disk-io.c 20.0 KB
Newer Older
C
Chris Mason 已提交
1 2
#include <linux/module.h>
#include <linux/fs.h>
3
#include <linux/blkdev.h>
C
Chris Mason 已提交
4 5
#include <linux/crypto.h>
#include <linux/scatterlist.h>
C
Chris Mason 已提交
6
#include <linux/swap.h>
7
#include <linux/radix-tree.h>
C
Chris Mason 已提交
8
#include <linux/writeback.h>
9 10
#include "ctree.h"
#include "disk-io.h"
11
#include "transaction.h"
12
#include "btrfs_inode.h"
13

14 15 16
struct dev_lookup {
	u64 block_start;
	u64 num_blocks;
17
	u64 device_id;
18 19 20
	struct block_device *bdev;
};

21 22
int btrfs_insert_dev_radix(struct btrfs_root *root,
			   struct block_device *bdev,
23
			   u64 device_id,
24 25 26 27 28 29 30 31 32 33 34 35
			   u64 block_start,
			   u64 num_blocks)
{
	struct dev_lookup *lookup;
	int ret;

	lookup = kmalloc(sizeof(*lookup), GFP_NOFS);
	if (!lookup)
		return -ENOMEM;
	lookup->block_start = block_start;
	lookup->num_blocks = num_blocks;
	lookup->bdev = bdev;
36
	lookup->device_id = device_id;
37 38 39 40 41 42

	ret = radix_tree_insert(&root->fs_info->dev_radix, block_start +
				num_blocks - 1, lookup);
	return ret;
}

43 44 45 46 47 48 49 50 51 52 53 54 55 56
u64 bh_blocknr(struct buffer_head *bh)
{
	int blkbits = bh->b_page->mapping->host->i_blkbits;
	u64 blocknr = bh->b_page->index << (PAGE_CACHE_SHIFT - blkbits);
	unsigned long offset;

	if (PageHighMem(bh->b_page))
		offset = (unsigned long)bh->b_data;
	else
		offset = bh->b_data - (char *)page_address(bh->b_page);
	blocknr += offset >> (PAGE_CACHE_SHIFT - blkbits);
	return blocknr;
}

C
Chris Mason 已提交
57
static int check_tree_block(struct btrfs_root *root, struct buffer_head *buf)
58
{
C
Chris Mason 已提交
59
	struct btrfs_node *node = btrfs_buffer_node(buf);
60
	if (bh_blocknr(buf) != btrfs_header_blocknr(&node->header)) {
61 62
		printk(KERN_CRIT "bh_blocknr(buf) is %Lu, header is %Lu\n",
		       bh_blocknr(buf), btrfs_header_blocknr(&node->header));
63
		BUG();
64
	}
65
	return 0;
66 67
}

68 69 70 71 72 73 74 75 76 77
struct buffer_head *btrfs_find_tree_block(struct btrfs_root *root, u64 blocknr)
{
	struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
	int blockbits = root->fs_info->sb->s_blocksize_bits;
	unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
	struct page *page;
	struct buffer_head *bh;
	struct buffer_head *head;
	struct buffer_head *ret = NULL;

C
Chris Mason 已提交
78

79 80 81 82 83 84 85 86 87 88
	page = find_lock_page(mapping, index);
	if (!page)
		return NULL;

	if (!page_has_buffers(page))
		goto out_unlock;

	head = page_buffers(page);
	bh = head;
	do {
89
		if (buffer_mapped(bh) && bh_blocknr(bh) == blocknr) {
90 91 92 93 94 95 96 97 98 99 100 101
			ret = bh;
			get_bh(bh);
			goto out_unlock;
		}
		bh = bh->b_this_page;
	} while (bh != head);
out_unlock:
	unlock_page(page);
	page_cache_release(page);
	return ret;
}

102
int btrfs_map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
103 104 105 106 107 108
			     u64 logical)
{
	struct dev_lookup *lookup[2];

	int ret;

109 110 111 112 113 114
	if (logical == 0) {
		bh->b_bdev = NULL;
		bh->b_blocknr = 0;
		set_buffer_mapped(bh);
		return 0;
	}
115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132
	root = root->fs_info->dev_root;
	ret = radix_tree_gang_lookup(&root->fs_info->dev_radix,
				     (void **)lookup,
				     (unsigned long)logical,
				     ARRAY_SIZE(lookup));
	if (ret == 0 || lookup[0]->block_start > logical ||
	    lookup[0]->block_start + lookup[0]->num_blocks <= logical) {
		ret = -ENOENT;
		goto out;
	}
	bh->b_bdev = lookup[0]->bdev;
	bh->b_blocknr = logical - lookup[0]->block_start;
	set_buffer_mapped(bh);
	ret = 0;
out:
	return ret;
}

133 134 135 136 137 138 139 140 141 142
struct buffer_head *btrfs_find_create_tree_block(struct btrfs_root *root,
						 u64 blocknr)
{
	struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
	int blockbits = root->fs_info->sb->s_blocksize_bits;
	unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
	struct page *page;
	struct buffer_head *bh;
	struct buffer_head *head;
	struct buffer_head *ret = NULL;
143
	int err;
144
	u64 first_block = index << (PAGE_CACHE_SHIFT - blockbits);
C
Chris Mason 已提交
145

146 147 148 149 150 151 152 153 154 155
	page = grab_cache_page(mapping, index);
	if (!page)
		return NULL;

	if (!page_has_buffers(page))
		create_empty_buffers(page, root->fs_info->sb->s_blocksize, 0);
	head = page_buffers(page);
	bh = head;
	do {
		if (!buffer_mapped(bh)) {
156
			err = btrfs_map_bh_to_logical(root, bh, first_block);
157
			BUG_ON(err);
158
		}
159
		if (bh_blocknr(bh) == blocknr) {
160 161 162 163 164 165 166 167 168
			ret = bh;
			get_bh(bh);
			goto out_unlock;
		}
		bh = bh->b_this_page;
		first_block++;
	} while (bh != head);
out_unlock:
	unlock_page(page);
C
Chris Mason 已提交
169 170
	if (ret)
		touch_buffer(ret);
171 172 173 174 175 176 177
	page_cache_release(page);
	return ret;
}

static int btree_get_block(struct inode *inode, sector_t iblock,
			   struct buffer_head *bh, int create)
{
178 179
	int err;
	struct btrfs_root *root = BTRFS_I(bh->b_page->mapping->host)->root;
180
	err = btrfs_map_bh_to_logical(root, bh, iblock);
181
	return err;
182 183
}

C
Chris Mason 已提交
184 185
int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
		    char *result)
C
Chris Mason 已提交
186 187 188 189 190 191 192 193
{
	struct scatterlist sg;
	struct crypto_hash *tfm = root->fs_info->hash_tfm;
	struct hash_desc desc;
	int ret;

	desc.tfm = tfm;
	desc.flags = 0;
C
Chris Mason 已提交
194
	sg_init_one(&sg, data, len);
C
Chris Mason 已提交
195
	spin_lock(&root->fs_info->hash_lock);
C
Chris Mason 已提交
196
	ret = crypto_hash_digest(&desc, &sg, 1, result);
C
Chris Mason 已提交
197 198
	spin_unlock(&root->fs_info->hash_lock);
	if (ret) {
199
		printk("digest failed\n");
C
Chris Mason 已提交
200
	}
C
Chris Mason 已提交
201 202 203 204 205
	return ret;
}
static int csum_tree_block(struct btrfs_root *root, struct buffer_head *bh,
			   int verify)
{
206
	char result[BTRFS_CRC32_SIZE];
C
Chris Mason 已提交
207 208 209 210 211 212 213
	int ret;
	struct btrfs_node *node;

	ret = btrfs_csum_data(root, bh->b_data + BTRFS_CSUM_SIZE,
			      bh->b_size - BTRFS_CSUM_SIZE, result);
	if (ret)
		return ret;
C
Chris Mason 已提交
214
	if (verify) {
215
		if (memcmp(bh->b_data, result, BTRFS_CRC32_SIZE)) {
216 217
			printk("checksum verify failed on %Lu\n",
			       bh_blocknr(bh));
C
Chris Mason 已提交
218 219 220 221
			return 1;
		}
	} else {
		node = btrfs_buffer_node(bh);
222
		memcpy(node->header.csum, result, BTRFS_CRC32_SIZE);
C
Chris Mason 已提交
223
	}
C
Chris Mason 已提交
224 225 226
	return 0;
}

227
static int btree_writepage(struct page *page, struct writeback_control *wbc)
228
{
C
Chris Mason 已提交
229
	struct buffer_head *bh;
230
	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
C
Chris Mason 已提交
231 232 233 234 235 236 237 238 239 240 241 242
	struct buffer_head *head;
	if (!page_has_buffers(page)) {
		create_empty_buffers(page, root->fs_info->sb->s_blocksize,
					(1 << BH_Dirty)|(1 << BH_Uptodate));
	}
	head = page_buffers(page);
	bh = head;
	do {
		if (buffer_dirty(bh))
			csum_tree_block(root, bh, 0);
		bh = bh->b_this_page;
	} while (bh != head);
243
	return block_write_full_page(page, btree_get_block, wbc);
244 245
}

246
static int btree_readpage(struct file * file, struct page * page)
247
{
248
	return block_read_full_page(page, btree_get_block);
249 250
}

251 252 253 254 255 256
static struct address_space_operations btree_aops = {
	.readpage	= btree_readpage,
	.writepage	= btree_writepage,
	.sync_page	= block_sync_page,
};

C
Chris Mason 已提交
257 258 259
int readahead_tree_block(struct btrfs_root *root, u64 blocknr)
{
	struct buffer_head *bh = NULL;
260
	int ret = 0;
C
Chris Mason 已提交
261 262 263 264

	bh = btrfs_find_create_tree_block(root, blocknr);
	if (!bh)
		return 0;
265 266
	if (buffer_uptodate(bh)) {
		ret = 1;
C
Chris Mason 已提交
267
		goto done;
268 269 270
	}
	if (test_set_buffer_locked(bh)) {
		ret = 1;
C
Chris Mason 已提交
271
		goto done;
272
	}
C
Chris Mason 已提交
273 274 275 276 277 278
	if (!buffer_uptodate(bh)) {
		get_bh(bh);
		bh->b_end_io = end_buffer_read_sync;
		submit_bh(READ, bh);
	} else {
		unlock_buffer(bh);
279
		ret = 1;
C
Chris Mason 已提交
280 281 282
	}
done:
	brelse(bh);
283
	return ret;
C
Chris Mason 已提交
284 285
}

C
Chris Mason 已提交
286
struct buffer_head *read_tree_block(struct btrfs_root *root, u64 blocknr)
287
{
288
	struct buffer_head *bh = NULL;
289

290 291 292
	bh = btrfs_find_create_tree_block(root, blocknr);
	if (!bh)
		return bh;
293 294
	if (buffer_uptodate(bh))
		goto uptodate;
295 296 297 298 299 300 301 302 303 304 305
	lock_buffer(bh);
	if (!buffer_uptodate(bh)) {
		get_bh(bh);
		bh->b_end_io = end_buffer_read_sync;
		submit_bh(READ, bh);
		wait_on_buffer(bh);
		if (!buffer_uptodate(bh))
			goto fail;
	} else {
		unlock_buffer(bh);
	}
306
uptodate:
C
Chris Mason 已提交
307 308 309 310
	if (!buffer_checked(bh)) {
		csum_tree_block(root, bh, 1);
		set_buffer_checked(bh);
	}
311
	if (check_tree_block(root, bh))
C
Chris Mason 已提交
312
		BUG();
313 314 315 316
	return bh;
fail:
	brelse(bh);
	return NULL;
317 318
}

319
int dirty_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
C
Chris Mason 已提交
320
		     struct buffer_head *buf)
321
{
C
Chris Mason 已提交
322
	WARN_ON(atomic_read(&buf->b_count) == 0);
C
Chris Mason 已提交
323
	mark_buffer_dirty(buf);
324 325 326
	return 0;
}

327
int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
C
Chris Mason 已提交
328
		     struct buffer_head *buf)
329
{
C
Chris Mason 已提交
330
	WARN_ON(atomic_read(&buf->b_count) == 0);
C
Chris Mason 已提交
331
	clear_buffer_dirty(buf);
332 333 334
	return 0;
}

C
Chris Mason 已提交
335
static int __setup_root(int blocksize,
336 337
			struct btrfs_root *root,
			struct btrfs_fs_info *fs_info,
C
Chris Mason 已提交
338
			u64 objectid)
339
{
C
Chris Mason 已提交
340
	root->node = NULL;
341
	root->inode = NULL;
342
	root->commit_root = NULL;
C
Chris Mason 已提交
343
	root->blocksize = blocksize;
C
Chris Mason 已提交
344
	root->ref_cows = 0;
345
	root->fs_info = fs_info;
346 347
	root->objectid = objectid;
	root->last_trans = 0;
C
Chris Mason 已提交
348 349
	root->highest_inode = 0;
	root->last_inode_alloc = 0;
350 351
	memset(&root->root_key, 0, sizeof(root->root_key));
	memset(&root->root_item, 0, sizeof(root->root_item));
352
	root->root_key.objectid = objectid;
353 354 355
	return 0;
}

C
Chris Mason 已提交
356
static int find_and_setup_root(int blocksize,
357 358 359
			       struct btrfs_root *tree_root,
			       struct btrfs_fs_info *fs_info,
			       u64 objectid,
C
Chris Mason 已提交
360
			       struct btrfs_root *root)
361 362 363
{
	int ret;

C
Chris Mason 已提交
364
	__setup_root(blocksize, root, fs_info, objectid);
365 366 367 368 369 370 371
	ret = btrfs_find_last_root(tree_root, objectid,
				   &root->root_item, &root->root_key);
	BUG_ON(ret);

	root->node = read_tree_block(root,
				     btrfs_root_blocknr(&root->root_item));
	BUG_ON(!root->node);
372 373 374
	return 0;
}

375 376 377 378 379 380 381
struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
				      struct btrfs_key *location)
{
	struct btrfs_root *root;
	struct btrfs_root *tree_root = fs_info->tree_root;
	struct btrfs_path *path;
	struct btrfs_leaf *l;
C
Chris Mason 已提交
382
	u64 highest_inode;
383 384 385
	int ret = 0;

printk("read_fs_root looking for %Lu %Lu %u\n", location->objectid, location->offset, location->flags);
C
Chris Mason 已提交
386 387 388 389 390 391
	root = radix_tree_lookup(&fs_info->fs_roots_radix,
				 (unsigned long)location->objectid);
	if (root) {
printk("found %p in cache\n", root);
		return root;
	}
392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439
	root = kmalloc(sizeof(*root), GFP_NOFS);
	if (!root) {
printk("failed1\n");
		return ERR_PTR(-ENOMEM);
	}
	if (location->offset == (u64)-1) {
		ret = find_and_setup_root(fs_info->sb->s_blocksize,
					  fs_info->tree_root, fs_info,
					  location->objectid, root);
		if (ret) {
printk("failed2\n");
			kfree(root);
			return ERR_PTR(ret);
		}
		goto insert;
	}

	__setup_root(fs_info->sb->s_blocksize, root, fs_info,
		     location->objectid);

	path = btrfs_alloc_path();
	BUG_ON(!path);
	ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
	if (ret != 0) {
printk("internal search_slot gives us %d\n", ret);
		if (ret > 0)
			ret = -ENOENT;
		goto out;
	}
	l = btrfs_buffer_leaf(path->nodes[0]);
	memcpy(&root->root_item,
	       btrfs_item_ptr(l, path->slots[0], struct btrfs_root_item),
	       sizeof(root->root_item));
	memcpy(&root->root_key, location, sizeof(*location));
	ret = 0;
out:
	btrfs_release_path(root, path);
	btrfs_free_path(path);
	if (ret) {
		kfree(root);
		return ERR_PTR(ret);
	}
	root->node = read_tree_block(root,
				     btrfs_root_blocknr(&root->root_item));
	BUG_ON(!root->node);
insert:
printk("inserting %p\n", root);
	root->ref_cows = 1;
C
Chris Mason 已提交
440 441
	ret = radix_tree_insert(&fs_info->fs_roots_radix,
				(unsigned long)root->root_key.objectid,
442 443 444 445 446 447 448
				root);
	if (ret) {
printk("radix_tree_insert gives us %d\n", ret);
		brelse(root->node);
		kfree(root);
		return ERR_PTR(ret);
	}
C
Chris Mason 已提交
449 450 451 452 453 454
	ret = btrfs_find_highest_inode(root, &highest_inode);
	if (ret == 0) {
		root->highest_inode = highest_inode;
		root->last_inode_alloc = highest_inode;
printk("highest inode is %Lu\n", highest_inode);
	}
455 456 457 458
printk("all worked\n");
	return root;
}

459 460 461
static int btrfs_open_disk(struct btrfs_root *root, u64 device_id,
			   u64 block_start, u64 num_blocks,
			   char *filename, int name_len)
462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478
{
	char *null_filename;
	struct block_device *bdev;
	int ret;

	null_filename = kmalloc(name_len + 1, GFP_NOFS);
	if (!null_filename)
		return -ENOMEM;
	memcpy(null_filename, filename, name_len);
	null_filename[name_len] = '\0';

	bdev = open_bdev_excl(null_filename, O_RDWR, root->fs_info->sb);
	if (IS_ERR(bdev)) {
		ret = PTR_ERR(bdev);
		goto out;
	}
	set_blocksize(bdev, root->fs_info->sb->s_blocksize);
479 480
	ret = btrfs_insert_dev_radix(root, bdev, device_id,
				     block_start, num_blocks);
481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528
	BUG_ON(ret);
	ret = 0;
out:
	kfree(null_filename);
	return ret;
}

static int read_device_info(struct btrfs_root *root)
{
	struct btrfs_path *path;
	int ret;
	struct btrfs_key key;
	struct btrfs_leaf *leaf;
	struct btrfs_device_item *dev_item;
	int nritems;
	int slot;

	root = root->fs_info->dev_root;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;
	key.objectid = 0;
	key.offset = 0;
	key.flags = 0;
	btrfs_set_key_type(&key, BTRFS_DEV_ITEM_KEY);

	mutex_lock(&root->fs_info->fs_mutex);
	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
	leaf = btrfs_buffer_leaf(path->nodes[0]);
	nritems = btrfs_header_nritems(&leaf->header);
	while(1) {
		slot = path->slots[0];
		if (slot >= nritems) {
			ret = btrfs_next_leaf(root, path);
			if (ret)
				break;
			leaf = btrfs_buffer_leaf(path->nodes[0]);
			nritems = btrfs_header_nritems(&leaf->header);
			slot = path->slots[0];
		}
		btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
		if (btrfs_key_type(&key) != BTRFS_DEV_ITEM_KEY) {
			path->slots[0]++;
			continue;
		}
		dev_item = btrfs_item_ptr(leaf, slot, struct btrfs_device_item);
printk("found key %Lu %Lu\n", key.objectid, key.offset);
529 530 531 532 533 534 535 536
		if (btrfs_device_id(dev_item) !=
		    btrfs_super_device_id(root->fs_info->disk_super)) {
			ret = btrfs_open_disk(root, btrfs_device_id(dev_item),
					      key.objectid, key.offset,
					      (char *)(dev_item + 1),
					      btrfs_device_pathlen(dev_item));
			BUG_ON(ret);
		}
537 538 539 540 541 542 543
		path->slots[0]++;
	}
	btrfs_free_path(path);
	mutex_unlock(&root->fs_info->fs_mutex);
	return 0;
}

C
Chris Mason 已提交
544
struct btrfs_root *open_ctree(struct super_block *sb)
545
{
C
Chris Mason 已提交
546 547
	struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
						 GFP_NOFS);
548 549
	struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
						 GFP_NOFS);
C
Chris Mason 已提交
550 551 552 553
	struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
					       GFP_NOFS);
	struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
						GFP_NOFS);
554
	int ret;
C
Chris Mason 已提交
555
	struct btrfs_super_block *disk_super;
556
	struct dev_lookup *dev_lookup;
557

C
Chris Mason 已提交
558 559
	init_bit_radix(&fs_info->pinned_radix);
	init_bit_radix(&fs_info->pending_del_radix);
560
	init_bit_radix(&fs_info->extent_map_radix);
561
	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
562
	INIT_RADIX_TREE(&fs_info->dev_radix, GFP_NOFS);
C
Chris Mason 已提交
563
	INIT_RADIX_TREE(&fs_info->block_group_radix, GFP_KERNEL);
C
Chris Mason 已提交
564
	INIT_RADIX_TREE(&fs_info->block_group_data_radix, GFP_KERNEL);
C
Chris Mason 已提交
565
	INIT_LIST_HEAD(&fs_info->trans_list);
C
Chris Mason 已提交
566
	sb_set_blocksize(sb, 4096);
567 568 569
	fs_info->running_transaction = NULL;
	fs_info->tree_root = tree_root;
	fs_info->extent_root = extent_root;
570
	fs_info->dev_root = dev_root;
C
Chris Mason 已提交
571
	fs_info->sb = sb;
572 573
	fs_info->btree_inode = new_inode(sb);
	fs_info->btree_inode->i_ino = 1;
C
Chris Mason 已提交
574
	fs_info->btree_inode->i_nlink = 1;
575 576
	fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
	fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
577
	fs_info->do_barriers = 1;
578 579
	fs_info->extent_tree_insert_nr = 0;
	fs_info->extent_tree_prealloc_nr = 0;
C
Chris Mason 已提交
580
	INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
581 582 583
	BTRFS_I(fs_info->btree_inode)->root = tree_root;
	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
	       sizeof(struct btrfs_key));
C
Chris Mason 已提交
584
	insert_inode_hash(fs_info->btree_inode);
585
	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
586
	fs_info->hash_tfm = crypto_alloc_hash("crc32c", 0, CRYPTO_ALG_ASYNC);
C
Chris Mason 已提交
587 588
	spin_lock_init(&fs_info->hash_lock);
	if (!fs_info->hash_tfm || IS_ERR(fs_info->hash_tfm)) {
589
		printk("failed to allocate digest hash\n");
C
Chris Mason 已提交
590 591
		return NULL;
	}
C
Chris Mason 已提交
592
	mutex_init(&fs_info->trans_mutex);
C
Chris Mason 已提交
593
	mutex_init(&fs_info->fs_mutex);
594

595 596 597
	__setup_root(sb->s_blocksize, dev_root,
		     fs_info, BTRFS_DEV_TREE_OBJECTID);

C
Chris Mason 已提交
598 599
	__setup_root(sb->s_blocksize, tree_root,
		     fs_info, BTRFS_ROOT_TREE_OBJECTID);
600 601 602 603 604

	dev_lookup = kmalloc(sizeof(*dev_lookup), GFP_NOFS);
	dev_lookup->block_start = 0;
	dev_lookup->num_blocks = (u32)-2;
	dev_lookup->bdev = sb->s_bdev;
605
	dev_lookup->device_id = 0;
606 607
	ret = radix_tree_insert(&fs_info->dev_radix, (u32)-2, dev_lookup);
	BUG_ON(ret);
C
Chris Mason 已提交
608 609 610
	fs_info->sb_buffer = read_tree_block(tree_root,
					     BTRFS_SUPER_INFO_OFFSET /
					     sb->s_blocksize);
611

612
	if (!fs_info->sb_buffer)
613 614
		return NULL;
	disk_super = (struct btrfs_super_block *)fs_info->sb_buffer->b_data;
615
	if (!btrfs_super_root(disk_super))
C
Chris Mason 已提交
616
		return NULL;
617

618 619 620 621
	i_size_write(fs_info->btree_inode,
		     btrfs_super_total_blocks(disk_super) <<
		     fs_info->btree_inode->i_blkbits);

622 623 624
	radix_tree_delete(&fs_info->dev_radix, (u32)-2);
	dev_lookup->block_start = btrfs_super_device_block_start(disk_super);
	dev_lookup->num_blocks = btrfs_super_device_num_blocks(disk_super);
625 626
	dev_lookup->device_id = btrfs_super_device_id(disk_super);

627 628
	ret = radix_tree_insert(&fs_info->dev_radix,
				dev_lookup->block_start +
629
				dev_lookup->num_blocks - 1, dev_lookup);
630 631
	BUG_ON(ret);

632
	fs_info->disk_super = disk_super;
633

634 635
	dev_root->node = read_tree_block(tree_root,
					  btrfs_super_device_root(disk_super));
636 637 638 639

	ret = read_device_info(dev_root);
	BUG_ON(ret);

C
Chris Mason 已提交
640 641
	tree_root->node = read_tree_block(tree_root,
					  btrfs_super_root(disk_super));
642 643
	BUG_ON(!tree_root->node);

C
Chris Mason 已提交
644 645
	mutex_lock(&fs_info->fs_mutex);
	ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
C
Chris Mason 已提交
646
				  BTRFS_EXTENT_TREE_OBJECTID, extent_root);
647 648
	BUG_ON(ret);

C
Chris Mason 已提交
649 650
	btrfs_read_block_groups(extent_root);

651
	fs_info->generation = btrfs_super_generation(disk_super) + 1;
C
Chris Mason 已提交
652
	mutex_unlock(&fs_info->fs_mutex);
653
	return tree_root;
654 655
}

656
int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
C
Chris Mason 已提交
657
		      *root)
658
{
659
	int ret;
660
	struct buffer_head *bh = root->fs_info->sb_buffer;
C
Chris Mason 已提交
661

662
	btrfs_set_super_root(root->fs_info->disk_super,
663
			     bh_blocknr(root->fs_info->tree_root->node));
664
	lock_buffer(bh);
C
Chris Mason 已提交
665
	WARN_ON(atomic_read(&bh->b_count) < 1);
666
	clear_buffer_dirty(bh);
C
Chris Mason 已提交
667
	csum_tree_block(root, bh, 0);
668 669
	bh->b_end_io = end_buffer_write_sync;
	get_bh(bh);
670 671 672 673 674 675 676 677 678
	if (root->fs_info->do_barriers)
		ret = submit_bh(WRITE_BARRIER, bh);
	else
		ret = submit_bh(WRITE, bh);
	if (ret == -EOPNOTSUPP) {
		set_buffer_uptodate(bh);
		root->fs_info->do_barriers = 0;
		ret = submit_bh(WRITE, bh);
	}
679 680 681 682
	wait_on_buffer(bh);
	if (!buffer_uptodate(bh)) {
		WARN_ON(1);
		return -EIO;
C
Chris Mason 已提交
683 684 685 686
	}
	return 0;
}

C
Chris Mason 已提交
687 688 689 690 691 692 693 694 695 696 697 698 699 700
static int free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
{
	radix_tree_delete(&fs_info->fs_roots_radix,
			  (unsigned long)root->root_key.objectid);
	if (root->inode)
		iput(root->inode);
	if (root->node)
		brelse(root->node);
	if (root->commit_root)
		brelse(root->commit_root);
	kfree(root);
	return 0;
}

C
Chris Mason 已提交
701
static int del_fs_roots(struct btrfs_fs_info *fs_info)
702 703 704 705 706 707 708 709 710 711 712
{
	int ret;
	struct btrfs_root *gang[8];
	int i;

	while(1) {
		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
					     (void **)gang, 0,
					     ARRAY_SIZE(gang));
		if (!ret)
			break;
C
Chris Mason 已提交
713 714
		for (i = 0; i < ret; i++)
			free_fs_root(fs_info, gang[i]);
715 716 717
	}
	return 0;
}
718

719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735
static int free_dev_radix(struct btrfs_fs_info *fs_info)
{
	struct dev_lookup *lookup[8];
	struct block_device *super_bdev = fs_info->sb->s_bdev;
	int ret;
	int i;
	while(1) {
		ret = radix_tree_gang_lookup(&fs_info->dev_radix,
					     (void **)lookup, 0,
					     ARRAY_SIZE(lookup));
		if (!ret)
			break;
		for (i = 0; i < ret; i++) {
			if (lookup[i]->bdev != super_bdev)
				close_bdev_excl(lookup[i]->bdev);
			radix_tree_delete(&fs_info->dev_radix,
					  lookup[i]->block_start +
736
					  lookup[i]->num_blocks - 1);
737 738 739 740 741
			kfree(lookup[i]);
		}
	}
	return 0;
}
742

C
Chris Mason 已提交
743
int close_ctree(struct btrfs_root *root)
C
Chris Mason 已提交
744
{
745
	int ret;
746
	struct btrfs_trans_handle *trans;
747
	struct btrfs_fs_info *fs_info = root->fs_info;
748

C
Chris Mason 已提交
749
	btrfs_transaction_flush_work(root);
750
	mutex_lock(&fs_info->fs_mutex);
C
Chris Mason 已提交
751 752 753 754 755 756
	trans = btrfs_start_transaction(root, 1);
	btrfs_commit_transaction(trans, root);
	/* run commit again to  drop the original snapshot */
	trans = btrfs_start_transaction(root, 1);
	btrfs_commit_transaction(trans, root);
	ret = btrfs_write_and_wait_transaction(NULL, root);
757
	BUG_ON(ret);
C
Chris Mason 已提交
758
	write_ctree_super(NULL, root);
759 760 761 762 763
	mutex_unlock(&fs_info->fs_mutex);

	if (fs_info->extent_root->node)
		btrfs_block_release(fs_info->extent_root,
				    fs_info->extent_root->node);
764 765 766
	if (fs_info->dev_root->node)
		btrfs_block_release(fs_info->dev_root,
				    fs_info->dev_root->node);
767 768 769 770 771 772 773
	if (fs_info->tree_root->node)
		btrfs_block_release(fs_info->tree_root,
				    fs_info->tree_root->node);
	btrfs_block_release(root, fs_info->sb_buffer);
	crypto_free_hash(fs_info->hash_tfm);
	truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
	iput(fs_info->btree_inode);
774 775

	free_dev_radix(fs_info);
C
Chris Mason 已提交
776
	btrfs_free_block_groups(root->fs_info);
777 778 779
	del_fs_roots(fs_info);
	kfree(fs_info->extent_root);
	kfree(fs_info->tree_root);
780 781 782
	return 0;
}

C
Chris Mason 已提交
783
void btrfs_block_release(struct btrfs_root *root, struct buffer_head *buf)
784
{
C
Chris Mason 已提交
785
	brelse(buf);
786 787
}

C
Chris Mason 已提交
788 789 790 791
void btrfs_btree_balance_dirty(struct btrfs_root *root)
{
	balance_dirty_pages_ratelimited(root->fs_info->btree_inode->i_mapping);
}