file-item.c 22.8 KB
Newer Older
C
Chris Mason 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 * Copyright (C) 2007 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

19
#include <linux/bio.h>
20
#include <linux/slab.h>
21 22
#include <linux/pagemap.h>
#include <linux/highmem.h>
C
Chris Mason 已提交
23
#include "ctree.h"
C
Chris Mason 已提交
24
#include "disk-io.h"
25
#include "transaction.h"
C
Chris Mason 已提交
26
#include "print-tree.h"
C
Chris Mason 已提交
27

28
#define __MAX_CSUM_ITEMS(r, size) ((((BTRFS_LEAF_DATA_SIZE(r) - \
29 30
				   sizeof(struct btrfs_item) * 2) / \
				  size) - 1))
Y
Yan Zheng 已提交
31

32 33
#define MAX_CSUM_ITEMS(r, size) (min(__MAX_CSUM_ITEMS(r, size), PAGE_CACHE_SIZE))

Y
Yan Zheng 已提交
34 35 36 37 38
#define MAX_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
				   sizeof(struct btrfs_ordered_sum)) / \
				   sizeof(struct btrfs_sector_sum) * \
				   (r)->sectorsize - (r)->sectorsize)

C
Chris Mason 已提交
39
int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
S
Sage Weil 已提交
40 41 42
			     struct btrfs_root *root,
			     u64 objectid, u64 pos,
			     u64 disk_offset, u64 disk_num_bytes,
C
Chris Mason 已提交
43 44
			     u64 num_bytes, u64 offset, u64 ram_bytes,
			     u8 compression, u8 encryption, u16 other_encoding)
45
{
C
Chris Mason 已提交
46 47 48
	int ret = 0;
	struct btrfs_file_extent_item *item;
	struct btrfs_key file_key;
49
	struct btrfs_path *path;
50
	struct extent_buffer *leaf;
C
Chris Mason 已提交
51

52
	path = btrfs_alloc_path();
T
Tsutomu Itoh 已提交
53 54
	if (!path)
		return -ENOMEM;
C
Chris Mason 已提交
55
	file_key.objectid = objectid;
C
Chris Mason 已提交
56
	file_key.offset = pos;
C
Chris Mason 已提交
57 58
	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);

59
	path->leave_spinning = 1;
60
	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
C
Chris Mason 已提交
61
				      sizeof(*item));
62 63
	if (ret < 0)
		goto out;
64
	BUG_ON(ret); /* Can't happen */
65 66
	leaf = path->nodes[0];
	item = btrfs_item_ptr(leaf, path->slots[0],
C
Chris Mason 已提交
67
			      struct btrfs_file_extent_item);
S
Sage Weil 已提交
68
	btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
69
	btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
S
Sage Weil 已提交
70
	btrfs_set_file_extent_offset(leaf, item, offset);
71
	btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
C
Chris Mason 已提交
72
	btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
73 74
	btrfs_set_file_extent_generation(leaf, item, trans->transid);
	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
C
Chris Mason 已提交
75 76 77 78
	btrfs_set_file_extent_compression(leaf, item, compression);
	btrfs_set_file_extent_encryption(leaf, item, encryption);
	btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);

79
	btrfs_mark_buffer_dirty(leaf);
80
out:
81
	btrfs_free_path(path);
82
	return ret;
83
}
C
Chris Mason 已提交
84

C
Chris Mason 已提交
85 86 87
struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
					  struct btrfs_root *root,
					  struct btrfs_path *path,
88
					  u64 bytenr, int cow)
89 90 91 92 93
{
	int ret;
	struct btrfs_key file_key;
	struct btrfs_key found_key;
	struct btrfs_csum_item *item;
94
	struct extent_buffer *leaf;
95
	u64 csum_offset = 0;
96
	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
97
	int csums_in_item;
98

99 100 101
	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	file_key.offset = bytenr;
	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
C
Chris Mason 已提交
102
	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
103 104
	if (ret < 0)
		goto fail;
105
	leaf = path->nodes[0];
106 107
	if (ret > 0) {
		ret = 1;
108
		if (path->slots[0] == 0)
109 110
			goto fail;
		path->slots[0]--;
111
		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
112
		if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY)
113
			goto fail;
114 115

		csum_offset = (bytenr - found_key.offset) >>
116
				root->fs_info->sb->s_blocksize_bits;
117
		csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
118
		csums_in_item /= csum_size;
119 120 121

		if (csum_offset >= csums_in_item) {
			ret = -EFBIG;
122 123 124 125
			goto fail;
		}
	}
	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
126
	item = (struct btrfs_csum_item *)((unsigned char *)item +
127
					  csum_offset * csum_size);
128 129 130
	return item;
fail:
	if (ret > 0)
C
Chris Mason 已提交
131
		ret = -ENOENT;
132 133 134 135
	return ERR_PTR(ret);
}


C
Chris Mason 已提交
136 137 138
int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
			     struct btrfs_root *root,
			     struct btrfs_path *path, u64 objectid,
C
Chris Mason 已提交
139
			     u64 offset, int mod)
C
Chris Mason 已提交
140 141 142 143 144 145 146
{
	int ret;
	struct btrfs_key file_key;
	int ins_len = mod < 0 ? -1 : 0;
	int cow = mod != 0;

	file_key.objectid = objectid;
147
	file_key.offset = offset;
C
Chris Mason 已提交
148 149 150 151
	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
	ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
	return ret;
}
C
Chris Mason 已提交
152

153

154 155 156
static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
				   struct inode *inode, struct bio *bio,
				   u64 logical_offset, u32 *dst, int dio)
157 158 159 160
{
	u32 sum;
	struct bio_vec *bvec = bio->bi_io_vec;
	int bio_index = 0;
161
	u64 offset = 0;
162 163
	u64 item_start_offset = 0;
	u64 item_last_offset = 0;
164
	u64 disk_bytenr;
165
	u32 diff;
166
	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
167 168 169 170 171 172
	int ret;
	struct btrfs_path *path;
	struct btrfs_csum_item *item = NULL;
	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;

	path = btrfs_alloc_path();
173 174
	if (!path)
		return -ENOMEM;
175 176
	if (bio->bi_size > PAGE_CACHE_SIZE * 8)
		path->reada = 2;
177 178 179

	WARN_ON(bio->bi_vcnt <= 0);

180 181 182 183 184 185
	/*
	 * the free space stuff is only read when it hasn't been
	 * updated in the current transaction.  So, we can safely
	 * read from the commit root and sidestep a nasty deadlock
	 * between reading the free space cache and updating the csum tree.
	 */
186
	if (btrfs_is_free_space_inode(root, inode)) {
187
		path->search_commit_root = 1;
188 189
		path->skip_locking = 1;
	}
190

191
	disk_bytenr = (u64)bio->bi_sector << 9;
192 193
	if (dio)
		offset = logical_offset;
C
Chris Mason 已提交
194
	while (bio_index < bio->bi_vcnt) {
195 196
		if (!dio)
			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
197
		ret = btrfs_find_ordered_sum(inode, offset, disk_bytenr, &sum);
198 199 200
		if (ret == 0)
			goto found;

201 202
		if (!item || disk_bytenr < item_start_offset ||
		    disk_bytenr >= item_last_offset) {
203 204 205 206
			struct btrfs_key found_key;
			u32 item_size;

			if (item)
207
				btrfs_release_path(path);
208 209
			item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
						 path, disk_bytenr, 0);
210 211 212 213 214
			if (IS_ERR(item)) {
				ret = PTR_ERR(item);
				if (ret == -ENOENT || ret == -EFBIG)
					ret = 0;
				sum = 0;
215 216 217 218 219 220
				if (BTRFS_I(inode)->root->root_key.objectid ==
				    BTRFS_DATA_RELOC_TREE_OBJECTID) {
					set_extent_bits(io_tree, offset,
						offset + bvec->bv_len - 1,
						EXTENT_NODATASUM, GFP_NOFS);
				} else {
C
Chris Mason 已提交
221
					printk(KERN_INFO "btrfs no csum found "
L
Li Zefan 已提交
222 223 224
					       "for inode %llu start %llu\n",
					       (unsigned long long)
					       btrfs_ino(inode),
225 226
					       (unsigned long long)offset);
				}
227
				item = NULL;
228
				btrfs_release_path(path);
229 230 231 232 233 234 235 236 237
				goto found;
			}
			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
					      path->slots[0]);

			item_start_offset = found_key.offset;
			item_size = btrfs_item_size_nr(path->nodes[0],
						       path->slots[0]);
			item_last_offset = item_start_offset +
238
				(item_size / csum_size) *
239 240 241 242 243 244 245 246
				root->sectorsize;
			item = btrfs_item_ptr(path->nodes[0], path->slots[0],
					      struct btrfs_csum_item);
		}
		/*
		 * this byte range must be able to fit inside
		 * a single leaf so it will also fit inside a u32
		 */
247
		diff = disk_bytenr - item_start_offset;
248
		diff = diff / root->sectorsize;
249
		diff = diff * csum_size;
250 251

		read_extent_buffer(path->nodes[0], &sum,
252
				   ((unsigned long)item) + diff,
253
				   csum_size);
254
found:
255 256 257 258 259
		if (dst)
			*dst++ = sum;
		else
			set_state_private(io_tree, offset, sum);
		disk_bytenr += bvec->bv_len;
260
		offset += bvec->bv_len;
261 262 263 264 265 266 267
		bio_index++;
		bvec++;
	}
	btrfs_free_path(path);
	return 0;
}

268 269 270 271 272 273 274 275 276 277 278 279
int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
			  struct bio *bio, u32 *dst)
{
	return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
}

int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
			      struct bio *bio, u64 offset, u32 *dst)
{
	return __btrfs_lookup_bio_sums(root, inode, bio, offset, dst, 1);
}

280
int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
A
Arne Jansen 已提交
281
			     struct list_head *list, int search_commit)
282 283 284 285 286 287 288
{
	struct btrfs_key key;
	struct btrfs_path *path;
	struct extent_buffer *leaf;
	struct btrfs_ordered_sum *sums;
	struct btrfs_sector_sum *sector_sum;
	struct btrfs_csum_item *item;
289
	LIST_HEAD(tmplist);
290 291 292 293
	unsigned long offset;
	int ret;
	size_t size;
	u64 csum_end;
294
	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
295 296

	path = btrfs_alloc_path();
297 298
	if (!path)
		return -ENOMEM;
299

A
Arne Jansen 已提交
300 301 302 303 304 305
	if (search_commit) {
		path->skip_locking = 1;
		path->reada = 2;
		path->search_commit_root = 1;
	}

306 307 308 309
	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	key.offset = start;
	key.type = BTRFS_EXTENT_CSUM_KEY;

Y
Yan Zheng 已提交
310
	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328
	if (ret < 0)
		goto fail;
	if (ret > 0 && path->slots[0] > 0) {
		leaf = path->nodes[0];
		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
		    key.type == BTRFS_EXTENT_CSUM_KEY) {
			offset = (start - key.offset) >>
				 root->fs_info->sb->s_blocksize_bits;
			if (offset * csum_size <
			    btrfs_item_size_nr(leaf, path->slots[0] - 1))
				path->slots[0]--;
		}
	}

	while (start <= end) {
		leaf = path->nodes[0];
		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
Y
Yan Zheng 已提交
329
			ret = btrfs_next_leaf(root, path);
330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350
			if (ret < 0)
				goto fail;
			if (ret > 0)
				break;
			leaf = path->nodes[0];
		}

		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
		    key.type != BTRFS_EXTENT_CSUM_KEY)
			break;

		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
		if (key.offset > end)
			break;

		if (key.offset > start)
			start = key.offset;

		size = btrfs_item_size_nr(leaf, path->slots[0]);
		csum_end = key.offset + (size / csum_size) * root->sectorsize;
351 352 353 354
		if (csum_end <= start) {
			path->slots[0]++;
			continue;
		}
355

Y
Yan Zheng 已提交
356
		csum_end = min(csum_end, end + 1);
357 358
		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
				      struct btrfs_csum_item);
Y
Yan Zheng 已提交
359 360 361 362 363
		while (start < csum_end) {
			size = min_t(size_t, csum_end - start,
					MAX_ORDERED_SUM_BYTES(root));
			sums = kzalloc(btrfs_ordered_sum_size(root, size),
					GFP_NOFS);
364 365 366 367
			if (!sums) {
				ret = -ENOMEM;
				goto fail;
			}
368

Y
Yan Zheng 已提交
369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388
			sector_sum = sums->sums;
			sums->bytenr = start;
			sums->len = size;

			offset = (start - key.offset) >>
				root->fs_info->sb->s_blocksize_bits;
			offset *= csum_size;

			while (size > 0) {
				read_extent_buffer(path->nodes[0],
						&sector_sum->sum,
						((unsigned long)item) +
						offset, csum_size);
				sector_sum->bytenr = start;

				size -= root->sectorsize;
				start += root->sectorsize;
				offset += csum_size;
				sector_sum++;
			}
389
			list_add_tail(&sums->list, &tmplist);
Y
Yan Zheng 已提交
390
		}
391 392 393 394
		path->slots[0]++;
	}
	ret = 0;
fail:
395 396 397 398 399 400 401
	while (ret < 0 && !list_empty(&tmplist)) {
		sums = list_entry(&tmplist, struct btrfs_ordered_sum, list);
		list_del(&sums->list);
		kfree(sums);
	}
	list_splice_tail(&tmplist, list);

402 403 404 405
	btrfs_free_path(path);
	return ret;
}

406
int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
407
		       struct bio *bio, u64 file_start, int contig)
408
{
409 410
	struct btrfs_ordered_sum *sums;
	struct btrfs_sector_sum *sector_sum;
411
	struct btrfs_ordered_extent *ordered;
412 413 414
	char *data;
	struct bio_vec *bvec = bio->bi_io_vec;
	int bio_index = 0;
415 416 417
	unsigned long total_bytes = 0;
	unsigned long this_sum_bytes = 0;
	u64 offset;
418
	u64 disk_bytenr;
419

420 421
	WARN_ON(bio->bi_vcnt <= 0);
	sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
422 423
	if (!sums)
		return -ENOMEM;
424

425
	sector_sum = sums->sums;
426
	disk_bytenr = (u64)bio->bi_sector << 9;
427 428
	sums->len = bio->bi_size;
	INIT_LIST_HEAD(&sums->list);
429 430 431 432 433 434 435

	if (contig)
		offset = file_start;
	else
		offset = page_offset(bvec->bv_page) + bvec->bv_offset;

	ordered = btrfs_lookup_ordered_extent(inode, offset);
436
	BUG_ON(!ordered); /* Logic error */
437
	sums->bytenr = ordered->start;
438

C
Chris Mason 已提交
439
	while (bio_index < bio->bi_vcnt) {
440 441 442 443 444
		if (!contig)
			offset = page_offset(bvec->bv_page) + bvec->bv_offset;

		if (!contig && (offset >= ordered->file_offset + ordered->len ||
		    offset < ordered->file_offset)) {
445 446 447 448 449 450 451 452 453 454
			unsigned long bytes_left;
			sums->len = this_sum_bytes;
			this_sum_bytes = 0;
			btrfs_add_ordered_sum(inode, ordered, sums);
			btrfs_put_ordered_extent(ordered);

			bytes_left = bio->bi_size - total_bytes;

			sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
				       GFP_NOFS);
455
			BUG_ON(!sums); /* -ENOMEM */
456
			sector_sum = sums->sums;
457
			sums->len = bytes_left;
458
			ordered = btrfs_lookup_ordered_extent(inode, offset);
459
			BUG_ON(!ordered); /* Logic error */
460
			sums->bytenr = ordered->start;
461 462
		}

463
		data = kmap_atomic(bvec->bv_page);
464 465 466 467 468
		sector_sum->sum = ~(u32)0;
		sector_sum->sum = btrfs_csum_data(root,
						  data + bvec->bv_offset,
						  sector_sum->sum,
						  bvec->bv_len);
469
		kunmap_atomic(data);
470 471
		btrfs_csum_final(sector_sum->sum,
				 (char *)&sector_sum->sum);
472
		sector_sum->bytenr = disk_bytenr;
473

474
		sector_sum++;
475
		bio_index++;
476 477
		total_bytes += bvec->bv_len;
		this_sum_bytes += bvec->bv_len;
478 479
		disk_bytenr += bvec->bv_len;
		offset += bvec->bv_len;
480 481
		bvec++;
	}
482
	this_sum_bytes = 0;
483 484
	btrfs_add_ordered_sum(inode, ordered, sums);
	btrfs_put_ordered_extent(ordered);
485 486 487
	return 0;
}

488 489 490 491 492 493 494 495 496 497 498
/*
 * helper function for csum removal, this expects the
 * key to describe the csum pointed to by the path, and it expects
 * the csum to overlap the range [bytenr, len]
 *
 * The csum should not be entirely contained in the range and the
 * range should not be entirely contained in the csum.
 *
 * This calls btrfs_truncate_item with the correct args based on the
 * overlap, and fixes up the key as required.
 */
499 500 501 502 503
static noinline void truncate_one_csum(struct btrfs_trans_handle *trans,
				       struct btrfs_root *root,
				       struct btrfs_path *path,
				       struct btrfs_key *key,
				       u64 bytenr, u64 len)
504 505
{
	struct extent_buffer *leaf;
506
	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524
	u64 csum_end;
	u64 end_byte = bytenr + len;
	u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;

	leaf = path->nodes[0];
	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
	csum_end <<= root->fs_info->sb->s_blocksize_bits;
	csum_end += key->offset;

	if (key->offset < bytenr && csum_end <= end_byte) {
		/*
		 *         [ bytenr - len ]
		 *         [   ]
		 *   [csum     ]
		 *   A simple truncate off the end of the item
		 */
		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
		new_size *= csum_size;
525
		btrfs_truncate_item(trans, root, path, new_size, 1);
526 527 528 529 530 531 532 533 534 535 536
	} else if (key->offset >= bytenr && csum_end > end_byte &&
		   end_byte > key->offset) {
		/*
		 *         [ bytenr - len ]
		 *                 [ ]
		 *                 [csum     ]
		 * we need to truncate from the beginning of the csum
		 */
		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
		new_size *= csum_size;

537
		btrfs_truncate_item(trans, root, path, new_size, 0);
538 539

		key->offset = end_byte;
540
		btrfs_set_item_key_safe(trans, root, path, key);
541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558
	} else {
		BUG();
	}
}

/*
 * deletes the csum items from the csum tree for a given
 * range of bytes.
 */
int btrfs_del_csums(struct btrfs_trans_handle *trans,
		    struct btrfs_root *root, u64 bytenr, u64 len)
{
	struct btrfs_path *path;
	struct btrfs_key key;
	u64 end_byte = bytenr + len;
	u64 csum_end;
	struct extent_buffer *leaf;
	int ret;
559
	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
560 561 562 563 564
	int blocksize_bits = root->fs_info->sb->s_blocksize_bits;

	root = root->fs_info->csum_root;

	path = btrfs_alloc_path();
565 566
	if (!path)
		return -ENOMEM;
567

C
Chris Mason 已提交
568
	while (1) {
569 570 571 572
		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
		key.offset = end_byte - 1;
		key.type = BTRFS_EXTENT_CSUM_KEY;

573
		path->leave_spinning = 1;
574 575 576
		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
		if (ret > 0) {
			if (path->slots[0] == 0)
577
				break;
578
			path->slots[0]--;
579
		} else if (ret < 0) {
580
			break;
581
		}
582

583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604
		leaf = path->nodes[0];
		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);

		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
		    key.type != BTRFS_EXTENT_CSUM_KEY) {
			break;
		}

		if (key.offset >= end_byte)
			break;

		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
		csum_end <<= blocksize_bits;
		csum_end += key.offset;

		/* this csum ends before we start, we're done */
		if (csum_end <= bytenr)
			break;

		/* delete the entire item, it is inside our range */
		if (key.offset >= bytenr && csum_end <= end_byte) {
			ret = btrfs_del_item(trans, root, path);
605 606
			if (ret)
				goto out;
607 608
			if (key.offset == bytenr)
				break;
609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647
		} else if (key.offset < bytenr && csum_end > end_byte) {
			unsigned long offset;
			unsigned long shift_len;
			unsigned long item_offset;
			/*
			 *        [ bytenr - len ]
			 *     [csum                ]
			 *
			 * Our bytes are in the middle of the csum,
			 * we need to split this item and insert a new one.
			 *
			 * But we can't drop the path because the
			 * csum could change, get removed, extended etc.
			 *
			 * The trick here is the max size of a csum item leaves
			 * enough room in the tree block for a single
			 * item header.  So, we split the item in place,
			 * adding a new header pointing to the existing
			 * bytes.  Then we loop around again and we have
			 * a nicely formed csum item that we can neatly
			 * truncate.
			 */
			offset = (bytenr - key.offset) >> blocksize_bits;
			offset *= csum_size;

			shift_len = (len >> blocksize_bits) * csum_size;

			item_offset = btrfs_item_ptr_offset(leaf,
							    path->slots[0]);

			memset_extent_buffer(leaf, 0, item_offset + offset,
					     shift_len);
			key.offset = bytenr;

			/*
			 * btrfs_split_item returns -EAGAIN when the
			 * item changed size or key
			 */
			ret = btrfs_split_item(trans, root, path, &key, offset);
648 649 650 651
			if (ret && ret != -EAGAIN) {
				btrfs_abort_transaction(trans, root, ret);
				goto out;
			}
652 653 654

			key.offset = end_byte - 1;
		} else {
655
			truncate_one_csum(trans, root, path, &key, bytenr, len);
656 657
			if (key.offset < bytenr)
				break;
658
		}
659
		btrfs_release_path(path);
660
	}
661
	ret = 0;
662 663
out:
	btrfs_free_path(path);
664
	return ret;
665 666
}

667
int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
668
			   struct btrfs_root *root,
669
			   struct btrfs_ordered_sum *sums)
C
Chris Mason 已提交
670
{
671
	u64 bytenr;
C
Chris Mason 已提交
672 673
	int ret;
	struct btrfs_key file_key;
674
	struct btrfs_key found_key;
675
	u64 next_offset;
676
	u64 total_bytes = 0;
677
	int found_next;
678
	struct btrfs_path *path;
C
Chris Mason 已提交
679
	struct btrfs_csum_item *item;
680
	struct btrfs_csum_item *item_end;
681
	struct extent_buffer *leaf = NULL;
682
	u64 csum_offset;
683
	struct btrfs_sector_sum *sector_sum;
684 685
	u32 nritems;
	u32 ins_size;
686
	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
687

688
	path = btrfs_alloc_path();
689 690 691
	if (!path)
		return -ENOMEM;

692
	sector_sum = sums->sums;
693
	trans->adding_csums = 1;
694 695 696
again:
	next_offset = (u64)-1;
	found_next = 0;
697 698 699 700
	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	file_key.offset = sector_sum->bytenr;
	bytenr = sector_sum->bytenr;
	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
701

702
	item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
703 704
	if (!IS_ERR(item)) {
		leaf = path->nodes[0];
705
		ret = 0;
706
		goto found;
707
	}
708
	ret = PTR_ERR(item);
709 710 711
	if (ret != -EFBIG && ret != -ENOENT)
		goto fail_unlock;

712 713 714
	if (ret == -EFBIG) {
		u32 item_size;
		/* we found one, but it isn't big enough yet */
715 716
		leaf = path->nodes[0];
		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
717 718
		if ((item_size / csum_size) >=
		    MAX_CSUM_ITEMS(root, csum_size)) {
719 720 721 722
			/* already at max size, make a new one */
			goto insert;
		}
	} else {
723
		int slot = path->slots[0] + 1;
724
		/* we didn't find a csum item, insert one */
725 726 727
		nritems = btrfs_header_nritems(path->nodes[0]);
		if (path->slots[0] >= nritems - 1) {
			ret = btrfs_next_leaf(root, path);
Y
Yan 已提交
728
			if (ret == 1)
729
				found_next = 1;
Y
Yan 已提交
730
			if (ret != 0)
731
				goto insert;
Y
Yan 已提交
732
			slot = 0;
733 734
		}
		btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
735 736
		if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
		    found_key.type != BTRFS_EXTENT_CSUM_KEY) {
737 738 739 740 741
			found_next = 1;
			goto insert;
		}
		next_offset = found_key.offset;
		found_next = 1;
742 743 744 745 746 747 748
		goto insert;
	}

	/*
	 * at this point, we know the tree has an item, but it isn't big
	 * enough yet to put our csum in.  Grow it
	 */
749
	btrfs_release_path(path);
750
	ret = btrfs_search_slot(trans, root, &file_key, path,
751
				csum_size, 1);
752
	if (ret < 0)
753
		goto fail_unlock;
754 755 756 757 758

	if (ret > 0) {
		if (path->slots[0] == 0)
			goto insert;
		path->slots[0]--;
759
	}
760

761 762
	leaf = path->nodes[0];
	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
763
	csum_offset = (bytenr - found_key.offset) >>
764
			root->fs_info->sb->s_blocksize_bits;
765

766 767
	if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
768
	    csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
769 770
		goto insert;
	}
771

772
	if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
773 774
	    csum_size) {
		u32 diff = (csum_offset + 1) * csum_size;
775 776 777 778 779 780 781 782

		/*
		 * is the item big enough already?  we dropped our lock
		 * before and need to recheck
		 */
		if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
			goto csum;

783
		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
C
Chris Mason 已提交
784
		if (diff != csum_size)
C
Chris Mason 已提交
785
			goto insert;
786

787
		btrfs_extend_item(trans, root, path, diff);
788 789 790 791
		goto csum;
	}

insert:
792
	btrfs_release_path(path);
793
	csum_offset = 0;
794
	if (found_next) {
795 796 797 798
		u64 tmp = total_bytes + root->sectorsize;
		u64 next_sector = sector_sum->bytenr;
		struct btrfs_sector_sum *next = sector_sum + 1;

C
Chris Mason 已提交
799
		while (tmp < sums->len) {
800 801 802 803 804 805 806
			if (next_sector + root->sectorsize != next->bytenr)
				break;
			tmp += root->sectorsize;
			next_sector = next->bytenr;
			next++;
		}
		tmp = min(tmp, next_offset - file_key.offset);
807 808
		tmp >>= root->fs_info->sb->s_blocksize_bits;
		tmp = max((u64)1, tmp);
809 810
		tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
		ins_size = csum_size * tmp;
811
	} else {
812
		ins_size = csum_size;
813
	}
814
	path->leave_spinning = 1;
815
	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
816
				      ins_size);
817
	path->leave_spinning = 0;
818
	if (ret < 0)
819
		goto fail_unlock;
820 821
	if (ret != 0) {
		WARN_ON(1);
822
		goto fail_unlock;
823
	}
824
csum:
825 826
	leaf = path->nodes[0];
	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
C
Chris Mason 已提交
827
	ret = 0;
828
	item = (struct btrfs_csum_item *)((unsigned char *)item +
829
					  csum_offset * csum_size);
C
Chris Mason 已提交
830
found:
831 832 833
	item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
	item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
				      btrfs_item_size_nr(leaf, path->slots[0]));
834
next_sector:
835

836
	write_extent_buffer(leaf, &sector_sum->sum, (unsigned long)item, csum_size);
837

838 839 840
	total_bytes += root->sectorsize;
	sector_sum++;
	if (total_bytes < sums->len) {
841
		item = (struct btrfs_csum_item *)((char *)item +
842
						  csum_size);
843 844 845
		if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
		    sector_sum->bytenr) {
			bytenr = sector_sum->bytenr;
846
			goto next_sector;
847
		}
848
	}
849

850
	btrfs_mark_buffer_dirty(path->nodes[0]);
851
	if (total_bytes < sums->len) {
852
		btrfs_release_path(path);
853
		cond_resched();
854 855
		goto again;
	}
856
out:
857
	trans->adding_csums = 0;
858
	btrfs_free_path(path);
C
Chris Mason 已提交
859
	return ret;
860 861 862

fail_unlock:
	goto out;
C
Chris Mason 已提交
863
}