xref: /linux/fs/btrfs/file-item.c (revision f9c41a62bba3f3f7ef3541b2a025e3371bcbba97)
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) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
29 				   sizeof(struct btrfs_item) * 2) / \
30 				  size) - 1))
31 
32 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
33 				       PAGE_CACHE_SIZE))
34 
35 #define MAX_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
36 				   sizeof(struct btrfs_ordered_sum)) / \
37 				   sizeof(struct btrfs_sector_sum) * \
38 				   (r)->sectorsize - (r)->sectorsize)
39 
40 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
41 			     struct btrfs_root *root,
42 			     u64 objectid, u64 pos,
43 			     u64 disk_offset, u64 disk_num_bytes,
44 			     u64 num_bytes, u64 offset, u64 ram_bytes,
45 			     u8 compression, u8 encryption, u16 other_encoding)
46 {
47 	int ret = 0;
48 	struct btrfs_file_extent_item *item;
49 	struct btrfs_key file_key;
50 	struct btrfs_path *path;
51 	struct extent_buffer *leaf;
52 
53 	path = btrfs_alloc_path();
54 	if (!path)
55 		return -ENOMEM;
56 	file_key.objectid = objectid;
57 	file_key.offset = pos;
58 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
59 
60 	path->leave_spinning = 1;
61 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
62 				      sizeof(*item));
63 	if (ret < 0)
64 		goto out;
65 	BUG_ON(ret); /* Can't happen */
66 	leaf = path->nodes[0];
67 	item = btrfs_item_ptr(leaf, path->slots[0],
68 			      struct btrfs_file_extent_item);
69 	btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
70 	btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
71 	btrfs_set_file_extent_offset(leaf, item, offset);
72 	btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
73 	btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
74 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
75 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
76 	btrfs_set_file_extent_compression(leaf, item, compression);
77 	btrfs_set_file_extent_encryption(leaf, item, encryption);
78 	btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
79 
80 	btrfs_mark_buffer_dirty(leaf);
81 out:
82 	btrfs_free_path(path);
83 	return ret;
84 }
85 
86 struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
87 					  struct btrfs_root *root,
88 					  struct btrfs_path *path,
89 					  u64 bytenr, int cow)
90 {
91 	int ret;
92 	struct btrfs_key file_key;
93 	struct btrfs_key found_key;
94 	struct btrfs_csum_item *item;
95 	struct extent_buffer *leaf;
96 	u64 csum_offset = 0;
97 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
98 	int csums_in_item;
99 
100 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
101 	file_key.offset = bytenr;
102 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
103 	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
104 	if (ret < 0)
105 		goto fail;
106 	leaf = path->nodes[0];
107 	if (ret > 0) {
108 		ret = 1;
109 		if (path->slots[0] == 0)
110 			goto fail;
111 		path->slots[0]--;
112 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
113 		if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY)
114 			goto fail;
115 
116 		csum_offset = (bytenr - found_key.offset) >>
117 				root->fs_info->sb->s_blocksize_bits;
118 		csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
119 		csums_in_item /= csum_size;
120 
121 		if (csum_offset == csums_in_item) {
122 			ret = -EFBIG;
123 			goto fail;
124 		} else if (csum_offset > csums_in_item) {
125 			goto fail;
126 		}
127 	}
128 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
129 	item = (struct btrfs_csum_item *)((unsigned char *)item +
130 					  csum_offset * csum_size);
131 	return item;
132 fail:
133 	if (ret > 0)
134 		ret = -ENOENT;
135 	return ERR_PTR(ret);
136 }
137 
138 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
139 			     struct btrfs_root *root,
140 			     struct btrfs_path *path, u64 objectid,
141 			     u64 offset, int mod)
142 {
143 	int ret;
144 	struct btrfs_key file_key;
145 	int ins_len = mod < 0 ? -1 : 0;
146 	int cow = mod != 0;
147 
148 	file_key.objectid = objectid;
149 	file_key.offset = offset;
150 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
151 	ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
152 	return ret;
153 }
154 
155 u64 btrfs_file_extent_length(struct btrfs_path *path)
156 {
157 	int extent_type;
158 	struct btrfs_file_extent_item *fi;
159 	u64 len;
160 
161 	fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
162 			    struct btrfs_file_extent_item);
163 	extent_type = btrfs_file_extent_type(path->nodes[0], fi);
164 
165 	if (extent_type == BTRFS_FILE_EXTENT_REG ||
166 	    extent_type == BTRFS_FILE_EXTENT_PREALLOC)
167 		len = btrfs_file_extent_num_bytes(path->nodes[0], fi);
168 	else if (extent_type == BTRFS_FILE_EXTENT_INLINE)
169 		len = btrfs_file_extent_inline_len(path->nodes[0], fi);
170 	else
171 		BUG();
172 
173 	return len;
174 }
175 
176 static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
177 				   struct inode *inode, struct bio *bio,
178 				   u64 logical_offset, u32 *dst, int dio)
179 {
180 	u32 sum;
181 	struct bio_vec *bvec = bio->bi_io_vec;
182 	int bio_index = 0;
183 	u64 offset = 0;
184 	u64 item_start_offset = 0;
185 	u64 item_last_offset = 0;
186 	u64 disk_bytenr;
187 	u32 diff;
188 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
189 	int ret;
190 	struct btrfs_path *path;
191 	struct btrfs_csum_item *item = NULL;
192 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
193 
194 	path = btrfs_alloc_path();
195 	if (!path)
196 		return -ENOMEM;
197 	if (bio->bi_size > PAGE_CACHE_SIZE * 8)
198 		path->reada = 2;
199 
200 	WARN_ON(bio->bi_vcnt <= 0);
201 
202 	/*
203 	 * the free space stuff is only read when it hasn't been
204 	 * updated in the current transaction.  So, we can safely
205 	 * read from the commit root and sidestep a nasty deadlock
206 	 * between reading the free space cache and updating the csum tree.
207 	 */
208 	if (btrfs_is_free_space_inode(inode)) {
209 		path->search_commit_root = 1;
210 		path->skip_locking = 1;
211 	}
212 
213 	disk_bytenr = (u64)bio->bi_sector << 9;
214 	if (dio)
215 		offset = logical_offset;
216 	while (bio_index < bio->bi_vcnt) {
217 		if (!dio)
218 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
219 		ret = btrfs_find_ordered_sum(inode, offset, disk_bytenr, &sum);
220 		if (ret == 0)
221 			goto found;
222 
223 		if (!item || disk_bytenr < item_start_offset ||
224 		    disk_bytenr >= item_last_offset) {
225 			struct btrfs_key found_key;
226 			u32 item_size;
227 
228 			if (item)
229 				btrfs_release_path(path);
230 			item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
231 						 path, disk_bytenr, 0);
232 			if (IS_ERR(item)) {
233 				ret = PTR_ERR(item);
234 				if (ret == -ENOENT || ret == -EFBIG)
235 					ret = 0;
236 				sum = 0;
237 				if (BTRFS_I(inode)->root->root_key.objectid ==
238 				    BTRFS_DATA_RELOC_TREE_OBJECTID) {
239 					set_extent_bits(io_tree, offset,
240 						offset + bvec->bv_len - 1,
241 						EXTENT_NODATASUM, GFP_NOFS);
242 				} else {
243 					printk(KERN_INFO "btrfs no csum found "
244 					       "for inode %llu start %llu\n",
245 					       (unsigned long long)
246 					       btrfs_ino(inode),
247 					       (unsigned long long)offset);
248 				}
249 				item = NULL;
250 				btrfs_release_path(path);
251 				goto found;
252 			}
253 			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
254 					      path->slots[0]);
255 
256 			item_start_offset = found_key.offset;
257 			item_size = btrfs_item_size_nr(path->nodes[0],
258 						       path->slots[0]);
259 			item_last_offset = item_start_offset +
260 				(item_size / csum_size) *
261 				root->sectorsize;
262 			item = btrfs_item_ptr(path->nodes[0], path->slots[0],
263 					      struct btrfs_csum_item);
264 		}
265 		/*
266 		 * this byte range must be able to fit inside
267 		 * a single leaf so it will also fit inside a u32
268 		 */
269 		diff = disk_bytenr - item_start_offset;
270 		diff = diff / root->sectorsize;
271 		diff = diff * csum_size;
272 
273 		read_extent_buffer(path->nodes[0], &sum,
274 				   ((unsigned long)item) + diff,
275 				   csum_size);
276 found:
277 		if (dst)
278 			*dst++ = sum;
279 		else
280 			set_state_private(io_tree, offset, sum);
281 		disk_bytenr += bvec->bv_len;
282 		offset += bvec->bv_len;
283 		bio_index++;
284 		bvec++;
285 	}
286 	btrfs_free_path(path);
287 	return 0;
288 }
289 
290 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
291 			  struct bio *bio, u32 *dst)
292 {
293 	return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
294 }
295 
296 int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
297 			      struct bio *bio, u64 offset)
298 {
299 	return __btrfs_lookup_bio_sums(root, inode, bio, offset, NULL, 1);
300 }
301 
302 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
303 			     struct list_head *list, int search_commit)
304 {
305 	struct btrfs_key key;
306 	struct btrfs_path *path;
307 	struct extent_buffer *leaf;
308 	struct btrfs_ordered_sum *sums;
309 	struct btrfs_sector_sum *sector_sum;
310 	struct btrfs_csum_item *item;
311 	LIST_HEAD(tmplist);
312 	unsigned long offset;
313 	int ret;
314 	size_t size;
315 	u64 csum_end;
316 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
317 
318 	path = btrfs_alloc_path();
319 	if (!path)
320 		return -ENOMEM;
321 
322 	if (search_commit) {
323 		path->skip_locking = 1;
324 		path->reada = 2;
325 		path->search_commit_root = 1;
326 	}
327 
328 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
329 	key.offset = start;
330 	key.type = BTRFS_EXTENT_CSUM_KEY;
331 
332 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
333 	if (ret < 0)
334 		goto fail;
335 	if (ret > 0 && path->slots[0] > 0) {
336 		leaf = path->nodes[0];
337 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
338 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
339 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
340 			offset = (start - key.offset) >>
341 				 root->fs_info->sb->s_blocksize_bits;
342 			if (offset * csum_size <
343 			    btrfs_item_size_nr(leaf, path->slots[0] - 1))
344 				path->slots[0]--;
345 		}
346 	}
347 
348 	while (start <= end) {
349 		leaf = path->nodes[0];
350 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
351 			ret = btrfs_next_leaf(root, path);
352 			if (ret < 0)
353 				goto fail;
354 			if (ret > 0)
355 				break;
356 			leaf = path->nodes[0];
357 		}
358 
359 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
360 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
361 		    key.type != BTRFS_EXTENT_CSUM_KEY)
362 			break;
363 
364 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
365 		if (key.offset > end)
366 			break;
367 
368 		if (key.offset > start)
369 			start = key.offset;
370 
371 		size = btrfs_item_size_nr(leaf, path->slots[0]);
372 		csum_end = key.offset + (size / csum_size) * root->sectorsize;
373 		if (csum_end <= start) {
374 			path->slots[0]++;
375 			continue;
376 		}
377 
378 		csum_end = min(csum_end, end + 1);
379 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
380 				      struct btrfs_csum_item);
381 		while (start < csum_end) {
382 			size = min_t(size_t, csum_end - start,
383 					MAX_ORDERED_SUM_BYTES(root));
384 			sums = kzalloc(btrfs_ordered_sum_size(root, size),
385 					GFP_NOFS);
386 			if (!sums) {
387 				ret = -ENOMEM;
388 				goto fail;
389 			}
390 
391 			sector_sum = sums->sums;
392 			sums->bytenr = start;
393 			sums->len = size;
394 
395 			offset = (start - key.offset) >>
396 				root->fs_info->sb->s_blocksize_bits;
397 			offset *= csum_size;
398 
399 			while (size > 0) {
400 				read_extent_buffer(path->nodes[0],
401 						&sector_sum->sum,
402 						((unsigned long)item) +
403 						offset, csum_size);
404 				sector_sum->bytenr = start;
405 
406 				size -= root->sectorsize;
407 				start += root->sectorsize;
408 				offset += csum_size;
409 				sector_sum++;
410 			}
411 			list_add_tail(&sums->list, &tmplist);
412 		}
413 		path->slots[0]++;
414 	}
415 	ret = 0;
416 fail:
417 	while (ret < 0 && !list_empty(&tmplist)) {
418 		sums = list_entry(&tmplist, struct btrfs_ordered_sum, list);
419 		list_del(&sums->list);
420 		kfree(sums);
421 	}
422 	list_splice_tail(&tmplist, list);
423 
424 	btrfs_free_path(path);
425 	return ret;
426 }
427 
428 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
429 		       struct bio *bio, u64 file_start, int contig)
430 {
431 	struct btrfs_ordered_sum *sums;
432 	struct btrfs_sector_sum *sector_sum;
433 	struct btrfs_ordered_extent *ordered;
434 	char *data;
435 	struct bio_vec *bvec = bio->bi_io_vec;
436 	int bio_index = 0;
437 	unsigned long total_bytes = 0;
438 	unsigned long this_sum_bytes = 0;
439 	u64 offset;
440 	u64 disk_bytenr;
441 
442 	WARN_ON(bio->bi_vcnt <= 0);
443 	sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
444 	if (!sums)
445 		return -ENOMEM;
446 
447 	sector_sum = sums->sums;
448 	disk_bytenr = (u64)bio->bi_sector << 9;
449 	sums->len = bio->bi_size;
450 	INIT_LIST_HEAD(&sums->list);
451 
452 	if (contig)
453 		offset = file_start;
454 	else
455 		offset = page_offset(bvec->bv_page) + bvec->bv_offset;
456 
457 	ordered = btrfs_lookup_ordered_extent(inode, offset);
458 	BUG_ON(!ordered); /* Logic error */
459 	sums->bytenr = ordered->start;
460 
461 	while (bio_index < bio->bi_vcnt) {
462 		if (!contig)
463 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
464 
465 		if (offset >= ordered->file_offset + ordered->len ||
466 		    offset < ordered->file_offset) {
467 			unsigned long bytes_left;
468 			sums->len = this_sum_bytes;
469 			this_sum_bytes = 0;
470 			btrfs_add_ordered_sum(inode, ordered, sums);
471 			btrfs_put_ordered_extent(ordered);
472 
473 			bytes_left = bio->bi_size - total_bytes;
474 
475 			sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
476 				       GFP_NOFS);
477 			BUG_ON(!sums); /* -ENOMEM */
478 			sector_sum = sums->sums;
479 			sums->len = bytes_left;
480 			ordered = btrfs_lookup_ordered_extent(inode, offset);
481 			BUG_ON(!ordered); /* Logic error */
482 			sums->bytenr = ordered->start;
483 		}
484 
485 		data = kmap_atomic(bvec->bv_page);
486 		sector_sum->sum = ~(u32)0;
487 		sector_sum->sum = btrfs_csum_data(root,
488 						  data + bvec->bv_offset,
489 						  sector_sum->sum,
490 						  bvec->bv_len);
491 		kunmap_atomic(data);
492 		btrfs_csum_final(sector_sum->sum,
493 				 (char *)&sector_sum->sum);
494 		sector_sum->bytenr = disk_bytenr;
495 
496 		sector_sum++;
497 		bio_index++;
498 		total_bytes += bvec->bv_len;
499 		this_sum_bytes += bvec->bv_len;
500 		disk_bytenr += bvec->bv_len;
501 		offset += bvec->bv_len;
502 		bvec++;
503 	}
504 	this_sum_bytes = 0;
505 	btrfs_add_ordered_sum(inode, ordered, sums);
506 	btrfs_put_ordered_extent(ordered);
507 	return 0;
508 }
509 
510 /*
511  * helper function for csum removal, this expects the
512  * key to describe the csum pointed to by the path, and it expects
513  * the csum to overlap the range [bytenr, len]
514  *
515  * The csum should not be entirely contained in the range and the
516  * range should not be entirely contained in the csum.
517  *
518  * This calls btrfs_truncate_item with the correct args based on the
519  * overlap, and fixes up the key as required.
520  */
521 static noinline void truncate_one_csum(struct btrfs_trans_handle *trans,
522 				       struct btrfs_root *root,
523 				       struct btrfs_path *path,
524 				       struct btrfs_key *key,
525 				       u64 bytenr, u64 len)
526 {
527 	struct extent_buffer *leaf;
528 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
529 	u64 csum_end;
530 	u64 end_byte = bytenr + len;
531 	u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
532 
533 	leaf = path->nodes[0];
534 	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
535 	csum_end <<= root->fs_info->sb->s_blocksize_bits;
536 	csum_end += key->offset;
537 
538 	if (key->offset < bytenr && csum_end <= end_byte) {
539 		/*
540 		 *         [ bytenr - len ]
541 		 *         [   ]
542 		 *   [csum     ]
543 		 *   A simple truncate off the end of the item
544 		 */
545 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
546 		new_size *= csum_size;
547 		btrfs_truncate_item(trans, root, path, new_size, 1);
548 	} else if (key->offset >= bytenr && csum_end > end_byte &&
549 		   end_byte > key->offset) {
550 		/*
551 		 *         [ bytenr - len ]
552 		 *                 [ ]
553 		 *                 [csum     ]
554 		 * we need to truncate from the beginning of the csum
555 		 */
556 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
557 		new_size *= csum_size;
558 
559 		btrfs_truncate_item(trans, root, path, new_size, 0);
560 
561 		key->offset = end_byte;
562 		btrfs_set_item_key_safe(trans, root, path, key);
563 	} else {
564 		BUG();
565 	}
566 }
567 
568 /*
569  * deletes the csum items from the csum tree for a given
570  * range of bytes.
571  */
572 int btrfs_del_csums(struct btrfs_trans_handle *trans,
573 		    struct btrfs_root *root, u64 bytenr, u64 len)
574 {
575 	struct btrfs_path *path;
576 	struct btrfs_key key;
577 	u64 end_byte = bytenr + len;
578 	u64 csum_end;
579 	struct extent_buffer *leaf;
580 	int ret;
581 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
582 	int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
583 
584 	root = root->fs_info->csum_root;
585 
586 	path = btrfs_alloc_path();
587 	if (!path)
588 		return -ENOMEM;
589 
590 	while (1) {
591 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
592 		key.offset = end_byte - 1;
593 		key.type = BTRFS_EXTENT_CSUM_KEY;
594 
595 		path->leave_spinning = 1;
596 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
597 		if (ret > 0) {
598 			if (path->slots[0] == 0)
599 				break;
600 			path->slots[0]--;
601 		} else if (ret < 0) {
602 			break;
603 		}
604 
605 		leaf = path->nodes[0];
606 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
607 
608 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
609 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
610 			break;
611 		}
612 
613 		if (key.offset >= end_byte)
614 			break;
615 
616 		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
617 		csum_end <<= blocksize_bits;
618 		csum_end += key.offset;
619 
620 		/* this csum ends before we start, we're done */
621 		if (csum_end <= bytenr)
622 			break;
623 
624 		/* delete the entire item, it is inside our range */
625 		if (key.offset >= bytenr && csum_end <= end_byte) {
626 			ret = btrfs_del_item(trans, root, path);
627 			if (ret)
628 				goto out;
629 			if (key.offset == bytenr)
630 				break;
631 		} else if (key.offset < bytenr && csum_end > end_byte) {
632 			unsigned long offset;
633 			unsigned long shift_len;
634 			unsigned long item_offset;
635 			/*
636 			 *        [ bytenr - len ]
637 			 *     [csum                ]
638 			 *
639 			 * Our bytes are in the middle of the csum,
640 			 * we need to split this item and insert a new one.
641 			 *
642 			 * But we can't drop the path because the
643 			 * csum could change, get removed, extended etc.
644 			 *
645 			 * The trick here is the max size of a csum item leaves
646 			 * enough room in the tree block for a single
647 			 * item header.  So, we split the item in place,
648 			 * adding a new header pointing to the existing
649 			 * bytes.  Then we loop around again and we have
650 			 * a nicely formed csum item that we can neatly
651 			 * truncate.
652 			 */
653 			offset = (bytenr - key.offset) >> blocksize_bits;
654 			offset *= csum_size;
655 
656 			shift_len = (len >> blocksize_bits) * csum_size;
657 
658 			item_offset = btrfs_item_ptr_offset(leaf,
659 							    path->slots[0]);
660 
661 			memset_extent_buffer(leaf, 0, item_offset + offset,
662 					     shift_len);
663 			key.offset = bytenr;
664 
665 			/*
666 			 * btrfs_split_item returns -EAGAIN when the
667 			 * item changed size or key
668 			 */
669 			ret = btrfs_split_item(trans, root, path, &key, offset);
670 			if (ret && ret != -EAGAIN) {
671 				btrfs_abort_transaction(trans, root, ret);
672 				goto out;
673 			}
674 
675 			key.offset = end_byte - 1;
676 		} else {
677 			truncate_one_csum(trans, root, path, &key, bytenr, len);
678 			if (key.offset < bytenr)
679 				break;
680 		}
681 		btrfs_release_path(path);
682 	}
683 	ret = 0;
684 out:
685 	btrfs_free_path(path);
686 	return ret;
687 }
688 
689 static u64 btrfs_sector_sum_left(struct btrfs_ordered_sum *sums,
690 				 struct btrfs_sector_sum *sector_sum,
691 				 u64 total_bytes, u64 sectorsize)
692 {
693 	u64 tmp = sectorsize;
694 	u64 next_sector = sector_sum->bytenr;
695 	struct btrfs_sector_sum *next = sector_sum + 1;
696 
697 	while ((tmp + total_bytes) < sums->len) {
698 		if (next_sector + sectorsize != next->bytenr)
699 			break;
700 		tmp += sectorsize;
701 		next_sector = next->bytenr;
702 		next++;
703 	}
704 	return tmp;
705 }
706 
707 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
708 			   struct btrfs_root *root,
709 			   struct btrfs_ordered_sum *sums)
710 {
711 	u64 bytenr;
712 	int ret;
713 	struct btrfs_key file_key;
714 	struct btrfs_key found_key;
715 	u64 next_offset;
716 	u64 total_bytes = 0;
717 	int found_next;
718 	struct btrfs_path *path;
719 	struct btrfs_csum_item *item;
720 	struct btrfs_csum_item *item_end;
721 	struct extent_buffer *leaf = NULL;
722 	u64 csum_offset;
723 	struct btrfs_sector_sum *sector_sum;
724 	u32 nritems;
725 	u32 ins_size;
726 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
727 
728 	path = btrfs_alloc_path();
729 	if (!path)
730 		return -ENOMEM;
731 
732 	sector_sum = sums->sums;
733 again:
734 	next_offset = (u64)-1;
735 	found_next = 0;
736 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
737 	file_key.offset = sector_sum->bytenr;
738 	bytenr = sector_sum->bytenr;
739 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
740 
741 	item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
742 	if (!IS_ERR(item)) {
743 		leaf = path->nodes[0];
744 		ret = 0;
745 		goto found;
746 	}
747 	ret = PTR_ERR(item);
748 	if (ret != -EFBIG && ret != -ENOENT)
749 		goto fail_unlock;
750 
751 	if (ret == -EFBIG) {
752 		u32 item_size;
753 		/* we found one, but it isn't big enough yet */
754 		leaf = path->nodes[0];
755 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
756 		if ((item_size / csum_size) >=
757 		    MAX_CSUM_ITEMS(root, csum_size)) {
758 			/* already at max size, make a new one */
759 			goto insert;
760 		}
761 	} else {
762 		int slot = path->slots[0] + 1;
763 		/* we didn't find a csum item, insert one */
764 		nritems = btrfs_header_nritems(path->nodes[0]);
765 		if (path->slots[0] >= nritems - 1) {
766 			ret = btrfs_next_leaf(root, path);
767 			if (ret == 1)
768 				found_next = 1;
769 			if (ret != 0)
770 				goto insert;
771 			slot = 0;
772 		}
773 		btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
774 		if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
775 		    found_key.type != BTRFS_EXTENT_CSUM_KEY) {
776 			found_next = 1;
777 			goto insert;
778 		}
779 		next_offset = found_key.offset;
780 		found_next = 1;
781 		goto insert;
782 	}
783 
784 	/*
785 	 * at this point, we know the tree has an item, but it isn't big
786 	 * enough yet to put our csum in.  Grow it
787 	 */
788 	btrfs_release_path(path);
789 	ret = btrfs_search_slot(trans, root, &file_key, path,
790 				csum_size, 1);
791 	if (ret < 0)
792 		goto fail_unlock;
793 
794 	if (ret > 0) {
795 		if (path->slots[0] == 0)
796 			goto insert;
797 		path->slots[0]--;
798 	}
799 
800 	leaf = path->nodes[0];
801 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
802 	csum_offset = (bytenr - found_key.offset) >>
803 			root->fs_info->sb->s_blocksize_bits;
804 
805 	if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
806 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
807 	    csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
808 		goto insert;
809 	}
810 
811 	if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
812 	    csum_size) {
813 		int extend_nr;
814 		u64 tmp;
815 		u32 diff;
816 		u32 free_space;
817 
818 		if (btrfs_leaf_free_space(root, leaf) <
819 				 sizeof(struct btrfs_item) + csum_size * 2)
820 			goto insert;
821 
822 		free_space = btrfs_leaf_free_space(root, leaf) -
823 					 sizeof(struct btrfs_item) - csum_size;
824 		tmp = btrfs_sector_sum_left(sums, sector_sum, total_bytes,
825 					    root->sectorsize);
826 		tmp >>= root->fs_info->sb->s_blocksize_bits;
827 		WARN_ON(tmp < 1);
828 
829 		extend_nr = max_t(int, 1, (int)tmp);
830 		diff = (csum_offset + extend_nr) * csum_size;
831 		diff = min(diff, MAX_CSUM_ITEMS(root, csum_size) * csum_size);
832 
833 		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
834 		diff = min(free_space, diff);
835 		diff /= csum_size;
836 		diff *= csum_size;
837 
838 		btrfs_extend_item(trans, root, path, diff);
839 		goto csum;
840 	}
841 
842 insert:
843 	btrfs_release_path(path);
844 	csum_offset = 0;
845 	if (found_next) {
846 		u64 tmp;
847 
848 		tmp = btrfs_sector_sum_left(sums, sector_sum, total_bytes,
849 					    root->sectorsize);
850 		tmp >>= root->fs_info->sb->s_blocksize_bits;
851 		tmp = min(tmp, (next_offset - file_key.offset) >>
852 					 root->fs_info->sb->s_blocksize_bits);
853 
854 		tmp = max((u64)1, tmp);
855 		tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
856 		ins_size = csum_size * tmp;
857 	} else {
858 		ins_size = csum_size;
859 	}
860 	path->leave_spinning = 1;
861 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
862 				      ins_size);
863 	path->leave_spinning = 0;
864 	if (ret < 0)
865 		goto fail_unlock;
866 	if (ret != 0) {
867 		WARN_ON(1);
868 		goto fail_unlock;
869 	}
870 csum:
871 	leaf = path->nodes[0];
872 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
873 	ret = 0;
874 	item = (struct btrfs_csum_item *)((unsigned char *)item +
875 					  csum_offset * csum_size);
876 found:
877 	item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
878 	item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
879 				      btrfs_item_size_nr(leaf, path->slots[0]));
880 next_sector:
881 
882 	write_extent_buffer(leaf, &sector_sum->sum, (unsigned long)item, csum_size);
883 
884 	total_bytes += root->sectorsize;
885 	sector_sum++;
886 	if (total_bytes < sums->len) {
887 		item = (struct btrfs_csum_item *)((char *)item +
888 						  csum_size);
889 		if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
890 		    sector_sum->bytenr) {
891 			bytenr = sector_sum->bytenr;
892 			goto next_sector;
893 		}
894 	}
895 
896 	btrfs_mark_buffer_dirty(path->nodes[0]);
897 	if (total_bytes < sums->len) {
898 		btrfs_release_path(path);
899 		cond_resched();
900 		goto again;
901 	}
902 out:
903 	btrfs_free_path(path);
904 	return ret;
905 
906 fail_unlock:
907 	goto out;
908 }
909