xref: /openbmc/linux/fs/btrfs/file-item.c (revision 7e035230)
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 
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/pagemap.h>
22 #include <linux/highmem.h>
23 #include "ctree.h"
24 #include "disk-io.h"
25 #include "transaction.h"
26 #include "print-tree.h"
27 
28 #define __MAX_CSUM_ITEMS(r, size) ((((BTRFS_LEAF_DATA_SIZE(r) - \
29 				   sizeof(struct btrfs_item) * 2) / \
30 				  size) - 1))
31 
32 #define MAX_CSUM_ITEMS(r, size) (min(__MAX_CSUM_ITEMS(r, size), PAGE_CACHE_SIZE))
33 
34 #define MAX_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
35 				   sizeof(struct btrfs_ordered_sum)) / \
36 				   sizeof(struct btrfs_sector_sum) * \
37 				   (r)->sectorsize - (r)->sectorsize)
38 
39 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
40 			     struct btrfs_root *root,
41 			     u64 objectid, u64 pos,
42 			     u64 disk_offset, u64 disk_num_bytes,
43 			     u64 num_bytes, u64 offset, u64 ram_bytes,
44 			     u8 compression, u8 encryption, u16 other_encoding)
45 {
46 	int ret = 0;
47 	struct btrfs_file_extent_item *item;
48 	struct btrfs_key file_key;
49 	struct btrfs_path *path;
50 	struct extent_buffer *leaf;
51 
52 	path = btrfs_alloc_path();
53 	if (!path)
54 		return -ENOMEM;
55 	file_key.objectid = objectid;
56 	file_key.offset = pos;
57 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
58 
59 	path->leave_spinning = 1;
60 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
61 				      sizeof(*item));
62 	if (ret < 0)
63 		goto out;
64 	BUG_ON(ret); /* Can't happen */
65 	leaf = path->nodes[0];
66 	item = btrfs_item_ptr(leaf, path->slots[0],
67 			      struct btrfs_file_extent_item);
68 	btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
69 	btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
70 	btrfs_set_file_extent_offset(leaf, item, offset);
71 	btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
72 	btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
73 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
74 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
75 	btrfs_set_file_extent_compression(leaf, item, compression);
76 	btrfs_set_file_extent_encryption(leaf, item, encryption);
77 	btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
78 
79 	btrfs_mark_buffer_dirty(leaf);
80 out:
81 	btrfs_free_path(path);
82 	return ret;
83 }
84 
85 struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
86 					  struct btrfs_root *root,
87 					  struct btrfs_path *path,
88 					  u64 bytenr, int cow)
89 {
90 	int ret;
91 	struct btrfs_key file_key;
92 	struct btrfs_key found_key;
93 	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 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
100 	file_key.offset = bytenr;
101 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
102 	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
103 	if (ret < 0)
104 		goto fail;
105 	leaf = path->nodes[0];
106 	if (ret > 0) {
107 		ret = 1;
108 		if (path->slots[0] == 0)
109 			goto fail;
110 		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 		if (csum_offset >= csums_in_item) {
121 			ret = -EFBIG;
122 			goto fail;
123 		}
124 	}
125 	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 	return item;
129 fail:
130 	if (ret > 0)
131 		ret = -ENOENT;
132 	return ERR_PTR(ret);
133 }
134 
135 
136 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
137 			     struct btrfs_root *root,
138 			     struct btrfs_path *path, u64 objectid,
139 			     u64 offset, int mod)
140 {
141 	int ret;
142 	struct btrfs_key file_key;
143 	int ins_len = mod < 0 ? -1 : 0;
144 	int cow = mod != 0;
145 
146 	file_key.objectid = objectid;
147 	file_key.offset = offset;
148 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
149 	ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
150 	return ret;
151 }
152 
153 
154 static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
155 				   struct inode *inode, struct bio *bio,
156 				   u64 logical_offset, u32 *dst, int dio)
157 {
158 	u32 sum;
159 	struct bio_vec *bvec = bio->bi_io_vec;
160 	int bio_index = 0;
161 	u64 offset = 0;
162 	u64 item_start_offset = 0;
163 	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 	int ret;
168 	struct btrfs_path *path;
169 	struct btrfs_csum_item *item = NULL;
170 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
171 
172 	path = btrfs_alloc_path();
173 	if (!path)
174 		return -ENOMEM;
175 	if (bio->bi_size > PAGE_CACHE_SIZE * 8)
176 		path->reada = 2;
177 
178 	WARN_ON(bio->bi_vcnt <= 0);
179 
180 	/*
181 	 * the free space stuff is only read when it hasn't been
182 	 * updated in the current transaction.  So, we can safely
183 	 * read from the commit root and sidestep a nasty deadlock
184 	 * between reading the free space cache and updating the csum tree.
185 	 */
186 	if (btrfs_is_free_space_inode(inode)) {
187 		path->search_commit_root = 1;
188 		path->skip_locking = 1;
189 	}
190 
191 	disk_bytenr = (u64)bio->bi_sector << 9;
192 	if (dio)
193 		offset = logical_offset;
194 	while (bio_index < bio->bi_vcnt) {
195 		if (!dio)
196 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
197 		ret = btrfs_find_ordered_sum(inode, offset, disk_bytenr, &sum);
198 		if (ret == 0)
199 			goto found;
200 
201 		if (!item || disk_bytenr < item_start_offset ||
202 		    disk_bytenr >= item_last_offset) {
203 			struct btrfs_key found_key;
204 			u32 item_size;
205 
206 			if (item)
207 				btrfs_release_path(path);
208 			item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
209 						 path, disk_bytenr, 0);
210 			if (IS_ERR(item)) {
211 				ret = PTR_ERR(item);
212 				if (ret == -ENOENT || ret == -EFBIG)
213 					ret = 0;
214 				sum = 0;
215 				if (BTRFS_I(inode)->root->root_key.objectid ==
216 				    BTRFS_DATA_RELOC_TREE_OBJECTID) {
217 					set_extent_bits(io_tree, offset,
218 						offset + bvec->bv_len - 1,
219 						EXTENT_NODATASUM, GFP_NOFS);
220 				} else {
221 					printk(KERN_INFO "btrfs no csum found "
222 					       "for inode %llu start %llu\n",
223 					       (unsigned long long)
224 					       btrfs_ino(inode),
225 					       (unsigned long long)offset);
226 				}
227 				item = NULL;
228 				btrfs_release_path(path);
229 				goto found;
230 			}
231 			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
232 					      path->slots[0]);
233 
234 			item_start_offset = found_key.offset;
235 			item_size = btrfs_item_size_nr(path->nodes[0],
236 						       path->slots[0]);
237 			item_last_offset = item_start_offset +
238 				(item_size / csum_size) *
239 				root->sectorsize;
240 			item = btrfs_item_ptr(path->nodes[0], path->slots[0],
241 					      struct btrfs_csum_item);
242 		}
243 		/*
244 		 * this byte range must be able to fit inside
245 		 * a single leaf so it will also fit inside a u32
246 		 */
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 		if (dst)
256 			*dst++ = sum;
257 		else
258 			set_state_private(io_tree, offset, sum);
259 		disk_bytenr += bvec->bv_len;
260 		offset += bvec->bv_len;
261 		bio_index++;
262 		bvec++;
263 	}
264 	btrfs_free_path(path);
265 	return 0;
266 }
267 
268 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
269 			  struct bio *bio, u32 *dst)
270 {
271 	return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
272 }
273 
274 int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
275 			      struct bio *bio, u64 offset, u32 *dst)
276 {
277 	return __btrfs_lookup_bio_sums(root, inode, bio, offset, dst, 1);
278 }
279 
280 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
281 			     struct list_head *list, int search_commit)
282 {
283 	struct btrfs_key key;
284 	struct btrfs_path *path;
285 	struct extent_buffer *leaf;
286 	struct btrfs_ordered_sum *sums;
287 	struct btrfs_sector_sum *sector_sum;
288 	struct btrfs_csum_item *item;
289 	LIST_HEAD(tmplist);
290 	unsigned long offset;
291 	int ret;
292 	size_t size;
293 	u64 csum_end;
294 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
295 
296 	path = btrfs_alloc_path();
297 	if (!path)
298 		return -ENOMEM;
299 
300 	if (search_commit) {
301 		path->skip_locking = 1;
302 		path->reada = 2;
303 		path->search_commit_root = 1;
304 	}
305 
306 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
307 	key.offset = start;
308 	key.type = BTRFS_EXTENT_CSUM_KEY;
309 
310 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
311 	if (ret < 0)
312 		goto fail;
313 	if (ret > 0 && path->slots[0] > 0) {
314 		leaf = path->nodes[0];
315 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
316 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
317 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
318 			offset = (start - key.offset) >>
319 				 root->fs_info->sb->s_blocksize_bits;
320 			if (offset * csum_size <
321 			    btrfs_item_size_nr(leaf, path->slots[0] - 1))
322 				path->slots[0]--;
323 		}
324 	}
325 
326 	while (start <= end) {
327 		leaf = path->nodes[0];
328 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
329 			ret = btrfs_next_leaf(root, path);
330 			if (ret < 0)
331 				goto fail;
332 			if (ret > 0)
333 				break;
334 			leaf = path->nodes[0];
335 		}
336 
337 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
338 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
339 		    key.type != BTRFS_EXTENT_CSUM_KEY)
340 			break;
341 
342 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
343 		if (key.offset > end)
344 			break;
345 
346 		if (key.offset > start)
347 			start = key.offset;
348 
349 		size = btrfs_item_size_nr(leaf, path->slots[0]);
350 		csum_end = key.offset + (size / csum_size) * root->sectorsize;
351 		if (csum_end <= start) {
352 			path->slots[0]++;
353 			continue;
354 		}
355 
356 		csum_end = min(csum_end, end + 1);
357 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
358 				      struct btrfs_csum_item);
359 		while (start < csum_end) {
360 			size = min_t(size_t, csum_end - start,
361 					MAX_ORDERED_SUM_BYTES(root));
362 			sums = kzalloc(btrfs_ordered_sum_size(root, size),
363 					GFP_NOFS);
364 			if (!sums) {
365 				ret = -ENOMEM;
366 				goto fail;
367 			}
368 
369 			sector_sum = sums->sums;
370 			sums->bytenr = start;
371 			sums->len = size;
372 
373 			offset = (start - key.offset) >>
374 				root->fs_info->sb->s_blocksize_bits;
375 			offset *= csum_size;
376 
377 			while (size > 0) {
378 				read_extent_buffer(path->nodes[0],
379 						&sector_sum->sum,
380 						((unsigned long)item) +
381 						offset, csum_size);
382 				sector_sum->bytenr = start;
383 
384 				size -= root->sectorsize;
385 				start += root->sectorsize;
386 				offset += csum_size;
387 				sector_sum++;
388 			}
389 			list_add_tail(&sums->list, &tmplist);
390 		}
391 		path->slots[0]++;
392 	}
393 	ret = 0;
394 fail:
395 	while (ret < 0 && !list_empty(&tmplist)) {
396 		sums = list_entry(&tmplist, struct btrfs_ordered_sum, list);
397 		list_del(&sums->list);
398 		kfree(sums);
399 	}
400 	list_splice_tail(&tmplist, list);
401 
402 	btrfs_free_path(path);
403 	return ret;
404 }
405 
406 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
407 		       struct bio *bio, u64 file_start, int contig)
408 {
409 	struct btrfs_ordered_sum *sums;
410 	struct btrfs_sector_sum *sector_sum;
411 	struct btrfs_ordered_extent *ordered;
412 	char *data;
413 	struct bio_vec *bvec = bio->bi_io_vec;
414 	int bio_index = 0;
415 	unsigned long total_bytes = 0;
416 	unsigned long this_sum_bytes = 0;
417 	u64 offset;
418 	u64 disk_bytenr;
419 
420 	WARN_ON(bio->bi_vcnt <= 0);
421 	sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
422 	if (!sums)
423 		return -ENOMEM;
424 
425 	sector_sum = sums->sums;
426 	disk_bytenr = (u64)bio->bi_sector << 9;
427 	sums->len = bio->bi_size;
428 	INIT_LIST_HEAD(&sums->list);
429 
430 	if (contig)
431 		offset = file_start;
432 	else
433 		offset = page_offset(bvec->bv_page) + bvec->bv_offset;
434 
435 	ordered = btrfs_lookup_ordered_extent(inode, offset);
436 	BUG_ON(!ordered); /* Logic error */
437 	sums->bytenr = ordered->start;
438 
439 	while (bio_index < bio->bi_vcnt) {
440 		if (!contig)
441 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
442 
443 		if (!contig && (offset >= ordered->file_offset + ordered->len ||
444 		    offset < ordered->file_offset)) {
445 			unsigned long bytes_left;
446 			sums->len = this_sum_bytes;
447 			this_sum_bytes = 0;
448 			btrfs_add_ordered_sum(inode, ordered, sums);
449 			btrfs_put_ordered_extent(ordered);
450 
451 			bytes_left = bio->bi_size - total_bytes;
452 
453 			sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
454 				       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 		sector_sum->sum = ~(u32)0;
465 		sector_sum->sum = btrfs_csum_data(root,
466 						  data + bvec->bv_offset,
467 						  sector_sum->sum,
468 						  bvec->bv_len);
469 		kunmap_atomic(data);
470 		btrfs_csum_final(sector_sum->sum,
471 				 (char *)&sector_sum->sum);
472 		sector_sum->bytenr = disk_bytenr;
473 
474 		sector_sum++;
475 		bio_index++;
476 		total_bytes += bvec->bv_len;
477 		this_sum_bytes += bvec->bv_len;
478 		disk_bytenr += bvec->bv_len;
479 		offset += bvec->bv_len;
480 		bvec++;
481 	}
482 	this_sum_bytes = 0;
483 	btrfs_add_ordered_sum(inode, ordered, sums);
484 	btrfs_put_ordered_extent(ordered);
485 	return 0;
486 }
487 
488 /*
489  * helper function for csum removal, this expects the
490  * key to describe the csum pointed to by the path, and it expects
491  * the csum to overlap the range [bytenr, len]
492  *
493  * The csum should not be entirely contained in the range and the
494  * range should not be entirely contained in the csum.
495  *
496  * This calls btrfs_truncate_item with the correct args based on the
497  * overlap, and fixes up the key as required.
498  */
499 static noinline void truncate_one_csum(struct btrfs_trans_handle *trans,
500 				       struct btrfs_root *root,
501 				       struct btrfs_path *path,
502 				       struct btrfs_key *key,
503 				       u64 bytenr, u64 len)
504 {
505 	struct extent_buffer *leaf;
506 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
507 	u64 csum_end;
508 	u64 end_byte = bytenr + len;
509 	u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
510 
511 	leaf = path->nodes[0];
512 	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
513 	csum_end <<= root->fs_info->sb->s_blocksize_bits;
514 	csum_end += key->offset;
515 
516 	if (key->offset < bytenr && csum_end <= end_byte) {
517 		/*
518 		 *         [ bytenr - len ]
519 		 *         [   ]
520 		 *   [csum     ]
521 		 *   A simple truncate off the end of the item
522 		 */
523 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
524 		new_size *= csum_size;
525 		btrfs_truncate_item(trans, root, path, new_size, 1);
526 	} else if (key->offset >= bytenr && csum_end > end_byte &&
527 		   end_byte > key->offset) {
528 		/*
529 		 *         [ bytenr - len ]
530 		 *                 [ ]
531 		 *                 [csum     ]
532 		 * we need to truncate from the beginning of the csum
533 		 */
534 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
535 		new_size *= csum_size;
536 
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 	} else {
542 		BUG();
543 	}
544 }
545 
546 /*
547  * deletes the csum items from the csum tree for a given
548  * range of bytes.
549  */
550 int btrfs_del_csums(struct btrfs_trans_handle *trans,
551 		    struct btrfs_root *root, u64 bytenr, u64 len)
552 {
553 	struct btrfs_path *path;
554 	struct btrfs_key key;
555 	u64 end_byte = bytenr + len;
556 	u64 csum_end;
557 	struct extent_buffer *leaf;
558 	int ret;
559 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
560 	int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
561 
562 	root = root->fs_info->csum_root;
563 
564 	path = btrfs_alloc_path();
565 	if (!path)
566 		return -ENOMEM;
567 
568 	while (1) {
569 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
570 		key.offset = end_byte - 1;
571 		key.type = BTRFS_EXTENT_CSUM_KEY;
572 
573 		path->leave_spinning = 1;
574 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
575 		if (ret > 0) {
576 			if (path->slots[0] == 0)
577 				break;
578 			path->slots[0]--;
579 		} else if (ret < 0) {
580 			break;
581 		}
582 
583 		leaf = path->nodes[0];
584 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
585 
586 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
587 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
588 			break;
589 		}
590 
591 		if (key.offset >= end_byte)
592 			break;
593 
594 		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
595 		csum_end <<= blocksize_bits;
596 		csum_end += key.offset;
597 
598 		/* this csum ends before we start, we're done */
599 		if (csum_end <= bytenr)
600 			break;
601 
602 		/* delete the entire item, it is inside our range */
603 		if (key.offset >= bytenr && csum_end <= end_byte) {
604 			ret = btrfs_del_item(trans, root, path);
605 			if (ret)
606 				goto out;
607 			if (key.offset == bytenr)
608 				break;
609 		} else if (key.offset < bytenr && csum_end > end_byte) {
610 			unsigned long offset;
611 			unsigned long shift_len;
612 			unsigned long item_offset;
613 			/*
614 			 *        [ bytenr - len ]
615 			 *     [csum                ]
616 			 *
617 			 * Our bytes are in the middle of the csum,
618 			 * we need to split this item and insert a new one.
619 			 *
620 			 * But we can't drop the path because the
621 			 * csum could change, get removed, extended etc.
622 			 *
623 			 * The trick here is the max size of a csum item leaves
624 			 * enough room in the tree block for a single
625 			 * item header.  So, we split the item in place,
626 			 * adding a new header pointing to the existing
627 			 * bytes.  Then we loop around again and we have
628 			 * a nicely formed csum item that we can neatly
629 			 * truncate.
630 			 */
631 			offset = (bytenr - key.offset) >> blocksize_bits;
632 			offset *= csum_size;
633 
634 			shift_len = (len >> blocksize_bits) * csum_size;
635 
636 			item_offset = btrfs_item_ptr_offset(leaf,
637 							    path->slots[0]);
638 
639 			memset_extent_buffer(leaf, 0, item_offset + offset,
640 					     shift_len);
641 			key.offset = bytenr;
642 
643 			/*
644 			 * btrfs_split_item returns -EAGAIN when the
645 			 * item changed size or key
646 			 */
647 			ret = btrfs_split_item(trans, root, path, &key, offset);
648 			if (ret && ret != -EAGAIN) {
649 				btrfs_abort_transaction(trans, root, ret);
650 				goto out;
651 			}
652 
653 			key.offset = end_byte - 1;
654 		} else {
655 			truncate_one_csum(trans, root, path, &key, bytenr, len);
656 			if (key.offset < bytenr)
657 				break;
658 		}
659 		btrfs_release_path(path);
660 	}
661 	ret = 0;
662 out:
663 	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)
670 {
671 	u64 bytenr;
672 	int ret;
673 	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;
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 	u32 nritems;
685 	u32 ins_size;
686 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
687 
688 	path = btrfs_alloc_path();
689 	if (!path)
690 		return -ENOMEM;
691 
692 	sector_sum = sums->sums;
693 	trans->adding_csums = 1;
694 again:
695 	next_offset = (u64)-1;
696 	found_next = 0;
697 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
698 	file_key.offset = sector_sum->bytenr;
699 	bytenr = sector_sum->bytenr;
700 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
701 
702 	item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
703 	if (!IS_ERR(item)) {
704 		leaf = path->nodes[0];
705 		ret = 0;
706 		goto found;
707 	}
708 	ret = PTR_ERR(item);
709 	if (ret != -EFBIG && ret != -ENOENT)
710 		goto fail_unlock;
711 
712 	if (ret == -EFBIG) {
713 		u32 item_size;
714 		/* we found one, but it isn't big enough yet */
715 		leaf = path->nodes[0];
716 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
717 		if ((item_size / csum_size) >=
718 		    MAX_CSUM_ITEMS(root, csum_size)) {
719 			/* already at max size, make a new one */
720 			goto insert;
721 		}
722 	} else {
723 		int slot = path->slots[0] + 1;
724 		/* we didn't find a csum item, insert one */
725 		nritems = btrfs_header_nritems(path->nodes[0]);
726 		if (path->slots[0] >= nritems - 1) {
727 			ret = btrfs_next_leaf(root, path);
728 			if (ret == 1)
729 				found_next = 1;
730 			if (ret != 0)
731 				goto insert;
732 			slot = 0;
733 		}
734 		btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
735 		if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
736 		    found_key.type != BTRFS_EXTENT_CSUM_KEY) {
737 			found_next = 1;
738 			goto insert;
739 		}
740 		next_offset = found_key.offset;
741 		found_next = 1;
742 		goto insert;
743 	}
744 
745 	/*
746 	 * at this point, we know the tree has an item, but it isn't big
747 	 * enough yet to put our csum in.  Grow it
748 	 */
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 	if (ret > 0) {
756 		if (path->slots[0] == 0)
757 			goto insert;
758 		path->slots[0]--;
759 	}
760 
761 	leaf = path->nodes[0];
762 	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 	if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
767 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
768 	    csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
769 		goto insert;
770 	}
771 
772 	if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
773 	    csum_size) {
774 		u32 diff = (csum_offset + 1) * csum_size;
775 
776 		/*
777 		 * is the item big enough already?  we dropped our lock
778 		 * before and need to recheck
779 		 */
780 		if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
781 			goto csum;
782 
783 		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
784 		if (diff != csum_size)
785 			goto insert;
786 
787 		btrfs_extend_item(trans, root, path, diff);
788 		goto csum;
789 	}
790 
791 insert:
792 	btrfs_release_path(path);
793 	csum_offset = 0;
794 	if (found_next) {
795 		u64 tmp = total_bytes + root->sectorsize;
796 		u64 next_sector = sector_sum->bytenr;
797 		struct btrfs_sector_sum *next = sector_sum + 1;
798 
799 		while (tmp < sums->len) {
800 			if (next_sector + root->sectorsize != next->bytenr)
801 				break;
802 			tmp += root->sectorsize;
803 			next_sector = next->bytenr;
804 			next++;
805 		}
806 		tmp = min(tmp, next_offset - file_key.offset);
807 		tmp >>= root->fs_info->sb->s_blocksize_bits;
808 		tmp = max((u64)1, tmp);
809 		tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
810 		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 	if (ret != 0) {
821 		WARN_ON(1);
822 		goto fail_unlock;
823 	}
824 csum:
825 	leaf = path->nodes[0];
826 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
827 	ret = 0;
828 	item = (struct btrfs_csum_item *)((unsigned char *)item +
829 					  csum_offset * csum_size);
830 found:
831 	item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
832 	item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
833 				      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 	total_bytes += root->sectorsize;
839 	sector_sum++;
840 	if (total_bytes < sums->len) {
841 		item = (struct btrfs_csum_item *)((char *)item +
842 						  csum_size);
843 		if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
844 		    sector_sum->bytenr) {
845 			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 		goto again;
855 	}
856 out:
857 	trans->adding_csums = 0;
858 	btrfs_free_path(path);
859 	return ret;
860 
861 fail_unlock:
862 	goto out;
863 }
864