xref: /openbmc/linux/fs/btrfs/volumes.c (revision 8fdff1dc)
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 "compat.h"
29 #include "ctree.h"
30 #include "extent_map.h"
31 #include "disk-io.h"
32 #include "transaction.h"
33 #include "print-tree.h"
34 #include "volumes.h"
35 #include "async-thread.h"
36 #include "check-integrity.h"
37 #include "rcu-string.h"
38 #include "math.h"
39 #include "dev-replace.h"
40 
41 static int init_first_rw_device(struct btrfs_trans_handle *trans,
42 				struct btrfs_root *root,
43 				struct btrfs_device *device);
44 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
45 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
46 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
47 
48 static DEFINE_MUTEX(uuid_mutex);
49 static LIST_HEAD(fs_uuids);
50 
51 static void lock_chunks(struct btrfs_root *root)
52 {
53 	mutex_lock(&root->fs_info->chunk_mutex);
54 }
55 
56 static void unlock_chunks(struct btrfs_root *root)
57 {
58 	mutex_unlock(&root->fs_info->chunk_mutex);
59 }
60 
61 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
62 {
63 	struct btrfs_device *device;
64 	WARN_ON(fs_devices->opened);
65 	while (!list_empty(&fs_devices->devices)) {
66 		device = list_entry(fs_devices->devices.next,
67 				    struct btrfs_device, dev_list);
68 		list_del(&device->dev_list);
69 		rcu_string_free(device->name);
70 		kfree(device);
71 	}
72 	kfree(fs_devices);
73 }
74 
75 static void btrfs_kobject_uevent(struct block_device *bdev,
76 				 enum kobject_action action)
77 {
78 	int ret;
79 
80 	ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
81 	if (ret)
82 		pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
83 			action,
84 			kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
85 			&disk_to_dev(bdev->bd_disk)->kobj);
86 }
87 
88 void btrfs_cleanup_fs_uuids(void)
89 {
90 	struct btrfs_fs_devices *fs_devices;
91 
92 	while (!list_empty(&fs_uuids)) {
93 		fs_devices = list_entry(fs_uuids.next,
94 					struct btrfs_fs_devices, list);
95 		list_del(&fs_devices->list);
96 		free_fs_devices(fs_devices);
97 	}
98 }
99 
100 static noinline struct btrfs_device *__find_device(struct list_head *head,
101 						   u64 devid, u8 *uuid)
102 {
103 	struct btrfs_device *dev;
104 
105 	list_for_each_entry(dev, head, dev_list) {
106 		if (dev->devid == devid &&
107 		    (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
108 			return dev;
109 		}
110 	}
111 	return NULL;
112 }
113 
114 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
115 {
116 	struct btrfs_fs_devices *fs_devices;
117 
118 	list_for_each_entry(fs_devices, &fs_uuids, list) {
119 		if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
120 			return fs_devices;
121 	}
122 	return NULL;
123 }
124 
125 static int
126 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
127 		      int flush, struct block_device **bdev,
128 		      struct buffer_head **bh)
129 {
130 	int ret;
131 
132 	*bdev = blkdev_get_by_path(device_path, flags, holder);
133 
134 	if (IS_ERR(*bdev)) {
135 		ret = PTR_ERR(*bdev);
136 		printk(KERN_INFO "btrfs: open %s failed\n", device_path);
137 		goto error;
138 	}
139 
140 	if (flush)
141 		filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
142 	ret = set_blocksize(*bdev, 4096);
143 	if (ret) {
144 		blkdev_put(*bdev, flags);
145 		goto error;
146 	}
147 	invalidate_bdev(*bdev);
148 	*bh = btrfs_read_dev_super(*bdev);
149 	if (!*bh) {
150 		ret = -EINVAL;
151 		blkdev_put(*bdev, flags);
152 		goto error;
153 	}
154 
155 	return 0;
156 
157 error:
158 	*bdev = NULL;
159 	*bh = NULL;
160 	return ret;
161 }
162 
163 static void requeue_list(struct btrfs_pending_bios *pending_bios,
164 			struct bio *head, struct bio *tail)
165 {
166 
167 	struct bio *old_head;
168 
169 	old_head = pending_bios->head;
170 	pending_bios->head = head;
171 	if (pending_bios->tail)
172 		tail->bi_next = old_head;
173 	else
174 		pending_bios->tail = tail;
175 }
176 
177 /*
178  * we try to collect pending bios for a device so we don't get a large
179  * number of procs sending bios down to the same device.  This greatly
180  * improves the schedulers ability to collect and merge the bios.
181  *
182  * But, it also turns into a long list of bios to process and that is sure
183  * to eventually make the worker thread block.  The solution here is to
184  * make some progress and then put this work struct back at the end of
185  * the list if the block device is congested.  This way, multiple devices
186  * can make progress from a single worker thread.
187  */
188 static noinline void run_scheduled_bios(struct btrfs_device *device)
189 {
190 	struct bio *pending;
191 	struct backing_dev_info *bdi;
192 	struct btrfs_fs_info *fs_info;
193 	struct btrfs_pending_bios *pending_bios;
194 	struct bio *tail;
195 	struct bio *cur;
196 	int again = 0;
197 	unsigned long num_run;
198 	unsigned long batch_run = 0;
199 	unsigned long limit;
200 	unsigned long last_waited = 0;
201 	int force_reg = 0;
202 	int sync_pending = 0;
203 	struct blk_plug plug;
204 
205 	/*
206 	 * this function runs all the bios we've collected for
207 	 * a particular device.  We don't want to wander off to
208 	 * another device without first sending all of these down.
209 	 * So, setup a plug here and finish it off before we return
210 	 */
211 	blk_start_plug(&plug);
212 
213 	bdi = blk_get_backing_dev_info(device->bdev);
214 	fs_info = device->dev_root->fs_info;
215 	limit = btrfs_async_submit_limit(fs_info);
216 	limit = limit * 2 / 3;
217 
218 loop:
219 	spin_lock(&device->io_lock);
220 
221 loop_lock:
222 	num_run = 0;
223 
224 	/* take all the bios off the list at once and process them
225 	 * later on (without the lock held).  But, remember the
226 	 * tail and other pointers so the bios can be properly reinserted
227 	 * into the list if we hit congestion
228 	 */
229 	if (!force_reg && device->pending_sync_bios.head) {
230 		pending_bios = &device->pending_sync_bios;
231 		force_reg = 1;
232 	} else {
233 		pending_bios = &device->pending_bios;
234 		force_reg = 0;
235 	}
236 
237 	pending = pending_bios->head;
238 	tail = pending_bios->tail;
239 	WARN_ON(pending && !tail);
240 
241 	/*
242 	 * if pending was null this time around, no bios need processing
243 	 * at all and we can stop.  Otherwise it'll loop back up again
244 	 * and do an additional check so no bios are missed.
245 	 *
246 	 * device->running_pending is used to synchronize with the
247 	 * schedule_bio code.
248 	 */
249 	if (device->pending_sync_bios.head == NULL &&
250 	    device->pending_bios.head == NULL) {
251 		again = 0;
252 		device->running_pending = 0;
253 	} else {
254 		again = 1;
255 		device->running_pending = 1;
256 	}
257 
258 	pending_bios->head = NULL;
259 	pending_bios->tail = NULL;
260 
261 	spin_unlock(&device->io_lock);
262 
263 	while (pending) {
264 
265 		rmb();
266 		/* we want to work on both lists, but do more bios on the
267 		 * sync list than the regular list
268 		 */
269 		if ((num_run > 32 &&
270 		    pending_bios != &device->pending_sync_bios &&
271 		    device->pending_sync_bios.head) ||
272 		   (num_run > 64 && pending_bios == &device->pending_sync_bios &&
273 		    device->pending_bios.head)) {
274 			spin_lock(&device->io_lock);
275 			requeue_list(pending_bios, pending, tail);
276 			goto loop_lock;
277 		}
278 
279 		cur = pending;
280 		pending = pending->bi_next;
281 		cur->bi_next = NULL;
282 
283 		if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
284 		    waitqueue_active(&fs_info->async_submit_wait))
285 			wake_up(&fs_info->async_submit_wait);
286 
287 		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
288 
289 		/*
290 		 * if we're doing the sync list, record that our
291 		 * plug has some sync requests on it
292 		 *
293 		 * If we're doing the regular list and there are
294 		 * sync requests sitting around, unplug before
295 		 * we add more
296 		 */
297 		if (pending_bios == &device->pending_sync_bios) {
298 			sync_pending = 1;
299 		} else if (sync_pending) {
300 			blk_finish_plug(&plug);
301 			blk_start_plug(&plug);
302 			sync_pending = 0;
303 		}
304 
305 		btrfsic_submit_bio(cur->bi_rw, cur);
306 		num_run++;
307 		batch_run++;
308 		if (need_resched())
309 			cond_resched();
310 
311 		/*
312 		 * we made progress, there is more work to do and the bdi
313 		 * is now congested.  Back off and let other work structs
314 		 * run instead
315 		 */
316 		if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
317 		    fs_info->fs_devices->open_devices > 1) {
318 			struct io_context *ioc;
319 
320 			ioc = current->io_context;
321 
322 			/*
323 			 * the main goal here is that we don't want to
324 			 * block if we're going to be able to submit
325 			 * more requests without blocking.
326 			 *
327 			 * This code does two great things, it pokes into
328 			 * the elevator code from a filesystem _and_
329 			 * it makes assumptions about how batching works.
330 			 */
331 			if (ioc && ioc->nr_batch_requests > 0 &&
332 			    time_before(jiffies, ioc->last_waited + HZ/50UL) &&
333 			    (last_waited == 0 ||
334 			     ioc->last_waited == last_waited)) {
335 				/*
336 				 * we want to go through our batch of
337 				 * requests and stop.  So, we copy out
338 				 * the ioc->last_waited time and test
339 				 * against it before looping
340 				 */
341 				last_waited = ioc->last_waited;
342 				if (need_resched())
343 					cond_resched();
344 				continue;
345 			}
346 			spin_lock(&device->io_lock);
347 			requeue_list(pending_bios, pending, tail);
348 			device->running_pending = 1;
349 
350 			spin_unlock(&device->io_lock);
351 			btrfs_requeue_work(&device->work);
352 			goto done;
353 		}
354 		/* unplug every 64 requests just for good measure */
355 		if (batch_run % 64 == 0) {
356 			blk_finish_plug(&plug);
357 			blk_start_plug(&plug);
358 			sync_pending = 0;
359 		}
360 	}
361 
362 	cond_resched();
363 	if (again)
364 		goto loop;
365 
366 	spin_lock(&device->io_lock);
367 	if (device->pending_bios.head || device->pending_sync_bios.head)
368 		goto loop_lock;
369 	spin_unlock(&device->io_lock);
370 
371 done:
372 	blk_finish_plug(&plug);
373 }
374 
375 static void pending_bios_fn(struct btrfs_work *work)
376 {
377 	struct btrfs_device *device;
378 
379 	device = container_of(work, struct btrfs_device, work);
380 	run_scheduled_bios(device);
381 }
382 
383 static noinline int device_list_add(const char *path,
384 			   struct btrfs_super_block *disk_super,
385 			   u64 devid, struct btrfs_fs_devices **fs_devices_ret)
386 {
387 	struct btrfs_device *device;
388 	struct btrfs_fs_devices *fs_devices;
389 	struct rcu_string *name;
390 	u64 found_transid = btrfs_super_generation(disk_super);
391 
392 	fs_devices = find_fsid(disk_super->fsid);
393 	if (!fs_devices) {
394 		fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
395 		if (!fs_devices)
396 			return -ENOMEM;
397 		INIT_LIST_HEAD(&fs_devices->devices);
398 		INIT_LIST_HEAD(&fs_devices->alloc_list);
399 		list_add(&fs_devices->list, &fs_uuids);
400 		memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
401 		fs_devices->latest_devid = devid;
402 		fs_devices->latest_trans = found_transid;
403 		mutex_init(&fs_devices->device_list_mutex);
404 		device = NULL;
405 	} else {
406 		device = __find_device(&fs_devices->devices, devid,
407 				       disk_super->dev_item.uuid);
408 	}
409 	if (!device) {
410 		if (fs_devices->opened)
411 			return -EBUSY;
412 
413 		device = kzalloc(sizeof(*device), GFP_NOFS);
414 		if (!device) {
415 			/* we can safely leave the fs_devices entry around */
416 			return -ENOMEM;
417 		}
418 		device->devid = devid;
419 		device->dev_stats_valid = 0;
420 		device->work.func = pending_bios_fn;
421 		memcpy(device->uuid, disk_super->dev_item.uuid,
422 		       BTRFS_UUID_SIZE);
423 		spin_lock_init(&device->io_lock);
424 
425 		name = rcu_string_strdup(path, GFP_NOFS);
426 		if (!name) {
427 			kfree(device);
428 			return -ENOMEM;
429 		}
430 		rcu_assign_pointer(device->name, name);
431 		INIT_LIST_HEAD(&device->dev_alloc_list);
432 
433 		/* init readahead state */
434 		spin_lock_init(&device->reada_lock);
435 		device->reada_curr_zone = NULL;
436 		atomic_set(&device->reada_in_flight, 0);
437 		device->reada_next = 0;
438 		INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
439 		INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
440 
441 		mutex_lock(&fs_devices->device_list_mutex);
442 		list_add_rcu(&device->dev_list, &fs_devices->devices);
443 		mutex_unlock(&fs_devices->device_list_mutex);
444 
445 		device->fs_devices = fs_devices;
446 		fs_devices->num_devices++;
447 	} else if (!device->name || strcmp(device->name->str, path)) {
448 		name = rcu_string_strdup(path, GFP_NOFS);
449 		if (!name)
450 			return -ENOMEM;
451 		rcu_string_free(device->name);
452 		rcu_assign_pointer(device->name, name);
453 		if (device->missing) {
454 			fs_devices->missing_devices--;
455 			device->missing = 0;
456 		}
457 	}
458 
459 	if (found_transid > fs_devices->latest_trans) {
460 		fs_devices->latest_devid = devid;
461 		fs_devices->latest_trans = found_transid;
462 	}
463 	*fs_devices_ret = fs_devices;
464 	return 0;
465 }
466 
467 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
468 {
469 	struct btrfs_fs_devices *fs_devices;
470 	struct btrfs_device *device;
471 	struct btrfs_device *orig_dev;
472 
473 	fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
474 	if (!fs_devices)
475 		return ERR_PTR(-ENOMEM);
476 
477 	INIT_LIST_HEAD(&fs_devices->devices);
478 	INIT_LIST_HEAD(&fs_devices->alloc_list);
479 	INIT_LIST_HEAD(&fs_devices->list);
480 	mutex_init(&fs_devices->device_list_mutex);
481 	fs_devices->latest_devid = orig->latest_devid;
482 	fs_devices->latest_trans = orig->latest_trans;
483 	fs_devices->total_devices = orig->total_devices;
484 	memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
485 
486 	/* We have held the volume lock, it is safe to get the devices. */
487 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
488 		struct rcu_string *name;
489 
490 		device = kzalloc(sizeof(*device), GFP_NOFS);
491 		if (!device)
492 			goto error;
493 
494 		/*
495 		 * This is ok to do without rcu read locked because we hold the
496 		 * uuid mutex so nothing we touch in here is going to disappear.
497 		 */
498 		name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
499 		if (!name) {
500 			kfree(device);
501 			goto error;
502 		}
503 		rcu_assign_pointer(device->name, name);
504 
505 		device->devid = orig_dev->devid;
506 		device->work.func = pending_bios_fn;
507 		memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
508 		spin_lock_init(&device->io_lock);
509 		INIT_LIST_HEAD(&device->dev_list);
510 		INIT_LIST_HEAD(&device->dev_alloc_list);
511 
512 		list_add(&device->dev_list, &fs_devices->devices);
513 		device->fs_devices = fs_devices;
514 		fs_devices->num_devices++;
515 	}
516 	return fs_devices;
517 error:
518 	free_fs_devices(fs_devices);
519 	return ERR_PTR(-ENOMEM);
520 }
521 
522 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
523 			       struct btrfs_fs_devices *fs_devices, int step)
524 {
525 	struct btrfs_device *device, *next;
526 
527 	struct block_device *latest_bdev = NULL;
528 	u64 latest_devid = 0;
529 	u64 latest_transid = 0;
530 
531 	mutex_lock(&uuid_mutex);
532 again:
533 	/* This is the initialized path, it is safe to release the devices. */
534 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
535 		if (device->in_fs_metadata) {
536 			if (!device->is_tgtdev_for_dev_replace &&
537 			    (!latest_transid ||
538 			     device->generation > latest_transid)) {
539 				latest_devid = device->devid;
540 				latest_transid = device->generation;
541 				latest_bdev = device->bdev;
542 			}
543 			continue;
544 		}
545 
546 		if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
547 			/*
548 			 * In the first step, keep the device which has
549 			 * the correct fsid and the devid that is used
550 			 * for the dev_replace procedure.
551 			 * In the second step, the dev_replace state is
552 			 * read from the device tree and it is known
553 			 * whether the procedure is really active or
554 			 * not, which means whether this device is
555 			 * used or whether it should be removed.
556 			 */
557 			if (step == 0 || device->is_tgtdev_for_dev_replace) {
558 				continue;
559 			}
560 		}
561 		if (device->bdev) {
562 			blkdev_put(device->bdev, device->mode);
563 			device->bdev = NULL;
564 			fs_devices->open_devices--;
565 		}
566 		if (device->writeable) {
567 			list_del_init(&device->dev_alloc_list);
568 			device->writeable = 0;
569 			if (!device->is_tgtdev_for_dev_replace)
570 				fs_devices->rw_devices--;
571 		}
572 		list_del_init(&device->dev_list);
573 		fs_devices->num_devices--;
574 		rcu_string_free(device->name);
575 		kfree(device);
576 	}
577 
578 	if (fs_devices->seed) {
579 		fs_devices = fs_devices->seed;
580 		goto again;
581 	}
582 
583 	fs_devices->latest_bdev = latest_bdev;
584 	fs_devices->latest_devid = latest_devid;
585 	fs_devices->latest_trans = latest_transid;
586 
587 	mutex_unlock(&uuid_mutex);
588 }
589 
590 static void __free_device(struct work_struct *work)
591 {
592 	struct btrfs_device *device;
593 
594 	device = container_of(work, struct btrfs_device, rcu_work);
595 
596 	if (device->bdev)
597 		blkdev_put(device->bdev, device->mode);
598 
599 	rcu_string_free(device->name);
600 	kfree(device);
601 }
602 
603 static void free_device(struct rcu_head *head)
604 {
605 	struct btrfs_device *device;
606 
607 	device = container_of(head, struct btrfs_device, rcu);
608 
609 	INIT_WORK(&device->rcu_work, __free_device);
610 	schedule_work(&device->rcu_work);
611 }
612 
613 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
614 {
615 	struct btrfs_device *device;
616 
617 	if (--fs_devices->opened > 0)
618 		return 0;
619 
620 	mutex_lock(&fs_devices->device_list_mutex);
621 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
622 		struct btrfs_device *new_device;
623 		struct rcu_string *name;
624 
625 		if (device->bdev)
626 			fs_devices->open_devices--;
627 
628 		if (device->writeable && !device->is_tgtdev_for_dev_replace) {
629 			list_del_init(&device->dev_alloc_list);
630 			fs_devices->rw_devices--;
631 		}
632 
633 		if (device->can_discard)
634 			fs_devices->num_can_discard--;
635 
636 		new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
637 		BUG_ON(!new_device); /* -ENOMEM */
638 		memcpy(new_device, device, sizeof(*new_device));
639 
640 		/* Safe because we are under uuid_mutex */
641 		if (device->name) {
642 			name = rcu_string_strdup(device->name->str, GFP_NOFS);
643 			BUG_ON(device->name && !name); /* -ENOMEM */
644 			rcu_assign_pointer(new_device->name, name);
645 		}
646 		new_device->bdev = NULL;
647 		new_device->writeable = 0;
648 		new_device->in_fs_metadata = 0;
649 		new_device->can_discard = 0;
650 		list_replace_rcu(&device->dev_list, &new_device->dev_list);
651 
652 		call_rcu(&device->rcu, free_device);
653 	}
654 	mutex_unlock(&fs_devices->device_list_mutex);
655 
656 	WARN_ON(fs_devices->open_devices);
657 	WARN_ON(fs_devices->rw_devices);
658 	fs_devices->opened = 0;
659 	fs_devices->seeding = 0;
660 
661 	return 0;
662 }
663 
664 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
665 {
666 	struct btrfs_fs_devices *seed_devices = NULL;
667 	int ret;
668 
669 	mutex_lock(&uuid_mutex);
670 	ret = __btrfs_close_devices(fs_devices);
671 	if (!fs_devices->opened) {
672 		seed_devices = fs_devices->seed;
673 		fs_devices->seed = NULL;
674 	}
675 	mutex_unlock(&uuid_mutex);
676 
677 	while (seed_devices) {
678 		fs_devices = seed_devices;
679 		seed_devices = fs_devices->seed;
680 		__btrfs_close_devices(fs_devices);
681 		free_fs_devices(fs_devices);
682 	}
683 	return ret;
684 }
685 
686 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
687 				fmode_t flags, void *holder)
688 {
689 	struct request_queue *q;
690 	struct block_device *bdev;
691 	struct list_head *head = &fs_devices->devices;
692 	struct btrfs_device *device;
693 	struct block_device *latest_bdev = NULL;
694 	struct buffer_head *bh;
695 	struct btrfs_super_block *disk_super;
696 	u64 latest_devid = 0;
697 	u64 latest_transid = 0;
698 	u64 devid;
699 	int seeding = 1;
700 	int ret = 0;
701 
702 	flags |= FMODE_EXCL;
703 
704 	list_for_each_entry(device, head, dev_list) {
705 		if (device->bdev)
706 			continue;
707 		if (!device->name)
708 			continue;
709 
710 		ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
711 					    &bdev, &bh);
712 		if (ret)
713 			continue;
714 
715 		disk_super = (struct btrfs_super_block *)bh->b_data;
716 		devid = btrfs_stack_device_id(&disk_super->dev_item);
717 		if (devid != device->devid)
718 			goto error_brelse;
719 
720 		if (memcmp(device->uuid, disk_super->dev_item.uuid,
721 			   BTRFS_UUID_SIZE))
722 			goto error_brelse;
723 
724 		device->generation = btrfs_super_generation(disk_super);
725 		if (!latest_transid || device->generation > latest_transid) {
726 			latest_devid = devid;
727 			latest_transid = device->generation;
728 			latest_bdev = bdev;
729 		}
730 
731 		if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
732 			device->writeable = 0;
733 		} else {
734 			device->writeable = !bdev_read_only(bdev);
735 			seeding = 0;
736 		}
737 
738 		q = bdev_get_queue(bdev);
739 		if (blk_queue_discard(q)) {
740 			device->can_discard = 1;
741 			fs_devices->num_can_discard++;
742 		}
743 
744 		device->bdev = bdev;
745 		device->in_fs_metadata = 0;
746 		device->mode = flags;
747 
748 		if (!blk_queue_nonrot(bdev_get_queue(bdev)))
749 			fs_devices->rotating = 1;
750 
751 		fs_devices->open_devices++;
752 		if (device->writeable && !device->is_tgtdev_for_dev_replace) {
753 			fs_devices->rw_devices++;
754 			list_add(&device->dev_alloc_list,
755 				 &fs_devices->alloc_list);
756 		}
757 		brelse(bh);
758 		continue;
759 
760 error_brelse:
761 		brelse(bh);
762 		blkdev_put(bdev, flags);
763 		continue;
764 	}
765 	if (fs_devices->open_devices == 0) {
766 		ret = -EINVAL;
767 		goto out;
768 	}
769 	fs_devices->seeding = seeding;
770 	fs_devices->opened = 1;
771 	fs_devices->latest_bdev = latest_bdev;
772 	fs_devices->latest_devid = latest_devid;
773 	fs_devices->latest_trans = latest_transid;
774 	fs_devices->total_rw_bytes = 0;
775 out:
776 	return ret;
777 }
778 
779 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
780 		       fmode_t flags, void *holder)
781 {
782 	int ret;
783 
784 	mutex_lock(&uuid_mutex);
785 	if (fs_devices->opened) {
786 		fs_devices->opened++;
787 		ret = 0;
788 	} else {
789 		ret = __btrfs_open_devices(fs_devices, flags, holder);
790 	}
791 	mutex_unlock(&uuid_mutex);
792 	return ret;
793 }
794 
795 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
796 			  struct btrfs_fs_devices **fs_devices_ret)
797 {
798 	struct btrfs_super_block *disk_super;
799 	struct block_device *bdev;
800 	struct buffer_head *bh;
801 	int ret;
802 	u64 devid;
803 	u64 transid;
804 	u64 total_devices;
805 
806 	flags |= FMODE_EXCL;
807 	mutex_lock(&uuid_mutex);
808 	ret = btrfs_get_bdev_and_sb(path, flags, holder, 0, &bdev, &bh);
809 	if (ret)
810 		goto error;
811 	disk_super = (struct btrfs_super_block *)bh->b_data;
812 	devid = btrfs_stack_device_id(&disk_super->dev_item);
813 	transid = btrfs_super_generation(disk_super);
814 	total_devices = btrfs_super_num_devices(disk_super);
815 	if (disk_super->label[0]) {
816 		if (disk_super->label[BTRFS_LABEL_SIZE - 1])
817 			disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
818 		printk(KERN_INFO "device label %s ", disk_super->label);
819 	} else {
820 		printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
821 	}
822 	printk(KERN_CONT "devid %llu transid %llu %s\n",
823 	       (unsigned long long)devid, (unsigned long long)transid, path);
824 	ret = device_list_add(path, disk_super, devid, fs_devices_ret);
825 	if (!ret && fs_devices_ret)
826 		(*fs_devices_ret)->total_devices = total_devices;
827 	brelse(bh);
828 	blkdev_put(bdev, flags);
829 error:
830 	mutex_unlock(&uuid_mutex);
831 	return ret;
832 }
833 
834 /* helper to account the used device space in the range */
835 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
836 				   u64 end, u64 *length)
837 {
838 	struct btrfs_key key;
839 	struct btrfs_root *root = device->dev_root;
840 	struct btrfs_dev_extent *dev_extent;
841 	struct btrfs_path *path;
842 	u64 extent_end;
843 	int ret;
844 	int slot;
845 	struct extent_buffer *l;
846 
847 	*length = 0;
848 
849 	if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
850 		return 0;
851 
852 	path = btrfs_alloc_path();
853 	if (!path)
854 		return -ENOMEM;
855 	path->reada = 2;
856 
857 	key.objectid = device->devid;
858 	key.offset = start;
859 	key.type = BTRFS_DEV_EXTENT_KEY;
860 
861 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
862 	if (ret < 0)
863 		goto out;
864 	if (ret > 0) {
865 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
866 		if (ret < 0)
867 			goto out;
868 	}
869 
870 	while (1) {
871 		l = path->nodes[0];
872 		slot = path->slots[0];
873 		if (slot >= btrfs_header_nritems(l)) {
874 			ret = btrfs_next_leaf(root, path);
875 			if (ret == 0)
876 				continue;
877 			if (ret < 0)
878 				goto out;
879 
880 			break;
881 		}
882 		btrfs_item_key_to_cpu(l, &key, slot);
883 
884 		if (key.objectid < device->devid)
885 			goto next;
886 
887 		if (key.objectid > device->devid)
888 			break;
889 
890 		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
891 			goto next;
892 
893 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
894 		extent_end = key.offset + btrfs_dev_extent_length(l,
895 								  dev_extent);
896 		if (key.offset <= start && extent_end > end) {
897 			*length = end - start + 1;
898 			break;
899 		} else if (key.offset <= start && extent_end > start)
900 			*length += extent_end - start;
901 		else if (key.offset > start && extent_end <= end)
902 			*length += extent_end - key.offset;
903 		else if (key.offset > start && key.offset <= end) {
904 			*length += end - key.offset + 1;
905 			break;
906 		} else if (key.offset > end)
907 			break;
908 
909 next:
910 		path->slots[0]++;
911 	}
912 	ret = 0;
913 out:
914 	btrfs_free_path(path);
915 	return ret;
916 }
917 
918 /*
919  * find_free_dev_extent - find free space in the specified device
920  * @device:	the device which we search the free space in
921  * @num_bytes:	the size of the free space that we need
922  * @start:	store the start of the free space.
923  * @len:	the size of the free space. that we find, or the size of the max
924  * 		free space if we don't find suitable free space
925  *
926  * this uses a pretty simple search, the expectation is that it is
927  * called very infrequently and that a given device has a small number
928  * of extents
929  *
930  * @start is used to store the start of the free space if we find. But if we
931  * don't find suitable free space, it will be used to store the start position
932  * of the max free space.
933  *
934  * @len is used to store the size of the free space that we find.
935  * But if we don't find suitable free space, it is used to store the size of
936  * the max free space.
937  */
938 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
939 			 u64 *start, u64 *len)
940 {
941 	struct btrfs_key key;
942 	struct btrfs_root *root = device->dev_root;
943 	struct btrfs_dev_extent *dev_extent;
944 	struct btrfs_path *path;
945 	u64 hole_size;
946 	u64 max_hole_start;
947 	u64 max_hole_size;
948 	u64 extent_end;
949 	u64 search_start;
950 	u64 search_end = device->total_bytes;
951 	int ret;
952 	int slot;
953 	struct extent_buffer *l;
954 
955 	/* FIXME use last free of some kind */
956 
957 	/* we don't want to overwrite the superblock on the drive,
958 	 * so we make sure to start at an offset of at least 1MB
959 	 */
960 	search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
961 
962 	max_hole_start = search_start;
963 	max_hole_size = 0;
964 	hole_size = 0;
965 
966 	if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
967 		ret = -ENOSPC;
968 		goto error;
969 	}
970 
971 	path = btrfs_alloc_path();
972 	if (!path) {
973 		ret = -ENOMEM;
974 		goto error;
975 	}
976 	path->reada = 2;
977 
978 	key.objectid = device->devid;
979 	key.offset = search_start;
980 	key.type = BTRFS_DEV_EXTENT_KEY;
981 
982 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
983 	if (ret < 0)
984 		goto out;
985 	if (ret > 0) {
986 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
987 		if (ret < 0)
988 			goto out;
989 	}
990 
991 	while (1) {
992 		l = path->nodes[0];
993 		slot = path->slots[0];
994 		if (slot >= btrfs_header_nritems(l)) {
995 			ret = btrfs_next_leaf(root, path);
996 			if (ret == 0)
997 				continue;
998 			if (ret < 0)
999 				goto out;
1000 
1001 			break;
1002 		}
1003 		btrfs_item_key_to_cpu(l, &key, slot);
1004 
1005 		if (key.objectid < device->devid)
1006 			goto next;
1007 
1008 		if (key.objectid > device->devid)
1009 			break;
1010 
1011 		if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1012 			goto next;
1013 
1014 		if (key.offset > search_start) {
1015 			hole_size = key.offset - search_start;
1016 
1017 			if (hole_size > max_hole_size) {
1018 				max_hole_start = search_start;
1019 				max_hole_size = hole_size;
1020 			}
1021 
1022 			/*
1023 			 * If this free space is greater than which we need,
1024 			 * it must be the max free space that we have found
1025 			 * until now, so max_hole_start must point to the start
1026 			 * of this free space and the length of this free space
1027 			 * is stored in max_hole_size. Thus, we return
1028 			 * max_hole_start and max_hole_size and go back to the
1029 			 * caller.
1030 			 */
1031 			if (hole_size >= num_bytes) {
1032 				ret = 0;
1033 				goto out;
1034 			}
1035 		}
1036 
1037 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1038 		extent_end = key.offset + btrfs_dev_extent_length(l,
1039 								  dev_extent);
1040 		if (extent_end > search_start)
1041 			search_start = extent_end;
1042 next:
1043 		path->slots[0]++;
1044 		cond_resched();
1045 	}
1046 
1047 	/*
1048 	 * At this point, search_start should be the end of
1049 	 * allocated dev extents, and when shrinking the device,
1050 	 * search_end may be smaller than search_start.
1051 	 */
1052 	if (search_end > search_start)
1053 		hole_size = search_end - search_start;
1054 
1055 	if (hole_size > max_hole_size) {
1056 		max_hole_start = search_start;
1057 		max_hole_size = hole_size;
1058 	}
1059 
1060 	/* See above. */
1061 	if (hole_size < num_bytes)
1062 		ret = -ENOSPC;
1063 	else
1064 		ret = 0;
1065 
1066 out:
1067 	btrfs_free_path(path);
1068 error:
1069 	*start = max_hole_start;
1070 	if (len)
1071 		*len = max_hole_size;
1072 	return ret;
1073 }
1074 
1075 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1076 			  struct btrfs_device *device,
1077 			  u64 start)
1078 {
1079 	int ret;
1080 	struct btrfs_path *path;
1081 	struct btrfs_root *root = device->dev_root;
1082 	struct btrfs_key key;
1083 	struct btrfs_key found_key;
1084 	struct extent_buffer *leaf = NULL;
1085 	struct btrfs_dev_extent *extent = NULL;
1086 
1087 	path = btrfs_alloc_path();
1088 	if (!path)
1089 		return -ENOMEM;
1090 
1091 	key.objectid = device->devid;
1092 	key.offset = start;
1093 	key.type = BTRFS_DEV_EXTENT_KEY;
1094 again:
1095 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1096 	if (ret > 0) {
1097 		ret = btrfs_previous_item(root, path, key.objectid,
1098 					  BTRFS_DEV_EXTENT_KEY);
1099 		if (ret)
1100 			goto out;
1101 		leaf = path->nodes[0];
1102 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1103 		extent = btrfs_item_ptr(leaf, path->slots[0],
1104 					struct btrfs_dev_extent);
1105 		BUG_ON(found_key.offset > start || found_key.offset +
1106 		       btrfs_dev_extent_length(leaf, extent) < start);
1107 		key = found_key;
1108 		btrfs_release_path(path);
1109 		goto again;
1110 	} else if (ret == 0) {
1111 		leaf = path->nodes[0];
1112 		extent = btrfs_item_ptr(leaf, path->slots[0],
1113 					struct btrfs_dev_extent);
1114 	} else {
1115 		btrfs_error(root->fs_info, ret, "Slot search failed");
1116 		goto out;
1117 	}
1118 
1119 	if (device->bytes_used > 0) {
1120 		u64 len = btrfs_dev_extent_length(leaf, extent);
1121 		device->bytes_used -= len;
1122 		spin_lock(&root->fs_info->free_chunk_lock);
1123 		root->fs_info->free_chunk_space += len;
1124 		spin_unlock(&root->fs_info->free_chunk_lock);
1125 	}
1126 	ret = btrfs_del_item(trans, root, path);
1127 	if (ret) {
1128 		btrfs_error(root->fs_info, ret,
1129 			    "Failed to remove dev extent item");
1130 	}
1131 out:
1132 	btrfs_free_path(path);
1133 	return ret;
1134 }
1135 
1136 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1137 			   struct btrfs_device *device,
1138 			   u64 chunk_tree, u64 chunk_objectid,
1139 			   u64 chunk_offset, u64 start, u64 num_bytes)
1140 {
1141 	int ret;
1142 	struct btrfs_path *path;
1143 	struct btrfs_root *root = device->dev_root;
1144 	struct btrfs_dev_extent *extent;
1145 	struct extent_buffer *leaf;
1146 	struct btrfs_key key;
1147 
1148 	WARN_ON(!device->in_fs_metadata);
1149 	WARN_ON(device->is_tgtdev_for_dev_replace);
1150 	path = btrfs_alloc_path();
1151 	if (!path)
1152 		return -ENOMEM;
1153 
1154 	key.objectid = device->devid;
1155 	key.offset = start;
1156 	key.type = BTRFS_DEV_EXTENT_KEY;
1157 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1158 				      sizeof(*extent));
1159 	if (ret)
1160 		goto out;
1161 
1162 	leaf = path->nodes[0];
1163 	extent = btrfs_item_ptr(leaf, path->slots[0],
1164 				struct btrfs_dev_extent);
1165 	btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1166 	btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1167 	btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1168 
1169 	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1170 		    (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1171 		    BTRFS_UUID_SIZE);
1172 
1173 	btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1174 	btrfs_mark_buffer_dirty(leaf);
1175 out:
1176 	btrfs_free_path(path);
1177 	return ret;
1178 }
1179 
1180 static noinline int find_next_chunk(struct btrfs_root *root,
1181 				    u64 objectid, u64 *offset)
1182 {
1183 	struct btrfs_path *path;
1184 	int ret;
1185 	struct btrfs_key key;
1186 	struct btrfs_chunk *chunk;
1187 	struct btrfs_key found_key;
1188 
1189 	path = btrfs_alloc_path();
1190 	if (!path)
1191 		return -ENOMEM;
1192 
1193 	key.objectid = objectid;
1194 	key.offset = (u64)-1;
1195 	key.type = BTRFS_CHUNK_ITEM_KEY;
1196 
1197 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1198 	if (ret < 0)
1199 		goto error;
1200 
1201 	BUG_ON(ret == 0); /* Corruption */
1202 
1203 	ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1204 	if (ret) {
1205 		*offset = 0;
1206 	} else {
1207 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1208 				      path->slots[0]);
1209 		if (found_key.objectid != objectid)
1210 			*offset = 0;
1211 		else {
1212 			chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1213 					       struct btrfs_chunk);
1214 			*offset = found_key.offset +
1215 				btrfs_chunk_length(path->nodes[0], chunk);
1216 		}
1217 	}
1218 	ret = 0;
1219 error:
1220 	btrfs_free_path(path);
1221 	return ret;
1222 }
1223 
1224 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1225 {
1226 	int ret;
1227 	struct btrfs_key key;
1228 	struct btrfs_key found_key;
1229 	struct btrfs_path *path;
1230 
1231 	root = root->fs_info->chunk_root;
1232 
1233 	path = btrfs_alloc_path();
1234 	if (!path)
1235 		return -ENOMEM;
1236 
1237 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1238 	key.type = BTRFS_DEV_ITEM_KEY;
1239 	key.offset = (u64)-1;
1240 
1241 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1242 	if (ret < 0)
1243 		goto error;
1244 
1245 	BUG_ON(ret == 0); /* Corruption */
1246 
1247 	ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1248 				  BTRFS_DEV_ITEM_KEY);
1249 	if (ret) {
1250 		*objectid = 1;
1251 	} else {
1252 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1253 				      path->slots[0]);
1254 		*objectid = found_key.offset + 1;
1255 	}
1256 	ret = 0;
1257 error:
1258 	btrfs_free_path(path);
1259 	return ret;
1260 }
1261 
1262 /*
1263  * the device information is stored in the chunk root
1264  * the btrfs_device struct should be fully filled in
1265  */
1266 int btrfs_add_device(struct btrfs_trans_handle *trans,
1267 		     struct btrfs_root *root,
1268 		     struct btrfs_device *device)
1269 {
1270 	int ret;
1271 	struct btrfs_path *path;
1272 	struct btrfs_dev_item *dev_item;
1273 	struct extent_buffer *leaf;
1274 	struct btrfs_key key;
1275 	unsigned long ptr;
1276 
1277 	root = root->fs_info->chunk_root;
1278 
1279 	path = btrfs_alloc_path();
1280 	if (!path)
1281 		return -ENOMEM;
1282 
1283 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1284 	key.type = BTRFS_DEV_ITEM_KEY;
1285 	key.offset = device->devid;
1286 
1287 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1288 				      sizeof(*dev_item));
1289 	if (ret)
1290 		goto out;
1291 
1292 	leaf = path->nodes[0];
1293 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1294 
1295 	btrfs_set_device_id(leaf, dev_item, device->devid);
1296 	btrfs_set_device_generation(leaf, dev_item, 0);
1297 	btrfs_set_device_type(leaf, dev_item, device->type);
1298 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1299 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1300 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1301 	btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1302 	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1303 	btrfs_set_device_group(leaf, dev_item, 0);
1304 	btrfs_set_device_seek_speed(leaf, dev_item, 0);
1305 	btrfs_set_device_bandwidth(leaf, dev_item, 0);
1306 	btrfs_set_device_start_offset(leaf, dev_item, 0);
1307 
1308 	ptr = (unsigned long)btrfs_device_uuid(dev_item);
1309 	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1310 	ptr = (unsigned long)btrfs_device_fsid(dev_item);
1311 	write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1312 	btrfs_mark_buffer_dirty(leaf);
1313 
1314 	ret = 0;
1315 out:
1316 	btrfs_free_path(path);
1317 	return ret;
1318 }
1319 
1320 static int btrfs_rm_dev_item(struct btrfs_root *root,
1321 			     struct btrfs_device *device)
1322 {
1323 	int ret;
1324 	struct btrfs_path *path;
1325 	struct btrfs_key key;
1326 	struct btrfs_trans_handle *trans;
1327 
1328 	root = root->fs_info->chunk_root;
1329 
1330 	path = btrfs_alloc_path();
1331 	if (!path)
1332 		return -ENOMEM;
1333 
1334 	trans = btrfs_start_transaction(root, 0);
1335 	if (IS_ERR(trans)) {
1336 		btrfs_free_path(path);
1337 		return PTR_ERR(trans);
1338 	}
1339 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1340 	key.type = BTRFS_DEV_ITEM_KEY;
1341 	key.offset = device->devid;
1342 	lock_chunks(root);
1343 
1344 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1345 	if (ret < 0)
1346 		goto out;
1347 
1348 	if (ret > 0) {
1349 		ret = -ENOENT;
1350 		goto out;
1351 	}
1352 
1353 	ret = btrfs_del_item(trans, root, path);
1354 	if (ret)
1355 		goto out;
1356 out:
1357 	btrfs_free_path(path);
1358 	unlock_chunks(root);
1359 	btrfs_commit_transaction(trans, root);
1360 	return ret;
1361 }
1362 
1363 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1364 {
1365 	struct btrfs_device *device;
1366 	struct btrfs_device *next_device;
1367 	struct block_device *bdev;
1368 	struct buffer_head *bh = NULL;
1369 	struct btrfs_super_block *disk_super;
1370 	struct btrfs_fs_devices *cur_devices;
1371 	u64 all_avail;
1372 	u64 devid;
1373 	u64 num_devices;
1374 	u8 *dev_uuid;
1375 	int ret = 0;
1376 	bool clear_super = false;
1377 
1378 	mutex_lock(&uuid_mutex);
1379 
1380 	all_avail = root->fs_info->avail_data_alloc_bits |
1381 		root->fs_info->avail_system_alloc_bits |
1382 		root->fs_info->avail_metadata_alloc_bits;
1383 
1384 	num_devices = root->fs_info->fs_devices->num_devices;
1385 	btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1386 	if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1387 		WARN_ON(num_devices < 1);
1388 		num_devices--;
1389 	}
1390 	btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1391 
1392 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1393 		printk(KERN_ERR "btrfs: unable to go below four devices "
1394 		       "on raid10\n");
1395 		ret = -EINVAL;
1396 		goto out;
1397 	}
1398 
1399 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1400 		printk(KERN_ERR "btrfs: unable to go below two "
1401 		       "devices on raid1\n");
1402 		ret = -EINVAL;
1403 		goto out;
1404 	}
1405 
1406 	if (strcmp(device_path, "missing") == 0) {
1407 		struct list_head *devices;
1408 		struct btrfs_device *tmp;
1409 
1410 		device = NULL;
1411 		devices = &root->fs_info->fs_devices->devices;
1412 		/*
1413 		 * It is safe to read the devices since the volume_mutex
1414 		 * is held.
1415 		 */
1416 		list_for_each_entry(tmp, devices, dev_list) {
1417 			if (tmp->in_fs_metadata &&
1418 			    !tmp->is_tgtdev_for_dev_replace &&
1419 			    !tmp->bdev) {
1420 				device = tmp;
1421 				break;
1422 			}
1423 		}
1424 		bdev = NULL;
1425 		bh = NULL;
1426 		disk_super = NULL;
1427 		if (!device) {
1428 			printk(KERN_ERR "btrfs: no missing devices found to "
1429 			       "remove\n");
1430 			goto out;
1431 		}
1432 	} else {
1433 		ret = btrfs_get_bdev_and_sb(device_path,
1434 					    FMODE_READ | FMODE_EXCL,
1435 					    root->fs_info->bdev_holder, 0,
1436 					    &bdev, &bh);
1437 		if (ret)
1438 			goto out;
1439 		disk_super = (struct btrfs_super_block *)bh->b_data;
1440 		devid = btrfs_stack_device_id(&disk_super->dev_item);
1441 		dev_uuid = disk_super->dev_item.uuid;
1442 		device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1443 					   disk_super->fsid);
1444 		if (!device) {
1445 			ret = -ENOENT;
1446 			goto error_brelse;
1447 		}
1448 	}
1449 
1450 	if (device->is_tgtdev_for_dev_replace) {
1451 		pr_err("btrfs: unable to remove the dev_replace target dev\n");
1452 		ret = -EINVAL;
1453 		goto error_brelse;
1454 	}
1455 
1456 	if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1457 		printk(KERN_ERR "btrfs: unable to remove the only writeable "
1458 		       "device\n");
1459 		ret = -EINVAL;
1460 		goto error_brelse;
1461 	}
1462 
1463 	if (device->writeable) {
1464 		lock_chunks(root);
1465 		list_del_init(&device->dev_alloc_list);
1466 		unlock_chunks(root);
1467 		root->fs_info->fs_devices->rw_devices--;
1468 		clear_super = true;
1469 	}
1470 
1471 	ret = btrfs_shrink_device(device, 0);
1472 	if (ret)
1473 		goto error_undo;
1474 
1475 	/*
1476 	 * TODO: the superblock still includes this device in its num_devices
1477 	 * counter although write_all_supers() is not locked out. This
1478 	 * could give a filesystem state which requires a degraded mount.
1479 	 */
1480 	ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1481 	if (ret)
1482 		goto error_undo;
1483 
1484 	spin_lock(&root->fs_info->free_chunk_lock);
1485 	root->fs_info->free_chunk_space = device->total_bytes -
1486 		device->bytes_used;
1487 	spin_unlock(&root->fs_info->free_chunk_lock);
1488 
1489 	device->in_fs_metadata = 0;
1490 	btrfs_scrub_cancel_dev(root->fs_info, device);
1491 
1492 	/*
1493 	 * the device list mutex makes sure that we don't change
1494 	 * the device list while someone else is writing out all
1495 	 * the device supers.
1496 	 */
1497 
1498 	cur_devices = device->fs_devices;
1499 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1500 	list_del_rcu(&device->dev_list);
1501 
1502 	device->fs_devices->num_devices--;
1503 	device->fs_devices->total_devices--;
1504 
1505 	if (device->missing)
1506 		root->fs_info->fs_devices->missing_devices--;
1507 
1508 	next_device = list_entry(root->fs_info->fs_devices->devices.next,
1509 				 struct btrfs_device, dev_list);
1510 	if (device->bdev == root->fs_info->sb->s_bdev)
1511 		root->fs_info->sb->s_bdev = next_device->bdev;
1512 	if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1513 		root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1514 
1515 	if (device->bdev)
1516 		device->fs_devices->open_devices--;
1517 
1518 	call_rcu(&device->rcu, free_device);
1519 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1520 
1521 	num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1522 	btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1523 
1524 	if (cur_devices->open_devices == 0) {
1525 		struct btrfs_fs_devices *fs_devices;
1526 		fs_devices = root->fs_info->fs_devices;
1527 		while (fs_devices) {
1528 			if (fs_devices->seed == cur_devices)
1529 				break;
1530 			fs_devices = fs_devices->seed;
1531 		}
1532 		fs_devices->seed = cur_devices->seed;
1533 		cur_devices->seed = NULL;
1534 		lock_chunks(root);
1535 		__btrfs_close_devices(cur_devices);
1536 		unlock_chunks(root);
1537 		free_fs_devices(cur_devices);
1538 	}
1539 
1540 	root->fs_info->num_tolerated_disk_barrier_failures =
1541 		btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1542 
1543 	/*
1544 	 * at this point, the device is zero sized.  We want to
1545 	 * remove it from the devices list and zero out the old super
1546 	 */
1547 	if (clear_super && disk_super) {
1548 		/* make sure this device isn't detected as part of
1549 		 * the FS anymore
1550 		 */
1551 		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1552 		set_buffer_dirty(bh);
1553 		sync_dirty_buffer(bh);
1554 	}
1555 
1556 	ret = 0;
1557 
1558 	/* Notify udev that device has changed */
1559 	btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1560 
1561 error_brelse:
1562 	brelse(bh);
1563 	if (bdev)
1564 		blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1565 out:
1566 	mutex_unlock(&uuid_mutex);
1567 	return ret;
1568 error_undo:
1569 	if (device->writeable) {
1570 		lock_chunks(root);
1571 		list_add(&device->dev_alloc_list,
1572 			 &root->fs_info->fs_devices->alloc_list);
1573 		unlock_chunks(root);
1574 		root->fs_info->fs_devices->rw_devices++;
1575 	}
1576 	goto error_brelse;
1577 }
1578 
1579 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1580 				 struct btrfs_device *srcdev)
1581 {
1582 	WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1583 	list_del_rcu(&srcdev->dev_list);
1584 	list_del_rcu(&srcdev->dev_alloc_list);
1585 	fs_info->fs_devices->num_devices--;
1586 	if (srcdev->missing) {
1587 		fs_info->fs_devices->missing_devices--;
1588 		fs_info->fs_devices->rw_devices++;
1589 	}
1590 	if (srcdev->can_discard)
1591 		fs_info->fs_devices->num_can_discard--;
1592 	if (srcdev->bdev)
1593 		fs_info->fs_devices->open_devices--;
1594 
1595 	call_rcu(&srcdev->rcu, free_device);
1596 }
1597 
1598 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1599 				      struct btrfs_device *tgtdev)
1600 {
1601 	struct btrfs_device *next_device;
1602 
1603 	WARN_ON(!tgtdev);
1604 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
1605 	if (tgtdev->bdev) {
1606 		btrfs_scratch_superblock(tgtdev);
1607 		fs_info->fs_devices->open_devices--;
1608 	}
1609 	fs_info->fs_devices->num_devices--;
1610 	if (tgtdev->can_discard)
1611 		fs_info->fs_devices->num_can_discard++;
1612 
1613 	next_device = list_entry(fs_info->fs_devices->devices.next,
1614 				 struct btrfs_device, dev_list);
1615 	if (tgtdev->bdev == fs_info->sb->s_bdev)
1616 		fs_info->sb->s_bdev = next_device->bdev;
1617 	if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1618 		fs_info->fs_devices->latest_bdev = next_device->bdev;
1619 	list_del_rcu(&tgtdev->dev_list);
1620 
1621 	call_rcu(&tgtdev->rcu, free_device);
1622 
1623 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1624 }
1625 
1626 int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1627 			      struct btrfs_device **device)
1628 {
1629 	int ret = 0;
1630 	struct btrfs_super_block *disk_super;
1631 	u64 devid;
1632 	u8 *dev_uuid;
1633 	struct block_device *bdev;
1634 	struct buffer_head *bh;
1635 
1636 	*device = NULL;
1637 	ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1638 				    root->fs_info->bdev_holder, 0, &bdev, &bh);
1639 	if (ret)
1640 		return ret;
1641 	disk_super = (struct btrfs_super_block *)bh->b_data;
1642 	devid = btrfs_stack_device_id(&disk_super->dev_item);
1643 	dev_uuid = disk_super->dev_item.uuid;
1644 	*device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1645 				    disk_super->fsid);
1646 	brelse(bh);
1647 	if (!*device)
1648 		ret = -ENOENT;
1649 	blkdev_put(bdev, FMODE_READ);
1650 	return ret;
1651 }
1652 
1653 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1654 					 char *device_path,
1655 					 struct btrfs_device **device)
1656 {
1657 	*device = NULL;
1658 	if (strcmp(device_path, "missing") == 0) {
1659 		struct list_head *devices;
1660 		struct btrfs_device *tmp;
1661 
1662 		devices = &root->fs_info->fs_devices->devices;
1663 		/*
1664 		 * It is safe to read the devices since the volume_mutex
1665 		 * is held by the caller.
1666 		 */
1667 		list_for_each_entry(tmp, devices, dev_list) {
1668 			if (tmp->in_fs_metadata && !tmp->bdev) {
1669 				*device = tmp;
1670 				break;
1671 			}
1672 		}
1673 
1674 		if (!*device) {
1675 			pr_err("btrfs: no missing device found\n");
1676 			return -ENOENT;
1677 		}
1678 
1679 		return 0;
1680 	} else {
1681 		return btrfs_find_device_by_path(root, device_path, device);
1682 	}
1683 }
1684 
1685 /*
1686  * does all the dirty work required for changing file system's UUID.
1687  */
1688 static int btrfs_prepare_sprout(struct btrfs_root *root)
1689 {
1690 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1691 	struct btrfs_fs_devices *old_devices;
1692 	struct btrfs_fs_devices *seed_devices;
1693 	struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1694 	struct btrfs_device *device;
1695 	u64 super_flags;
1696 
1697 	BUG_ON(!mutex_is_locked(&uuid_mutex));
1698 	if (!fs_devices->seeding)
1699 		return -EINVAL;
1700 
1701 	seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1702 	if (!seed_devices)
1703 		return -ENOMEM;
1704 
1705 	old_devices = clone_fs_devices(fs_devices);
1706 	if (IS_ERR(old_devices)) {
1707 		kfree(seed_devices);
1708 		return PTR_ERR(old_devices);
1709 	}
1710 
1711 	list_add(&old_devices->list, &fs_uuids);
1712 
1713 	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1714 	seed_devices->opened = 1;
1715 	INIT_LIST_HEAD(&seed_devices->devices);
1716 	INIT_LIST_HEAD(&seed_devices->alloc_list);
1717 	mutex_init(&seed_devices->device_list_mutex);
1718 
1719 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1720 	list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1721 			      synchronize_rcu);
1722 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1723 
1724 	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1725 	list_for_each_entry(device, &seed_devices->devices, dev_list) {
1726 		device->fs_devices = seed_devices;
1727 	}
1728 
1729 	fs_devices->seeding = 0;
1730 	fs_devices->num_devices = 0;
1731 	fs_devices->open_devices = 0;
1732 	fs_devices->total_devices = 0;
1733 	fs_devices->seed = seed_devices;
1734 
1735 	generate_random_uuid(fs_devices->fsid);
1736 	memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1737 	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1738 	super_flags = btrfs_super_flags(disk_super) &
1739 		      ~BTRFS_SUPER_FLAG_SEEDING;
1740 	btrfs_set_super_flags(disk_super, super_flags);
1741 
1742 	return 0;
1743 }
1744 
1745 /*
1746  * strore the expected generation for seed devices in device items.
1747  */
1748 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1749 			       struct btrfs_root *root)
1750 {
1751 	struct btrfs_path *path;
1752 	struct extent_buffer *leaf;
1753 	struct btrfs_dev_item *dev_item;
1754 	struct btrfs_device *device;
1755 	struct btrfs_key key;
1756 	u8 fs_uuid[BTRFS_UUID_SIZE];
1757 	u8 dev_uuid[BTRFS_UUID_SIZE];
1758 	u64 devid;
1759 	int ret;
1760 
1761 	path = btrfs_alloc_path();
1762 	if (!path)
1763 		return -ENOMEM;
1764 
1765 	root = root->fs_info->chunk_root;
1766 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1767 	key.offset = 0;
1768 	key.type = BTRFS_DEV_ITEM_KEY;
1769 
1770 	while (1) {
1771 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1772 		if (ret < 0)
1773 			goto error;
1774 
1775 		leaf = path->nodes[0];
1776 next_slot:
1777 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1778 			ret = btrfs_next_leaf(root, path);
1779 			if (ret > 0)
1780 				break;
1781 			if (ret < 0)
1782 				goto error;
1783 			leaf = path->nodes[0];
1784 			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1785 			btrfs_release_path(path);
1786 			continue;
1787 		}
1788 
1789 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1790 		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1791 		    key.type != BTRFS_DEV_ITEM_KEY)
1792 			break;
1793 
1794 		dev_item = btrfs_item_ptr(leaf, path->slots[0],
1795 					  struct btrfs_dev_item);
1796 		devid = btrfs_device_id(leaf, dev_item);
1797 		read_extent_buffer(leaf, dev_uuid,
1798 				   (unsigned long)btrfs_device_uuid(dev_item),
1799 				   BTRFS_UUID_SIZE);
1800 		read_extent_buffer(leaf, fs_uuid,
1801 				   (unsigned long)btrfs_device_fsid(dev_item),
1802 				   BTRFS_UUID_SIZE);
1803 		device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1804 					   fs_uuid);
1805 		BUG_ON(!device); /* Logic error */
1806 
1807 		if (device->fs_devices->seeding) {
1808 			btrfs_set_device_generation(leaf, dev_item,
1809 						    device->generation);
1810 			btrfs_mark_buffer_dirty(leaf);
1811 		}
1812 
1813 		path->slots[0]++;
1814 		goto next_slot;
1815 	}
1816 	ret = 0;
1817 error:
1818 	btrfs_free_path(path);
1819 	return ret;
1820 }
1821 
1822 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1823 {
1824 	struct request_queue *q;
1825 	struct btrfs_trans_handle *trans;
1826 	struct btrfs_device *device;
1827 	struct block_device *bdev;
1828 	struct list_head *devices;
1829 	struct super_block *sb = root->fs_info->sb;
1830 	struct rcu_string *name;
1831 	u64 total_bytes;
1832 	int seeding_dev = 0;
1833 	int ret = 0;
1834 
1835 	if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1836 		return -EROFS;
1837 
1838 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1839 				  root->fs_info->bdev_holder);
1840 	if (IS_ERR(bdev))
1841 		return PTR_ERR(bdev);
1842 
1843 	if (root->fs_info->fs_devices->seeding) {
1844 		seeding_dev = 1;
1845 		down_write(&sb->s_umount);
1846 		mutex_lock(&uuid_mutex);
1847 	}
1848 
1849 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
1850 
1851 	devices = &root->fs_info->fs_devices->devices;
1852 
1853 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1854 	list_for_each_entry(device, devices, dev_list) {
1855 		if (device->bdev == bdev) {
1856 			ret = -EEXIST;
1857 			mutex_unlock(
1858 				&root->fs_info->fs_devices->device_list_mutex);
1859 			goto error;
1860 		}
1861 	}
1862 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1863 
1864 	device = kzalloc(sizeof(*device), GFP_NOFS);
1865 	if (!device) {
1866 		/* we can safely leave the fs_devices entry around */
1867 		ret = -ENOMEM;
1868 		goto error;
1869 	}
1870 
1871 	name = rcu_string_strdup(device_path, GFP_NOFS);
1872 	if (!name) {
1873 		kfree(device);
1874 		ret = -ENOMEM;
1875 		goto error;
1876 	}
1877 	rcu_assign_pointer(device->name, name);
1878 
1879 	ret = find_next_devid(root, &device->devid);
1880 	if (ret) {
1881 		rcu_string_free(device->name);
1882 		kfree(device);
1883 		goto error;
1884 	}
1885 
1886 	trans = btrfs_start_transaction(root, 0);
1887 	if (IS_ERR(trans)) {
1888 		rcu_string_free(device->name);
1889 		kfree(device);
1890 		ret = PTR_ERR(trans);
1891 		goto error;
1892 	}
1893 
1894 	lock_chunks(root);
1895 
1896 	q = bdev_get_queue(bdev);
1897 	if (blk_queue_discard(q))
1898 		device->can_discard = 1;
1899 	device->writeable = 1;
1900 	device->work.func = pending_bios_fn;
1901 	generate_random_uuid(device->uuid);
1902 	spin_lock_init(&device->io_lock);
1903 	device->generation = trans->transid;
1904 	device->io_width = root->sectorsize;
1905 	device->io_align = root->sectorsize;
1906 	device->sector_size = root->sectorsize;
1907 	device->total_bytes = i_size_read(bdev->bd_inode);
1908 	device->disk_total_bytes = device->total_bytes;
1909 	device->dev_root = root->fs_info->dev_root;
1910 	device->bdev = bdev;
1911 	device->in_fs_metadata = 1;
1912 	device->is_tgtdev_for_dev_replace = 0;
1913 	device->mode = FMODE_EXCL;
1914 	set_blocksize(device->bdev, 4096);
1915 
1916 	if (seeding_dev) {
1917 		sb->s_flags &= ~MS_RDONLY;
1918 		ret = btrfs_prepare_sprout(root);
1919 		BUG_ON(ret); /* -ENOMEM */
1920 	}
1921 
1922 	device->fs_devices = root->fs_info->fs_devices;
1923 
1924 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1925 	list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1926 	list_add(&device->dev_alloc_list,
1927 		 &root->fs_info->fs_devices->alloc_list);
1928 	root->fs_info->fs_devices->num_devices++;
1929 	root->fs_info->fs_devices->open_devices++;
1930 	root->fs_info->fs_devices->rw_devices++;
1931 	root->fs_info->fs_devices->total_devices++;
1932 	if (device->can_discard)
1933 		root->fs_info->fs_devices->num_can_discard++;
1934 	root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1935 
1936 	spin_lock(&root->fs_info->free_chunk_lock);
1937 	root->fs_info->free_chunk_space += device->total_bytes;
1938 	spin_unlock(&root->fs_info->free_chunk_lock);
1939 
1940 	if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1941 		root->fs_info->fs_devices->rotating = 1;
1942 
1943 	total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1944 	btrfs_set_super_total_bytes(root->fs_info->super_copy,
1945 				    total_bytes + device->total_bytes);
1946 
1947 	total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1948 	btrfs_set_super_num_devices(root->fs_info->super_copy,
1949 				    total_bytes + 1);
1950 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1951 
1952 	if (seeding_dev) {
1953 		ret = init_first_rw_device(trans, root, device);
1954 		if (ret) {
1955 			btrfs_abort_transaction(trans, root, ret);
1956 			goto error_trans;
1957 		}
1958 		ret = btrfs_finish_sprout(trans, root);
1959 		if (ret) {
1960 			btrfs_abort_transaction(trans, root, ret);
1961 			goto error_trans;
1962 		}
1963 	} else {
1964 		ret = btrfs_add_device(trans, root, device);
1965 		if (ret) {
1966 			btrfs_abort_transaction(trans, root, ret);
1967 			goto error_trans;
1968 		}
1969 	}
1970 
1971 	/*
1972 	 * we've got more storage, clear any full flags on the space
1973 	 * infos
1974 	 */
1975 	btrfs_clear_space_info_full(root->fs_info);
1976 
1977 	unlock_chunks(root);
1978 	root->fs_info->num_tolerated_disk_barrier_failures =
1979 		btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1980 	ret = btrfs_commit_transaction(trans, root);
1981 
1982 	if (seeding_dev) {
1983 		mutex_unlock(&uuid_mutex);
1984 		up_write(&sb->s_umount);
1985 
1986 		if (ret) /* transaction commit */
1987 			return ret;
1988 
1989 		ret = btrfs_relocate_sys_chunks(root);
1990 		if (ret < 0)
1991 			btrfs_error(root->fs_info, ret,
1992 				    "Failed to relocate sys chunks after "
1993 				    "device initialization. This can be fixed "
1994 				    "using the \"btrfs balance\" command.");
1995 		trans = btrfs_attach_transaction(root);
1996 		if (IS_ERR(trans)) {
1997 			if (PTR_ERR(trans) == -ENOENT)
1998 				return 0;
1999 			return PTR_ERR(trans);
2000 		}
2001 		ret = btrfs_commit_transaction(trans, root);
2002 	}
2003 
2004 	return ret;
2005 
2006 error_trans:
2007 	unlock_chunks(root);
2008 	btrfs_end_transaction(trans, root);
2009 	rcu_string_free(device->name);
2010 	kfree(device);
2011 error:
2012 	blkdev_put(bdev, FMODE_EXCL);
2013 	if (seeding_dev) {
2014 		mutex_unlock(&uuid_mutex);
2015 		up_write(&sb->s_umount);
2016 	}
2017 	return ret;
2018 }
2019 
2020 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2021 				  struct btrfs_device **device_out)
2022 {
2023 	struct request_queue *q;
2024 	struct btrfs_device *device;
2025 	struct block_device *bdev;
2026 	struct btrfs_fs_info *fs_info = root->fs_info;
2027 	struct list_head *devices;
2028 	struct rcu_string *name;
2029 	int ret = 0;
2030 
2031 	*device_out = NULL;
2032 	if (fs_info->fs_devices->seeding)
2033 		return -EINVAL;
2034 
2035 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2036 				  fs_info->bdev_holder);
2037 	if (IS_ERR(bdev))
2038 		return PTR_ERR(bdev);
2039 
2040 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
2041 
2042 	devices = &fs_info->fs_devices->devices;
2043 	list_for_each_entry(device, devices, dev_list) {
2044 		if (device->bdev == bdev) {
2045 			ret = -EEXIST;
2046 			goto error;
2047 		}
2048 	}
2049 
2050 	device = kzalloc(sizeof(*device), GFP_NOFS);
2051 	if (!device) {
2052 		ret = -ENOMEM;
2053 		goto error;
2054 	}
2055 
2056 	name = rcu_string_strdup(device_path, GFP_NOFS);
2057 	if (!name) {
2058 		kfree(device);
2059 		ret = -ENOMEM;
2060 		goto error;
2061 	}
2062 	rcu_assign_pointer(device->name, name);
2063 
2064 	q = bdev_get_queue(bdev);
2065 	if (blk_queue_discard(q))
2066 		device->can_discard = 1;
2067 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2068 	device->writeable = 1;
2069 	device->work.func = pending_bios_fn;
2070 	generate_random_uuid(device->uuid);
2071 	device->devid = BTRFS_DEV_REPLACE_DEVID;
2072 	spin_lock_init(&device->io_lock);
2073 	device->generation = 0;
2074 	device->io_width = root->sectorsize;
2075 	device->io_align = root->sectorsize;
2076 	device->sector_size = root->sectorsize;
2077 	device->total_bytes = i_size_read(bdev->bd_inode);
2078 	device->disk_total_bytes = device->total_bytes;
2079 	device->dev_root = fs_info->dev_root;
2080 	device->bdev = bdev;
2081 	device->in_fs_metadata = 1;
2082 	device->is_tgtdev_for_dev_replace = 1;
2083 	device->mode = FMODE_EXCL;
2084 	set_blocksize(device->bdev, 4096);
2085 	device->fs_devices = fs_info->fs_devices;
2086 	list_add(&device->dev_list, &fs_info->fs_devices->devices);
2087 	fs_info->fs_devices->num_devices++;
2088 	fs_info->fs_devices->open_devices++;
2089 	if (device->can_discard)
2090 		fs_info->fs_devices->num_can_discard++;
2091 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2092 
2093 	*device_out = device;
2094 	return ret;
2095 
2096 error:
2097 	blkdev_put(bdev, FMODE_EXCL);
2098 	return ret;
2099 }
2100 
2101 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2102 					      struct btrfs_device *tgtdev)
2103 {
2104 	WARN_ON(fs_info->fs_devices->rw_devices == 0);
2105 	tgtdev->io_width = fs_info->dev_root->sectorsize;
2106 	tgtdev->io_align = fs_info->dev_root->sectorsize;
2107 	tgtdev->sector_size = fs_info->dev_root->sectorsize;
2108 	tgtdev->dev_root = fs_info->dev_root;
2109 	tgtdev->in_fs_metadata = 1;
2110 }
2111 
2112 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2113 					struct btrfs_device *device)
2114 {
2115 	int ret;
2116 	struct btrfs_path *path;
2117 	struct btrfs_root *root;
2118 	struct btrfs_dev_item *dev_item;
2119 	struct extent_buffer *leaf;
2120 	struct btrfs_key key;
2121 
2122 	root = device->dev_root->fs_info->chunk_root;
2123 
2124 	path = btrfs_alloc_path();
2125 	if (!path)
2126 		return -ENOMEM;
2127 
2128 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2129 	key.type = BTRFS_DEV_ITEM_KEY;
2130 	key.offset = device->devid;
2131 
2132 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2133 	if (ret < 0)
2134 		goto out;
2135 
2136 	if (ret > 0) {
2137 		ret = -ENOENT;
2138 		goto out;
2139 	}
2140 
2141 	leaf = path->nodes[0];
2142 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2143 
2144 	btrfs_set_device_id(leaf, dev_item, device->devid);
2145 	btrfs_set_device_type(leaf, dev_item, device->type);
2146 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2147 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2148 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2149 	btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2150 	btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2151 	btrfs_mark_buffer_dirty(leaf);
2152 
2153 out:
2154 	btrfs_free_path(path);
2155 	return ret;
2156 }
2157 
2158 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2159 		      struct btrfs_device *device, u64 new_size)
2160 {
2161 	struct btrfs_super_block *super_copy =
2162 		device->dev_root->fs_info->super_copy;
2163 	u64 old_total = btrfs_super_total_bytes(super_copy);
2164 	u64 diff = new_size - device->total_bytes;
2165 
2166 	if (!device->writeable)
2167 		return -EACCES;
2168 	if (new_size <= device->total_bytes ||
2169 	    device->is_tgtdev_for_dev_replace)
2170 		return -EINVAL;
2171 
2172 	btrfs_set_super_total_bytes(super_copy, old_total + diff);
2173 	device->fs_devices->total_rw_bytes += diff;
2174 
2175 	device->total_bytes = new_size;
2176 	device->disk_total_bytes = new_size;
2177 	btrfs_clear_space_info_full(device->dev_root->fs_info);
2178 
2179 	return btrfs_update_device(trans, device);
2180 }
2181 
2182 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2183 		      struct btrfs_device *device, u64 new_size)
2184 {
2185 	int ret;
2186 	lock_chunks(device->dev_root);
2187 	ret = __btrfs_grow_device(trans, device, new_size);
2188 	unlock_chunks(device->dev_root);
2189 	return ret;
2190 }
2191 
2192 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2193 			    struct btrfs_root *root,
2194 			    u64 chunk_tree, u64 chunk_objectid,
2195 			    u64 chunk_offset)
2196 {
2197 	int ret;
2198 	struct btrfs_path *path;
2199 	struct btrfs_key key;
2200 
2201 	root = root->fs_info->chunk_root;
2202 	path = btrfs_alloc_path();
2203 	if (!path)
2204 		return -ENOMEM;
2205 
2206 	key.objectid = chunk_objectid;
2207 	key.offset = chunk_offset;
2208 	key.type = BTRFS_CHUNK_ITEM_KEY;
2209 
2210 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2211 	if (ret < 0)
2212 		goto out;
2213 	else if (ret > 0) { /* Logic error or corruption */
2214 		btrfs_error(root->fs_info, -ENOENT,
2215 			    "Failed lookup while freeing chunk.");
2216 		ret = -ENOENT;
2217 		goto out;
2218 	}
2219 
2220 	ret = btrfs_del_item(trans, root, path);
2221 	if (ret < 0)
2222 		btrfs_error(root->fs_info, ret,
2223 			    "Failed to delete chunk item.");
2224 out:
2225 	btrfs_free_path(path);
2226 	return ret;
2227 }
2228 
2229 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2230 			chunk_offset)
2231 {
2232 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2233 	struct btrfs_disk_key *disk_key;
2234 	struct btrfs_chunk *chunk;
2235 	u8 *ptr;
2236 	int ret = 0;
2237 	u32 num_stripes;
2238 	u32 array_size;
2239 	u32 len = 0;
2240 	u32 cur;
2241 	struct btrfs_key key;
2242 
2243 	array_size = btrfs_super_sys_array_size(super_copy);
2244 
2245 	ptr = super_copy->sys_chunk_array;
2246 	cur = 0;
2247 
2248 	while (cur < array_size) {
2249 		disk_key = (struct btrfs_disk_key *)ptr;
2250 		btrfs_disk_key_to_cpu(&key, disk_key);
2251 
2252 		len = sizeof(*disk_key);
2253 
2254 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2255 			chunk = (struct btrfs_chunk *)(ptr + len);
2256 			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2257 			len += btrfs_chunk_item_size(num_stripes);
2258 		} else {
2259 			ret = -EIO;
2260 			break;
2261 		}
2262 		if (key.objectid == chunk_objectid &&
2263 		    key.offset == chunk_offset) {
2264 			memmove(ptr, ptr + len, array_size - (cur + len));
2265 			array_size -= len;
2266 			btrfs_set_super_sys_array_size(super_copy, array_size);
2267 		} else {
2268 			ptr += len;
2269 			cur += len;
2270 		}
2271 	}
2272 	return ret;
2273 }
2274 
2275 static int btrfs_relocate_chunk(struct btrfs_root *root,
2276 			 u64 chunk_tree, u64 chunk_objectid,
2277 			 u64 chunk_offset)
2278 {
2279 	struct extent_map_tree *em_tree;
2280 	struct btrfs_root *extent_root;
2281 	struct btrfs_trans_handle *trans;
2282 	struct extent_map *em;
2283 	struct map_lookup *map;
2284 	int ret;
2285 	int i;
2286 
2287 	root = root->fs_info->chunk_root;
2288 	extent_root = root->fs_info->extent_root;
2289 	em_tree = &root->fs_info->mapping_tree.map_tree;
2290 
2291 	ret = btrfs_can_relocate(extent_root, chunk_offset);
2292 	if (ret)
2293 		return -ENOSPC;
2294 
2295 	/* step one, relocate all the extents inside this chunk */
2296 	ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2297 	if (ret)
2298 		return ret;
2299 
2300 	trans = btrfs_start_transaction(root, 0);
2301 	BUG_ON(IS_ERR(trans));
2302 
2303 	lock_chunks(root);
2304 
2305 	/*
2306 	 * step two, delete the device extents and the
2307 	 * chunk tree entries
2308 	 */
2309 	read_lock(&em_tree->lock);
2310 	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2311 	read_unlock(&em_tree->lock);
2312 
2313 	BUG_ON(!em || em->start > chunk_offset ||
2314 	       em->start + em->len < chunk_offset);
2315 	map = (struct map_lookup *)em->bdev;
2316 
2317 	for (i = 0; i < map->num_stripes; i++) {
2318 		ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2319 					    map->stripes[i].physical);
2320 		BUG_ON(ret);
2321 
2322 		if (map->stripes[i].dev) {
2323 			ret = btrfs_update_device(trans, map->stripes[i].dev);
2324 			BUG_ON(ret);
2325 		}
2326 	}
2327 	ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2328 			       chunk_offset);
2329 
2330 	BUG_ON(ret);
2331 
2332 	trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2333 
2334 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2335 		ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2336 		BUG_ON(ret);
2337 	}
2338 
2339 	ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2340 	BUG_ON(ret);
2341 
2342 	write_lock(&em_tree->lock);
2343 	remove_extent_mapping(em_tree, em);
2344 	write_unlock(&em_tree->lock);
2345 
2346 	kfree(map);
2347 	em->bdev = NULL;
2348 
2349 	/* once for the tree */
2350 	free_extent_map(em);
2351 	/* once for us */
2352 	free_extent_map(em);
2353 
2354 	unlock_chunks(root);
2355 	btrfs_end_transaction(trans, root);
2356 	return 0;
2357 }
2358 
2359 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2360 {
2361 	struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2362 	struct btrfs_path *path;
2363 	struct extent_buffer *leaf;
2364 	struct btrfs_chunk *chunk;
2365 	struct btrfs_key key;
2366 	struct btrfs_key found_key;
2367 	u64 chunk_tree = chunk_root->root_key.objectid;
2368 	u64 chunk_type;
2369 	bool retried = false;
2370 	int failed = 0;
2371 	int ret;
2372 
2373 	path = btrfs_alloc_path();
2374 	if (!path)
2375 		return -ENOMEM;
2376 
2377 again:
2378 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2379 	key.offset = (u64)-1;
2380 	key.type = BTRFS_CHUNK_ITEM_KEY;
2381 
2382 	while (1) {
2383 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2384 		if (ret < 0)
2385 			goto error;
2386 		BUG_ON(ret == 0); /* Corruption */
2387 
2388 		ret = btrfs_previous_item(chunk_root, path, key.objectid,
2389 					  key.type);
2390 		if (ret < 0)
2391 			goto error;
2392 		if (ret > 0)
2393 			break;
2394 
2395 		leaf = path->nodes[0];
2396 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2397 
2398 		chunk = btrfs_item_ptr(leaf, path->slots[0],
2399 				       struct btrfs_chunk);
2400 		chunk_type = btrfs_chunk_type(leaf, chunk);
2401 		btrfs_release_path(path);
2402 
2403 		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2404 			ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2405 						   found_key.objectid,
2406 						   found_key.offset);
2407 			if (ret == -ENOSPC)
2408 				failed++;
2409 			else if (ret)
2410 				BUG();
2411 		}
2412 
2413 		if (found_key.offset == 0)
2414 			break;
2415 		key.offset = found_key.offset - 1;
2416 	}
2417 	ret = 0;
2418 	if (failed && !retried) {
2419 		failed = 0;
2420 		retried = true;
2421 		goto again;
2422 	} else if (failed && retried) {
2423 		WARN_ON(1);
2424 		ret = -ENOSPC;
2425 	}
2426 error:
2427 	btrfs_free_path(path);
2428 	return ret;
2429 }
2430 
2431 static int insert_balance_item(struct btrfs_root *root,
2432 			       struct btrfs_balance_control *bctl)
2433 {
2434 	struct btrfs_trans_handle *trans;
2435 	struct btrfs_balance_item *item;
2436 	struct btrfs_disk_balance_args disk_bargs;
2437 	struct btrfs_path *path;
2438 	struct extent_buffer *leaf;
2439 	struct btrfs_key key;
2440 	int ret, err;
2441 
2442 	path = btrfs_alloc_path();
2443 	if (!path)
2444 		return -ENOMEM;
2445 
2446 	trans = btrfs_start_transaction(root, 0);
2447 	if (IS_ERR(trans)) {
2448 		btrfs_free_path(path);
2449 		return PTR_ERR(trans);
2450 	}
2451 
2452 	key.objectid = BTRFS_BALANCE_OBJECTID;
2453 	key.type = BTRFS_BALANCE_ITEM_KEY;
2454 	key.offset = 0;
2455 
2456 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2457 				      sizeof(*item));
2458 	if (ret)
2459 		goto out;
2460 
2461 	leaf = path->nodes[0];
2462 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2463 
2464 	memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2465 
2466 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2467 	btrfs_set_balance_data(leaf, item, &disk_bargs);
2468 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2469 	btrfs_set_balance_meta(leaf, item, &disk_bargs);
2470 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2471 	btrfs_set_balance_sys(leaf, item, &disk_bargs);
2472 
2473 	btrfs_set_balance_flags(leaf, item, bctl->flags);
2474 
2475 	btrfs_mark_buffer_dirty(leaf);
2476 out:
2477 	btrfs_free_path(path);
2478 	err = btrfs_commit_transaction(trans, root);
2479 	if (err && !ret)
2480 		ret = err;
2481 	return ret;
2482 }
2483 
2484 static int del_balance_item(struct btrfs_root *root)
2485 {
2486 	struct btrfs_trans_handle *trans;
2487 	struct btrfs_path *path;
2488 	struct btrfs_key key;
2489 	int ret, err;
2490 
2491 	path = btrfs_alloc_path();
2492 	if (!path)
2493 		return -ENOMEM;
2494 
2495 	trans = btrfs_start_transaction(root, 0);
2496 	if (IS_ERR(trans)) {
2497 		btrfs_free_path(path);
2498 		return PTR_ERR(trans);
2499 	}
2500 
2501 	key.objectid = BTRFS_BALANCE_OBJECTID;
2502 	key.type = BTRFS_BALANCE_ITEM_KEY;
2503 	key.offset = 0;
2504 
2505 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2506 	if (ret < 0)
2507 		goto out;
2508 	if (ret > 0) {
2509 		ret = -ENOENT;
2510 		goto out;
2511 	}
2512 
2513 	ret = btrfs_del_item(trans, root, path);
2514 out:
2515 	btrfs_free_path(path);
2516 	err = btrfs_commit_transaction(trans, root);
2517 	if (err && !ret)
2518 		ret = err;
2519 	return ret;
2520 }
2521 
2522 /*
2523  * This is a heuristic used to reduce the number of chunks balanced on
2524  * resume after balance was interrupted.
2525  */
2526 static void update_balance_args(struct btrfs_balance_control *bctl)
2527 {
2528 	/*
2529 	 * Turn on soft mode for chunk types that were being converted.
2530 	 */
2531 	if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2532 		bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2533 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2534 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2535 	if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2536 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2537 
2538 	/*
2539 	 * Turn on usage filter if is not already used.  The idea is
2540 	 * that chunks that we have already balanced should be
2541 	 * reasonably full.  Don't do it for chunks that are being
2542 	 * converted - that will keep us from relocating unconverted
2543 	 * (albeit full) chunks.
2544 	 */
2545 	if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2546 	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2547 		bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2548 		bctl->data.usage = 90;
2549 	}
2550 	if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2551 	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2552 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2553 		bctl->sys.usage = 90;
2554 	}
2555 	if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2556 	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2557 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2558 		bctl->meta.usage = 90;
2559 	}
2560 }
2561 
2562 /*
2563  * Should be called with both balance and volume mutexes held to
2564  * serialize other volume operations (add_dev/rm_dev/resize) with
2565  * restriper.  Same goes for unset_balance_control.
2566  */
2567 static void set_balance_control(struct btrfs_balance_control *bctl)
2568 {
2569 	struct btrfs_fs_info *fs_info = bctl->fs_info;
2570 
2571 	BUG_ON(fs_info->balance_ctl);
2572 
2573 	spin_lock(&fs_info->balance_lock);
2574 	fs_info->balance_ctl = bctl;
2575 	spin_unlock(&fs_info->balance_lock);
2576 }
2577 
2578 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2579 {
2580 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2581 
2582 	BUG_ON(!fs_info->balance_ctl);
2583 
2584 	spin_lock(&fs_info->balance_lock);
2585 	fs_info->balance_ctl = NULL;
2586 	spin_unlock(&fs_info->balance_lock);
2587 
2588 	kfree(bctl);
2589 }
2590 
2591 /*
2592  * Balance filters.  Return 1 if chunk should be filtered out
2593  * (should not be balanced).
2594  */
2595 static int chunk_profiles_filter(u64 chunk_type,
2596 				 struct btrfs_balance_args *bargs)
2597 {
2598 	chunk_type = chunk_to_extended(chunk_type) &
2599 				BTRFS_EXTENDED_PROFILE_MASK;
2600 
2601 	if (bargs->profiles & chunk_type)
2602 		return 0;
2603 
2604 	return 1;
2605 }
2606 
2607 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2608 			      struct btrfs_balance_args *bargs)
2609 {
2610 	struct btrfs_block_group_cache *cache;
2611 	u64 chunk_used, user_thresh;
2612 	int ret = 1;
2613 
2614 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2615 	chunk_used = btrfs_block_group_used(&cache->item);
2616 
2617 	user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2618 	if (chunk_used < user_thresh)
2619 		ret = 0;
2620 
2621 	btrfs_put_block_group(cache);
2622 	return ret;
2623 }
2624 
2625 static int chunk_devid_filter(struct extent_buffer *leaf,
2626 			      struct btrfs_chunk *chunk,
2627 			      struct btrfs_balance_args *bargs)
2628 {
2629 	struct btrfs_stripe *stripe;
2630 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2631 	int i;
2632 
2633 	for (i = 0; i < num_stripes; i++) {
2634 		stripe = btrfs_stripe_nr(chunk, i);
2635 		if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2636 			return 0;
2637 	}
2638 
2639 	return 1;
2640 }
2641 
2642 /* [pstart, pend) */
2643 static int chunk_drange_filter(struct extent_buffer *leaf,
2644 			       struct btrfs_chunk *chunk,
2645 			       u64 chunk_offset,
2646 			       struct btrfs_balance_args *bargs)
2647 {
2648 	struct btrfs_stripe *stripe;
2649 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2650 	u64 stripe_offset;
2651 	u64 stripe_length;
2652 	int factor;
2653 	int i;
2654 
2655 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2656 		return 0;
2657 
2658 	if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2659 	     BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2660 		factor = 2;
2661 	else
2662 		factor = 1;
2663 	factor = num_stripes / factor;
2664 
2665 	for (i = 0; i < num_stripes; i++) {
2666 		stripe = btrfs_stripe_nr(chunk, i);
2667 		if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2668 			continue;
2669 
2670 		stripe_offset = btrfs_stripe_offset(leaf, stripe);
2671 		stripe_length = btrfs_chunk_length(leaf, chunk);
2672 		do_div(stripe_length, factor);
2673 
2674 		if (stripe_offset < bargs->pend &&
2675 		    stripe_offset + stripe_length > bargs->pstart)
2676 			return 0;
2677 	}
2678 
2679 	return 1;
2680 }
2681 
2682 /* [vstart, vend) */
2683 static int chunk_vrange_filter(struct extent_buffer *leaf,
2684 			       struct btrfs_chunk *chunk,
2685 			       u64 chunk_offset,
2686 			       struct btrfs_balance_args *bargs)
2687 {
2688 	if (chunk_offset < bargs->vend &&
2689 	    chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2690 		/* at least part of the chunk is inside this vrange */
2691 		return 0;
2692 
2693 	return 1;
2694 }
2695 
2696 static int chunk_soft_convert_filter(u64 chunk_type,
2697 				     struct btrfs_balance_args *bargs)
2698 {
2699 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2700 		return 0;
2701 
2702 	chunk_type = chunk_to_extended(chunk_type) &
2703 				BTRFS_EXTENDED_PROFILE_MASK;
2704 
2705 	if (bargs->target == chunk_type)
2706 		return 1;
2707 
2708 	return 0;
2709 }
2710 
2711 static int should_balance_chunk(struct btrfs_root *root,
2712 				struct extent_buffer *leaf,
2713 				struct btrfs_chunk *chunk, u64 chunk_offset)
2714 {
2715 	struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2716 	struct btrfs_balance_args *bargs = NULL;
2717 	u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2718 
2719 	/* type filter */
2720 	if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2721 	      (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2722 		return 0;
2723 	}
2724 
2725 	if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2726 		bargs = &bctl->data;
2727 	else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2728 		bargs = &bctl->sys;
2729 	else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2730 		bargs = &bctl->meta;
2731 
2732 	/* profiles filter */
2733 	if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2734 	    chunk_profiles_filter(chunk_type, bargs)) {
2735 		return 0;
2736 	}
2737 
2738 	/* usage filter */
2739 	if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2740 	    chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2741 		return 0;
2742 	}
2743 
2744 	/* devid filter */
2745 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2746 	    chunk_devid_filter(leaf, chunk, bargs)) {
2747 		return 0;
2748 	}
2749 
2750 	/* drange filter, makes sense only with devid filter */
2751 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2752 	    chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2753 		return 0;
2754 	}
2755 
2756 	/* vrange filter */
2757 	if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2758 	    chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2759 		return 0;
2760 	}
2761 
2762 	/* soft profile changing mode */
2763 	if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2764 	    chunk_soft_convert_filter(chunk_type, bargs)) {
2765 		return 0;
2766 	}
2767 
2768 	return 1;
2769 }
2770 
2771 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2772 {
2773 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2774 	struct btrfs_root *chunk_root = fs_info->chunk_root;
2775 	struct btrfs_root *dev_root = fs_info->dev_root;
2776 	struct list_head *devices;
2777 	struct btrfs_device *device;
2778 	u64 old_size;
2779 	u64 size_to_free;
2780 	struct btrfs_chunk *chunk;
2781 	struct btrfs_path *path;
2782 	struct btrfs_key key;
2783 	struct btrfs_key found_key;
2784 	struct btrfs_trans_handle *trans;
2785 	struct extent_buffer *leaf;
2786 	int slot;
2787 	int ret;
2788 	int enospc_errors = 0;
2789 	bool counting = true;
2790 
2791 	/* step one make some room on all the devices */
2792 	devices = &fs_info->fs_devices->devices;
2793 	list_for_each_entry(device, devices, dev_list) {
2794 		old_size = device->total_bytes;
2795 		size_to_free = div_factor(old_size, 1);
2796 		size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2797 		if (!device->writeable ||
2798 		    device->total_bytes - device->bytes_used > size_to_free ||
2799 		    device->is_tgtdev_for_dev_replace)
2800 			continue;
2801 
2802 		ret = btrfs_shrink_device(device, old_size - size_to_free);
2803 		if (ret == -ENOSPC)
2804 			break;
2805 		BUG_ON(ret);
2806 
2807 		trans = btrfs_start_transaction(dev_root, 0);
2808 		BUG_ON(IS_ERR(trans));
2809 
2810 		ret = btrfs_grow_device(trans, device, old_size);
2811 		BUG_ON(ret);
2812 
2813 		btrfs_end_transaction(trans, dev_root);
2814 	}
2815 
2816 	/* step two, relocate all the chunks */
2817 	path = btrfs_alloc_path();
2818 	if (!path) {
2819 		ret = -ENOMEM;
2820 		goto error;
2821 	}
2822 
2823 	/* zero out stat counters */
2824 	spin_lock(&fs_info->balance_lock);
2825 	memset(&bctl->stat, 0, sizeof(bctl->stat));
2826 	spin_unlock(&fs_info->balance_lock);
2827 again:
2828 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2829 	key.offset = (u64)-1;
2830 	key.type = BTRFS_CHUNK_ITEM_KEY;
2831 
2832 	while (1) {
2833 		if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2834 		    atomic_read(&fs_info->balance_cancel_req)) {
2835 			ret = -ECANCELED;
2836 			goto error;
2837 		}
2838 
2839 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2840 		if (ret < 0)
2841 			goto error;
2842 
2843 		/*
2844 		 * this shouldn't happen, it means the last relocate
2845 		 * failed
2846 		 */
2847 		if (ret == 0)
2848 			BUG(); /* FIXME break ? */
2849 
2850 		ret = btrfs_previous_item(chunk_root, path, 0,
2851 					  BTRFS_CHUNK_ITEM_KEY);
2852 		if (ret) {
2853 			ret = 0;
2854 			break;
2855 		}
2856 
2857 		leaf = path->nodes[0];
2858 		slot = path->slots[0];
2859 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
2860 
2861 		if (found_key.objectid != key.objectid)
2862 			break;
2863 
2864 		/* chunk zero is special */
2865 		if (found_key.offset == 0)
2866 			break;
2867 
2868 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2869 
2870 		if (!counting) {
2871 			spin_lock(&fs_info->balance_lock);
2872 			bctl->stat.considered++;
2873 			spin_unlock(&fs_info->balance_lock);
2874 		}
2875 
2876 		ret = should_balance_chunk(chunk_root, leaf, chunk,
2877 					   found_key.offset);
2878 		btrfs_release_path(path);
2879 		if (!ret)
2880 			goto loop;
2881 
2882 		if (counting) {
2883 			spin_lock(&fs_info->balance_lock);
2884 			bctl->stat.expected++;
2885 			spin_unlock(&fs_info->balance_lock);
2886 			goto loop;
2887 		}
2888 
2889 		ret = btrfs_relocate_chunk(chunk_root,
2890 					   chunk_root->root_key.objectid,
2891 					   found_key.objectid,
2892 					   found_key.offset);
2893 		if (ret && ret != -ENOSPC)
2894 			goto error;
2895 		if (ret == -ENOSPC) {
2896 			enospc_errors++;
2897 		} else {
2898 			spin_lock(&fs_info->balance_lock);
2899 			bctl->stat.completed++;
2900 			spin_unlock(&fs_info->balance_lock);
2901 		}
2902 loop:
2903 		key.offset = found_key.offset - 1;
2904 	}
2905 
2906 	if (counting) {
2907 		btrfs_release_path(path);
2908 		counting = false;
2909 		goto again;
2910 	}
2911 error:
2912 	btrfs_free_path(path);
2913 	if (enospc_errors) {
2914 		printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2915 		       enospc_errors);
2916 		if (!ret)
2917 			ret = -ENOSPC;
2918 	}
2919 
2920 	return ret;
2921 }
2922 
2923 /**
2924  * alloc_profile_is_valid - see if a given profile is valid and reduced
2925  * @flags: profile to validate
2926  * @extended: if true @flags is treated as an extended profile
2927  */
2928 static int alloc_profile_is_valid(u64 flags, int extended)
2929 {
2930 	u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2931 			       BTRFS_BLOCK_GROUP_PROFILE_MASK);
2932 
2933 	flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2934 
2935 	/* 1) check that all other bits are zeroed */
2936 	if (flags & ~mask)
2937 		return 0;
2938 
2939 	/* 2) see if profile is reduced */
2940 	if (flags == 0)
2941 		return !extended; /* "0" is valid for usual profiles */
2942 
2943 	/* true if exactly one bit set */
2944 	return (flags & (flags - 1)) == 0;
2945 }
2946 
2947 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2948 {
2949 	/* cancel requested || normal exit path */
2950 	return atomic_read(&fs_info->balance_cancel_req) ||
2951 		(atomic_read(&fs_info->balance_pause_req) == 0 &&
2952 		 atomic_read(&fs_info->balance_cancel_req) == 0);
2953 }
2954 
2955 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2956 {
2957 	int ret;
2958 
2959 	unset_balance_control(fs_info);
2960 	ret = del_balance_item(fs_info->tree_root);
2961 	BUG_ON(ret);
2962 }
2963 
2964 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2965 			       struct btrfs_ioctl_balance_args *bargs);
2966 
2967 /*
2968  * Should be called with both balance and volume mutexes held
2969  */
2970 int btrfs_balance(struct btrfs_balance_control *bctl,
2971 		  struct btrfs_ioctl_balance_args *bargs)
2972 {
2973 	struct btrfs_fs_info *fs_info = bctl->fs_info;
2974 	u64 allowed;
2975 	int mixed = 0;
2976 	int ret;
2977 	u64 num_devices;
2978 
2979 	if (btrfs_fs_closing(fs_info) ||
2980 	    atomic_read(&fs_info->balance_pause_req) ||
2981 	    atomic_read(&fs_info->balance_cancel_req)) {
2982 		ret = -EINVAL;
2983 		goto out;
2984 	}
2985 
2986 	allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2987 	if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2988 		mixed = 1;
2989 
2990 	/*
2991 	 * In case of mixed groups both data and meta should be picked,
2992 	 * and identical options should be given for both of them.
2993 	 */
2994 	allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
2995 	if (mixed && (bctl->flags & allowed)) {
2996 		if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2997 		    !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2998 		    memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2999 			printk(KERN_ERR "btrfs: with mixed groups data and "
3000 			       "metadata balance options must be the same\n");
3001 			ret = -EINVAL;
3002 			goto out;
3003 		}
3004 	}
3005 
3006 	num_devices = fs_info->fs_devices->num_devices;
3007 	btrfs_dev_replace_lock(&fs_info->dev_replace);
3008 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3009 		BUG_ON(num_devices < 1);
3010 		num_devices--;
3011 	}
3012 	btrfs_dev_replace_unlock(&fs_info->dev_replace);
3013 	allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3014 	if (num_devices == 1)
3015 		allowed |= BTRFS_BLOCK_GROUP_DUP;
3016 	else if (num_devices < 4)
3017 		allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3018 	else
3019 		allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3020 				BTRFS_BLOCK_GROUP_RAID10);
3021 
3022 	if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3023 	    (!alloc_profile_is_valid(bctl->data.target, 1) ||
3024 	     (bctl->data.target & ~allowed))) {
3025 		printk(KERN_ERR "btrfs: unable to start balance with target "
3026 		       "data profile %llu\n",
3027 		       (unsigned long long)bctl->data.target);
3028 		ret = -EINVAL;
3029 		goto out;
3030 	}
3031 	if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3032 	    (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3033 	     (bctl->meta.target & ~allowed))) {
3034 		printk(KERN_ERR "btrfs: unable to start balance with target "
3035 		       "metadata profile %llu\n",
3036 		       (unsigned long long)bctl->meta.target);
3037 		ret = -EINVAL;
3038 		goto out;
3039 	}
3040 	if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3041 	    (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3042 	     (bctl->sys.target & ~allowed))) {
3043 		printk(KERN_ERR "btrfs: unable to start balance with target "
3044 		       "system profile %llu\n",
3045 		       (unsigned long long)bctl->sys.target);
3046 		ret = -EINVAL;
3047 		goto out;
3048 	}
3049 
3050 	/* allow dup'ed data chunks only in mixed mode */
3051 	if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3052 	    (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3053 		printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3054 		ret = -EINVAL;
3055 		goto out;
3056 	}
3057 
3058 	/* allow to reduce meta or sys integrity only if force set */
3059 	allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3060 			BTRFS_BLOCK_GROUP_RAID10;
3061 	if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3062 	     (fs_info->avail_system_alloc_bits & allowed) &&
3063 	     !(bctl->sys.target & allowed)) ||
3064 	    ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3065 	     (fs_info->avail_metadata_alloc_bits & allowed) &&
3066 	     !(bctl->meta.target & allowed))) {
3067 		if (bctl->flags & BTRFS_BALANCE_FORCE) {
3068 			printk(KERN_INFO "btrfs: force reducing metadata "
3069 			       "integrity\n");
3070 		} else {
3071 			printk(KERN_ERR "btrfs: balance will reduce metadata "
3072 			       "integrity, use force if you want this\n");
3073 			ret = -EINVAL;
3074 			goto out;
3075 		}
3076 	}
3077 
3078 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3079 		int num_tolerated_disk_barrier_failures;
3080 		u64 target = bctl->sys.target;
3081 
3082 		num_tolerated_disk_barrier_failures =
3083 			btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3084 		if (num_tolerated_disk_barrier_failures > 0 &&
3085 		    (target &
3086 		     (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3087 		      BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3088 			num_tolerated_disk_barrier_failures = 0;
3089 		else if (num_tolerated_disk_barrier_failures > 1 &&
3090 			 (target &
3091 			  (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3092 			num_tolerated_disk_barrier_failures = 1;
3093 
3094 		fs_info->num_tolerated_disk_barrier_failures =
3095 			num_tolerated_disk_barrier_failures;
3096 	}
3097 
3098 	ret = insert_balance_item(fs_info->tree_root, bctl);
3099 	if (ret && ret != -EEXIST)
3100 		goto out;
3101 
3102 	if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3103 		BUG_ON(ret == -EEXIST);
3104 		set_balance_control(bctl);
3105 	} else {
3106 		BUG_ON(ret != -EEXIST);
3107 		spin_lock(&fs_info->balance_lock);
3108 		update_balance_args(bctl);
3109 		spin_unlock(&fs_info->balance_lock);
3110 	}
3111 
3112 	atomic_inc(&fs_info->balance_running);
3113 	mutex_unlock(&fs_info->balance_mutex);
3114 
3115 	ret = __btrfs_balance(fs_info);
3116 
3117 	mutex_lock(&fs_info->balance_mutex);
3118 	atomic_dec(&fs_info->balance_running);
3119 
3120 	if (bargs) {
3121 		memset(bargs, 0, sizeof(*bargs));
3122 		update_ioctl_balance_args(fs_info, 0, bargs);
3123 	}
3124 
3125 	if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3126 	    balance_need_close(fs_info)) {
3127 		__cancel_balance(fs_info);
3128 	}
3129 
3130 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3131 		fs_info->num_tolerated_disk_barrier_failures =
3132 			btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3133 	}
3134 
3135 	wake_up(&fs_info->balance_wait_q);
3136 
3137 	return ret;
3138 out:
3139 	if (bctl->flags & BTRFS_BALANCE_RESUME)
3140 		__cancel_balance(fs_info);
3141 	else
3142 		kfree(bctl);
3143 	return ret;
3144 }
3145 
3146 static int balance_kthread(void *data)
3147 {
3148 	struct btrfs_fs_info *fs_info = data;
3149 	int ret = 0;
3150 
3151 	mutex_lock(&fs_info->volume_mutex);
3152 	mutex_lock(&fs_info->balance_mutex);
3153 
3154 	if (fs_info->balance_ctl) {
3155 		printk(KERN_INFO "btrfs: continuing balance\n");
3156 		ret = btrfs_balance(fs_info->balance_ctl, NULL);
3157 	}
3158 
3159 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3160 	mutex_unlock(&fs_info->balance_mutex);
3161 	mutex_unlock(&fs_info->volume_mutex);
3162 
3163 	return ret;
3164 }
3165 
3166 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3167 {
3168 	struct task_struct *tsk;
3169 
3170 	spin_lock(&fs_info->balance_lock);
3171 	if (!fs_info->balance_ctl) {
3172 		spin_unlock(&fs_info->balance_lock);
3173 		return 0;
3174 	}
3175 	spin_unlock(&fs_info->balance_lock);
3176 
3177 	if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3178 		printk(KERN_INFO "btrfs: force skipping balance\n");
3179 		return 0;
3180 	}
3181 
3182 	WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3183 	tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3184 	if (IS_ERR(tsk))
3185 		return PTR_ERR(tsk);
3186 
3187 	return 0;
3188 }
3189 
3190 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3191 {
3192 	struct btrfs_balance_control *bctl;
3193 	struct btrfs_balance_item *item;
3194 	struct btrfs_disk_balance_args disk_bargs;
3195 	struct btrfs_path *path;
3196 	struct extent_buffer *leaf;
3197 	struct btrfs_key key;
3198 	int ret;
3199 
3200 	path = btrfs_alloc_path();
3201 	if (!path)
3202 		return -ENOMEM;
3203 
3204 	key.objectid = BTRFS_BALANCE_OBJECTID;
3205 	key.type = BTRFS_BALANCE_ITEM_KEY;
3206 	key.offset = 0;
3207 
3208 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3209 	if (ret < 0)
3210 		goto out;
3211 	if (ret > 0) { /* ret = -ENOENT; */
3212 		ret = 0;
3213 		goto out;
3214 	}
3215 
3216 	bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3217 	if (!bctl) {
3218 		ret = -ENOMEM;
3219 		goto out;
3220 	}
3221 
3222 	leaf = path->nodes[0];
3223 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3224 
3225 	bctl->fs_info = fs_info;
3226 	bctl->flags = btrfs_balance_flags(leaf, item);
3227 	bctl->flags |= BTRFS_BALANCE_RESUME;
3228 
3229 	btrfs_balance_data(leaf, item, &disk_bargs);
3230 	btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3231 	btrfs_balance_meta(leaf, item, &disk_bargs);
3232 	btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3233 	btrfs_balance_sys(leaf, item, &disk_bargs);
3234 	btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3235 
3236 	mutex_lock(&fs_info->volume_mutex);
3237 	mutex_lock(&fs_info->balance_mutex);
3238 
3239 	set_balance_control(bctl);
3240 
3241 	mutex_unlock(&fs_info->balance_mutex);
3242 	mutex_unlock(&fs_info->volume_mutex);
3243 out:
3244 	btrfs_free_path(path);
3245 	return ret;
3246 }
3247 
3248 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3249 {
3250 	int ret = 0;
3251 
3252 	mutex_lock(&fs_info->balance_mutex);
3253 	if (!fs_info->balance_ctl) {
3254 		mutex_unlock(&fs_info->balance_mutex);
3255 		return -ENOTCONN;
3256 	}
3257 
3258 	if (atomic_read(&fs_info->balance_running)) {
3259 		atomic_inc(&fs_info->balance_pause_req);
3260 		mutex_unlock(&fs_info->balance_mutex);
3261 
3262 		wait_event(fs_info->balance_wait_q,
3263 			   atomic_read(&fs_info->balance_running) == 0);
3264 
3265 		mutex_lock(&fs_info->balance_mutex);
3266 		/* we are good with balance_ctl ripped off from under us */
3267 		BUG_ON(atomic_read(&fs_info->balance_running));
3268 		atomic_dec(&fs_info->balance_pause_req);
3269 	} else {
3270 		ret = -ENOTCONN;
3271 	}
3272 
3273 	mutex_unlock(&fs_info->balance_mutex);
3274 	return ret;
3275 }
3276 
3277 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3278 {
3279 	mutex_lock(&fs_info->balance_mutex);
3280 	if (!fs_info->balance_ctl) {
3281 		mutex_unlock(&fs_info->balance_mutex);
3282 		return -ENOTCONN;
3283 	}
3284 
3285 	atomic_inc(&fs_info->balance_cancel_req);
3286 	/*
3287 	 * if we are running just wait and return, balance item is
3288 	 * deleted in btrfs_balance in this case
3289 	 */
3290 	if (atomic_read(&fs_info->balance_running)) {
3291 		mutex_unlock(&fs_info->balance_mutex);
3292 		wait_event(fs_info->balance_wait_q,
3293 			   atomic_read(&fs_info->balance_running) == 0);
3294 		mutex_lock(&fs_info->balance_mutex);
3295 	} else {
3296 		/* __cancel_balance needs volume_mutex */
3297 		mutex_unlock(&fs_info->balance_mutex);
3298 		mutex_lock(&fs_info->volume_mutex);
3299 		mutex_lock(&fs_info->balance_mutex);
3300 
3301 		if (fs_info->balance_ctl)
3302 			__cancel_balance(fs_info);
3303 
3304 		mutex_unlock(&fs_info->volume_mutex);
3305 	}
3306 
3307 	BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3308 	atomic_dec(&fs_info->balance_cancel_req);
3309 	mutex_unlock(&fs_info->balance_mutex);
3310 	return 0;
3311 }
3312 
3313 /*
3314  * shrinking a device means finding all of the device extents past
3315  * the new size, and then following the back refs to the chunks.
3316  * The chunk relocation code actually frees the device extent
3317  */
3318 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3319 {
3320 	struct btrfs_trans_handle *trans;
3321 	struct btrfs_root *root = device->dev_root;
3322 	struct btrfs_dev_extent *dev_extent = NULL;
3323 	struct btrfs_path *path;
3324 	u64 length;
3325 	u64 chunk_tree;
3326 	u64 chunk_objectid;
3327 	u64 chunk_offset;
3328 	int ret;
3329 	int slot;
3330 	int failed = 0;
3331 	bool retried = false;
3332 	struct extent_buffer *l;
3333 	struct btrfs_key key;
3334 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3335 	u64 old_total = btrfs_super_total_bytes(super_copy);
3336 	u64 old_size = device->total_bytes;
3337 	u64 diff = device->total_bytes - new_size;
3338 
3339 	if (device->is_tgtdev_for_dev_replace)
3340 		return -EINVAL;
3341 
3342 	path = btrfs_alloc_path();
3343 	if (!path)
3344 		return -ENOMEM;
3345 
3346 	path->reada = 2;
3347 
3348 	lock_chunks(root);
3349 
3350 	device->total_bytes = new_size;
3351 	if (device->writeable) {
3352 		device->fs_devices->total_rw_bytes -= diff;
3353 		spin_lock(&root->fs_info->free_chunk_lock);
3354 		root->fs_info->free_chunk_space -= diff;
3355 		spin_unlock(&root->fs_info->free_chunk_lock);
3356 	}
3357 	unlock_chunks(root);
3358 
3359 again:
3360 	key.objectid = device->devid;
3361 	key.offset = (u64)-1;
3362 	key.type = BTRFS_DEV_EXTENT_KEY;
3363 
3364 	do {
3365 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3366 		if (ret < 0)
3367 			goto done;
3368 
3369 		ret = btrfs_previous_item(root, path, 0, key.type);
3370 		if (ret < 0)
3371 			goto done;
3372 		if (ret) {
3373 			ret = 0;
3374 			btrfs_release_path(path);
3375 			break;
3376 		}
3377 
3378 		l = path->nodes[0];
3379 		slot = path->slots[0];
3380 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3381 
3382 		if (key.objectid != device->devid) {
3383 			btrfs_release_path(path);
3384 			break;
3385 		}
3386 
3387 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3388 		length = btrfs_dev_extent_length(l, dev_extent);
3389 
3390 		if (key.offset + length <= new_size) {
3391 			btrfs_release_path(path);
3392 			break;
3393 		}
3394 
3395 		chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3396 		chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3397 		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3398 		btrfs_release_path(path);
3399 
3400 		ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3401 					   chunk_offset);
3402 		if (ret && ret != -ENOSPC)
3403 			goto done;
3404 		if (ret == -ENOSPC)
3405 			failed++;
3406 	} while (key.offset-- > 0);
3407 
3408 	if (failed && !retried) {
3409 		failed = 0;
3410 		retried = true;
3411 		goto again;
3412 	} else if (failed && retried) {
3413 		ret = -ENOSPC;
3414 		lock_chunks(root);
3415 
3416 		device->total_bytes = old_size;
3417 		if (device->writeable)
3418 			device->fs_devices->total_rw_bytes += diff;
3419 		spin_lock(&root->fs_info->free_chunk_lock);
3420 		root->fs_info->free_chunk_space += diff;
3421 		spin_unlock(&root->fs_info->free_chunk_lock);
3422 		unlock_chunks(root);
3423 		goto done;
3424 	}
3425 
3426 	/* Shrinking succeeded, else we would be at "done". */
3427 	trans = btrfs_start_transaction(root, 0);
3428 	if (IS_ERR(trans)) {
3429 		ret = PTR_ERR(trans);
3430 		goto done;
3431 	}
3432 
3433 	lock_chunks(root);
3434 
3435 	device->disk_total_bytes = new_size;
3436 	/* Now btrfs_update_device() will change the on-disk size. */
3437 	ret = btrfs_update_device(trans, device);
3438 	if (ret) {
3439 		unlock_chunks(root);
3440 		btrfs_end_transaction(trans, root);
3441 		goto done;
3442 	}
3443 	WARN_ON(diff > old_total);
3444 	btrfs_set_super_total_bytes(super_copy, old_total - diff);
3445 	unlock_chunks(root);
3446 	btrfs_end_transaction(trans, root);
3447 done:
3448 	btrfs_free_path(path);
3449 	return ret;
3450 }
3451 
3452 static int btrfs_add_system_chunk(struct btrfs_root *root,
3453 			   struct btrfs_key *key,
3454 			   struct btrfs_chunk *chunk, int item_size)
3455 {
3456 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3457 	struct btrfs_disk_key disk_key;
3458 	u32 array_size;
3459 	u8 *ptr;
3460 
3461 	array_size = btrfs_super_sys_array_size(super_copy);
3462 	if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3463 		return -EFBIG;
3464 
3465 	ptr = super_copy->sys_chunk_array + array_size;
3466 	btrfs_cpu_key_to_disk(&disk_key, key);
3467 	memcpy(ptr, &disk_key, sizeof(disk_key));
3468 	ptr += sizeof(disk_key);
3469 	memcpy(ptr, chunk, item_size);
3470 	item_size += sizeof(disk_key);
3471 	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3472 	return 0;
3473 }
3474 
3475 /*
3476  * sort the devices in descending order by max_avail, total_avail
3477  */
3478 static int btrfs_cmp_device_info(const void *a, const void *b)
3479 {
3480 	const struct btrfs_device_info *di_a = a;
3481 	const struct btrfs_device_info *di_b = b;
3482 
3483 	if (di_a->max_avail > di_b->max_avail)
3484 		return -1;
3485 	if (di_a->max_avail < di_b->max_avail)
3486 		return 1;
3487 	if (di_a->total_avail > di_b->total_avail)
3488 		return -1;
3489 	if (di_a->total_avail < di_b->total_avail)
3490 		return 1;
3491 	return 0;
3492 }
3493 
3494 struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3495 	{ 2, 1, 0, 4, 2, 2 /* raid10 */ },
3496 	{ 1, 1, 2, 2, 2, 2 /* raid1 */ },
3497 	{ 1, 2, 1, 1, 1, 2 /* dup */ },
3498 	{ 1, 1, 0, 2, 1, 1 /* raid0 */ },
3499 	{ 1, 1, 0, 1, 1, 1 /* single */ },
3500 };
3501 
3502 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3503 			       struct btrfs_root *extent_root,
3504 			       struct map_lookup **map_ret,
3505 			       u64 *num_bytes_out, u64 *stripe_size_out,
3506 			       u64 start, u64 type)
3507 {
3508 	struct btrfs_fs_info *info = extent_root->fs_info;
3509 	struct btrfs_fs_devices *fs_devices = info->fs_devices;
3510 	struct list_head *cur;
3511 	struct map_lookup *map = NULL;
3512 	struct extent_map_tree *em_tree;
3513 	struct extent_map *em;
3514 	struct btrfs_device_info *devices_info = NULL;
3515 	u64 total_avail;
3516 	int num_stripes;	/* total number of stripes to allocate */
3517 	int sub_stripes;	/* sub_stripes info for map */
3518 	int dev_stripes;	/* stripes per dev */
3519 	int devs_max;		/* max devs to use */
3520 	int devs_min;		/* min devs needed */
3521 	int devs_increment;	/* ndevs has to be a multiple of this */
3522 	int ncopies;		/* how many copies to data has */
3523 	int ret;
3524 	u64 max_stripe_size;
3525 	u64 max_chunk_size;
3526 	u64 stripe_size;
3527 	u64 num_bytes;
3528 	int ndevs;
3529 	int i;
3530 	int j;
3531 	int index;
3532 
3533 	BUG_ON(!alloc_profile_is_valid(type, 0));
3534 
3535 	if (list_empty(&fs_devices->alloc_list))
3536 		return -ENOSPC;
3537 
3538 	index = __get_raid_index(type);
3539 
3540 	sub_stripes = btrfs_raid_array[index].sub_stripes;
3541 	dev_stripes = btrfs_raid_array[index].dev_stripes;
3542 	devs_max = btrfs_raid_array[index].devs_max;
3543 	devs_min = btrfs_raid_array[index].devs_min;
3544 	devs_increment = btrfs_raid_array[index].devs_increment;
3545 	ncopies = btrfs_raid_array[index].ncopies;
3546 
3547 	if (type & BTRFS_BLOCK_GROUP_DATA) {
3548 		max_stripe_size = 1024 * 1024 * 1024;
3549 		max_chunk_size = 10 * max_stripe_size;
3550 	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3551 		/* for larger filesystems, use larger metadata chunks */
3552 		if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3553 			max_stripe_size = 1024 * 1024 * 1024;
3554 		else
3555 			max_stripe_size = 256 * 1024 * 1024;
3556 		max_chunk_size = max_stripe_size;
3557 	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3558 		max_stripe_size = 32 * 1024 * 1024;
3559 		max_chunk_size = 2 * max_stripe_size;
3560 	} else {
3561 		printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3562 		       type);
3563 		BUG_ON(1);
3564 	}
3565 
3566 	/* we don't want a chunk larger than 10% of writeable space */
3567 	max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3568 			     max_chunk_size);
3569 
3570 	devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3571 			       GFP_NOFS);
3572 	if (!devices_info)
3573 		return -ENOMEM;
3574 
3575 	cur = fs_devices->alloc_list.next;
3576 
3577 	/*
3578 	 * in the first pass through the devices list, we gather information
3579 	 * about the available holes on each device.
3580 	 */
3581 	ndevs = 0;
3582 	while (cur != &fs_devices->alloc_list) {
3583 		struct btrfs_device *device;
3584 		u64 max_avail;
3585 		u64 dev_offset;
3586 
3587 		device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3588 
3589 		cur = cur->next;
3590 
3591 		if (!device->writeable) {
3592 			WARN(1, KERN_ERR
3593 			       "btrfs: read-only device in alloc_list\n");
3594 			continue;
3595 		}
3596 
3597 		if (!device->in_fs_metadata ||
3598 		    device->is_tgtdev_for_dev_replace)
3599 			continue;
3600 
3601 		if (device->total_bytes > device->bytes_used)
3602 			total_avail = device->total_bytes - device->bytes_used;
3603 		else
3604 			total_avail = 0;
3605 
3606 		/* If there is no space on this device, skip it. */
3607 		if (total_avail == 0)
3608 			continue;
3609 
3610 		ret = find_free_dev_extent(device,
3611 					   max_stripe_size * dev_stripes,
3612 					   &dev_offset, &max_avail);
3613 		if (ret && ret != -ENOSPC)
3614 			goto error;
3615 
3616 		if (ret == 0)
3617 			max_avail = max_stripe_size * dev_stripes;
3618 
3619 		if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3620 			continue;
3621 
3622 		devices_info[ndevs].dev_offset = dev_offset;
3623 		devices_info[ndevs].max_avail = max_avail;
3624 		devices_info[ndevs].total_avail = total_avail;
3625 		devices_info[ndevs].dev = device;
3626 		++ndevs;
3627 		WARN_ON(ndevs > fs_devices->rw_devices);
3628 	}
3629 
3630 	/*
3631 	 * now sort the devices by hole size / available space
3632 	 */
3633 	sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3634 	     btrfs_cmp_device_info, NULL);
3635 
3636 	/* round down to number of usable stripes */
3637 	ndevs -= ndevs % devs_increment;
3638 
3639 	if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3640 		ret = -ENOSPC;
3641 		goto error;
3642 	}
3643 
3644 	if (devs_max && ndevs > devs_max)
3645 		ndevs = devs_max;
3646 	/*
3647 	 * the primary goal is to maximize the number of stripes, so use as many
3648 	 * devices as possible, even if the stripes are not maximum sized.
3649 	 */
3650 	stripe_size = devices_info[ndevs-1].max_avail;
3651 	num_stripes = ndevs * dev_stripes;
3652 
3653 	if (stripe_size * ndevs > max_chunk_size * ncopies) {
3654 		stripe_size = max_chunk_size * ncopies;
3655 		do_div(stripe_size, ndevs);
3656 	}
3657 
3658 	do_div(stripe_size, dev_stripes);
3659 
3660 	/* align to BTRFS_STRIPE_LEN */
3661 	do_div(stripe_size, BTRFS_STRIPE_LEN);
3662 	stripe_size *= BTRFS_STRIPE_LEN;
3663 
3664 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3665 	if (!map) {
3666 		ret = -ENOMEM;
3667 		goto error;
3668 	}
3669 	map->num_stripes = num_stripes;
3670 
3671 	for (i = 0; i < ndevs; ++i) {
3672 		for (j = 0; j < dev_stripes; ++j) {
3673 			int s = i * dev_stripes + j;
3674 			map->stripes[s].dev = devices_info[i].dev;
3675 			map->stripes[s].physical = devices_info[i].dev_offset +
3676 						   j * stripe_size;
3677 		}
3678 	}
3679 	map->sector_size = extent_root->sectorsize;
3680 	map->stripe_len = BTRFS_STRIPE_LEN;
3681 	map->io_align = BTRFS_STRIPE_LEN;
3682 	map->io_width = BTRFS_STRIPE_LEN;
3683 	map->type = type;
3684 	map->sub_stripes = sub_stripes;
3685 
3686 	*map_ret = map;
3687 	num_bytes = stripe_size * (num_stripes / ncopies);
3688 
3689 	*stripe_size_out = stripe_size;
3690 	*num_bytes_out = num_bytes;
3691 
3692 	trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3693 
3694 	em = alloc_extent_map();
3695 	if (!em) {
3696 		ret = -ENOMEM;
3697 		goto error;
3698 	}
3699 	em->bdev = (struct block_device *)map;
3700 	em->start = start;
3701 	em->len = num_bytes;
3702 	em->block_start = 0;
3703 	em->block_len = em->len;
3704 
3705 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3706 	write_lock(&em_tree->lock);
3707 	ret = add_extent_mapping(em_tree, em);
3708 	write_unlock(&em_tree->lock);
3709 	free_extent_map(em);
3710 	if (ret)
3711 		goto error;
3712 
3713 	ret = btrfs_make_block_group(trans, extent_root, 0, type,
3714 				     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3715 				     start, num_bytes);
3716 	if (ret)
3717 		goto error;
3718 
3719 	for (i = 0; i < map->num_stripes; ++i) {
3720 		struct btrfs_device *device;
3721 		u64 dev_offset;
3722 
3723 		device = map->stripes[i].dev;
3724 		dev_offset = map->stripes[i].physical;
3725 
3726 		ret = btrfs_alloc_dev_extent(trans, device,
3727 				info->chunk_root->root_key.objectid,
3728 				BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3729 				start, dev_offset, stripe_size);
3730 		if (ret) {
3731 			btrfs_abort_transaction(trans, extent_root, ret);
3732 			goto error;
3733 		}
3734 	}
3735 
3736 	kfree(devices_info);
3737 	return 0;
3738 
3739 error:
3740 	kfree(map);
3741 	kfree(devices_info);
3742 	return ret;
3743 }
3744 
3745 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3746 				struct btrfs_root *extent_root,
3747 				struct map_lookup *map, u64 chunk_offset,
3748 				u64 chunk_size, u64 stripe_size)
3749 {
3750 	u64 dev_offset;
3751 	struct btrfs_key key;
3752 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3753 	struct btrfs_device *device;
3754 	struct btrfs_chunk *chunk;
3755 	struct btrfs_stripe *stripe;
3756 	size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3757 	int index = 0;
3758 	int ret;
3759 
3760 	chunk = kzalloc(item_size, GFP_NOFS);
3761 	if (!chunk)
3762 		return -ENOMEM;
3763 
3764 	index = 0;
3765 	while (index < map->num_stripes) {
3766 		device = map->stripes[index].dev;
3767 		device->bytes_used += stripe_size;
3768 		ret = btrfs_update_device(trans, device);
3769 		if (ret)
3770 			goto out_free;
3771 		index++;
3772 	}
3773 
3774 	spin_lock(&extent_root->fs_info->free_chunk_lock);
3775 	extent_root->fs_info->free_chunk_space -= (stripe_size *
3776 						   map->num_stripes);
3777 	spin_unlock(&extent_root->fs_info->free_chunk_lock);
3778 
3779 	index = 0;
3780 	stripe = &chunk->stripe;
3781 	while (index < map->num_stripes) {
3782 		device = map->stripes[index].dev;
3783 		dev_offset = map->stripes[index].physical;
3784 
3785 		btrfs_set_stack_stripe_devid(stripe, device->devid);
3786 		btrfs_set_stack_stripe_offset(stripe, dev_offset);
3787 		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3788 		stripe++;
3789 		index++;
3790 	}
3791 
3792 	btrfs_set_stack_chunk_length(chunk, chunk_size);
3793 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3794 	btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3795 	btrfs_set_stack_chunk_type(chunk, map->type);
3796 	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3797 	btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3798 	btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3799 	btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3800 	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3801 
3802 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3803 	key.type = BTRFS_CHUNK_ITEM_KEY;
3804 	key.offset = chunk_offset;
3805 
3806 	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3807 
3808 	if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3809 		/*
3810 		 * TODO: Cleanup of inserted chunk root in case of
3811 		 * failure.
3812 		 */
3813 		ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3814 					     item_size);
3815 	}
3816 
3817 out_free:
3818 	kfree(chunk);
3819 	return ret;
3820 }
3821 
3822 /*
3823  * Chunk allocation falls into two parts. The first part does works
3824  * that make the new allocated chunk useable, but not do any operation
3825  * that modifies the chunk tree. The second part does the works that
3826  * require modifying the chunk tree. This division is important for the
3827  * bootstrap process of adding storage to a seed btrfs.
3828  */
3829 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3830 		      struct btrfs_root *extent_root, u64 type)
3831 {
3832 	u64 chunk_offset;
3833 	u64 chunk_size;
3834 	u64 stripe_size;
3835 	struct map_lookup *map;
3836 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3837 	int ret;
3838 
3839 	ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3840 			      &chunk_offset);
3841 	if (ret)
3842 		return ret;
3843 
3844 	ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3845 				  &stripe_size, chunk_offset, type);
3846 	if (ret)
3847 		return ret;
3848 
3849 	ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3850 				   chunk_size, stripe_size);
3851 	if (ret)
3852 		return ret;
3853 	return 0;
3854 }
3855 
3856 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3857 					 struct btrfs_root *root,
3858 					 struct btrfs_device *device)
3859 {
3860 	u64 chunk_offset;
3861 	u64 sys_chunk_offset;
3862 	u64 chunk_size;
3863 	u64 sys_chunk_size;
3864 	u64 stripe_size;
3865 	u64 sys_stripe_size;
3866 	u64 alloc_profile;
3867 	struct map_lookup *map;
3868 	struct map_lookup *sys_map;
3869 	struct btrfs_fs_info *fs_info = root->fs_info;
3870 	struct btrfs_root *extent_root = fs_info->extent_root;
3871 	int ret;
3872 
3873 	ret = find_next_chunk(fs_info->chunk_root,
3874 			      BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3875 	if (ret)
3876 		return ret;
3877 
3878 	alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3879 				fs_info->avail_metadata_alloc_bits;
3880 	alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3881 
3882 	ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3883 				  &stripe_size, chunk_offset, alloc_profile);
3884 	if (ret)
3885 		return ret;
3886 
3887 	sys_chunk_offset = chunk_offset + chunk_size;
3888 
3889 	alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3890 				fs_info->avail_system_alloc_bits;
3891 	alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3892 
3893 	ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3894 				  &sys_chunk_size, &sys_stripe_size,
3895 				  sys_chunk_offset, alloc_profile);
3896 	if (ret) {
3897 		btrfs_abort_transaction(trans, root, ret);
3898 		goto out;
3899 	}
3900 
3901 	ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3902 	if (ret) {
3903 		btrfs_abort_transaction(trans, root, ret);
3904 		goto out;
3905 	}
3906 
3907 	/*
3908 	 * Modifying chunk tree needs allocating new blocks from both
3909 	 * system block group and metadata block group. So we only can
3910 	 * do operations require modifying the chunk tree after both
3911 	 * block groups were created.
3912 	 */
3913 	ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3914 				   chunk_size, stripe_size);
3915 	if (ret) {
3916 		btrfs_abort_transaction(trans, root, ret);
3917 		goto out;
3918 	}
3919 
3920 	ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3921 				   sys_chunk_offset, sys_chunk_size,
3922 				   sys_stripe_size);
3923 	if (ret)
3924 		btrfs_abort_transaction(trans, root, ret);
3925 
3926 out:
3927 
3928 	return ret;
3929 }
3930 
3931 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3932 {
3933 	struct extent_map *em;
3934 	struct map_lookup *map;
3935 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3936 	int readonly = 0;
3937 	int i;
3938 
3939 	read_lock(&map_tree->map_tree.lock);
3940 	em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3941 	read_unlock(&map_tree->map_tree.lock);
3942 	if (!em)
3943 		return 1;
3944 
3945 	if (btrfs_test_opt(root, DEGRADED)) {
3946 		free_extent_map(em);
3947 		return 0;
3948 	}
3949 
3950 	map = (struct map_lookup *)em->bdev;
3951 	for (i = 0; i < map->num_stripes; i++) {
3952 		if (!map->stripes[i].dev->writeable) {
3953 			readonly = 1;
3954 			break;
3955 		}
3956 	}
3957 	free_extent_map(em);
3958 	return readonly;
3959 }
3960 
3961 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3962 {
3963 	extent_map_tree_init(&tree->map_tree);
3964 }
3965 
3966 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3967 {
3968 	struct extent_map *em;
3969 
3970 	while (1) {
3971 		write_lock(&tree->map_tree.lock);
3972 		em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3973 		if (em)
3974 			remove_extent_mapping(&tree->map_tree, em);
3975 		write_unlock(&tree->map_tree.lock);
3976 		if (!em)
3977 			break;
3978 		kfree(em->bdev);
3979 		/* once for us */
3980 		free_extent_map(em);
3981 		/* once for the tree */
3982 		free_extent_map(em);
3983 	}
3984 }
3985 
3986 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
3987 {
3988 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
3989 	struct extent_map *em;
3990 	struct map_lookup *map;
3991 	struct extent_map_tree *em_tree = &map_tree->map_tree;
3992 	int ret;
3993 
3994 	read_lock(&em_tree->lock);
3995 	em = lookup_extent_mapping(em_tree, logical, len);
3996 	read_unlock(&em_tree->lock);
3997 	BUG_ON(!em);
3998 
3999 	BUG_ON(em->start > logical || em->start + em->len < logical);
4000 	map = (struct map_lookup *)em->bdev;
4001 	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4002 		ret = map->num_stripes;
4003 	else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4004 		ret = map->sub_stripes;
4005 	else
4006 		ret = 1;
4007 	free_extent_map(em);
4008 
4009 	btrfs_dev_replace_lock(&fs_info->dev_replace);
4010 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4011 		ret++;
4012 	btrfs_dev_replace_unlock(&fs_info->dev_replace);
4013 
4014 	return ret;
4015 }
4016 
4017 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4018 			    struct map_lookup *map, int first, int num,
4019 			    int optimal, int dev_replace_is_ongoing)
4020 {
4021 	int i;
4022 	int tolerance;
4023 	struct btrfs_device *srcdev;
4024 
4025 	if (dev_replace_is_ongoing &&
4026 	    fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4027 	     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4028 		srcdev = fs_info->dev_replace.srcdev;
4029 	else
4030 		srcdev = NULL;
4031 
4032 	/*
4033 	 * try to avoid the drive that is the source drive for a
4034 	 * dev-replace procedure, only choose it if no other non-missing
4035 	 * mirror is available
4036 	 */
4037 	for (tolerance = 0; tolerance < 2; tolerance++) {
4038 		if (map->stripes[optimal].dev->bdev &&
4039 		    (tolerance || map->stripes[optimal].dev != srcdev))
4040 			return optimal;
4041 		for (i = first; i < first + num; i++) {
4042 			if (map->stripes[i].dev->bdev &&
4043 			    (tolerance || map->stripes[i].dev != srcdev))
4044 				return i;
4045 		}
4046 	}
4047 
4048 	/* we couldn't find one that doesn't fail.  Just return something
4049 	 * and the io error handling code will clean up eventually
4050 	 */
4051 	return optimal;
4052 }
4053 
4054 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4055 			     u64 logical, u64 *length,
4056 			     struct btrfs_bio **bbio_ret,
4057 			     int mirror_num)
4058 {
4059 	struct extent_map *em;
4060 	struct map_lookup *map;
4061 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4062 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4063 	u64 offset;
4064 	u64 stripe_offset;
4065 	u64 stripe_end_offset;
4066 	u64 stripe_nr;
4067 	u64 stripe_nr_orig;
4068 	u64 stripe_nr_end;
4069 	int stripe_index;
4070 	int i;
4071 	int ret = 0;
4072 	int num_stripes;
4073 	int max_errors = 0;
4074 	struct btrfs_bio *bbio = NULL;
4075 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4076 	int dev_replace_is_ongoing = 0;
4077 	int num_alloc_stripes;
4078 	int patch_the_first_stripe_for_dev_replace = 0;
4079 	u64 physical_to_patch_in_first_stripe = 0;
4080 
4081 	read_lock(&em_tree->lock);
4082 	em = lookup_extent_mapping(em_tree, logical, *length);
4083 	read_unlock(&em_tree->lock);
4084 
4085 	if (!em) {
4086 		printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
4087 		       (unsigned long long)logical,
4088 		       (unsigned long long)*length);
4089 		BUG();
4090 	}
4091 
4092 	BUG_ON(em->start > logical || em->start + em->len < logical);
4093 	map = (struct map_lookup *)em->bdev;
4094 	offset = logical - em->start;
4095 
4096 	stripe_nr = offset;
4097 	/*
4098 	 * stripe_nr counts the total number of stripes we have to stride
4099 	 * to get to this block
4100 	 */
4101 	do_div(stripe_nr, map->stripe_len);
4102 
4103 	stripe_offset = stripe_nr * map->stripe_len;
4104 	BUG_ON(offset < stripe_offset);
4105 
4106 	/* stripe_offset is the offset of this block in its stripe*/
4107 	stripe_offset = offset - stripe_offset;
4108 
4109 	if (rw & REQ_DISCARD)
4110 		*length = min_t(u64, em->len - offset, *length);
4111 	else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4112 		/* we limit the length of each bio to what fits in a stripe */
4113 		*length = min_t(u64, em->len - offset,
4114 				map->stripe_len - stripe_offset);
4115 	} else {
4116 		*length = em->len - offset;
4117 	}
4118 
4119 	if (!bbio_ret)
4120 		goto out;
4121 
4122 	btrfs_dev_replace_lock(dev_replace);
4123 	dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4124 	if (!dev_replace_is_ongoing)
4125 		btrfs_dev_replace_unlock(dev_replace);
4126 
4127 	if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4128 	    !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4129 	    dev_replace->tgtdev != NULL) {
4130 		/*
4131 		 * in dev-replace case, for repair case (that's the only
4132 		 * case where the mirror is selected explicitly when
4133 		 * calling btrfs_map_block), blocks left of the left cursor
4134 		 * can also be read from the target drive.
4135 		 * For REQ_GET_READ_MIRRORS, the target drive is added as
4136 		 * the last one to the array of stripes. For READ, it also
4137 		 * needs to be supported using the same mirror number.
4138 		 * If the requested block is not left of the left cursor,
4139 		 * EIO is returned. This can happen because btrfs_num_copies()
4140 		 * returns one more in the dev-replace case.
4141 		 */
4142 		u64 tmp_length = *length;
4143 		struct btrfs_bio *tmp_bbio = NULL;
4144 		int tmp_num_stripes;
4145 		u64 srcdev_devid = dev_replace->srcdev->devid;
4146 		int index_srcdev = 0;
4147 		int found = 0;
4148 		u64 physical_of_found = 0;
4149 
4150 		ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4151 			     logical, &tmp_length, &tmp_bbio, 0);
4152 		if (ret) {
4153 			WARN_ON(tmp_bbio != NULL);
4154 			goto out;
4155 		}
4156 
4157 		tmp_num_stripes = tmp_bbio->num_stripes;
4158 		if (mirror_num > tmp_num_stripes) {
4159 			/*
4160 			 * REQ_GET_READ_MIRRORS does not contain this
4161 			 * mirror, that means that the requested area
4162 			 * is not left of the left cursor
4163 			 */
4164 			ret = -EIO;
4165 			kfree(tmp_bbio);
4166 			goto out;
4167 		}
4168 
4169 		/*
4170 		 * process the rest of the function using the mirror_num
4171 		 * of the source drive. Therefore look it up first.
4172 		 * At the end, patch the device pointer to the one of the
4173 		 * target drive.
4174 		 */
4175 		for (i = 0; i < tmp_num_stripes; i++) {
4176 			if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4177 				/*
4178 				 * In case of DUP, in order to keep it
4179 				 * simple, only add the mirror with the
4180 				 * lowest physical address
4181 				 */
4182 				if (found &&
4183 				    physical_of_found <=
4184 				     tmp_bbio->stripes[i].physical)
4185 					continue;
4186 				index_srcdev = i;
4187 				found = 1;
4188 				physical_of_found =
4189 					tmp_bbio->stripes[i].physical;
4190 			}
4191 		}
4192 
4193 		if (found) {
4194 			mirror_num = index_srcdev + 1;
4195 			patch_the_first_stripe_for_dev_replace = 1;
4196 			physical_to_patch_in_first_stripe = physical_of_found;
4197 		} else {
4198 			WARN_ON(1);
4199 			ret = -EIO;
4200 			kfree(tmp_bbio);
4201 			goto out;
4202 		}
4203 
4204 		kfree(tmp_bbio);
4205 	} else if (mirror_num > map->num_stripes) {
4206 		mirror_num = 0;
4207 	}
4208 
4209 	num_stripes = 1;
4210 	stripe_index = 0;
4211 	stripe_nr_orig = stripe_nr;
4212 	stripe_nr_end = (offset + *length + map->stripe_len - 1) &
4213 			(~(map->stripe_len - 1));
4214 	do_div(stripe_nr_end, map->stripe_len);
4215 	stripe_end_offset = stripe_nr_end * map->stripe_len -
4216 			    (offset + *length);
4217 	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4218 		if (rw & REQ_DISCARD)
4219 			num_stripes = min_t(u64, map->num_stripes,
4220 					    stripe_nr_end - stripe_nr_orig);
4221 		stripe_index = do_div(stripe_nr, map->num_stripes);
4222 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4223 		if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4224 			num_stripes = map->num_stripes;
4225 		else if (mirror_num)
4226 			stripe_index = mirror_num - 1;
4227 		else {
4228 			stripe_index = find_live_mirror(fs_info, map, 0,
4229 					    map->num_stripes,
4230 					    current->pid % map->num_stripes,
4231 					    dev_replace_is_ongoing);
4232 			mirror_num = stripe_index + 1;
4233 		}
4234 
4235 	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4236 		if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4237 			num_stripes = map->num_stripes;
4238 		} else if (mirror_num) {
4239 			stripe_index = mirror_num - 1;
4240 		} else {
4241 			mirror_num = 1;
4242 		}
4243 
4244 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4245 		int factor = map->num_stripes / map->sub_stripes;
4246 
4247 		stripe_index = do_div(stripe_nr, factor);
4248 		stripe_index *= map->sub_stripes;
4249 
4250 		if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4251 			num_stripes = map->sub_stripes;
4252 		else if (rw & REQ_DISCARD)
4253 			num_stripes = min_t(u64, map->sub_stripes *
4254 					    (stripe_nr_end - stripe_nr_orig),
4255 					    map->num_stripes);
4256 		else if (mirror_num)
4257 			stripe_index += mirror_num - 1;
4258 		else {
4259 			int old_stripe_index = stripe_index;
4260 			stripe_index = find_live_mirror(fs_info, map,
4261 					      stripe_index,
4262 					      map->sub_stripes, stripe_index +
4263 					      current->pid % map->sub_stripes,
4264 					      dev_replace_is_ongoing);
4265 			mirror_num = stripe_index - old_stripe_index + 1;
4266 		}
4267 	} else {
4268 		/*
4269 		 * after this do_div call, stripe_nr is the number of stripes
4270 		 * on this device we have to walk to find the data, and
4271 		 * stripe_index is the number of our device in the stripe array
4272 		 */
4273 		stripe_index = do_div(stripe_nr, map->num_stripes);
4274 		mirror_num = stripe_index + 1;
4275 	}
4276 	BUG_ON(stripe_index >= map->num_stripes);
4277 
4278 	num_alloc_stripes = num_stripes;
4279 	if (dev_replace_is_ongoing) {
4280 		if (rw & (REQ_WRITE | REQ_DISCARD))
4281 			num_alloc_stripes <<= 1;
4282 		if (rw & REQ_GET_READ_MIRRORS)
4283 			num_alloc_stripes++;
4284 	}
4285 	bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4286 	if (!bbio) {
4287 		ret = -ENOMEM;
4288 		goto out;
4289 	}
4290 	atomic_set(&bbio->error, 0);
4291 
4292 	if (rw & REQ_DISCARD) {
4293 		int factor = 0;
4294 		int sub_stripes = 0;
4295 		u64 stripes_per_dev = 0;
4296 		u32 remaining_stripes = 0;
4297 		u32 last_stripe = 0;
4298 
4299 		if (map->type &
4300 		    (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4301 			if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4302 				sub_stripes = 1;
4303 			else
4304 				sub_stripes = map->sub_stripes;
4305 
4306 			factor = map->num_stripes / sub_stripes;
4307 			stripes_per_dev = div_u64_rem(stripe_nr_end -
4308 						      stripe_nr_orig,
4309 						      factor,
4310 						      &remaining_stripes);
4311 			div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4312 			last_stripe *= sub_stripes;
4313 		}
4314 
4315 		for (i = 0; i < num_stripes; i++) {
4316 			bbio->stripes[i].physical =
4317 				map->stripes[stripe_index].physical +
4318 				stripe_offset + stripe_nr * map->stripe_len;
4319 			bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4320 
4321 			if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4322 					 BTRFS_BLOCK_GROUP_RAID10)) {
4323 				bbio->stripes[i].length = stripes_per_dev *
4324 							  map->stripe_len;
4325 
4326 				if (i / sub_stripes < remaining_stripes)
4327 					bbio->stripes[i].length +=
4328 						map->stripe_len;
4329 
4330 				/*
4331 				 * Special for the first stripe and
4332 				 * the last stripe:
4333 				 *
4334 				 * |-------|...|-------|
4335 				 *     |----------|
4336 				 *    off     end_off
4337 				 */
4338 				if (i < sub_stripes)
4339 					bbio->stripes[i].length -=
4340 						stripe_offset;
4341 
4342 				if (stripe_index >= last_stripe &&
4343 				    stripe_index <= (last_stripe +
4344 						     sub_stripes - 1))
4345 					bbio->stripes[i].length -=
4346 						stripe_end_offset;
4347 
4348 				if (i == sub_stripes - 1)
4349 					stripe_offset = 0;
4350 			} else
4351 				bbio->stripes[i].length = *length;
4352 
4353 			stripe_index++;
4354 			if (stripe_index == map->num_stripes) {
4355 				/* This could only happen for RAID0/10 */
4356 				stripe_index = 0;
4357 				stripe_nr++;
4358 			}
4359 		}
4360 	} else {
4361 		for (i = 0; i < num_stripes; i++) {
4362 			bbio->stripes[i].physical =
4363 				map->stripes[stripe_index].physical +
4364 				stripe_offset +
4365 				stripe_nr * map->stripe_len;
4366 			bbio->stripes[i].dev =
4367 				map->stripes[stripe_index].dev;
4368 			stripe_index++;
4369 		}
4370 	}
4371 
4372 	if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4373 		if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4374 				 BTRFS_BLOCK_GROUP_RAID10 |
4375 				 BTRFS_BLOCK_GROUP_DUP)) {
4376 			max_errors = 1;
4377 		}
4378 	}
4379 
4380 	if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4381 	    dev_replace->tgtdev != NULL) {
4382 		int index_where_to_add;
4383 		u64 srcdev_devid = dev_replace->srcdev->devid;
4384 
4385 		/*
4386 		 * duplicate the write operations while the dev replace
4387 		 * procedure is running. Since the copying of the old disk
4388 		 * to the new disk takes place at run time while the
4389 		 * filesystem is mounted writable, the regular write
4390 		 * operations to the old disk have to be duplicated to go
4391 		 * to the new disk as well.
4392 		 * Note that device->missing is handled by the caller, and
4393 		 * that the write to the old disk is already set up in the
4394 		 * stripes array.
4395 		 */
4396 		index_where_to_add = num_stripes;
4397 		for (i = 0; i < num_stripes; i++) {
4398 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
4399 				/* write to new disk, too */
4400 				struct btrfs_bio_stripe *new =
4401 					bbio->stripes + index_where_to_add;
4402 				struct btrfs_bio_stripe *old =
4403 					bbio->stripes + i;
4404 
4405 				new->physical = old->physical;
4406 				new->length = old->length;
4407 				new->dev = dev_replace->tgtdev;
4408 				index_where_to_add++;
4409 				max_errors++;
4410 			}
4411 		}
4412 		num_stripes = index_where_to_add;
4413 	} else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4414 		   dev_replace->tgtdev != NULL) {
4415 		u64 srcdev_devid = dev_replace->srcdev->devid;
4416 		int index_srcdev = 0;
4417 		int found = 0;
4418 		u64 physical_of_found = 0;
4419 
4420 		/*
4421 		 * During the dev-replace procedure, the target drive can
4422 		 * also be used to read data in case it is needed to repair
4423 		 * a corrupt block elsewhere. This is possible if the
4424 		 * requested area is left of the left cursor. In this area,
4425 		 * the target drive is a full copy of the source drive.
4426 		 */
4427 		for (i = 0; i < num_stripes; i++) {
4428 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
4429 				/*
4430 				 * In case of DUP, in order to keep it
4431 				 * simple, only add the mirror with the
4432 				 * lowest physical address
4433 				 */
4434 				if (found &&
4435 				    physical_of_found <=
4436 				     bbio->stripes[i].physical)
4437 					continue;
4438 				index_srcdev = i;
4439 				found = 1;
4440 				physical_of_found = bbio->stripes[i].physical;
4441 			}
4442 		}
4443 		if (found) {
4444 			u64 length = map->stripe_len;
4445 
4446 			if (physical_of_found + length <=
4447 			    dev_replace->cursor_left) {
4448 				struct btrfs_bio_stripe *tgtdev_stripe =
4449 					bbio->stripes + num_stripes;
4450 
4451 				tgtdev_stripe->physical = physical_of_found;
4452 				tgtdev_stripe->length =
4453 					bbio->stripes[index_srcdev].length;
4454 				tgtdev_stripe->dev = dev_replace->tgtdev;
4455 
4456 				num_stripes++;
4457 			}
4458 		}
4459 	}
4460 
4461 	*bbio_ret = bbio;
4462 	bbio->num_stripes = num_stripes;
4463 	bbio->max_errors = max_errors;
4464 	bbio->mirror_num = mirror_num;
4465 
4466 	/*
4467 	 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4468 	 * mirror_num == num_stripes + 1 && dev_replace target drive is
4469 	 * available as a mirror
4470 	 */
4471 	if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4472 		WARN_ON(num_stripes > 1);
4473 		bbio->stripes[0].dev = dev_replace->tgtdev;
4474 		bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4475 		bbio->mirror_num = map->num_stripes + 1;
4476 	}
4477 out:
4478 	if (dev_replace_is_ongoing)
4479 		btrfs_dev_replace_unlock(dev_replace);
4480 	free_extent_map(em);
4481 	return ret;
4482 }
4483 
4484 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4485 		      u64 logical, u64 *length,
4486 		      struct btrfs_bio **bbio_ret, int mirror_num)
4487 {
4488 	return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4489 				 mirror_num);
4490 }
4491 
4492 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4493 		     u64 chunk_start, u64 physical, u64 devid,
4494 		     u64 **logical, int *naddrs, int *stripe_len)
4495 {
4496 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4497 	struct extent_map *em;
4498 	struct map_lookup *map;
4499 	u64 *buf;
4500 	u64 bytenr;
4501 	u64 length;
4502 	u64 stripe_nr;
4503 	int i, j, nr = 0;
4504 
4505 	read_lock(&em_tree->lock);
4506 	em = lookup_extent_mapping(em_tree, chunk_start, 1);
4507 	read_unlock(&em_tree->lock);
4508 
4509 	BUG_ON(!em || em->start != chunk_start);
4510 	map = (struct map_lookup *)em->bdev;
4511 
4512 	length = em->len;
4513 	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4514 		do_div(length, map->num_stripes / map->sub_stripes);
4515 	else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4516 		do_div(length, map->num_stripes);
4517 
4518 	buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4519 	BUG_ON(!buf); /* -ENOMEM */
4520 
4521 	for (i = 0; i < map->num_stripes; i++) {
4522 		if (devid && map->stripes[i].dev->devid != devid)
4523 			continue;
4524 		if (map->stripes[i].physical > physical ||
4525 		    map->stripes[i].physical + length <= physical)
4526 			continue;
4527 
4528 		stripe_nr = physical - map->stripes[i].physical;
4529 		do_div(stripe_nr, map->stripe_len);
4530 
4531 		if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4532 			stripe_nr = stripe_nr * map->num_stripes + i;
4533 			do_div(stripe_nr, map->sub_stripes);
4534 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4535 			stripe_nr = stripe_nr * map->num_stripes + i;
4536 		}
4537 		bytenr = chunk_start + stripe_nr * map->stripe_len;
4538 		WARN_ON(nr >= map->num_stripes);
4539 		for (j = 0; j < nr; j++) {
4540 			if (buf[j] == bytenr)
4541 				break;
4542 		}
4543 		if (j == nr) {
4544 			WARN_ON(nr >= map->num_stripes);
4545 			buf[nr++] = bytenr;
4546 		}
4547 	}
4548 
4549 	*logical = buf;
4550 	*naddrs = nr;
4551 	*stripe_len = map->stripe_len;
4552 
4553 	free_extent_map(em);
4554 	return 0;
4555 }
4556 
4557 static void *merge_stripe_index_into_bio_private(void *bi_private,
4558 						 unsigned int stripe_index)
4559 {
4560 	/*
4561 	 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4562 	 * at most 1.
4563 	 * The alternative solution (instead of stealing bits from the
4564 	 * pointer) would be to allocate an intermediate structure
4565 	 * that contains the old private pointer plus the stripe_index.
4566 	 */
4567 	BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4568 	BUG_ON(stripe_index > 3);
4569 	return (void *)(((uintptr_t)bi_private) | stripe_index);
4570 }
4571 
4572 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4573 {
4574 	return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4575 }
4576 
4577 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4578 {
4579 	return (unsigned int)((uintptr_t)bi_private) & 3;
4580 }
4581 
4582 static void btrfs_end_bio(struct bio *bio, int err)
4583 {
4584 	struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4585 	int is_orig_bio = 0;
4586 
4587 	if (err) {
4588 		atomic_inc(&bbio->error);
4589 		if (err == -EIO || err == -EREMOTEIO) {
4590 			unsigned int stripe_index =
4591 				extract_stripe_index_from_bio_private(
4592 					bio->bi_private);
4593 			struct btrfs_device *dev;
4594 
4595 			BUG_ON(stripe_index >= bbio->num_stripes);
4596 			dev = bbio->stripes[stripe_index].dev;
4597 			if (dev->bdev) {
4598 				if (bio->bi_rw & WRITE)
4599 					btrfs_dev_stat_inc(dev,
4600 						BTRFS_DEV_STAT_WRITE_ERRS);
4601 				else
4602 					btrfs_dev_stat_inc(dev,
4603 						BTRFS_DEV_STAT_READ_ERRS);
4604 				if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4605 					btrfs_dev_stat_inc(dev,
4606 						BTRFS_DEV_STAT_FLUSH_ERRS);
4607 				btrfs_dev_stat_print_on_error(dev);
4608 			}
4609 		}
4610 	}
4611 
4612 	if (bio == bbio->orig_bio)
4613 		is_orig_bio = 1;
4614 
4615 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
4616 		if (!is_orig_bio) {
4617 			bio_put(bio);
4618 			bio = bbio->orig_bio;
4619 		}
4620 		bio->bi_private = bbio->private;
4621 		bio->bi_end_io = bbio->end_io;
4622 		bio->bi_bdev = (struct block_device *)
4623 					(unsigned long)bbio->mirror_num;
4624 		/* only send an error to the higher layers if it is
4625 		 * beyond the tolerance of the multi-bio
4626 		 */
4627 		if (atomic_read(&bbio->error) > bbio->max_errors) {
4628 			err = -EIO;
4629 		} else {
4630 			/*
4631 			 * this bio is actually up to date, we didn't
4632 			 * go over the max number of errors
4633 			 */
4634 			set_bit(BIO_UPTODATE, &bio->bi_flags);
4635 			err = 0;
4636 		}
4637 		kfree(bbio);
4638 
4639 		bio_endio(bio, err);
4640 	} else if (!is_orig_bio) {
4641 		bio_put(bio);
4642 	}
4643 }
4644 
4645 struct async_sched {
4646 	struct bio *bio;
4647 	int rw;
4648 	struct btrfs_fs_info *info;
4649 	struct btrfs_work work;
4650 };
4651 
4652 /*
4653  * see run_scheduled_bios for a description of why bios are collected for
4654  * async submit.
4655  *
4656  * This will add one bio to the pending list for a device and make sure
4657  * the work struct is scheduled.
4658  */
4659 static noinline void schedule_bio(struct btrfs_root *root,
4660 				 struct btrfs_device *device,
4661 				 int rw, struct bio *bio)
4662 {
4663 	int should_queue = 1;
4664 	struct btrfs_pending_bios *pending_bios;
4665 
4666 	/* don't bother with additional async steps for reads, right now */
4667 	if (!(rw & REQ_WRITE)) {
4668 		bio_get(bio);
4669 		btrfsic_submit_bio(rw, bio);
4670 		bio_put(bio);
4671 		return;
4672 	}
4673 
4674 	/*
4675 	 * nr_async_bios allows us to reliably return congestion to the
4676 	 * higher layers.  Otherwise, the async bio makes it appear we have
4677 	 * made progress against dirty pages when we've really just put it
4678 	 * on a queue for later
4679 	 */
4680 	atomic_inc(&root->fs_info->nr_async_bios);
4681 	WARN_ON(bio->bi_next);
4682 	bio->bi_next = NULL;
4683 	bio->bi_rw |= rw;
4684 
4685 	spin_lock(&device->io_lock);
4686 	if (bio->bi_rw & REQ_SYNC)
4687 		pending_bios = &device->pending_sync_bios;
4688 	else
4689 		pending_bios = &device->pending_bios;
4690 
4691 	if (pending_bios->tail)
4692 		pending_bios->tail->bi_next = bio;
4693 
4694 	pending_bios->tail = bio;
4695 	if (!pending_bios->head)
4696 		pending_bios->head = bio;
4697 	if (device->running_pending)
4698 		should_queue = 0;
4699 
4700 	spin_unlock(&device->io_lock);
4701 
4702 	if (should_queue)
4703 		btrfs_queue_worker(&root->fs_info->submit_workers,
4704 				   &device->work);
4705 }
4706 
4707 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
4708 		       sector_t sector)
4709 {
4710 	struct bio_vec *prev;
4711 	struct request_queue *q = bdev_get_queue(bdev);
4712 	unsigned short max_sectors = queue_max_sectors(q);
4713 	struct bvec_merge_data bvm = {
4714 		.bi_bdev = bdev,
4715 		.bi_sector = sector,
4716 		.bi_rw = bio->bi_rw,
4717 	};
4718 
4719 	if (bio->bi_vcnt == 0) {
4720 		WARN_ON(1);
4721 		return 1;
4722 	}
4723 
4724 	prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
4725 	if ((bio->bi_size >> 9) > max_sectors)
4726 		return 0;
4727 
4728 	if (!q->merge_bvec_fn)
4729 		return 1;
4730 
4731 	bvm.bi_size = bio->bi_size - prev->bv_len;
4732 	if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
4733 		return 0;
4734 	return 1;
4735 }
4736 
4737 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4738 			      struct bio *bio, u64 physical, int dev_nr,
4739 			      int rw, int async)
4740 {
4741 	struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
4742 
4743 	bio->bi_private = bbio;
4744 	bio->bi_private = merge_stripe_index_into_bio_private(
4745 			bio->bi_private, (unsigned int)dev_nr);
4746 	bio->bi_end_io = btrfs_end_bio;
4747 	bio->bi_sector = physical >> 9;
4748 #ifdef DEBUG
4749 	{
4750 		struct rcu_string *name;
4751 
4752 		rcu_read_lock();
4753 		name = rcu_dereference(dev->name);
4754 		pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
4755 			 "(%s id %llu), size=%u\n", rw,
4756 			 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4757 			 name->str, dev->devid, bio->bi_size);
4758 		rcu_read_unlock();
4759 	}
4760 #endif
4761 	bio->bi_bdev = dev->bdev;
4762 	if (async)
4763 		schedule_bio(root, dev, rw, bio);
4764 	else
4765 		btrfsic_submit_bio(rw, bio);
4766 }
4767 
4768 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4769 			      struct bio *first_bio, struct btrfs_device *dev,
4770 			      int dev_nr, int rw, int async)
4771 {
4772 	struct bio_vec *bvec = first_bio->bi_io_vec;
4773 	struct bio *bio;
4774 	int nr_vecs = bio_get_nr_vecs(dev->bdev);
4775 	u64 physical = bbio->stripes[dev_nr].physical;
4776 
4777 again:
4778 	bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
4779 	if (!bio)
4780 		return -ENOMEM;
4781 
4782 	while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
4783 		if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
4784 				 bvec->bv_offset) < bvec->bv_len) {
4785 			u64 len = bio->bi_size;
4786 
4787 			atomic_inc(&bbio->stripes_pending);
4788 			submit_stripe_bio(root, bbio, bio, physical, dev_nr,
4789 					  rw, async);
4790 			physical += len;
4791 			goto again;
4792 		}
4793 		bvec++;
4794 	}
4795 
4796 	submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
4797 	return 0;
4798 }
4799 
4800 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
4801 {
4802 	atomic_inc(&bbio->error);
4803 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
4804 		bio->bi_private = bbio->private;
4805 		bio->bi_end_io = bbio->end_io;
4806 		bio->bi_bdev = (struct block_device *)
4807 			(unsigned long)bbio->mirror_num;
4808 		bio->bi_sector = logical >> 9;
4809 		kfree(bbio);
4810 		bio_endio(bio, -EIO);
4811 	}
4812 }
4813 
4814 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4815 		  int mirror_num, int async_submit)
4816 {
4817 	struct btrfs_device *dev;
4818 	struct bio *first_bio = bio;
4819 	u64 logical = (u64)bio->bi_sector << 9;
4820 	u64 length = 0;
4821 	u64 map_length;
4822 	int ret;
4823 	int dev_nr = 0;
4824 	int total_devs = 1;
4825 	struct btrfs_bio *bbio = NULL;
4826 
4827 	length = bio->bi_size;
4828 	map_length = length;
4829 
4830 	ret = btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
4831 			      mirror_num);
4832 	if (ret)
4833 		return ret;
4834 
4835 	total_devs = bbio->num_stripes;
4836 	if (map_length < length) {
4837 		printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
4838 		       "len %llu\n", (unsigned long long)logical,
4839 		       (unsigned long long)length,
4840 		       (unsigned long long)map_length);
4841 		BUG();
4842 	}
4843 
4844 	bbio->orig_bio = first_bio;
4845 	bbio->private = first_bio->bi_private;
4846 	bbio->end_io = first_bio->bi_end_io;
4847 	atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4848 
4849 	while (dev_nr < total_devs) {
4850 		dev = bbio->stripes[dev_nr].dev;
4851 		if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
4852 			bbio_error(bbio, first_bio, logical);
4853 			dev_nr++;
4854 			continue;
4855 		}
4856 
4857 		/*
4858 		 * Check and see if we're ok with this bio based on it's size
4859 		 * and offset with the given device.
4860 		 */
4861 		if (!bio_size_ok(dev->bdev, first_bio,
4862 				 bbio->stripes[dev_nr].physical >> 9)) {
4863 			ret = breakup_stripe_bio(root, bbio, first_bio, dev,
4864 						 dev_nr, rw, async_submit);
4865 			BUG_ON(ret);
4866 			dev_nr++;
4867 			continue;
4868 		}
4869 
4870 		if (dev_nr < total_devs - 1) {
4871 			bio = bio_clone(first_bio, GFP_NOFS);
4872 			BUG_ON(!bio); /* -ENOMEM */
4873 		} else {
4874 			bio = first_bio;
4875 		}
4876 
4877 		submit_stripe_bio(root, bbio, bio,
4878 				  bbio->stripes[dev_nr].physical, dev_nr, rw,
4879 				  async_submit);
4880 		dev_nr++;
4881 	}
4882 	return 0;
4883 }
4884 
4885 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
4886 				       u8 *uuid, u8 *fsid)
4887 {
4888 	struct btrfs_device *device;
4889 	struct btrfs_fs_devices *cur_devices;
4890 
4891 	cur_devices = fs_info->fs_devices;
4892 	while (cur_devices) {
4893 		if (!fsid ||
4894 		    !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4895 			device = __find_device(&cur_devices->devices,
4896 					       devid, uuid);
4897 			if (device)
4898 				return device;
4899 		}
4900 		cur_devices = cur_devices->seed;
4901 	}
4902 	return NULL;
4903 }
4904 
4905 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4906 					    u64 devid, u8 *dev_uuid)
4907 {
4908 	struct btrfs_device *device;
4909 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4910 
4911 	device = kzalloc(sizeof(*device), GFP_NOFS);
4912 	if (!device)
4913 		return NULL;
4914 	list_add(&device->dev_list,
4915 		 &fs_devices->devices);
4916 	device->dev_root = root->fs_info->dev_root;
4917 	device->devid = devid;
4918 	device->work.func = pending_bios_fn;
4919 	device->fs_devices = fs_devices;
4920 	device->missing = 1;
4921 	fs_devices->num_devices++;
4922 	fs_devices->missing_devices++;
4923 	spin_lock_init(&device->io_lock);
4924 	INIT_LIST_HEAD(&device->dev_alloc_list);
4925 	memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4926 	return device;
4927 }
4928 
4929 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4930 			  struct extent_buffer *leaf,
4931 			  struct btrfs_chunk *chunk)
4932 {
4933 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4934 	struct map_lookup *map;
4935 	struct extent_map *em;
4936 	u64 logical;
4937 	u64 length;
4938 	u64 devid;
4939 	u8 uuid[BTRFS_UUID_SIZE];
4940 	int num_stripes;
4941 	int ret;
4942 	int i;
4943 
4944 	logical = key->offset;
4945 	length = btrfs_chunk_length(leaf, chunk);
4946 
4947 	read_lock(&map_tree->map_tree.lock);
4948 	em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4949 	read_unlock(&map_tree->map_tree.lock);
4950 
4951 	/* already mapped? */
4952 	if (em && em->start <= logical && em->start + em->len > logical) {
4953 		free_extent_map(em);
4954 		return 0;
4955 	} else if (em) {
4956 		free_extent_map(em);
4957 	}
4958 
4959 	em = alloc_extent_map();
4960 	if (!em)
4961 		return -ENOMEM;
4962 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4963 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4964 	if (!map) {
4965 		free_extent_map(em);
4966 		return -ENOMEM;
4967 	}
4968 
4969 	em->bdev = (struct block_device *)map;
4970 	em->start = logical;
4971 	em->len = length;
4972 	em->orig_start = 0;
4973 	em->block_start = 0;
4974 	em->block_len = em->len;
4975 
4976 	map->num_stripes = num_stripes;
4977 	map->io_width = btrfs_chunk_io_width(leaf, chunk);
4978 	map->io_align = btrfs_chunk_io_align(leaf, chunk);
4979 	map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4980 	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4981 	map->type = btrfs_chunk_type(leaf, chunk);
4982 	map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4983 	for (i = 0; i < num_stripes; i++) {
4984 		map->stripes[i].physical =
4985 			btrfs_stripe_offset_nr(leaf, chunk, i);
4986 		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4987 		read_extent_buffer(leaf, uuid, (unsigned long)
4988 				   btrfs_stripe_dev_uuid_nr(chunk, i),
4989 				   BTRFS_UUID_SIZE);
4990 		map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
4991 							uuid, NULL);
4992 		if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4993 			kfree(map);
4994 			free_extent_map(em);
4995 			return -EIO;
4996 		}
4997 		if (!map->stripes[i].dev) {
4998 			map->stripes[i].dev =
4999 				add_missing_dev(root, devid, uuid);
5000 			if (!map->stripes[i].dev) {
5001 				kfree(map);
5002 				free_extent_map(em);
5003 				return -EIO;
5004 			}
5005 		}
5006 		map->stripes[i].dev->in_fs_metadata = 1;
5007 	}
5008 
5009 	write_lock(&map_tree->map_tree.lock);
5010 	ret = add_extent_mapping(&map_tree->map_tree, em);
5011 	write_unlock(&map_tree->map_tree.lock);
5012 	BUG_ON(ret); /* Tree corruption */
5013 	free_extent_map(em);
5014 
5015 	return 0;
5016 }
5017 
5018 static void fill_device_from_item(struct extent_buffer *leaf,
5019 				 struct btrfs_dev_item *dev_item,
5020 				 struct btrfs_device *device)
5021 {
5022 	unsigned long ptr;
5023 
5024 	device->devid = btrfs_device_id(leaf, dev_item);
5025 	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5026 	device->total_bytes = device->disk_total_bytes;
5027 	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5028 	device->type = btrfs_device_type(leaf, dev_item);
5029 	device->io_align = btrfs_device_io_align(leaf, dev_item);
5030 	device->io_width = btrfs_device_io_width(leaf, dev_item);
5031 	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5032 	WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5033 	device->is_tgtdev_for_dev_replace = 0;
5034 
5035 	ptr = (unsigned long)btrfs_device_uuid(dev_item);
5036 	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5037 }
5038 
5039 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5040 {
5041 	struct btrfs_fs_devices *fs_devices;
5042 	int ret;
5043 
5044 	BUG_ON(!mutex_is_locked(&uuid_mutex));
5045 
5046 	fs_devices = root->fs_info->fs_devices->seed;
5047 	while (fs_devices) {
5048 		if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5049 			ret = 0;
5050 			goto out;
5051 		}
5052 		fs_devices = fs_devices->seed;
5053 	}
5054 
5055 	fs_devices = find_fsid(fsid);
5056 	if (!fs_devices) {
5057 		ret = -ENOENT;
5058 		goto out;
5059 	}
5060 
5061 	fs_devices = clone_fs_devices(fs_devices);
5062 	if (IS_ERR(fs_devices)) {
5063 		ret = PTR_ERR(fs_devices);
5064 		goto out;
5065 	}
5066 
5067 	ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5068 				   root->fs_info->bdev_holder);
5069 	if (ret) {
5070 		free_fs_devices(fs_devices);
5071 		goto out;
5072 	}
5073 
5074 	if (!fs_devices->seeding) {
5075 		__btrfs_close_devices(fs_devices);
5076 		free_fs_devices(fs_devices);
5077 		ret = -EINVAL;
5078 		goto out;
5079 	}
5080 
5081 	fs_devices->seed = root->fs_info->fs_devices->seed;
5082 	root->fs_info->fs_devices->seed = fs_devices;
5083 out:
5084 	return ret;
5085 }
5086 
5087 static int read_one_dev(struct btrfs_root *root,
5088 			struct extent_buffer *leaf,
5089 			struct btrfs_dev_item *dev_item)
5090 {
5091 	struct btrfs_device *device;
5092 	u64 devid;
5093 	int ret;
5094 	u8 fs_uuid[BTRFS_UUID_SIZE];
5095 	u8 dev_uuid[BTRFS_UUID_SIZE];
5096 
5097 	devid = btrfs_device_id(leaf, dev_item);
5098 	read_extent_buffer(leaf, dev_uuid,
5099 			   (unsigned long)btrfs_device_uuid(dev_item),
5100 			   BTRFS_UUID_SIZE);
5101 	read_extent_buffer(leaf, fs_uuid,
5102 			   (unsigned long)btrfs_device_fsid(dev_item),
5103 			   BTRFS_UUID_SIZE);
5104 
5105 	if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5106 		ret = open_seed_devices(root, fs_uuid);
5107 		if (ret && !btrfs_test_opt(root, DEGRADED))
5108 			return ret;
5109 	}
5110 
5111 	device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5112 	if (!device || !device->bdev) {
5113 		if (!btrfs_test_opt(root, DEGRADED))
5114 			return -EIO;
5115 
5116 		if (!device) {
5117 			printk(KERN_WARNING "warning devid %llu missing\n",
5118 			       (unsigned long long)devid);
5119 			device = add_missing_dev(root, devid, dev_uuid);
5120 			if (!device)
5121 				return -ENOMEM;
5122 		} else if (!device->missing) {
5123 			/*
5124 			 * this happens when a device that was properly setup
5125 			 * in the device info lists suddenly goes bad.
5126 			 * device->bdev is NULL, and so we have to set
5127 			 * device->missing to one here
5128 			 */
5129 			root->fs_info->fs_devices->missing_devices++;
5130 			device->missing = 1;
5131 		}
5132 	}
5133 
5134 	if (device->fs_devices != root->fs_info->fs_devices) {
5135 		BUG_ON(device->writeable);
5136 		if (device->generation !=
5137 		    btrfs_device_generation(leaf, dev_item))
5138 			return -EINVAL;
5139 	}
5140 
5141 	fill_device_from_item(leaf, dev_item, device);
5142 	device->dev_root = root->fs_info->dev_root;
5143 	device->in_fs_metadata = 1;
5144 	if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5145 		device->fs_devices->total_rw_bytes += device->total_bytes;
5146 		spin_lock(&root->fs_info->free_chunk_lock);
5147 		root->fs_info->free_chunk_space += device->total_bytes -
5148 			device->bytes_used;
5149 		spin_unlock(&root->fs_info->free_chunk_lock);
5150 	}
5151 	ret = 0;
5152 	return ret;
5153 }
5154 
5155 int btrfs_read_sys_array(struct btrfs_root *root)
5156 {
5157 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5158 	struct extent_buffer *sb;
5159 	struct btrfs_disk_key *disk_key;
5160 	struct btrfs_chunk *chunk;
5161 	u8 *ptr;
5162 	unsigned long sb_ptr;
5163 	int ret = 0;
5164 	u32 num_stripes;
5165 	u32 array_size;
5166 	u32 len = 0;
5167 	u32 cur;
5168 	struct btrfs_key key;
5169 
5170 	sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5171 					  BTRFS_SUPER_INFO_SIZE);
5172 	if (!sb)
5173 		return -ENOMEM;
5174 	btrfs_set_buffer_uptodate(sb);
5175 	btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5176 	/*
5177 	 * The sb extent buffer is artifical and just used to read the system array.
5178 	 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5179 	 * pages up-to-date when the page is larger: extent does not cover the
5180 	 * whole page and consequently check_page_uptodate does not find all
5181 	 * the page's extents up-to-date (the hole beyond sb),
5182 	 * write_extent_buffer then triggers a WARN_ON.
5183 	 *
5184 	 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5185 	 * but sb spans only this function. Add an explicit SetPageUptodate call
5186 	 * to silence the warning eg. on PowerPC 64.
5187 	 */
5188 	if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5189 		SetPageUptodate(sb->pages[0]);
5190 
5191 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5192 	array_size = btrfs_super_sys_array_size(super_copy);
5193 
5194 	ptr = super_copy->sys_chunk_array;
5195 	sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5196 	cur = 0;
5197 
5198 	while (cur < array_size) {
5199 		disk_key = (struct btrfs_disk_key *)ptr;
5200 		btrfs_disk_key_to_cpu(&key, disk_key);
5201 
5202 		len = sizeof(*disk_key); ptr += len;
5203 		sb_ptr += len;
5204 		cur += len;
5205 
5206 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5207 			chunk = (struct btrfs_chunk *)sb_ptr;
5208 			ret = read_one_chunk(root, &key, sb, chunk);
5209 			if (ret)
5210 				break;
5211 			num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5212 			len = btrfs_chunk_item_size(num_stripes);
5213 		} else {
5214 			ret = -EIO;
5215 			break;
5216 		}
5217 		ptr += len;
5218 		sb_ptr += len;
5219 		cur += len;
5220 	}
5221 	free_extent_buffer(sb);
5222 	return ret;
5223 }
5224 
5225 int btrfs_read_chunk_tree(struct btrfs_root *root)
5226 {
5227 	struct btrfs_path *path;
5228 	struct extent_buffer *leaf;
5229 	struct btrfs_key key;
5230 	struct btrfs_key found_key;
5231 	int ret;
5232 	int slot;
5233 
5234 	root = root->fs_info->chunk_root;
5235 
5236 	path = btrfs_alloc_path();
5237 	if (!path)
5238 		return -ENOMEM;
5239 
5240 	mutex_lock(&uuid_mutex);
5241 	lock_chunks(root);
5242 
5243 	/* first we search for all of the device items, and then we
5244 	 * read in all of the chunk items.  This way we can create chunk
5245 	 * mappings that reference all of the devices that are afound
5246 	 */
5247 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5248 	key.offset = 0;
5249 	key.type = 0;
5250 again:
5251 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5252 	if (ret < 0)
5253 		goto error;
5254 	while (1) {
5255 		leaf = path->nodes[0];
5256 		slot = path->slots[0];
5257 		if (slot >= btrfs_header_nritems(leaf)) {
5258 			ret = btrfs_next_leaf(root, path);
5259 			if (ret == 0)
5260 				continue;
5261 			if (ret < 0)
5262 				goto error;
5263 			break;
5264 		}
5265 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
5266 		if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5267 			if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5268 				break;
5269 			if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5270 				struct btrfs_dev_item *dev_item;
5271 				dev_item = btrfs_item_ptr(leaf, slot,
5272 						  struct btrfs_dev_item);
5273 				ret = read_one_dev(root, leaf, dev_item);
5274 				if (ret)
5275 					goto error;
5276 			}
5277 		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5278 			struct btrfs_chunk *chunk;
5279 			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5280 			ret = read_one_chunk(root, &found_key, leaf, chunk);
5281 			if (ret)
5282 				goto error;
5283 		}
5284 		path->slots[0]++;
5285 	}
5286 	if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5287 		key.objectid = 0;
5288 		btrfs_release_path(path);
5289 		goto again;
5290 	}
5291 	ret = 0;
5292 error:
5293 	unlock_chunks(root);
5294 	mutex_unlock(&uuid_mutex);
5295 
5296 	btrfs_free_path(path);
5297 	return ret;
5298 }
5299 
5300 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5301 {
5302 	int i;
5303 
5304 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5305 		btrfs_dev_stat_reset(dev, i);
5306 }
5307 
5308 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5309 {
5310 	struct btrfs_key key;
5311 	struct btrfs_key found_key;
5312 	struct btrfs_root *dev_root = fs_info->dev_root;
5313 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5314 	struct extent_buffer *eb;
5315 	int slot;
5316 	int ret = 0;
5317 	struct btrfs_device *device;
5318 	struct btrfs_path *path = NULL;
5319 	int i;
5320 
5321 	path = btrfs_alloc_path();
5322 	if (!path) {
5323 		ret = -ENOMEM;
5324 		goto out;
5325 	}
5326 
5327 	mutex_lock(&fs_devices->device_list_mutex);
5328 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
5329 		int item_size;
5330 		struct btrfs_dev_stats_item *ptr;
5331 
5332 		key.objectid = 0;
5333 		key.type = BTRFS_DEV_STATS_KEY;
5334 		key.offset = device->devid;
5335 		ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5336 		if (ret) {
5337 			__btrfs_reset_dev_stats(device);
5338 			device->dev_stats_valid = 1;
5339 			btrfs_release_path(path);
5340 			continue;
5341 		}
5342 		slot = path->slots[0];
5343 		eb = path->nodes[0];
5344 		btrfs_item_key_to_cpu(eb, &found_key, slot);
5345 		item_size = btrfs_item_size_nr(eb, slot);
5346 
5347 		ptr = btrfs_item_ptr(eb, slot,
5348 				     struct btrfs_dev_stats_item);
5349 
5350 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5351 			if (item_size >= (1 + i) * sizeof(__le64))
5352 				btrfs_dev_stat_set(device, i,
5353 					btrfs_dev_stats_value(eb, ptr, i));
5354 			else
5355 				btrfs_dev_stat_reset(device, i);
5356 		}
5357 
5358 		device->dev_stats_valid = 1;
5359 		btrfs_dev_stat_print_on_load(device);
5360 		btrfs_release_path(path);
5361 	}
5362 	mutex_unlock(&fs_devices->device_list_mutex);
5363 
5364 out:
5365 	btrfs_free_path(path);
5366 	return ret < 0 ? ret : 0;
5367 }
5368 
5369 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5370 				struct btrfs_root *dev_root,
5371 				struct btrfs_device *device)
5372 {
5373 	struct btrfs_path *path;
5374 	struct btrfs_key key;
5375 	struct extent_buffer *eb;
5376 	struct btrfs_dev_stats_item *ptr;
5377 	int ret;
5378 	int i;
5379 
5380 	key.objectid = 0;
5381 	key.type = BTRFS_DEV_STATS_KEY;
5382 	key.offset = device->devid;
5383 
5384 	path = btrfs_alloc_path();
5385 	BUG_ON(!path);
5386 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5387 	if (ret < 0) {
5388 		printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5389 			      ret, rcu_str_deref(device->name));
5390 		goto out;
5391 	}
5392 
5393 	if (ret == 0 &&
5394 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5395 		/* need to delete old one and insert a new one */
5396 		ret = btrfs_del_item(trans, dev_root, path);
5397 		if (ret != 0) {
5398 			printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5399 				      rcu_str_deref(device->name), ret);
5400 			goto out;
5401 		}
5402 		ret = 1;
5403 	}
5404 
5405 	if (ret == 1) {
5406 		/* need to insert a new item */
5407 		btrfs_release_path(path);
5408 		ret = btrfs_insert_empty_item(trans, dev_root, path,
5409 					      &key, sizeof(*ptr));
5410 		if (ret < 0) {
5411 			printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5412 				      rcu_str_deref(device->name), ret);
5413 			goto out;
5414 		}
5415 	}
5416 
5417 	eb = path->nodes[0];
5418 	ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5419 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5420 		btrfs_set_dev_stats_value(eb, ptr, i,
5421 					  btrfs_dev_stat_read(device, i));
5422 	btrfs_mark_buffer_dirty(eb);
5423 
5424 out:
5425 	btrfs_free_path(path);
5426 	return ret;
5427 }
5428 
5429 /*
5430  * called from commit_transaction. Writes all changed device stats to disk.
5431  */
5432 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5433 			struct btrfs_fs_info *fs_info)
5434 {
5435 	struct btrfs_root *dev_root = fs_info->dev_root;
5436 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5437 	struct btrfs_device *device;
5438 	int ret = 0;
5439 
5440 	mutex_lock(&fs_devices->device_list_mutex);
5441 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
5442 		if (!device->dev_stats_valid || !device->dev_stats_dirty)
5443 			continue;
5444 
5445 		ret = update_dev_stat_item(trans, dev_root, device);
5446 		if (!ret)
5447 			device->dev_stats_dirty = 0;
5448 	}
5449 	mutex_unlock(&fs_devices->device_list_mutex);
5450 
5451 	return ret;
5452 }
5453 
5454 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5455 {
5456 	btrfs_dev_stat_inc(dev, index);
5457 	btrfs_dev_stat_print_on_error(dev);
5458 }
5459 
5460 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5461 {
5462 	if (!dev->dev_stats_valid)
5463 		return;
5464 	printk_ratelimited_in_rcu(KERN_ERR
5465 			   "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5466 			   rcu_str_deref(dev->name),
5467 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5468 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5469 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5470 			   btrfs_dev_stat_read(dev,
5471 					       BTRFS_DEV_STAT_CORRUPTION_ERRS),
5472 			   btrfs_dev_stat_read(dev,
5473 					       BTRFS_DEV_STAT_GENERATION_ERRS));
5474 }
5475 
5476 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5477 {
5478 	int i;
5479 
5480 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5481 		if (btrfs_dev_stat_read(dev, i) != 0)
5482 			break;
5483 	if (i == BTRFS_DEV_STAT_VALUES_MAX)
5484 		return; /* all values == 0, suppress message */
5485 
5486 	printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5487 	       rcu_str_deref(dev->name),
5488 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5489 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5490 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5491 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5492 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5493 }
5494 
5495 int btrfs_get_dev_stats(struct btrfs_root *root,
5496 			struct btrfs_ioctl_get_dev_stats *stats)
5497 {
5498 	struct btrfs_device *dev;
5499 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5500 	int i;
5501 
5502 	mutex_lock(&fs_devices->device_list_mutex);
5503 	dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5504 	mutex_unlock(&fs_devices->device_list_mutex);
5505 
5506 	if (!dev) {
5507 		printk(KERN_WARNING
5508 		       "btrfs: get dev_stats failed, device not found\n");
5509 		return -ENODEV;
5510 	} else if (!dev->dev_stats_valid) {
5511 		printk(KERN_WARNING
5512 		       "btrfs: get dev_stats failed, not yet valid\n");
5513 		return -ENODEV;
5514 	} else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5515 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5516 			if (stats->nr_items > i)
5517 				stats->values[i] =
5518 					btrfs_dev_stat_read_and_reset(dev, i);
5519 			else
5520 				btrfs_dev_stat_reset(dev, i);
5521 		}
5522 	} else {
5523 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5524 			if (stats->nr_items > i)
5525 				stats->values[i] = btrfs_dev_stat_read(dev, i);
5526 	}
5527 	if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5528 		stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5529 	return 0;
5530 }
5531 
5532 int btrfs_scratch_superblock(struct btrfs_device *device)
5533 {
5534 	struct buffer_head *bh;
5535 	struct btrfs_super_block *disk_super;
5536 
5537 	bh = btrfs_read_dev_super(device->bdev);
5538 	if (!bh)
5539 		return -EINVAL;
5540 	disk_super = (struct btrfs_super_block *)bh->b_data;
5541 
5542 	memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5543 	set_buffer_dirty(bh);
5544 	sync_dirty_buffer(bh);
5545 	brelse(bh);
5546 
5547 	return 0;
5548 }
5549