xref: /openbmc/linux/fs/btrfs/file-item.c (revision 81d67439)
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_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
33 				   sizeof(struct btrfs_ordered_sum)) / \
34 				   sizeof(struct btrfs_sector_sum) * \
35 				   (r)->sectorsize - (r)->sectorsize)
36 
37 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
38 			     struct btrfs_root *root,
39 			     u64 objectid, u64 pos,
40 			     u64 disk_offset, u64 disk_num_bytes,
41 			     u64 num_bytes, u64 offset, u64 ram_bytes,
42 			     u8 compression, u8 encryption, u16 other_encoding)
43 {
44 	int ret = 0;
45 	struct btrfs_file_extent_item *item;
46 	struct btrfs_key file_key;
47 	struct btrfs_path *path;
48 	struct extent_buffer *leaf;
49 
50 	path = btrfs_alloc_path();
51 	if (!path)
52 		return -ENOMEM;
53 	file_key.objectid = objectid;
54 	file_key.offset = pos;
55 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
56 
57 	path->leave_spinning = 1;
58 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
59 				      sizeof(*item));
60 	if (ret < 0)
61 		goto out;
62 	BUG_ON(ret);
63 	leaf = path->nodes[0];
64 	item = btrfs_item_ptr(leaf, path->slots[0],
65 			      struct btrfs_file_extent_item);
66 	btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
67 	btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
68 	btrfs_set_file_extent_offset(leaf, item, offset);
69 	btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
70 	btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
71 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
72 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
73 	btrfs_set_file_extent_compression(leaf, item, compression);
74 	btrfs_set_file_extent_encryption(leaf, item, encryption);
75 	btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
76 
77 	btrfs_mark_buffer_dirty(leaf);
78 out:
79 	btrfs_free_path(path);
80 	return ret;
81 }
82 
83 struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
84 					  struct btrfs_root *root,
85 					  struct btrfs_path *path,
86 					  u64 bytenr, int cow)
87 {
88 	int ret;
89 	struct btrfs_key file_key;
90 	struct btrfs_key found_key;
91 	struct btrfs_csum_item *item;
92 	struct extent_buffer *leaf;
93 	u64 csum_offset = 0;
94 	u16 csum_size =
95 		btrfs_super_csum_size(&root->fs_info->super_copy);
96 	int csums_in_item;
97 
98 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
99 	file_key.offset = bytenr;
100 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
101 	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
102 	if (ret < 0)
103 		goto fail;
104 	leaf = path->nodes[0];
105 	if (ret > 0) {
106 		ret = 1;
107 		if (path->slots[0] == 0)
108 			goto fail;
109 		path->slots[0]--;
110 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
111 		if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY)
112 			goto fail;
113 
114 		csum_offset = (bytenr - found_key.offset) >>
115 				root->fs_info->sb->s_blocksize_bits;
116 		csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
117 		csums_in_item /= csum_size;
118 
119 		if (csum_offset >= csums_in_item) {
120 			ret = -EFBIG;
121 			goto fail;
122 		}
123 	}
124 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
125 	item = (struct btrfs_csum_item *)((unsigned char *)item +
126 					  csum_offset * csum_size);
127 	return item;
128 fail:
129 	if (ret > 0)
130 		ret = -ENOENT;
131 	return ERR_PTR(ret);
132 }
133 
134 
135 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
136 			     struct btrfs_root *root,
137 			     struct btrfs_path *path, u64 objectid,
138 			     u64 offset, int mod)
139 {
140 	int ret;
141 	struct btrfs_key file_key;
142 	int ins_len = mod < 0 ? -1 : 0;
143 	int cow = mod != 0;
144 
145 	file_key.objectid = objectid;
146 	file_key.offset = offset;
147 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
148 	ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
149 	return ret;
150 }
151 
152 
153 static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
154 				   struct inode *inode, struct bio *bio,
155 				   u64 logical_offset, u32 *dst, int dio)
156 {
157 	u32 sum;
158 	struct bio_vec *bvec = bio->bi_io_vec;
159 	int bio_index = 0;
160 	u64 offset = 0;
161 	u64 item_start_offset = 0;
162 	u64 item_last_offset = 0;
163 	u64 disk_bytenr;
164 	u32 diff;
165 	u16 csum_size =
166 		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 	disk_bytenr = (u64)bio->bi_sector << 9;
181 	if (dio)
182 		offset = logical_offset;
183 	while (bio_index < bio->bi_vcnt) {
184 		if (!dio)
185 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
186 		ret = btrfs_find_ordered_sum(inode, offset, disk_bytenr, &sum);
187 		if (ret == 0)
188 			goto found;
189 
190 		if (!item || disk_bytenr < item_start_offset ||
191 		    disk_bytenr >= item_last_offset) {
192 			struct btrfs_key found_key;
193 			u32 item_size;
194 
195 			if (item)
196 				btrfs_release_path(path);
197 			item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
198 						 path, disk_bytenr, 0);
199 			if (IS_ERR(item)) {
200 				ret = PTR_ERR(item);
201 				if (ret == -ENOENT || ret == -EFBIG)
202 					ret = 0;
203 				sum = 0;
204 				if (BTRFS_I(inode)->root->root_key.objectid ==
205 				    BTRFS_DATA_RELOC_TREE_OBJECTID) {
206 					set_extent_bits(io_tree, offset,
207 						offset + bvec->bv_len - 1,
208 						EXTENT_NODATASUM, GFP_NOFS);
209 				} else {
210 					printk(KERN_INFO "btrfs no csum found "
211 					       "for inode %llu start %llu\n",
212 					       (unsigned long long)
213 					       btrfs_ino(inode),
214 					       (unsigned long long)offset);
215 				}
216 				item = NULL;
217 				btrfs_release_path(path);
218 				goto found;
219 			}
220 			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
221 					      path->slots[0]);
222 
223 			item_start_offset = found_key.offset;
224 			item_size = btrfs_item_size_nr(path->nodes[0],
225 						       path->slots[0]);
226 			item_last_offset = item_start_offset +
227 				(item_size / csum_size) *
228 				root->sectorsize;
229 			item = btrfs_item_ptr(path->nodes[0], path->slots[0],
230 					      struct btrfs_csum_item);
231 		}
232 		/*
233 		 * this byte range must be able to fit inside
234 		 * a single leaf so it will also fit inside a u32
235 		 */
236 		diff = disk_bytenr - item_start_offset;
237 		diff = diff / root->sectorsize;
238 		diff = diff * csum_size;
239 
240 		read_extent_buffer(path->nodes[0], &sum,
241 				   ((unsigned long)item) + diff,
242 				   csum_size);
243 found:
244 		if (dst)
245 			*dst++ = sum;
246 		else
247 			set_state_private(io_tree, offset, sum);
248 		disk_bytenr += bvec->bv_len;
249 		offset += bvec->bv_len;
250 		bio_index++;
251 		bvec++;
252 	}
253 	btrfs_free_path(path);
254 	return 0;
255 }
256 
257 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
258 			  struct bio *bio, u32 *dst)
259 {
260 	return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
261 }
262 
263 int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
264 			      struct bio *bio, u64 offset, u32 *dst)
265 {
266 	return __btrfs_lookup_bio_sums(root, inode, bio, offset, dst, 1);
267 }
268 
269 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
270 			     struct list_head *list, int search_commit)
271 {
272 	struct btrfs_key key;
273 	struct btrfs_path *path;
274 	struct extent_buffer *leaf;
275 	struct btrfs_ordered_sum *sums;
276 	struct btrfs_sector_sum *sector_sum;
277 	struct btrfs_csum_item *item;
278 	unsigned long offset;
279 	int ret;
280 	size_t size;
281 	u64 csum_end;
282 	u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
283 
284 	path = btrfs_alloc_path();
285 	BUG_ON(!path);
286 
287 	if (search_commit) {
288 		path->skip_locking = 1;
289 		path->reada = 2;
290 		path->search_commit_root = 1;
291 	}
292 
293 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
294 	key.offset = start;
295 	key.type = BTRFS_EXTENT_CSUM_KEY;
296 
297 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
298 	if (ret < 0)
299 		goto fail;
300 	if (ret > 0 && path->slots[0] > 0) {
301 		leaf = path->nodes[0];
302 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
303 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
304 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
305 			offset = (start - key.offset) >>
306 				 root->fs_info->sb->s_blocksize_bits;
307 			if (offset * csum_size <
308 			    btrfs_item_size_nr(leaf, path->slots[0] - 1))
309 				path->slots[0]--;
310 		}
311 	}
312 
313 	while (start <= end) {
314 		leaf = path->nodes[0];
315 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
316 			ret = btrfs_next_leaf(root, path);
317 			if (ret < 0)
318 				goto fail;
319 			if (ret > 0)
320 				break;
321 			leaf = path->nodes[0];
322 		}
323 
324 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
325 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
326 		    key.type != BTRFS_EXTENT_CSUM_KEY)
327 			break;
328 
329 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
330 		if (key.offset > end)
331 			break;
332 
333 		if (key.offset > start)
334 			start = key.offset;
335 
336 		size = btrfs_item_size_nr(leaf, path->slots[0]);
337 		csum_end = key.offset + (size / csum_size) * root->sectorsize;
338 		if (csum_end <= start) {
339 			path->slots[0]++;
340 			continue;
341 		}
342 
343 		csum_end = min(csum_end, end + 1);
344 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
345 				      struct btrfs_csum_item);
346 		while (start < csum_end) {
347 			size = min_t(size_t, csum_end - start,
348 					MAX_ORDERED_SUM_BYTES(root));
349 			sums = kzalloc(btrfs_ordered_sum_size(root, size),
350 					GFP_NOFS);
351 			BUG_ON(!sums);
352 
353 			sector_sum = sums->sums;
354 			sums->bytenr = start;
355 			sums->len = size;
356 
357 			offset = (start - key.offset) >>
358 				root->fs_info->sb->s_blocksize_bits;
359 			offset *= csum_size;
360 
361 			while (size > 0) {
362 				read_extent_buffer(path->nodes[0],
363 						&sector_sum->sum,
364 						((unsigned long)item) +
365 						offset, csum_size);
366 				sector_sum->bytenr = start;
367 
368 				size -= root->sectorsize;
369 				start += root->sectorsize;
370 				offset += csum_size;
371 				sector_sum++;
372 			}
373 			list_add_tail(&sums->list, list);
374 		}
375 		path->slots[0]++;
376 	}
377 	ret = 0;
378 fail:
379 	btrfs_free_path(path);
380 	return ret;
381 }
382 
383 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
384 		       struct bio *bio, u64 file_start, int contig)
385 {
386 	struct btrfs_ordered_sum *sums;
387 	struct btrfs_sector_sum *sector_sum;
388 	struct btrfs_ordered_extent *ordered;
389 	char *data;
390 	struct bio_vec *bvec = bio->bi_io_vec;
391 	int bio_index = 0;
392 	unsigned long total_bytes = 0;
393 	unsigned long this_sum_bytes = 0;
394 	u64 offset;
395 	u64 disk_bytenr;
396 
397 	WARN_ON(bio->bi_vcnt <= 0);
398 	sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
399 	if (!sums)
400 		return -ENOMEM;
401 
402 	sector_sum = sums->sums;
403 	disk_bytenr = (u64)bio->bi_sector << 9;
404 	sums->len = bio->bi_size;
405 	INIT_LIST_HEAD(&sums->list);
406 
407 	if (contig)
408 		offset = file_start;
409 	else
410 		offset = page_offset(bvec->bv_page) + bvec->bv_offset;
411 
412 	ordered = btrfs_lookup_ordered_extent(inode, offset);
413 	BUG_ON(!ordered);
414 	sums->bytenr = ordered->start;
415 
416 	while (bio_index < bio->bi_vcnt) {
417 		if (!contig)
418 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
419 
420 		if (!contig && (offset >= ordered->file_offset + ordered->len ||
421 		    offset < ordered->file_offset)) {
422 			unsigned long bytes_left;
423 			sums->len = this_sum_bytes;
424 			this_sum_bytes = 0;
425 			btrfs_add_ordered_sum(inode, ordered, sums);
426 			btrfs_put_ordered_extent(ordered);
427 
428 			bytes_left = bio->bi_size - total_bytes;
429 
430 			sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
431 				       GFP_NOFS);
432 			BUG_ON(!sums);
433 			sector_sum = sums->sums;
434 			sums->len = bytes_left;
435 			ordered = btrfs_lookup_ordered_extent(inode, offset);
436 			BUG_ON(!ordered);
437 			sums->bytenr = ordered->start;
438 		}
439 
440 		data = kmap_atomic(bvec->bv_page, KM_USER0);
441 		sector_sum->sum = ~(u32)0;
442 		sector_sum->sum = btrfs_csum_data(root,
443 						  data + bvec->bv_offset,
444 						  sector_sum->sum,
445 						  bvec->bv_len);
446 		kunmap_atomic(data, KM_USER0);
447 		btrfs_csum_final(sector_sum->sum,
448 				 (char *)&sector_sum->sum);
449 		sector_sum->bytenr = disk_bytenr;
450 
451 		sector_sum++;
452 		bio_index++;
453 		total_bytes += bvec->bv_len;
454 		this_sum_bytes += bvec->bv_len;
455 		disk_bytenr += bvec->bv_len;
456 		offset += bvec->bv_len;
457 		bvec++;
458 	}
459 	this_sum_bytes = 0;
460 	btrfs_add_ordered_sum(inode, ordered, sums);
461 	btrfs_put_ordered_extent(ordered);
462 	return 0;
463 }
464 
465 /*
466  * helper function for csum removal, this expects the
467  * key to describe the csum pointed to by the path, and it expects
468  * the csum to overlap the range [bytenr, len]
469  *
470  * The csum should not be entirely contained in the range and the
471  * range should not be entirely contained in the csum.
472  *
473  * This calls btrfs_truncate_item with the correct args based on the
474  * overlap, and fixes up the key as required.
475  */
476 static noinline int truncate_one_csum(struct btrfs_trans_handle *trans,
477 				      struct btrfs_root *root,
478 				      struct btrfs_path *path,
479 				      struct btrfs_key *key,
480 				      u64 bytenr, u64 len)
481 {
482 	struct extent_buffer *leaf;
483 	u16 csum_size =
484 		btrfs_super_csum_size(&root->fs_info->super_copy);
485 	u64 csum_end;
486 	u64 end_byte = bytenr + len;
487 	u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
488 	int ret;
489 
490 	leaf = path->nodes[0];
491 	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
492 	csum_end <<= root->fs_info->sb->s_blocksize_bits;
493 	csum_end += key->offset;
494 
495 	if (key->offset < bytenr && csum_end <= end_byte) {
496 		/*
497 		 *         [ bytenr - len ]
498 		 *         [   ]
499 		 *   [csum     ]
500 		 *   A simple truncate off the end of the item
501 		 */
502 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
503 		new_size *= csum_size;
504 		ret = btrfs_truncate_item(trans, root, path, new_size, 1);
505 	} else if (key->offset >= bytenr && csum_end > end_byte &&
506 		   end_byte > key->offset) {
507 		/*
508 		 *         [ bytenr - len ]
509 		 *                 [ ]
510 		 *                 [csum     ]
511 		 * we need to truncate from the beginning of the csum
512 		 */
513 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
514 		new_size *= csum_size;
515 
516 		ret = btrfs_truncate_item(trans, root, path, new_size, 0);
517 
518 		key->offset = end_byte;
519 		ret = btrfs_set_item_key_safe(trans, root, path, key);
520 		BUG_ON(ret);
521 	} else {
522 		BUG();
523 	}
524 	return 0;
525 }
526 
527 /*
528  * deletes the csum items from the csum tree for a given
529  * range of bytes.
530  */
531 int btrfs_del_csums(struct btrfs_trans_handle *trans,
532 		    struct btrfs_root *root, u64 bytenr, u64 len)
533 {
534 	struct btrfs_path *path;
535 	struct btrfs_key key;
536 	u64 end_byte = bytenr + len;
537 	u64 csum_end;
538 	struct extent_buffer *leaf;
539 	int ret;
540 	u16 csum_size =
541 		btrfs_super_csum_size(&root->fs_info->super_copy);
542 	int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
543 
544 	root = root->fs_info->csum_root;
545 
546 	path = btrfs_alloc_path();
547 	if (!path)
548 		return -ENOMEM;
549 
550 	while (1) {
551 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
552 		key.offset = end_byte - 1;
553 		key.type = BTRFS_EXTENT_CSUM_KEY;
554 
555 		path->leave_spinning = 1;
556 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
557 		if (ret > 0) {
558 			if (path->slots[0] == 0)
559 				break;
560 			path->slots[0]--;
561 		} else if (ret < 0) {
562 			break;
563 		}
564 
565 		leaf = path->nodes[0];
566 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
567 
568 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
569 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
570 			break;
571 		}
572 
573 		if (key.offset >= end_byte)
574 			break;
575 
576 		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
577 		csum_end <<= blocksize_bits;
578 		csum_end += key.offset;
579 
580 		/* this csum ends before we start, we're done */
581 		if (csum_end <= bytenr)
582 			break;
583 
584 		/* delete the entire item, it is inside our range */
585 		if (key.offset >= bytenr && csum_end <= end_byte) {
586 			ret = btrfs_del_item(trans, root, path);
587 			if (ret)
588 				goto out;
589 			if (key.offset == bytenr)
590 				break;
591 		} else if (key.offset < bytenr && csum_end > end_byte) {
592 			unsigned long offset;
593 			unsigned long shift_len;
594 			unsigned long item_offset;
595 			/*
596 			 *        [ bytenr - len ]
597 			 *     [csum                ]
598 			 *
599 			 * Our bytes are in the middle of the csum,
600 			 * we need to split this item and insert a new one.
601 			 *
602 			 * But we can't drop the path because the
603 			 * csum could change, get removed, extended etc.
604 			 *
605 			 * The trick here is the max size of a csum item leaves
606 			 * enough room in the tree block for a single
607 			 * item header.  So, we split the item in place,
608 			 * adding a new header pointing to the existing
609 			 * bytes.  Then we loop around again and we have
610 			 * a nicely formed csum item that we can neatly
611 			 * truncate.
612 			 */
613 			offset = (bytenr - key.offset) >> blocksize_bits;
614 			offset *= csum_size;
615 
616 			shift_len = (len >> blocksize_bits) * csum_size;
617 
618 			item_offset = btrfs_item_ptr_offset(leaf,
619 							    path->slots[0]);
620 
621 			memset_extent_buffer(leaf, 0, item_offset + offset,
622 					     shift_len);
623 			key.offset = bytenr;
624 
625 			/*
626 			 * btrfs_split_item returns -EAGAIN when the
627 			 * item changed size or key
628 			 */
629 			ret = btrfs_split_item(trans, root, path, &key, offset);
630 			BUG_ON(ret && ret != -EAGAIN);
631 
632 			key.offset = end_byte - 1;
633 		} else {
634 			ret = truncate_one_csum(trans, root, path,
635 						&key, bytenr, len);
636 			BUG_ON(ret);
637 			if (key.offset < bytenr)
638 				break;
639 		}
640 		btrfs_release_path(path);
641 	}
642 	ret = 0;
643 out:
644 	btrfs_free_path(path);
645 	return ret;
646 }
647 
648 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
649 			   struct btrfs_root *root,
650 			   struct btrfs_ordered_sum *sums)
651 {
652 	u64 bytenr;
653 	int ret;
654 	struct btrfs_key file_key;
655 	struct btrfs_key found_key;
656 	u64 next_offset;
657 	u64 total_bytes = 0;
658 	int found_next;
659 	struct btrfs_path *path;
660 	struct btrfs_csum_item *item;
661 	struct btrfs_csum_item *item_end;
662 	struct extent_buffer *leaf = NULL;
663 	u64 csum_offset;
664 	struct btrfs_sector_sum *sector_sum;
665 	u32 nritems;
666 	u32 ins_size;
667 	char *eb_map;
668 	char *eb_token;
669 	unsigned long map_len;
670 	unsigned long map_start;
671 	u16 csum_size =
672 		btrfs_super_csum_size(&root->fs_info->super_copy);
673 
674 	path = btrfs_alloc_path();
675 	BUG_ON(!path);
676 	sector_sum = sums->sums;
677 again:
678 	next_offset = (u64)-1;
679 	found_next = 0;
680 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
681 	file_key.offset = sector_sum->bytenr;
682 	bytenr = sector_sum->bytenr;
683 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
684 
685 	item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
686 	if (!IS_ERR(item)) {
687 		leaf = path->nodes[0];
688 		ret = 0;
689 		goto found;
690 	}
691 	ret = PTR_ERR(item);
692 	if (ret != -EFBIG && ret != -ENOENT)
693 		goto fail_unlock;
694 
695 	if (ret == -EFBIG) {
696 		u32 item_size;
697 		/* we found one, but it isn't big enough yet */
698 		leaf = path->nodes[0];
699 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
700 		if ((item_size / csum_size) >=
701 		    MAX_CSUM_ITEMS(root, csum_size)) {
702 			/* already at max size, make a new one */
703 			goto insert;
704 		}
705 	} else {
706 		int slot = path->slots[0] + 1;
707 		/* we didn't find a csum item, insert one */
708 		nritems = btrfs_header_nritems(path->nodes[0]);
709 		if (path->slots[0] >= nritems - 1) {
710 			ret = btrfs_next_leaf(root, path);
711 			if (ret == 1)
712 				found_next = 1;
713 			if (ret != 0)
714 				goto insert;
715 			slot = 0;
716 		}
717 		btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
718 		if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
719 		    found_key.type != BTRFS_EXTENT_CSUM_KEY) {
720 			found_next = 1;
721 			goto insert;
722 		}
723 		next_offset = found_key.offset;
724 		found_next = 1;
725 		goto insert;
726 	}
727 
728 	/*
729 	 * at this point, we know the tree has an item, but it isn't big
730 	 * enough yet to put our csum in.  Grow it
731 	 */
732 	btrfs_release_path(path);
733 	ret = btrfs_search_slot(trans, root, &file_key, path,
734 				csum_size, 1);
735 	if (ret < 0)
736 		goto fail_unlock;
737 
738 	if (ret > 0) {
739 		if (path->slots[0] == 0)
740 			goto insert;
741 		path->slots[0]--;
742 	}
743 
744 	leaf = path->nodes[0];
745 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
746 	csum_offset = (bytenr - found_key.offset) >>
747 			root->fs_info->sb->s_blocksize_bits;
748 
749 	if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
750 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
751 	    csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
752 		goto insert;
753 	}
754 
755 	if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
756 	    csum_size) {
757 		u32 diff = (csum_offset + 1) * csum_size;
758 
759 		/*
760 		 * is the item big enough already?  we dropped our lock
761 		 * before and need to recheck
762 		 */
763 		if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
764 			goto csum;
765 
766 		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
767 		if (diff != csum_size)
768 			goto insert;
769 
770 		ret = btrfs_extend_item(trans, root, path, diff);
771 		goto csum;
772 	}
773 
774 insert:
775 	btrfs_release_path(path);
776 	csum_offset = 0;
777 	if (found_next) {
778 		u64 tmp = total_bytes + root->sectorsize;
779 		u64 next_sector = sector_sum->bytenr;
780 		struct btrfs_sector_sum *next = sector_sum + 1;
781 
782 		while (tmp < sums->len) {
783 			if (next_sector + root->sectorsize != next->bytenr)
784 				break;
785 			tmp += root->sectorsize;
786 			next_sector = next->bytenr;
787 			next++;
788 		}
789 		tmp = min(tmp, next_offset - file_key.offset);
790 		tmp >>= root->fs_info->sb->s_blocksize_bits;
791 		tmp = max((u64)1, tmp);
792 		tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
793 		ins_size = csum_size * tmp;
794 	} else {
795 		ins_size = csum_size;
796 	}
797 	path->leave_spinning = 1;
798 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
799 				      ins_size);
800 	path->leave_spinning = 0;
801 	if (ret < 0)
802 		goto fail_unlock;
803 	if (ret != 0) {
804 		WARN_ON(1);
805 		goto fail_unlock;
806 	}
807 csum:
808 	leaf = path->nodes[0];
809 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
810 	ret = 0;
811 	item = (struct btrfs_csum_item *)((unsigned char *)item +
812 					  csum_offset * csum_size);
813 found:
814 	item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
815 	item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
816 				      btrfs_item_size_nr(leaf, path->slots[0]));
817 	eb_token = NULL;
818 next_sector:
819 
820 	if (!eb_token ||
821 	   (unsigned long)item + csum_size >= map_start + map_len) {
822 		int err;
823 
824 		if (eb_token)
825 			unmap_extent_buffer(leaf, eb_token, KM_USER1);
826 		eb_token = NULL;
827 		err = map_private_extent_buffer(leaf, (unsigned long)item,
828 						csum_size,
829 						&eb_token, &eb_map,
830 						&map_start, &map_len, KM_USER1);
831 		if (err)
832 			eb_token = NULL;
833 	}
834 	if (eb_token) {
835 		memcpy(eb_token + ((unsigned long)item & (PAGE_CACHE_SIZE - 1)),
836 		       &sector_sum->sum, csum_size);
837 	} else {
838 		write_extent_buffer(leaf, &sector_sum->sum,
839 				    (unsigned long)item, csum_size);
840 	}
841 
842 	total_bytes += root->sectorsize;
843 	sector_sum++;
844 	if (total_bytes < sums->len) {
845 		item = (struct btrfs_csum_item *)((char *)item +
846 						  csum_size);
847 		if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
848 		    sector_sum->bytenr) {
849 			bytenr = sector_sum->bytenr;
850 			goto next_sector;
851 		}
852 	}
853 	if (eb_token) {
854 		unmap_extent_buffer(leaf, eb_token, KM_USER1);
855 		eb_token = NULL;
856 	}
857 	btrfs_mark_buffer_dirty(path->nodes[0]);
858 	if (total_bytes < sums->len) {
859 		btrfs_release_path(path);
860 		cond_resched();
861 		goto again;
862 	}
863 out:
864 	btrfs_free_path(path);
865 	return ret;
866 
867 fail_unlock:
868 	goto out;
869 }
870