsize;
3902 	map->stripe_len = raid_stripe_len;
3903 	map->io_align = raid_stripe_len;
3904 	map->io_width = raid_stripe_len;
3905 	map->type = type;
3906 	map->sub_stripes = sub_stripes;
3907 
3908 	num_bytes = stripe_size * data_stripes;
3909 
3910 	trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3911 
3912 	em = alloc_extent_map();
3913 	if (!em) {
3914 		ret = -ENOMEM;
3915 		goto error;
3916 	}
3917 	em->bdev = (struct block_device *)map;
3918 	em->start = start;
3919 	em->len = num_bytes;
3920 	em->block_start = 0;
3921 	em->block_len = em->len;
3922 	em->orig_block_len = stripe_size;
3923 
3924 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3925 	write_lock(&em_tree->lock);
3926 	ret = add_extent_mapping(em_tree, em, 0);
3927 	if (!ret) {
3928 		list_add_tail(&em->list, &trans->transaction->pending_chunks);
3929 		atomic_inc(&em->refs);
3930 	}
3931 	write_unlock(&em_tree->lock);
3932 	if (ret) {
3933 		free_extent_map(em);
3934 		goto error;
3935 	}
3936 
3937 	ret = btrfs_make_block_group(trans, extent_root, 0, type,
3938 				     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3939 				     start, num_bytes);
3940 	if (ret)
3941 		goto error_del_extent;
3942 
3943 	free_extent_map(em);
3944 	check_raid56_incompat_flag(extent_root->fs_info, type);
3945 
3946 	kfree(devices_info);
3947 	return 0;
3948 
3949 error_del_extent:
3950 	write_lock(&em_tree->lock);
3951 	remove_extent_mapping(em_tree, em);
3952 	write_unlock(&em_tree->lock);
3953 
3954 	/* One for our allocation */
3955 	free_extent_map(em);
3956 	/* One for the tree reference */
3957 	free_extent_map(em);
3958 error:
3959 	kfree(map);
3960 	kfree(devices_info);
3961 	return ret;
3962 }
3963 
3964 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
3965 				struct btrfs_root *extent_root,
3966 				u64 chunk_offset, u64 chunk_size)
3967 {
3968 	struct btrfs_key key;
3969 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3970 	struct btrfs_device *device;
3971 	struct btrfs_chunk *chunk;
3972 	struct btrfs_stripe *stripe;
3973 	struct extent_map_tree *em_tree;
3974 	struct extent_map *em;
3975 	struct map_lookup *map;
3976 	size_t item_size;
3977 	u64 dev_offset;
3978 	u64 stripe_size;
3979 	int i = 0;
3980 	int ret;
3981 
3982 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3983 	read_lock(&em_tree->lock);
3984 	em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
3985 	read_unlock(&em_tree->lock);
3986 
3987 	if (!em) {
3988 		btrfs_crit(extent_root->fs_info, "unable to find logical "
3989 			   "%Lu len %Lu", chunk_offset, chunk_size);
3990 		return -EINVAL;
3991 	}
3992 
3993 	if (em->start != chunk_offset || em->len != chunk_size) {
3994 		btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
3995 			  " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset,
3996 			  chunk_size, em->start, em->len);
3997 		free_extent_map(em);
3998 		return -EINVAL;
3999 	}
4000 
4001 	map = (struct map_lookup *)em->bdev;
4002 	item_size = btrfs_chunk_item_size(map->num_stripes);
4003 	stripe_size = em->orig_block_len;
4004 
4005 	chunk = kzalloc(item_size, GFP_NOFS);
4006 	if (!chunk) {
4007 		ret = -ENOMEM;
4008 		goto out;
4009 	}
4010 
4011 	for (i = 0; i < map->num_stripes; i++) {
4012 		device = map->stripes[i].dev;
4013 		dev_offset = map->stripes[i].physical;
4014 
4015 		device->bytes_used += stripe_size;
4016 		ret = btrfs_update_device(trans, device);
4017 		if (ret)
4018 			goto out;
4019 		ret = btrfs_alloc_dev_extent(trans, device,
4020 					     chunk_root->root_key.objectid,
4021 					     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4022 					     chunk_offset, dev_offset,
4023 					     stripe_size);
4024 		if (ret)
4025 			goto out;
4026 	}
4027 
4028 	spin_lock(&extent_root->fs_info->free_chunk_lock);
4029 	extent_root->fs_info->free_chunk_space -= (stripe_size *
4030 						   map->num_stripes);
4031 	spin_unlock(&extent_root->fs_info->free_chunk_lock);
4032 
4033 	stripe = &chunk->stripe;
4034 	for (i = 0; i < map->num_stripes; i++) {
4035 		device = map->stripes[i].dev;
4036 		dev_offset = map->stripes[i].physical;
4037 
4038 		btrfs_set_stack_stripe_devid(stripe, device->devid);
4039 		btrfs_set_stack_stripe_offset(stripe, dev_offset);
4040 		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4041 		stripe++;
4042 	}
4043 
4044 	btrfs_set_stack_chunk_length(chunk, chunk_size);
4045 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4046 	btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4047 	btrfs_set_stack_chunk_type(chunk, map->type);
4048 	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4049 	btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4050 	btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4051 	btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4052 	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4053 
4054 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4055 	key.type = BTRFS_CHUNK_ITEM_KEY;
4056 	key.offset = chunk_offset;
4057 
4058 	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4059 	if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4060 		/*
4061 		 * TODO: Cleanup of inserted chunk root in case of
4062 		 * failure.
4063 		 */
4064 		ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4065 					     item_size);
4066 	}
4067 
4068 out:
4069 	kfree(chunk);
4070 	free_extent_map(em);
4071 	return ret;
4072 }
4073 
4074 /*
4075  * Chunk allocation falls into two parts. The first part does works
4076  * that make the new allocated chunk useable, but not do any operation
4077  * that modifies the chunk tree. The second part does the works that
4078  * require modifying the chunk tree. This division is important for the
4079  * bootstrap process of adding storage to a seed btrfs.
4080  */
4081 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4082 		      struct btrfs_root *extent_root, u64 type)
4083 {
4084 	u64 chunk_offset;
4085 
4086 	chunk_offset = find_next_chunk(extent_root->fs_info);
4087 	return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4088 }
4089 
4090 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4091 					 struct btrfs_root *root,
4092 					 struct btrfs_device *device)
4093 {
4094 	u64 chunk_offset;
4095 	u64 sys_chunk_offset;
4096 	u64 alloc_profile;
4097 	struct btrfs_fs_info *fs_info = root->fs_info;
4098 	struct btrfs_root *extent_root = fs_info->extent_root;
4099 	int ret;
4100 
4101 	chunk_offset = find_next_chunk(fs_info);
4102 	alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4103 	ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4104 				  alloc_profile);
4105 	if (ret)
4106 		return ret;
4107 
4108 	sys_chunk_offset = find_next_chunk(root->fs_info);
4109 	alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4110 	ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4111 				  alloc_profile);
4112 	if (ret) {
4113 		btrfs_abort_transaction(trans, root, ret);
4114 		goto out;
4115 	}
4116 
4117 	ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4118 	if (ret)
4119 		btrfs_abort_transaction(trans, root, ret);
4120 out:
4121 	return ret;
4122 }
4123 
4124 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4125 {
4126 	struct extent_map *em;
4127 	struct map_lookup *map;
4128 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4129 	int readonly = 0;
4130 	int i;
4131 
4132 	read_lock(&map_tree->map_tree.lock);
4133 	em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4134 	read_unlock(&map_tree->map_tree.lock);
4135 	if (!em)
4136 		return 1;
4137 
4138 	if (btrfs_test_opt(root, DEGRADED)) {
4139 		free_extent_map(em);
4140 		return 0;
4141 	}
4142 
4143 	map = (struct map_lookup *)em->bdev;
4144 	for (i = 0; i < map->num_stripes; i++) {
4145 		if (!map->stripes[i].dev->writeable) {
4146 			readonly = 1;
4147 			break;
4148 		}
4149 	}
4150 	free_extent_map(em);
4151 	return readonly;
4152 }
4153 
4154 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4155 {
4156 	extent_map_tree_init(&tree->map_tree);
4157 }
4158 
4159 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4160 {
4161 	struct extent_map *em;
4162 
4163 	while (1) {
4164 		write_lock(&tree->map_tree.lock);
4165 		em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4166 		if (em)
4167 			remove_extent_mapping(&tree->map_tree, em);
4168 		write_unlock(&tree->map_tree.lock);
4169 		if (!em)
4170 			break;
4171 		kfree(em->bdev);
4172 		/* once for us */
4173 		free_extent_map(em);
4174 		/* once for the tree */
4175 		free_extent_map(em);
4176 	}
4177 }
4178 
4179 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4180 {
4181 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4182 	struct extent_map *em;
4183 	struct map_lookup *map;
4184 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4185 	int ret;
4186 
4187 	read_lock(&em_tree->lock);
4188 	em = lookup_extent_mapping(em_tree, logical, len);
4189 	read_unlock(&em_tree->lock);
4190 
4191 	/*
4192 	 * We could return errors for these cases, but that could get ugly and
4193 	 * we'd probably do the same thing which is just not do anything else
4194 	 * and exit, so return 1 so the callers don't try to use other copies.
4195 	 */
4196 	if (!em) {
4197 		btrfs_emerg(fs_info, "No mapping for %Lu-%Lu\n", logical,
4198 			    logical+len);
4199 		return 1;
4200 	}
4201 
4202 	if (em->start > logical || em->start + em->len < logical) {
4203 		btrfs_emerg(fs_info, "Invalid mapping for %Lu-%Lu, got "
4204 			    "%Lu-%Lu\n", logical, logical+len, em->start,
4205 			    em->start + em->len);
4206 		return 1;
4207 	}
4208 
4209 	map = (struct map_lookup *)em->bdev;
4210 	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4211 		ret = map->num_stripes;
4212 	else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4213 		ret = map->sub_stripes;
4214 	else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4215 		ret = 2;
4216 	else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4217 		ret = 3;
4218 	else
4219 		ret = 1;
4220 	free_extent_map(em);
4221 
4222 	btrfs_dev_replace_lock(&fs_info->dev_replace);
4223 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4224 		ret++;
4225 	btrfs_dev_replace_unlock(&fs_info->dev_replace);
4226 
4227 	return ret;
4228 }
4229 
4230 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4231 				    struct btrfs_mapping_tree *map_tree,
4232 				    u64 logical)
4233 {
4234 	struct extent_map *em;
4235 	struct map_lookup *map;
4236 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4237 	unsigned long len = root->sectorsize;
4238 
4239 	read_lock(&em_tree->lock);
4240 	em = lookup_extent_mapping(em_tree, logical, len);
4241 	read_unlock(&em_tree->lock);
4242 	BUG_ON(!em);
4243 
4244 	BUG_ON(em->start > logical || em->start + em->len < logical);
4245 	map = (struct map_lookup *)em->bdev;
4246 	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4247 			 BTRFS_BLOCK_GROUP_RAID6)) {
4248 		len = map->stripe_len * nr_data_stripes(map);
4249 	}
4250 	free_extent_map(em);
4251 	return len;
4252 }
4253 
4254 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4255 			   u64 logical, u64 len, int mirror_num)
4256 {
4257 	struct extent_map *em;
4258 	struct map_lookup *map;
4259 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4260 	int ret = 0;
4261 
4262 	read_lock(&em_tree->lock);
4263 	em = lookup_extent_mapping(em_tree, logical, len);
4264 	read_unlock(&em_tree->lock);
4265 	BUG_ON(!em);
4266 
4267 	BUG_ON(em->start > logical || em->start + em->len < logical);
4268 	map = (struct map_lookup *)em->bdev;
4269 	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4270 			 BTRFS_BLOCK_GROUP_RAID6))
4271 		ret = 1;
4272 	free_extent_map(em);
4273 	return ret;
4274 }
4275 
4276 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4277 			    struct map_lookup *map, int first, int num,
4278 			    int optimal, int dev_replace_is_ongoing)
4279 {
4280 	int i;
4281 	int tolerance;
4282 	struct btrfs_device *srcdev;
4283 
4284 	if (dev_replace_is_ongoing &&
4285 	    fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4286 	     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4287 		srcdev = fs_info->dev_replace.srcdev;
4288 	else
4289 		srcdev = NULL;
4290 
4291 	/*
4292 	 * try to avoid the drive that is the source drive for a
4293 	 * dev-replace procedure, only choose it if no other non-missing
4294 	 * mirror is available
4295 	 */
4296 	for (tolerance = 0; tolerance < 2; tolerance++) {
4297 		if (map->stripes[optimal].dev->bdev &&
4298 		    (tolerance || map->stripes[optimal].dev != srcdev))
4299 			return optimal;
4300 		for (i = first; i < first + num; i++) {
4301 			if (map->stripes[i].dev->bdev &&
4302 			    (tolerance || map->stripes[i].dev != srcdev))
4303 				return i;
4304 		}
4305 	}
4306 
4307 	/* we couldn't find one that doesn't fail.  Just return something
4308 	 * and the io error handling code will clean up eventually
4309 	 */
4310 	return optimal;
4311 }
4312 
4313 static inline int parity_smaller(u64 a, u64 b)
4314 {
4315 	return a > b;
4316 }
4317 
4318 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4319 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4320 {
4321 	struct btrfs_bio_stripe s;
4322 	int i;
4323 	u64 l;
4324 	int again = 1;
4325 
4326 	while (again) {
4327 		again = 0;
4328 		for (i = 0; i < bbio->num_stripes - 1; i++) {
4329 			if (parity_smaller(raid_map[i], raid_map[i+1])) {
4330 				s = bbio->stripes[i];
4331 				l = raid_map[i];
4332 				bbio->stripes[i] = bbio->stripes[i+1];
4333 				raid_map[i] = raid_map[i+1];
4334 				bbio->stripes[i+1] = s;
4335 				raid_map[i+1] = l;
4336 				again = 1;
4337 			}
4338 		}
4339 	}
4340 }
4341 
4342 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4343 			     u64 logical, u64 *length,
4344 			     struct btrfs_bio **bbio_ret,
4345 			     int mirror_num, u64 **raid_map_ret)
4346 {
4347 	struct extent_map *em;
4348 	struct map_lookup *map;
4349 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4350 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4351 	u64 offset;
4352 	u64 stripe_offset;
4353 	u64 stripe_end_offset;
4354 	u64 stripe_nr;
4355 	u64 stripe_nr_orig;
4356 	u64 stripe_nr_end;
4357 	u64 stripe_len;
4358 	u64 *raid_map = NULL;
4359 	int stripe_index;
4360 	int i;
4361 	int ret = 0;
4362 	int num_stripes;
4363 	int max_errors = 0;
4364 	struct btrfs_bio *bbio = NULL;
4365 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4366 	int dev_replace_is_ongoing = 0;
4367 	int num_alloc_stripes;
4368 	int patch_the_first_stripe_for_dev_replace = 0;
4369 	u64 physical_to_patch_in_first_stripe = 0;
4370 	u64 raid56_full_stripe_start = (u64)-1;
4371 
4372 	read_lock(&em_tree->lock);
4373 	em = lookup_extent_mapping(em_tree, logical, *length);
4374 	read_unlock(&em_tree->lock);
4375 
4376 	if (!em) {
4377 		btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4378 			(unsigned long long)logical,
4379 			(unsigned long long)*length);
4380 		return -EINVAL;
4381 	}
4382 
4383 	if (em->start > logical || em->start + em->len < logical) {
4384 		btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4385 			   "found %Lu-%Lu\n", logical, em->start,
4386 			   em->start + em->len);
4387 		return -EINVAL;
4388 	}
4389 
4390 	map = (struct map_lookup *)em->bdev;
4391 	offset = logical - em->start;
4392 
4393 	stripe_len = map->stripe_len;
4394 	stripe_nr = offset;
4395 	/*
4396 	 * stripe_nr counts the total number of stripes we have to stride
4397 	 * to get to this block
4398 	 */
4399 	do_div(stripe_nr, stripe_len);
4400 
4401 	stripe_offset = stripe_nr * stripe_len;
4402 	BUG_ON(offset < stripe_offset);
4403 
4404 	/* stripe_offset is the offset of this block in its stripe*/
4405 	stripe_offset = offset - stripe_offset;
4406 
4407 	/* if we're here for raid56, we need to know the stripe aligned start */
4408 	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4409 		unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4410 		raid56_full_stripe_start = offset;
4411 
4412 		/* allow a write of a full stripe, but make sure we don't
4413 		 * allow straddling of stripes
4414 		 */
4415 		do_div(raid56_full_stripe_start, full_stripe_len);
4416 		raid56_full_stripe_start *= full_stripe_len;
4417 	}
4418 
4419 	if (rw & REQ_DISCARD) {
4420 		/* we don't discard raid56 yet */
4421 		if (map->type &
4422 		    (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4423 			ret = -EOPNOTSUPP;
4424 			goto out;
4425 		}
4426 		*length = min_t(u64, em->len - offset, *length);
4427 	} else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4428 		u64 max_len;
4429 		/* For writes to RAID[56], allow a full stripeset across all disks.
4430 		   For other RAID types and for RAID[56] reads, just allow a single
4431 		   stripe (on a single disk). */
4432 		if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4433 		    (rw & REQ_WRITE)) {
4434 			max_len = stripe_len * nr_data_stripes(map) -
4435 				(offset - raid56_full_stripe_start);
4436 		} else {
4437 			/* we limit the length of each bio to what fits in a stripe */
4438 			max_len = stripe_len - stripe_offset;
4439 		}
4440 		*length = min_t(u64, em->len - offset, max_len);
4441 	} else {
4442 		*length = em->len - offset;
4443 	}
4444 
4445 	/* This is for when we're called from btrfs_merge_bio_hook() and all
4446 	   it cares about is the length */
4447 	if (!bbio_ret)
4448 		goto out;
4449 
4450 	btrfs_dev_replace_lock(dev_replace);
4451 	dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4452 	if (!dev_replace_is_ongoing)
4453 		btrfs_dev_replace_unlock(dev_replace);
4454 
4455 	if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4456 	    !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4457 	    dev_replace->tgtdev != NULL) {
4458 		/*
4459 		 * in dev-replace case, for repair case (that's the only
4460 		 * case where the mirror is selected explicitly when
4461 		 * calling btrfs_map_block), blocks left of the left cursor
4462 		 * can also be read from the target drive.
4463 		 * For REQ_GET_READ_MIRRORS, the target drive is added as
4464 		 * the last one to the array of stripes. For READ, it also
4465 		 * needs to be supported using the same mirror number.
4466 		 * If the requested block is not left of the left cursor,
4467 		 * EIO is returned. This can happen because btrfs_num_copies()
4468 		 * returns one more in the dev-replace case.
4469 		 */
4470 		u64 tmp_length = *length;
4471 		struct btrfs_bio *tmp_bbio = NULL;
4472 		int tmp_num_stripes;
4473 		u64 srcdev_devid = dev_replace->srcdev->devid;
4474 		int index_srcdev = 0;
4475 		int found = 0;
4476 		u64 physical_of_found = 0;
4477 
4478 		ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4479 			     logical, &tmp_length, &tmp_bbio, 0, NULL);
4480 		if (ret) {
4481 			WARN_ON(tmp_bbio != NULL);
4482 			goto out;
4483 		}
4484 
4485 		tmp_num_stripes = tmp_bbio->num_stripes;
4486 		if (mirror_num > tmp_num_stripes) {
4487 			/*
4488 			 * REQ_GET_READ_MIRRORS does not contain this
4489 			 * mirror, that means that the requested area
4490 			 * is not left of the left cursor
4491 			 */
4492 			ret = -EIO;
4493 			kfree(tmp_bbio);
4494 			goto out;
4495 		}
4496 
4497 		/*
4498 		 * process the rest of the function using the mirror_num
4499 		 * of the source drive. Therefore look it up first.
4500 		 * At the end, patch the device pointer to the one of the
4501 		 * target drive.
4502 		 */
4503 		for (i = 0; i < tmp_num_stripes; i++) {
4504 			if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4505 				/*
4506 				 * In case of DUP, in order to keep it
4507 				 * simple, only add the mirror with the
4508 				 * lowest physical address
4509 				 */
4510 				if (found &&
4511 				    physical_of_found <=
4512 				     tmp_bbio->stripes[i].physical)
4513 					continue;
4514 				index_srcdev = i;
4515 				found = 1;
4516 				physical_of_found =
4517 					tmp_bbio->stripes[i].physical;
4518 			}
4519 		}
4520 
4521 		if (found) {
4522 			mirror_num = index_srcdev + 1;
4523 			patch_the_first_stripe_for_dev_replace = 1;
4524 			physical_to_patch_in_first_stripe = physical_of_found;
4525 		} else {
4526 			WARN_ON(1);
4527 			ret = -EIO;
4528 			kfree(tmp_bbio);
4529 			goto out;
4530 		}
4531 
4532 		kfree(tmp_bbio);
4533 	} else if (mirror_num > map->num_stripes) {
4534 		mirror_num = 0;
4535 	}
4536 
4537 	num_stripes = 1;
4538 	stripe_index = 0;
4539 	stripe_nr_orig = stripe_nr;
4540 	stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
4541 	do_div(stripe_nr_end, map->stripe_len);
4542 	stripe_end_offset = stripe_nr_end * map->stripe_len -
4543 			    (offset + *length);
4544 
4545 	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4546 		if (rw & REQ_DISCARD)
4547 			num_stripes = min_t(u64, map->num_stripes,
4548 					    stripe_nr_end - stripe_nr_orig);
4549 		stripe_index = do_div(stripe_nr, map->num_stripes);
4550 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4551 		if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4552 			num_stripes = map->num_stripes;
4553 		else if (mirror_num)
4554 			stripe_index = mirror_num - 1;
4555 		else {
4556 			stripe_index = find_live_mirror(fs_info, map, 0,
4557 					    map->num_stripes,
4558 					    current->pid % map->num_stripes,
4559 					    dev_replace_is_ongoing);
4560 			mirror_num = stripe_index + 1;
4561 		}
4562 
4563 	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4564 		if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4565 			num_stripes = map->num_stripes;
4566 		} else if (mirror_num) {
4567 			stripe_index = mirror_num - 1;
4568 		} else {
4569 			mirror_num = 1;
4570 		}
4571 
4572 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4573 		int factor = map->num_stripes / map->sub_stripes;
4574 
4575 		stripe_index = do_div(stripe_nr, factor);
4576 		stripe_index *= map->sub_stripes;
4577 
4578 		if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4579 			num_stripes = map->sub_stripes;
4580 		else if (rw & REQ_DISCARD)
4581 			num_stripes = min_t(u64, map->sub_stripes *
4582 					    (stripe_nr_end - stripe_nr_orig),
4583 					    map->num_stripes);
4584 		else if (mirror_num)
4585 			stripe_index += mirror_num - 1;
4586 		else {
4587 			int old_stripe_index = stripe_index;
4588 			stripe_index = find_live_mirror(fs_info, map,
4589 					      stripe_index,
4590 					      map->sub_stripes, stripe_index +
4591 					      current->pid % map->sub_stripes,
4592 					      dev_replace_is_ongoing);
4593 			mirror_num = stripe_index - old_stripe_index + 1;
4594 		}
4595 
4596 	} else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4597 				BTRFS_BLOCK_GROUP_RAID6)) {
4598 		u64 tmp;
4599 
4600 		if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4601 		    && raid_map_ret) {
4602 			int i, rot;
4603 
4604 			/* push stripe_nr back to the start of the full stripe */
4605 			stripe_nr = raid56_full_stripe_start;
4606 			do_div(stripe_nr, stripe_len);
4607 
4608 			stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4609 
4610 			/* RAID[56] write or recovery. Return all stripes */
4611 			num_stripes = map->num_stripes;
4612 			max_errors = nr_parity_stripes(map);
4613 
4614 			raid_map = kmalloc(sizeof(u64) * num_stripes,
4615 					   GFP_NOFS);
4616 			if (!raid_map) {
4617 				ret = -ENOMEM;
4618 				goto out;
4619 			}
4620 
4621 			/* Work out the disk rotation on this stripe-set */
4622 			tmp = stripe_nr;
4623 			rot = do_div(tmp, num_stripes);
4624 
4625 			/* Fill in the logical address of each stripe */
4626 			tmp = stripe_nr * nr_data_stripes(map);
4627 			for (i = 0; i < nr_data_stripes(map); i++)
4628 				raid_map[(i+rot) % num_stripes] =
4629 					em->start + (tmp + i) * map->stripe_len;
4630 
4631 			raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4632 			if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4633 				raid_map[(i+rot+1) % num_stripes] =
4634 					RAID6_Q_STRIPE;
4635 
4636 			*length = map->stripe_len;
4637 			stripe_index = 0;
4638 			stripe_offset = 0;
4639 		} else {
4640 			/*
4641 			 * Mirror #0 or #1 means the original data block.
4642 			 * Mirror #2 is RAID5 parity block.
4643 			 * Mirror #3 is RAID6 Q block.
4644 			 */
4645 			stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4646 			if (mirror_num > 1)
4647 				stripe_index = nr_data_stripes(map) +
4648 						mirror_num - 2;
4649 
4650 			/* We distribute the parity blocks across stripes */
4651 			tmp = stripe_nr + stripe_index;
4652 			stripe_index = do_div(tmp, map->num_stripes);
4653 		}
4654 	} else {
4655 		/*
4656 		 * after this do_div call, stripe_nr is the number of stripes
4657 		 * on this device we have to walk to find the data, and
4658 		 * stripe_index is the number of our device in the stripe array
4659 		 */
4660 		stripe_index = do_div(stripe_nr, map->num_stripes);
4661 		mirror_num = stripe_index + 1;
4662 	}
4663 	BUG_ON(stripe_index >= map->num_stripes);
4664 
4665 	num_alloc_stripes = num_stripes;
4666 	if (dev_replace_is_ongoing) {
4667 		if (rw & (REQ_WRITE | REQ_DISCARD))
4668 			num_alloc_stripes <<= 1;
4669 		if (rw & REQ_GET_READ_MIRRORS)
4670 			num_alloc_stripes++;
4671 	}
4672 	bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4673 	if (!bbio) {
4674 		ret = -ENOMEM;
4675 		goto out;
4676 	}
4677 	atomic_set(&bbio->error, 0);
4678 
4679 	if (rw & REQ_DISCARD) {
4680 		int factor = 0;
4681 		int sub_stripes = 0;
4682 		u64 stripes_per_dev = 0;
4683 		u32 remaining_stripes = 0;
4684 		u32 last_stripe = 0;
4685 
4686 		if (map->type &
4687 		    (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4688 			if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4689 				sub_stripes = 1;
4690 			else
4691 				sub_stripes = map->sub_stripes;
4692 
4693 			factor = map->num_stripes / sub_stripes;
4694 			stripes_per_dev = div_u64_rem(stripe_nr_end -
4695 						      stripe_nr_orig,
4696 						      factor,
4697 						      &remaining_stripes);
4698 			div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4699 			last_stripe *= sub_stripes;
4700 		}
4701 
4702 		for (i = 0; i < num_stripes; i++) {
4703 			bbio->stripes[i].physical =
4704 				map->stripes[stripe_index].physical +
4705 				stripe_offset + stripe_nr * map->stripe_len;
4706 			bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4707 
4708 			if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4709 					 BTRFS_BLOCK_GROUP_RAID10)) {
4710 				bbio->stripes[i].length = stripes_per_dev *
4711 							  map->stripe_len;
4712 
4713 				if (i / sub_stripes < remaining_stripes)
4714 					bbio->stripes[i].length +=
4715 						map->stripe_len;
4716 
4717 				/*
4718 				 * Special for the first stripe and
4719 				 * the last stripe:
4720 				 *
4721 				 * |-------|...|-------|
4722 				 *     |----------|
4723 				 *    off     end_off
4724 				 */
4725 				if (i < sub_stripes)
4726 					bbio->stripes[i].length -=
4727 						stripe_offset;
4728 
4729 				if (stripe_index >= last_stripe &&
4730 				    stripe_index <= (last_stripe +
4731 						     sub_stripes - 1))
4732 					bbio->stripes[i].length -=
4733 						stripe_end_offset;
4734 
4735 				if (i == sub_stripes - 1)
4736 					stripe_offset = 0;
4737 			} else
4738 				bbio->stripes[i].length = *length;
4739 
4740 			stripe_index++;
4741 			if (stripe_index == map->num_stripes) {
4742 				/* This could only happen for RAID0/10 */
4743 				stripe_index = 0;
4744 				stripe_nr++;
4745 			}
4746 		}
4747 	} else {
4748 		for (i = 0; i < num_stripes; i++) {
4749 			bbio->stripes[i].physical =
4750 				map->stripes[stripe_index].physical +
4751 				stripe_offset +
4752 				stripe_nr * map->stripe_len;
4753 			bbio->stripes[i].dev =
4754 				map->stripes[stripe_index].dev;
4755 			stripe_index++;
4756 		}
4757 	}
4758 
4759 	if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4760 		if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4761 				 BTRFS_BLOCK_GROUP_RAID10 |
4762 				 BTRFS_BLOCK_GROUP_RAID5 |
4763 				 BTRFS_BLOCK_GROUP_DUP)) {
4764 			max_errors = 1;
4765 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4766 			max_errors = 2;
4767 		}
4768 	}
4769 
4770 	if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4771 	    dev_replace->tgtdev != NULL) {
4772 		int index_where_to_add;
4773 		u64 srcdev_devid = dev_replace->srcdev->devid;
4774 
4775 		/*
4776 		 * duplicate the write operations while the dev replace
4777 		 * procedure is running. Since the copying of the old disk
4778 		 * to the new disk takes place at run time while the
4779 		 * filesystem is mounted writable, the regular write
4780 		 * operations to the old disk have to be duplicated to go
4781 		 * to the new disk as well.
4782 		 * Note that device->missing is handled by the caller, and
4783 		 * that the write to the old disk is already set up in the
4784 		 * stripes array.
4785 		 */
4786 		index_where_to_add = num_stripes;
4787 		for (i = 0; i < num_stripes; i++) {
4788 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
4789 				/* write to new disk, too */
4790 				struct btrfs_bio_stripe *new =
4791 					bbio->stripes + index_where_to_add;
4792 				struct btrfs_bio_stripe *old =
4793 					bbio->stripes + i;
4794 
4795 				new->physical = old->physical;
4796 				new->length = old->length;
4797 				new->dev = dev_replace->tgtdev;
4798 				index_where_to_add++;
4799 				max_errors++;
4800 			}
4801 		}
4802 		num_stripes = index_where_to_add;
4803 	} else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4804 		   dev_replace->tgtdev != NULL) {
4805 		u64 srcdev_devid = dev_replace->srcdev->devid;
4806 		int index_srcdev = 0;
4807 		int found = 0;
4808 		u64 physical_of_found = 0;
4809 
4810 		/*
4811 		 * During the dev-replace procedure, the target drive can
4812 		 * also be used to read data in case it is needed to repair
4813 		 * a corrupt block elsewhere. This is possible if the
4814 		 * requested area is left of the left cursor. In this area,
4815 		 * the target drive is a full copy of the source drive.
4816 		 */
4817 		for (i = 0; i < num_stripes; i++) {
4818 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
4819 				/*
4820 				 * In case of DUP, in order to keep it
4821 				 * simple, only add the mirror with the
4822 				 * lowest physical address
4823 				 */
4824 				if (found &&
4825 				    physical_of_found <=
4826 				     bbio->stripes[i].physical)
4827 					continue;
4828 				index_srcdev = i;
4829 				found = 1;
4830 				physical_of_found = bbio->stripes[i].physical;
4831 			}
4832 		}
4833 		if (found) {
4834 			u64 length = map->stripe_len;
4835 
4836 			if (physical_of_found + length <=
4837 			    dev_replace->cursor_left) {
4838 				struct btrfs_bio_stripe *tgtdev_stripe =
4839 					bbio->stripes + num_stripes;
4840 
4841 				tgtdev_stripe->physical = physical_of_found;
4842 				tgtdev_stripe->length =
4843 					bbio->stripes[index_srcdev].length;
4844 				tgtdev_stripe->dev = dev_replace->tgtdev;
4845 
4846 				num_stripes++;
4847 			}
4848 		}
4849 	}
4850 
4851 	*bbio_ret = bbio;
4852 	bbio->num_stripes = num_stripes;
4853 	bbio->max_errors = max_errors;
4854 	bbio->mirror_num = mirror_num;
4855 
4856 	/*
4857 	 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4858 	 * mirror_num == num_stripes + 1 && dev_replace target drive is
4859 	 * available as a mirror
4860 	 */
4861 	if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4862 		WARN_ON(num_stripes > 1);
4863 		bbio->stripes[0].dev = dev_replace->tgtdev;
4864 		bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4865 		bbio->mirror_num = map->num_stripes + 1;
4866 	}
4867 	if (raid_map) {
4868 		sort_parity_stripes(bbio, raid_map);
4869 		*raid_map_ret = raid_map;
4870 	}
4871 out:
4872 	if (dev_replace_is_ongoing)
4873 		btrfs_dev_replace_unlock(dev_replace);
4874 	free_extent_map(em);
4875 	return ret;
4876 }
4877 
4878 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4879 		      u64 logical, u64 *length,
4880 		      struct btrfs_bio **bbio_ret, int mirror_num)
4881 {
4882 	return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4883 				 mirror_num, NULL);
4884 }
4885 
4886 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4887 		     u64 chunk_start, u64 physical, u64 devid,
4888 		     u64 **logical, int *naddrs, int *stripe_len)
4889 {
4890 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4891 	struct extent_map *em;
4892 	struct map_lookup *map;
4893 	u64 *buf;
4894 	u64 bytenr;
4895 	u64 length;
4896 	u64 stripe_nr;
4897 	u64 rmap_len;
4898 	int i, j, nr = 0;
4899 
4900 	read_lock(&em_tree->lock);
4901 	em = lookup_extent_mapping(em_tree, chunk_start, 1);
4902 	read_unlock(&em_tree->lock);
4903 
4904 	if (!em) {
4905 		printk(KERN_ERR "btrfs: couldn't find em for chunk %Lu\n",
4906 		       chunk_start);
4907 		return -EIO;
4908 	}
4909 
4910 	if (em->start != chunk_start) {
4911 		printk(KERN_ERR "btrfs: bad chunk start, em=%Lu, wanted=%Lu\n",
4912 		       em->start, chunk_start);
4913 		free_extent_map(em);
4914 		return -EIO;
4915 	}
4916 	map = (struct map_lookup *)em->bdev;
4917 
4918 	length = em->len;
4919 	rmap_len = map->stripe_len;
4920 
4921 	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4922 		do_div(length, map->num_stripes / map->sub_stripes);
4923 	else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4924 		do_div(length, map->num_stripes);
4925 	else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4926 			      BTRFS_BLOCK_GROUP_RAID6)) {
4927 		do_div(length, nr_data_stripes(map));
4928 		rmap_len = map->stripe_len * nr_data_stripes(map);
4929 	}
4930 
4931 	buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4932 	BUG_ON(!buf); /* -ENOMEM */
4933 
4934 	for (i = 0; i < map->num_stripes; i++) {
4935 		if (devid && map->stripes[i].dev->devid != devid)
4936 			continue;
4937 		if (map->stripes[i].physical > physical ||
4938 		    map->stripes[i].physical + length <= physical)
4939 			continue;
4940 
4941 		stripe_nr = physical - map->stripes[i].physical;
4942 		do_div(stripe_nr, map->stripe_len);
4943 
4944 		if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4945 			stripe_nr = stripe_nr * map->num_stripes + i;
4946 			do_div(stripe_nr, map->sub_stripes);
4947 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4948 			stripe_nr = stripe_nr * map->num_stripes + i;
4949 		} /* else if RAID[56], multiply by nr_data_stripes().
4950 		   * Alternatively, just use rmap_len below instead of
4951 		   * map->stripe_len */
4952 
4953 		bytenr = chunk_start + stripe_nr * rmap_len;
4954 		WARN_ON(nr >= map->num_stripes);
4955 		for (j = 0; j < nr; j++) {
4956 			if (buf[j] == bytenr)
4957 				break;
4958 		}
4959 		if (j == nr) {
4960 			WARN_ON(nr >= map->num_stripes);
4961 			buf[nr++] = bytenr;
4962 		}
4963 	}
4964 
4965 	*logical = buf;
4966 	*naddrs = nr;
4967 	*stripe_len = rmap_len;
4968 
4969 	free_extent_map(em);
4970 	return 0;
4971 }
4972 
4973 static void btrfs_end_bio(struct bio *bio, int err)
4974 {
4975 	struct btrfs_bio *bbio = bio->bi_private;
4976 	int is_orig_bio = 0;
4977 
4978 	if (err) {
4979 		atomic_inc(&bbio->error);
4980 		if (err == -EIO || err == -EREMOTEIO) {
4981 			unsigned int stripe_index =
4982 				btrfs_io_bio(bio)->stripe_index;
4983 			struct btrfs_device *dev;
4984 
4985 			BUG_ON(stripe_index >= bbio->num_stripes);
4986 			dev = bbio->stripes[stripe_index].dev;
4987 			if (dev->bdev) {
4988 				if (bio->bi_rw & WRITE)
4989 					btrfs_dev_stat_inc(dev,
4990 						BTRFS_DEV_STAT_WRITE_ERRS);
4991 				else
4992 					btrfs_dev_stat_inc(dev,
4993 						BTRFS_DEV_STAT_READ_ERRS);
4994 				if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4995 					btrfs_dev_stat_inc(dev,
4996 						BTRFS_DEV_STAT_FLUSH_ERRS);
4997 				btrfs_dev_stat_print_on_error(dev);
4998 			}
4999 		}
5000 	}
5001 
5002 	if (bio == bbio->orig_bio)
5003 		is_orig_bio = 1;
5004 
5005 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
5006 		if (!is_orig_bio) {
5007 			bio_put(bio);
5008 			bio = bbio->orig_bio;
5009 		}
5010 		bio->bi_private = bbio->private;
5011 		bio->bi_end_io = bbio->end_io;
5012 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5013 		/* only send an error to the higher layers if it is
5014 		 * beyond the tolerance of the btrfs bio
5015 		 */
5016 		if (atomic_read(&bbio->error) > bbio->max_errors) {
5017 			err = -EIO;
5018 		} else {
5019 			/*
5020 			 * this bio is actually up to date, we didn't
5021 			 * go over the max number of errors
5022 			 */
5023 			set_bit(BIO_UPTODATE, &bio->bi_flags);
5024 			err = 0;
5025 		}
5026 		kfree(bbio);
5027 
5028 		bio_endio(bio, err);
5029 	} else if (!is_orig_bio) {
5030 		bio_put(bio);
5031 	}
5032 }
5033 
5034 struct async_sched {
5035 	struct bio *bio;
5036 	int rw;
5037 	struct btrfs_fs_info *info;
5038 	struct btrfs_work work;
5039 };
5040 
5041 /*
5042  * see run_scheduled_bios for a description of why bios are collected for
5043  * async submit.
5044  *
5045  * This will add one bio to the pending list for a device and make sure
5046  * the work struct is scheduled.
5047  */
5048 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5049 					struct btrfs_device *device,
5050 					int rw, struct bio *bio)
5051 {
5052 	int should_queue = 1;
5053 	struct btrfs_pending_bios *pending_bios;
5054 
5055 	if (device->missing || !device->bdev) {
5056 		bio_endio(bio, -EIO);
5057 		return;
5058 	}
5059 
5060 	/* don't bother with additional async steps for reads, right now */
5061 	if (!(rw & REQ_WRITE)) {
5062 		bio_get(bio);
5063 		btrfsic_submit_bio(rw, bio);
5064 		bio_put(bio);
5065 		return;
5066 	}
5067 
5068 	/*
5069 	 * nr_async_bios allows us to reliably return congestion to the
5070 	 * higher layers.  Otherwise, the async bio makes it appear we have
5071 	 * made progress against dirty pages when we've really just put it
5072 	 * on a queue for later
5073 	 */
5074 	atomic_inc(&root->fs_info->nr_async_bios);
5075 	WARN_ON(bio->bi_next);
5076 	bio->bi_next = NULL;
5077 	bio->bi_rw |= rw;
5078 
5079 	spin_lock(&device->io_lock);
5080 	if (bio->bi_rw & REQ_SYNC)
5081 		pending_bios = &device->pending_sync_bios;
5082 	else
5083 		pending_bios = &device->pending_bios;
5084 
5085 	if (pending_bios->tail)
5086 		pending_bios->tail->bi_next = bio;
5087 
5088 	pending_bios->tail = bio;
5089 	if (!pending_bios->head)
5090 		pending_bios->head = bio;
5091 	if (device->running_pending)
5092 		should_queue = 0;
5093 
5094 	spin_unlock(&device->io_lock);
5095 
5096 	if (should_queue)
5097 		btrfs_queue_worker(&root->fs_info->submit_workers,
5098 				   &device->work);
5099 }
5100 
5101 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5102 		       sector_t sector)
5103 {
5104 	struct bio_vec *prev;
5105 	struct request_queue *q = bdev_get_queue(bdev);
5106 	unsigned short max_sectors = queue_max_sectors(q);
5107 	struct bvec_merge_data bvm = {
5108 		.bi_bdev = bdev,
5109 		.bi_sector = sector,
5110 		.bi_rw = bio->bi_rw,
5111 	};
5112 
5113 	if (bio->bi_vcnt == 0) {
5114 		WARN_ON(1);
5115 		return 1;
5116 	}
5117 
5118 	prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5119 	if (bio_sectors(bio) > max_sectors)
5120 		return 0;
5121 
5122 	if (!q->merge_bvec_fn)
5123 		return 1;
5124 
5125 	bvm.bi_size = bio->bi_size - prev->bv_len;
5126 	if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5127 		return 0;
5128 	return 1;
5129 }
5130 
5131 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5132 			      struct bio *bio, u64 physical, int dev_nr,
5133 			      int rw, int async)
5134 {
5135 	struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5136 
5137 	bio->bi_private = bbio;
5138 	btrfs_io_bio(bio)->stripe_index = dev_nr;
5139 	bio->bi_end_io = btrfs_end_bio;
5140 	bio->bi_sector = physical >> 9;
5141 #ifdef DEBUG
5142 	{
5143 		struct rcu_string *name;
5144 
5145 		rcu_read_lock();
5146 		name = rcu_dereference(dev->name);
5147 		pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5148 			 "(%s id %llu), size=%u\n", rw,
5149 			 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5150 			 name->str, dev->devid, bio->bi_size);
5151 		rcu_read_unlock();
5152 	}
5153 #endif
5154 	bio->bi_bdev = dev->bdev;
5155 	if (async)
5156 		btrfs_schedule_bio(root, dev, rw, bio);
5157 	else
5158 		btrfsic_submit_bio(rw, bio);
5159 }
5160 
5161 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5162 			      struct bio *first_bio, struct btrfs_device *dev,
5163 			      int dev_nr, int rw, int async)
5164 {
5165 	struct bio_vec *bvec = first_bio->bi_io_vec;
5166 	struct bio *bio;
5167 	int nr_vecs = bio_get_nr_vecs(dev->bdev);
5168 	u64 physical = bbio->stripes[dev_nr].physical;
5169 
5170 again:
5171 	bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5172 	if (!bio)
5173 		return -ENOMEM;
5174 
5175 	while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5176 		if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5177 				 bvec->bv_offset) < bvec->bv_len) {
5178 			u64 len = bio->bi_size;
5179 
5180 			atomic_inc(&bbio->stripes_pending);
5181 			submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5182 					  rw, async);
5183 			physical += len;
5184 			goto again;
5185 		}
5186 		bvec++;
5187 	}
5188 
5189 	submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5190 	return 0;
5191 }
5192 
5193 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5194 {
5195 	atomic_inc(&bbio->error);
5196 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
5197 		bio->bi_private = bbio->private;
5198 		bio->bi_end_io = bbio->end_io;
5199 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5200 		bio->bi_sector = logical >> 9;
5201 		kfree(bbio);
5202 		bio_endio(bio, -EIO);
5203 	}
5204 }
5205 
5206 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5207 		  int mirror_num, int async_submit)
5208 {
5209 	struct btrfs_device *dev;
5210 	struct bio *first_bio = bio;
5211 	u64 logical = (u64)bio->bi_sector << 9;
5212 	u64 length = 0;
5213 	u64 map_length;
5214 	u64 *raid_map = NULL;
5215 	int ret;
5216 	int dev_nr = 0;
5217 	int total_devs = 1;
5218 	struct btrfs_bio *bbio = NULL;
5219 
5220 	length = bio->bi_size;
5221 	map_length = length;
5222 
5223 	ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5224 			      mirror_num, &raid_map);
5225 	if (ret) /* -ENOMEM */
5226 		return ret;
5227 
5228 	total_devs = bbio->num_stripes;
5229 	bbio->orig_bio = first_bio;
5230 	bbio->private = first_bio->bi_private;
5231 	bbio->end_io = first_bio->bi_end_io;
5232 	atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5233 
5234 	if (raid_map) {
5235 		/* In this case, map_length has been set to the length of
5236 		   a single stripe; not the whole write */
5237 		if (rw & WRITE) {
5238 			return raid56_parity_write(root, bio, bbio,
5239 						   raid_map, map_length);
5240 		} else {
5241 			return raid56_parity_recover(root, bio, bbio,
5242 						     raid_map, map_length,
5243 						     mirror_num);
5244 		}
5245 	}
5246 
5247 	if (map_length < length) {
5248 		btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5249 			(unsigned long long)logical,
5250 			(unsigned long long)length,
5251 			(unsigned long long)map_length);
5252 		BUG();
5253 	}
5254 
5255 	while (dev_nr < total_devs) {
5256 		dev = bbio->stripes[dev_nr].dev;
5257 		if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5258 			bbio_error(bbio, first_bio, logical);
5259 			dev_nr++;
5260 			continue;
5261 		}
5262 
5263 		/*
5264 		 * Check and see if we're ok with this bio based on it's size
5265 		 * and offset with the given device.
5266 		 */
5267 		if (!bio_size_ok(dev->bdev, first_bio,
5268 				 bbio->stripes[dev_nr].physical >> 9)) {
5269 			ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5270 						 dev_nr, rw, async_submit);
5271 			BUG_ON(ret);
5272 			dev_nr++;
5273 			continue;
5274 		}
5275 
5276 		if (dev_nr < total_devs - 1) {
5277 			bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5278 			BUG_ON(!bio); /* -ENOMEM */
5279 		} else {
5280 			bio = first_bio;
5281 		}
5282 
5283 		submit_stripe_bio(root, bbio, bio,
5284 				  bbio->stripes[dev_nr].physical, dev_nr, rw,
5285 				  async_submit);
5286 		dev_nr++;
5287 	}
5288 	return 0;
5289 }
5290 
5291 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5292 				       u8 *uuid, u8 *fsid)
5293 {
5294 	struct btrfs_device *device;
5295 	struct btrfs_fs_devices *cur_devices;
5296 
5297 	cur_devices = fs_info->fs_devices;
5298 	while (cur_devices) {
5299 		if (!fsid ||
5300 		    !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5301 			device = __find_device(&cur_devices->devices,
5302 					       devid, uuid);
5303 			if (device)
5304 				return device;
5305 		}
5306 		cur_devices = cur_devices->seed;
5307 	}
5308 	return NULL;
5309 }
5310 
5311 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5312 					    u64 devid, u8 *dev_uuid)
5313 {
5314 	struct btrfs_device *device;
5315 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5316 
5317 	device = kzalloc(sizeof(*device), GFP_NOFS);
5318 	if (!device)
5319 		return NULL;
5320 	list_add(&device->dev_list,
5321 		 &fs_devices->devices);
5322 	device->devid = devid;
5323 	device->work.func = pending_bios_fn;
5324 	device->fs_devices = fs_devices;
5325 	device->missing = 1;
5326 	fs_devices->num_devices++;
5327 	fs_devices->missing_devices++;
5328 	spin_lock_init(&device->io_lock);
5329 	INIT_LIST_HEAD(&device->dev_alloc_list);
5330 	memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
5331 	return device;
5332 }
5333 
5334 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5335 			  struct extent_buffer *leaf,
5336 			  struct btrfs_chunk *chunk)
5337 {
5338 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5339 	struct map_lookup *map;
5340 	struct extent_map *em;
5341 	u64 logical;
5342 	u64 length;
5343 	u64 devid;
5344 	u8 uuid[BTRFS_UUID_SIZE];
5345 	int num_stripes;
5346 	int ret;
5347 	int i;
5348 
5349 	logical = key->offset;
5350 	length = btrfs_chunk_length(leaf, chunk);
5351 
5352 	read_lock(&map_tree->map_tree.lock);
5353 	em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5354 	read_unlock(&map_tree->map_tree.lock);
5355 
5356 	/* already mapped? */
5357 	if (em && em->start <= logical && em->start + em->len > logical) {
5358 		free_extent_map(em);
5359 		return 0;
5360 	} else if (em) {
5361 		free_extent_map(em);
5362 	}
5363 
5364 	em = alloc_extent_map();
5365 	if (!em)
5366 		return -ENOMEM;
5367 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5368 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5369 	if (!map) {
5370 		free_extent_map(em);
5371 		return -ENOMEM;
5372 	}
5373 
5374 	em->bdev = (struct block_device *)map;
5375 	em->start = logical;
5376 	em->len = length;
5377 	em->orig_start = 0;
5378 	em->block_start = 0;
5379 	em->block_len = em->len;
5380 
5381 	map->num_stripes = num_stripes;
5382 	map->io_width = btrfs_chunk_io_width(leaf, chunk);
5383 	map->io_align = btrfs_chunk_io_align(leaf, chunk);
5384 	map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5385 	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5386 	map->type = btrfs_chunk_type(leaf, chunk);
5387 	map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5388 	for (i = 0; i < num_stripes; i++) {
5389 		map->stripes[i].physical =
5390 			btrfs_stripe_offset_nr(leaf, chunk, i);
5391 		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5392 		read_extent_buffer(leaf, uuid, (unsigned long)
5393 				   btrfs_stripe_dev_uuid_nr(chunk, i),
5394 				   BTRFS_UUID_SIZE);
5395 		map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5396 							uuid, NULL);
5397 		if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5398 			kfree(map);
5399 			free_extent_map(em);
5400 			return -EIO;
5401 		}
5402 		if (!map->stripes[i].dev) {
5403 			map->stripes[i].dev =
5404 				add_missing_dev(root, devid, uuid);
5405 			if (!map->stripes[i].dev) {
5406 				kfree(map);
5407 				free_extent_map(em);
5408 				return -EIO;
5409 			}
5410 		}
5411 		map->stripes[i].dev->in_fs_metadata = 1;
5412 	}
5413 
5414 	write_lock(&map_tree->map_tree.lock);
5415 	ret = add_extent_mapping(&map_tree->map_tree, em, 0);
5416 	write_unlock(&map_tree->map_tree.lock);
5417 	BUG_ON(ret); /* Tree corruption */
5418 	free_extent_map(em);
5419 
5420 	return 0;
5421 }
5422 
5423 static void fill_device_from_item(struct extent_buffer *leaf,
5424 				 struct btrfs_dev_item *dev_item,
5425 				 struct btrfs_device *device)
5426 {
5427 	unsigned long ptr;
5428 
5429 	device->devid = btrfs_device_id(leaf, dev_item);
5430 	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5431 	device->total_bytes = device->disk_total_bytes;
5432 	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5433 	device->type = btrfs_device_type(leaf, dev_item);
5434 	device->io_align = btrfs_device_io_align(leaf, dev_item);
5435 	device->io_width = btrfs_device_io_width(leaf, dev_item);
5436 	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5437 	WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5438 	device->is_tgtdev_for_dev_replace = 0;
5439 
5440 	ptr = (unsigned long)btrfs_device_uuid(dev_item);
5441 	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5442 }
5443 
5444 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5445 {
5446 	struct btrfs_fs_devices *fs_devices;
5447 	int ret;
5448 
5449 	BUG_ON(!mutex_is_locked(&uuid_mutex));
5450 
5451 	fs_devices = root->fs_info->fs_devices->seed;
5452 	while (fs_devices) {
5453 		if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5454 			ret = 0;
5455 			goto out;
5456 		}
5457 		fs_devices = fs_devices->seed;
5458 	}
5459 
5460 	fs_devices = find_fsid(fsid);
5461 	if (!fs_devices) {
5462 		ret = -ENOENT;
5463 		goto out;
5464 	}
5465 
5466 	fs_devices = clone_fs_devices(fs_devices);
5467 	if (IS_ERR(fs_devices)) {
5468 		ret = PTR_ERR(fs_devices);
5469 		goto out;
5470 	}
5471 
5472 	ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5473 				   root->fs_info->bdev_holder);
5474 	if (ret) {
5475 		free_fs_devices(fs_devices);
5476 		goto out;
5477 	}
5478 
5479 	if (!fs_devices->seeding) {
5480 		__btrfs_close_devices(fs_devices);
5481 		free_fs_devices(fs_devices);
5482 		ret = -EINVAL;
5483 		goto out;
5484 	}
5485 
5486 	fs_devices->seed = root->fs_info->fs_devices->seed;
5487 	root->fs_info->fs_devices->seed = fs_devices;
5488 out:
5489 	return ret;
5490 }
5491 
5492 static int read_one_dev(struct btrfs_root *root,
5493 			struct extent_buffer *leaf,
5494 			struct btrfs_dev_item *dev_item)
5495 {
5496 	struct btrfs_device *device;
5497 	u64 devid;
5498 	int ret;
5499 	u8 fs_uuid[BTRFS_UUID_SIZE];
5500 	u8 dev_uuid[BTRFS_UUID_SIZE];
5501 
5502 	devid = btrfs_device_id(leaf, dev_item);
5503 	read_extent_buffer(leaf, dev_uuid,
5504 			   (unsigned long)btrfs_device_uuid(dev_item),
5505 			   BTRFS_UUID_SIZE);
5506 	read_extent_buffer(leaf, fs_uuid,
5507 			   (unsigned long)btrfs_device_fsid(dev_item),
5508 			   BTRFS_UUID_SIZE);
5509 
5510 	if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5511 		ret = open_seed_devices(root, fs_uuid);
5512 		if (ret && !btrfs_test_opt(root, DEGRADED))
5513 			return ret;
5514 	}
5515 
5516 	device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5517 	if (!device || !device->bdev) {
5518 		if (!btrfs_test_opt(root, DEGRADED))
5519 			return -EIO;
5520 
5521 		if (!device) {
5522 			btrfs_warn(root->fs_info, "devid %llu missing",
5523 				(unsigned long long)devid);
5524 			device = add_missing_dev(root, devid, dev_uuid);
5525 			if (!device)
5526 				return -ENOMEM;
5527 		} else if (!device->missing) {
5528 			/*
5529 			 * this happens when a device that was properly setup
5530 			 * in the device info lists suddenly goes bad.
5531 			 * device->bdev is NULL, and so we have to set
5532 			 * device->missing to one here
5533 			 */
5534 			root->fs_info->fs_devices->missing_devices++;
5535 			device->missing = 1;
5536 		}
5537 	}
5538 
5539 	if (device->fs_devices != root->fs_info->fs_devices) {
5540 		BUG_ON(device->writeable);
5541 		if (device->generation !=
5542 		    btrfs_device_generation(leaf, dev_item))
5543 			return -EINVAL;
5544 	}
5545 
5546 	fill_device_from_item(leaf, dev_item, device);
5547 	device->in_fs_metadata = 1;
5548 	if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5549 		device->fs_devices->total_rw_bytes += device->total_bytes;
5550 		spin_lock(&root->fs_info->free_chunk_lock);
5551 		root->fs_info->free_chunk_space += device->total_bytes -
5552 			device->bytes_used;
5553 		spin_unlock(&root->fs_info->free_chunk_lock);
5554 	}
5555 	ret = 0;
5556 	return ret;
5557 }
5558 
5559 int btrfs_read_sys_array(struct btrfs_root *root)
5560 {
5561 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5562 	struct extent_buffer *sb;
5563 	struct btrfs_disk_key *disk_key;
5564 	struct btrfs_chunk *chunk;
5565 	u8 *ptr;
5566 	unsigned long sb_ptr;
5567 	int ret = 0;
5568 	u32 num_stripes;
5569 	u32 array_size;
5570 	u32 len = 0;
5571 	u32 cur;
5572 	struct btrfs_key key;
5573 
5574 	sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5575 					  BTRFS_SUPER_INFO_SIZE);
5576 	if (!sb)
5577 		return -ENOMEM;
5578 	btrfs_set_buffer_uptodate(sb);
5579 	btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5580 	/*
5581 	 * The sb extent buffer is artifical and just used to read the system array.
5582 	 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5583 	 * pages up-to-date when the page is larger: extent does not cover the
5584 	 * whole page and consequently check_page_uptodate does not find all
5585 	 * the page's extents up-to-date (the hole beyond sb),
5586 	 * write_extent_buffer then triggers a WARN_ON.
5587 	 *
5588 	 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5589 	 * but sb spans only this function. Add an explicit SetPageUptodate call
5590 	 * to silence the warning eg. on PowerPC 64.
5591 	 */
5592 	if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5593 		SetPageUptodate(sb->pages[0]);
5594 
5595 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5596 	array_size = btrfs_super_sys_array_size(super_copy);
5597 
5598 	ptr = super_copy->sys_chunk_array;
5599 	sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5600 	cur = 0;
5601 
5602 	while (cur < array_size) {
5603 		disk_key = (struct btrfs_disk_key *)ptr;
5604 		btrfs_disk_key_to_cpu(&key, disk_key);
5605 
5606 		len = sizeof(*disk_key); ptr += len;
5607 		sb_ptr += len;
5608 		cur += len;
5609 
5610 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5611 			chunk = (struct btrfs_chunk *)sb_ptr;
5612 			ret = read_one_chunk(root, &key, sb, chunk);
5613 			if (ret)
5614 				break;
5615 			num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5616 			len = btrfs_chunk_item_size(num_stripes);
5617 		} else {
5618 			ret = -EIO;
5619 			break;
5620 		}
5621 		ptr += len;
5622 		sb_ptr += len;
5623 		cur += len;
5624 	}
5625 	free_extent_buffer(sb);
5626 	return ret;
5627 }
5628 
5629 int btrfs_read_chunk_tree(struct btrfs_root *root)
5630 {
5631 	struct btrfs_path *path;
5632 	struct extent_buffer *leaf;
5633 	struct btrfs_key key;
5634 	struct btrfs_key found_key;
5635 	int ret;
5636 	int slot;
5637 
5638 	root = root->fs_info->chunk_root;
5639 
5640 	path = btrfs_alloc_path();
5641 	if (!path)
5642 		return -ENOMEM;
5643 
5644 	mutex_lock(&uuid_mutex);
5645 	lock_chunks(root);
5646 
5647 	/* first we search for all of the device items, and then we
5648 	 * read in all of the chunk items.  This way we can create chunk
5649 	 * mappings that reference all of the devices that are afound
5650 	 */
5651 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5652 	key.offset = 0;
5653 	key.type = 0;
5654 again:
5655 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5656 	if (ret < 0)
5657 		goto error;
5658 	while (1) {
5659 		leaf = path->nodes[0];
5660 		slot = path->slots[0];
5661 		if (slot >= btrfs_header_nritems(leaf)) {
5662 			ret = btrfs_next_leaf(root, path);
5663 			if (ret == 0)
5664 				continue;
5665 			if (ret < 0)
5666 				goto error;
5667 			break;
5668 		}
5669 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
5670 		if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5671 			if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5672 				break;
5673 			if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5674 				struct btrfs_dev_item *dev_item;
5675 				dev_item = btrfs_item_ptr(leaf, slot,
5676 						  struct btrfs_dev_item);
5677 				ret = read_one_dev(root, leaf, dev_item);
5678 				if (ret)
5679 					goto error;
5680 			}
5681 		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5682 			struct btrfs_chunk *chunk;
5683 			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5684 			ret = read_one_chunk(root, &found_key, leaf, chunk);
5685 			if (ret)
5686 				goto error;
5687 		}
5688 		path->slots[0]++;
5689 	}
5690 	if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5691 		key.objectid = 0;
5692 		btrfs_release_path(path);
5693 		goto again;
5694 	}
5695 	ret = 0;
5696 error:
5697 	unlock_chunks(root);
5698 	mutex_unlock(&uuid_mutex);
5699 
5700 	btrfs_free_path(path);
5701 	return ret;
5702 }
5703 
5704 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
5705 {
5706 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5707 	struct btrfs_device *device;
5708 
5709 	mutex_lock(&fs_devices->device_list_mutex);
5710 	list_for_each_entry(device, &fs_devices->devices, dev_list)
5711 		device->dev_root = fs_info->dev_root;
5712 	mutex_unlock(&fs_devices->device_list_mutex);
5713 }
5714 
5715 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5716 {
5717 	int i;
5718 
5719 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5720 		btrfs_dev_stat_reset(dev, i);
5721 }
5722 
5723 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5724 {
5725 	struct btrfs_key key;
5726 	struct btrfs_key found_key;
5727 	struct btrfs_root *dev_root = fs_info->dev_root;
5728 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5729 	struct extent_buffer *eb;
5730 	int slot;
5731 	int ret = 0;
5732 	struct btrfs_device *device;
5733 	struct btrfs_path *path = NULL;
5734 	int i;
5735 
5736 	path = btrfs_alloc_path();
5737 	if (!path) {
5738 		ret = -ENOMEM;
5739 		goto out;
5740 	}
5741 
5742 	mutex_lock(&fs_devices->device_list_mutex);
5743 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
5744 		int item_size;
5745 		struct btrfs_dev_stats_item *ptr;
5746 
5747 		key.objectid = 0;
5748 		key.type = BTRFS_DEV_STATS_KEY;
5749 		key.offset = device->devid;
5750 		ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5751 		if (ret) {
5752 			__btrfs_reset_dev_stats(device);
5753 			device->dev_stats_valid = 1;
5754 			btrfs_release_path(path);
5755 			continue;
5756 		}
5757 		slot = path->slots[0];
5758 		eb = path->nodes[0];
5759 		btrfs_item_key_to_cpu(eb, &found_key, slot);
5760 		item_size = btrfs_item_size_nr(eb, slot);
5761 
5762 		ptr = btrfs_item_ptr(eb, slot,
5763 				     struct btrfs_dev_stats_item);
5764 
5765 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5766 			if (item_size >= (1 + i) * sizeof(__le64))
5767 				btrfs_dev_stat_set(device, i,
5768 					btrfs_dev_stats_value(eb, ptr, i));
5769 			else
5770 				btrfs_dev_stat_reset(device, i);
5771 		}
5772 
5773 		device->dev_stats_valid = 1;
5774 		btrfs_dev_stat_print_on_load(device);
5775 		btrfs_release_path(path);
5776 	}
5777 	mutex_unlock(&fs_devices->device_list_mutex);
5778 
5779 out:
5780 	btrfs_free_path(path);
5781 	return ret < 0 ? ret : 0;
5782 }
5783 
5784 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5785 				struct btrfs_root *dev_root,
5786 				struct btrfs_device *device)
5787 {
5788 	struct btrfs_path *path;
5789 	struct btrfs_key key;
5790 	struct extent_buffer *eb;
5791 	struct btrfs_dev_stats_item *ptr;
5792 	int ret;
5793 	int i;
5794 
5795 	key.objectid = 0;
5796 	key.type = BTRFS_DEV_STATS_KEY;
5797 	key.offset = device->devid;
5798 
5799 	path = btrfs_alloc_path();
5800 	BUG_ON(!path);
5801 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5802 	if (ret < 0) {
5803 		printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5804 			      ret, rcu_str_deref(device->name));
5805 		goto out;
5806 	}
5807 
5808 	if (ret == 0 &&
5809 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5810 		/* need to delete old one and insert a new one */
5811 		ret = btrfs_del_item(trans, dev_root, path);
5812 		if (ret != 0) {
5813 			printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5814 				      rcu_str_deref(device->name), ret);
5815 			goto out;
5816 		}
5817 		ret = 1;
5818 	}
5819 
5820 	if (ret == 1) {
5821 		/* need to insert a new item */
5822 		btrfs_release_path(path);
5823 		ret = btrfs_insert_empty_item(trans, dev_root, path,
5824 					      &key, sizeof(*ptr));
5825 		if (ret < 0) {
5826 			printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5827 				      rcu_str_deref(device->name), ret);
5828 			goto out;
5829 		}
5830 	}
5831 
5832 	eb = path->nodes[0];
5833 	ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5834 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5835 		btrfs_set_dev_stats_value(eb, ptr, i,
5836 					  btrfs_dev_stat_read(device, i));
5837 	btrfs_mark_buffer_dirty(eb);
5838 
5839 out:
5840 	btrfs_free_path(path);
5841 	return ret;
5842 }
5843 
5844 /*
5845  * called from commit_transaction. Writes all changed device stats to disk.
5846  */
5847 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5848 			struct btrfs_fs_info *fs_info)
5849 {
5850 	struct btrfs_root *dev_root = fs_info->dev_root;
5851 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5852 	struct btrfs_device *device;
5853 	int ret = 0;
5854 
5855 	mutex_lock(&fs_devices->device_list_mutex);
5856 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
5857 		if (!device->dev_stats_valid || !device->dev_stats_dirty)
5858 			continue;
5859 
5860 		ret = update_dev_stat_item(trans, dev_root, device);
5861 		if (!ret)
5862 			device->dev_stats_dirty = 0;
5863 	}
5864 	mutex_unlock(&fs_devices->device_list_mutex);
5865 
5866 	return ret;
5867 }
5868 
5869 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5870 {
5871 	btrfs_dev_stat_inc(dev, index);
5872 	btrfs_dev_stat_print_on_error(dev);
5873 }
5874 
5875 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5876 {
5877 	if (!dev->dev_stats_valid)
5878 		return;
5879 	printk_ratelimited_in_rcu(KERN_ERR
5880 			   "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5881 			   rcu_str_deref(dev->name),
5882 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5883 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5884 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5885 			   btrfs_dev_stat_read(dev,
5886 					       BTRFS_DEV_STAT_CORRUPTION_ERRS),
5887 			   btrfs_dev_stat_read(dev,
5888 					       BTRFS_DEV_STAT_GENERATION_ERRS));
5889 }
5890 
5891 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5892 {
5893 	int i;
5894 
5895 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5896 		if (btrfs_dev_stat_read(dev, i) != 0)
5897 			break;
5898 	if (i == BTRFS_DEV_STAT_VALUES_MAX)
5899 		return; /* all values == 0, suppress message */
5900 
5901 	printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5902 	       rcu_str_deref(dev->name),
5903 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5904 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5905 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5906 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5907 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5908 }
5909 
5910 int btrfs_get_dev_stats(struct btrfs_root *root,
5911 			struct btrfs_ioctl_get_dev_stats *stats)
5912 {
5913 	struct btrfs_device *dev;
5914 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5915 	int i;
5916 
5917 	mutex_lock(&fs_devices->device_list_mutex);
5918 	dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5919 	mutex_unlock(&fs_devices->device_list_mutex);
5920 
5921 	if (!dev) {
5922 		printk(KERN_WARNING
5923 		       "btrfs: get dev_stats failed, device not found\n");
5924 		return -ENODEV;
5925 	} else if (!dev->dev_stats_valid) {
5926 		printk(KERN_WARNING
5927 		       "btrfs: get dev_stats failed, not yet valid\n");
5928 		return -ENODEV;
5929 	} else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5930 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5931 			if (stats->nr_items > i)
5932 				stats->values[i] =
5933 					btrfs_dev_stat_read_and_reset(dev, i);
5934 			else
5935 				btrfs_dev_stat_reset(dev, i);
5936 		}
5937 	} else {
5938 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5939 			if (stats->nr_items > i)
5940 				stats->values[i] = btrfs_dev_stat_read(dev, i);
5941 	}
5942 	if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5943 		stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5944 	return 0;
5945 }
5946 
5947 int btrfs_scratch_superblock(struct btrfs_device *device)
5948 {
5949 	struct buffer_head *bh;
5950 	struct btrfs_super_block *disk_super;
5951 
5952 	bh = btrfs_read_dev_super(device->bdev);
5953 	if (!bh)
5954 		return -EINVAL;
5955 	disk_super = (struct btrfs_super_block *)bh->b_data;
5956 
5957 	memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5958 	set_buffer_dirty(bh);
5959 	sync_dirty_buffer(bh);
5960 	brelse(bh);
5961 
5962 	return 0;
5963 }
5964