xref: /openbmc/linux/fs/btrfs/file-item.c (revision 84ed8a99)
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 "volumes.h"
27 #include "print-tree.h"
28 
29 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
30 				   sizeof(struct btrfs_item) * 2) / \
31 				  size) - 1))
32 
33 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
34 				       PAGE_CACHE_SIZE))
35 
36 #define MAX_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
37 				   sizeof(struct btrfs_ordered_sum)) / \
38 				   sizeof(u32) * (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 static struct btrfs_csum_item *
87 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
88 		  struct btrfs_root *root,
89 		  struct btrfs_path *path,
90 		  u64 bytenr, int cow)
91 {
92 	int ret;
93 	struct btrfs_key file_key;
94 	struct btrfs_key found_key;
95 	struct btrfs_csum_item *item;
96 	struct extent_buffer *leaf;
97 	u64 csum_offset = 0;
98 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
99 	int csums_in_item;
100 
101 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
102 	file_key.offset = bytenr;
103 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
104 	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
105 	if (ret < 0)
106 		goto fail;
107 	leaf = path->nodes[0];
108 	if (ret > 0) {
109 		ret = 1;
110 		if (path->slots[0] == 0)
111 			goto fail;
112 		path->slots[0]--;
113 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
114 		if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY)
115 			goto fail;
116 
117 		csum_offset = (bytenr - found_key.offset) >>
118 				root->fs_info->sb->s_blocksize_bits;
119 		csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
120 		csums_in_item /= csum_size;
121 
122 		if (csum_offset == csums_in_item) {
123 			ret = -EFBIG;
124 			goto fail;
125 		} else if (csum_offset > csums_in_item) {
126 			goto fail;
127 		}
128 	}
129 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
130 	item = (struct btrfs_csum_item *)((unsigned char *)item +
131 					  csum_offset * csum_size);
132 	return item;
133 fail:
134 	if (ret > 0)
135 		ret = -ENOENT;
136 	return ERR_PTR(ret);
137 }
138 
139 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
140 			     struct btrfs_root *root,
141 			     struct btrfs_path *path, u64 objectid,
142 			     u64 offset, int mod)
143 {
144 	int ret;
145 	struct btrfs_key file_key;
146 	int ins_len = mod < 0 ? -1 : 0;
147 	int cow = mod != 0;
148 
149 	file_key.objectid = objectid;
150 	file_key.offset = offset;
151 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
152 	ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
153 	return ret;
154 }
155 
156 static void btrfs_io_bio_endio_readpage(struct btrfs_io_bio *bio, int err)
157 {
158 	kfree(bio->csum_allocated);
159 }
160 
161 static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
162 				   struct inode *inode, struct bio *bio,
163 				   u64 logical_offset, u32 *dst, int dio)
164 {
165 	struct bio_vec *bvec = bio->bi_io_vec;
166 	struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
167 	struct btrfs_csum_item *item = NULL;
168 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
169 	struct btrfs_path *path;
170 	u8 *csum;
171 	u64 offset = 0;
172 	u64 item_start_offset = 0;
173 	u64 item_last_offset = 0;
174 	u64 disk_bytenr;
175 	u32 diff;
176 	int nblocks;
177 	int bio_index = 0;
178 	int count;
179 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
180 
181 	path = btrfs_alloc_path();
182 	if (!path)
183 		return -ENOMEM;
184 
185 	nblocks = bio->bi_size >> inode->i_sb->s_blocksize_bits;
186 	if (!dst) {
187 		if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
188 			btrfs_bio->csum_allocated = kmalloc(nblocks * csum_size,
189 							    GFP_NOFS);
190 			if (!btrfs_bio->csum_allocated) {
191 				btrfs_free_path(path);
192 				return -ENOMEM;
193 			}
194 			btrfs_bio->csum = btrfs_bio->csum_allocated;
195 			btrfs_bio->end_io = btrfs_io_bio_endio_readpage;
196 		} else {
197 			btrfs_bio->csum = btrfs_bio->csum_inline;
198 		}
199 		csum = btrfs_bio->csum;
200 	} else {
201 		csum = (u8 *)dst;
202 	}
203 
204 	if (bio->bi_size > PAGE_CACHE_SIZE * 8)
205 		path->reada = 2;
206 
207 	WARN_ON(bio->bi_vcnt <= 0);
208 
209 	/*
210 	 * the free space stuff is only read when it hasn't been
211 	 * updated in the current transaction.  So, we can safely
212 	 * read from the commit root and sidestep a nasty deadlock
213 	 * between reading the free space cache and updating the csum tree.
214 	 */
215 	if (btrfs_is_free_space_inode(inode)) {
216 		path->search_commit_root = 1;
217 		path->skip_locking = 1;
218 	}
219 
220 	disk_bytenr = (u64)bio->bi_sector << 9;
221 	if (dio)
222 		offset = logical_offset;
223 	while (bio_index < bio->bi_vcnt) {
224 		if (!dio)
225 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
226 		count = btrfs_find_ordered_sum(inode, offset, disk_bytenr,
227 					       (u32 *)csum, nblocks);
228 		if (count)
229 			goto found;
230 
231 		if (!item || disk_bytenr < item_start_offset ||
232 		    disk_bytenr >= item_last_offset) {
233 			struct btrfs_key found_key;
234 			u32 item_size;
235 
236 			if (item)
237 				btrfs_release_path(path);
238 			item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
239 						 path, disk_bytenr, 0);
240 			if (IS_ERR(item)) {
241 				count = 1;
242 				memset(csum, 0, csum_size);
243 				if (BTRFS_I(inode)->root->root_key.objectid ==
244 				    BTRFS_DATA_RELOC_TREE_OBJECTID) {
245 					set_extent_bits(io_tree, offset,
246 						offset + bvec->bv_len - 1,
247 						EXTENT_NODATASUM, GFP_NOFS);
248 				} else {
249 					printk(KERN_INFO "btrfs no csum found "
250 					       "for inode %llu start %llu\n",
251 					       btrfs_ino(inode), offset);
252 				}
253 				item = NULL;
254 				btrfs_release_path(path);
255 				goto found;
256 			}
257 			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
258 					      path->slots[0]);
259 
260 			item_start_offset = found_key.offset;
261 			item_size = btrfs_item_size_nr(path->nodes[0],
262 						       path->slots[0]);
263 			item_last_offset = item_start_offset +
264 				(item_size / csum_size) *
265 				root->sectorsize;
266 			item = btrfs_item_ptr(path->nodes[0], path->slots[0],
267 					      struct btrfs_csum_item);
268 		}
269 		/*
270 		 * this byte range must be able to fit inside
271 		 * a single leaf so it will also fit inside a u32
272 		 */
273 		diff = disk_bytenr - item_start_offset;
274 		diff = diff / root->sectorsize;
275 		diff = diff * csum_size;
276 		count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
277 					    inode->i_sb->s_blocksize_bits);
278 		read_extent_buffer(path->nodes[0], csum,
279 				   ((unsigned long)item) + diff,
280 				   csum_size * count);
281 found:
282 		csum += count * csum_size;
283 		nblocks -= count;
284 		while (count--) {
285 			disk_bytenr += bvec->bv_len;
286 			offset += bvec->bv_len;
287 			bio_index++;
288 			bvec++;
289 		}
290 	}
291 	btrfs_free_path(path);
292 	return 0;
293 }
294 
295 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
296 			  struct bio *bio, u32 *dst)
297 {
298 	return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
299 }
300 
301 int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
302 			      struct btrfs_dio_private *dip, struct bio *bio,
303 			      u64 offset)
304 {
305 	int len = (bio->bi_sector << 9) - dip->disk_bytenr;
306 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
307 	int ret;
308 
309 	len >>= inode->i_sb->s_blocksize_bits;
310 	len *= csum_size;
311 
312 	ret = __btrfs_lookup_bio_sums(root, inode, bio, offset,
313 				      (u32 *)(dip->csum + len), 1);
314 	return ret;
315 }
316 
317 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
318 			     struct list_head *list, int search_commit)
319 {
320 	struct btrfs_key key;
321 	struct btrfs_path *path;
322 	struct extent_buffer *leaf;
323 	struct btrfs_ordered_sum *sums;
324 	struct btrfs_csum_item *item;
325 	LIST_HEAD(tmplist);
326 	unsigned long offset;
327 	int ret;
328 	size_t size;
329 	u64 csum_end;
330 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
331 
332 	ASSERT(start == ALIGN(start, root->sectorsize) &&
333 	       (end + 1) == ALIGN(end + 1, root->sectorsize));
334 
335 	path = btrfs_alloc_path();
336 	if (!path)
337 		return -ENOMEM;
338 
339 	if (search_commit) {
340 		path->skip_locking = 1;
341 		path->reada = 2;
342 		path->search_commit_root = 1;
343 	}
344 
345 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
346 	key.offset = start;
347 	key.type = BTRFS_EXTENT_CSUM_KEY;
348 
349 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
350 	if (ret < 0)
351 		goto fail;
352 	if (ret > 0 && path->slots[0] > 0) {
353 		leaf = path->nodes[0];
354 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
355 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
356 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
357 			offset = (start - key.offset) >>
358 				 root->fs_info->sb->s_blocksize_bits;
359 			if (offset * csum_size <
360 			    btrfs_item_size_nr(leaf, path->slots[0] - 1))
361 				path->slots[0]--;
362 		}
363 	}
364 
365 	while (start <= end) {
366 		leaf = path->nodes[0];
367 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
368 			ret = btrfs_next_leaf(root, path);
369 			if (ret < 0)
370 				goto fail;
371 			if (ret > 0)
372 				break;
373 			leaf = path->nodes[0];
374 		}
375 
376 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
377 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
378 		    key.type != BTRFS_EXTENT_CSUM_KEY ||
379 		    key.offset > end)
380 			break;
381 
382 		if (key.offset > start)
383 			start = key.offset;
384 
385 		size = btrfs_item_size_nr(leaf, path->slots[0]);
386 		csum_end = key.offset + (size / csum_size) * root->sectorsize;
387 		if (csum_end <= start) {
388 			path->slots[0]++;
389 			continue;
390 		}
391 
392 		csum_end = min(csum_end, end + 1);
393 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
394 				      struct btrfs_csum_item);
395 		while (start < csum_end) {
396 			size = min_t(size_t, csum_end - start,
397 				     MAX_ORDERED_SUM_BYTES(root));
398 			sums = kzalloc(btrfs_ordered_sum_size(root, size),
399 				       GFP_NOFS);
400 			if (!sums) {
401 				ret = -ENOMEM;
402 				goto fail;
403 			}
404 
405 			sums->bytenr = start;
406 			sums->len = (int)size;
407 
408 			offset = (start - key.offset) >>
409 				root->fs_info->sb->s_blocksize_bits;
410 			offset *= csum_size;
411 			size >>= root->fs_info->sb->s_blocksize_bits;
412 
413 			read_extent_buffer(path->nodes[0],
414 					   sums->sums,
415 					   ((unsigned long)item) + offset,
416 					   csum_size * size);
417 
418 			start += root->sectorsize * size;
419 			list_add_tail(&sums->list, &tmplist);
420 		}
421 		path->slots[0]++;
422 	}
423 	ret = 0;
424 fail:
425 	while (ret < 0 && !list_empty(&tmplist)) {
426 		sums = list_entry(&tmplist, struct btrfs_ordered_sum, list);
427 		list_del(&sums->list);
428 		kfree(sums);
429 	}
430 	list_splice_tail(&tmplist, list);
431 
432 	btrfs_free_path(path);
433 	return ret;
434 }
435 
436 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
437 		       struct bio *bio, u64 file_start, int contig)
438 {
439 	struct btrfs_ordered_sum *sums;
440 	struct btrfs_ordered_extent *ordered;
441 	char *data;
442 	struct bio_vec *bvec = bio->bi_io_vec;
443 	int bio_index = 0;
444 	int index;
445 	unsigned long total_bytes = 0;
446 	unsigned long this_sum_bytes = 0;
447 	u64 offset;
448 
449 	WARN_ON(bio->bi_vcnt <= 0);
450 	sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
451 	if (!sums)
452 		return -ENOMEM;
453 
454 	sums->len = bio->bi_size;
455 	INIT_LIST_HEAD(&sums->list);
456 
457 	if (contig)
458 		offset = file_start;
459 	else
460 		offset = page_offset(bvec->bv_page) + bvec->bv_offset;
461 
462 	ordered = btrfs_lookup_ordered_extent(inode, offset);
463 	BUG_ON(!ordered); /* Logic error */
464 	sums->bytenr = (u64)bio->bi_sector << 9;
465 	index = 0;
466 
467 	while (bio_index < bio->bi_vcnt) {
468 		if (!contig)
469 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
470 
471 		if (offset >= ordered->file_offset + ordered->len ||
472 		    offset < ordered->file_offset) {
473 			unsigned long bytes_left;
474 			sums->len = this_sum_bytes;
475 			this_sum_bytes = 0;
476 			btrfs_add_ordered_sum(inode, ordered, sums);
477 			btrfs_put_ordered_extent(ordered);
478 
479 			bytes_left = bio->bi_size - total_bytes;
480 
481 			sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
482 				       GFP_NOFS);
483 			BUG_ON(!sums); /* -ENOMEM */
484 			sums->len = bytes_left;
485 			ordered = btrfs_lookup_ordered_extent(inode, offset);
486 			BUG_ON(!ordered); /* Logic error */
487 			sums->bytenr = ((u64)bio->bi_sector << 9) +
488 				       total_bytes;
489 			index = 0;
490 		}
491 
492 		data = kmap_atomic(bvec->bv_page);
493 		sums->sums[index] = ~(u32)0;
494 		sums->sums[index] = btrfs_csum_data(data + bvec->bv_offset,
495 						    sums->sums[index],
496 						    bvec->bv_len);
497 		kunmap_atomic(data);
498 		btrfs_csum_final(sums->sums[index],
499 				 (char *)(sums->sums + index));
500 
501 		bio_index++;
502 		index++;
503 		total_bytes += bvec->bv_len;
504 		this_sum_bytes += bvec->bv_len;
505 		offset += bvec->bv_len;
506 		bvec++;
507 	}
508 	this_sum_bytes = 0;
509 	btrfs_add_ordered_sum(inode, ordered, sums);
510 	btrfs_put_ordered_extent(ordered);
511 	return 0;
512 }
513 
514 /*
515  * helper function for csum removal, this expects the
516  * key to describe the csum pointed to by the path, and it expects
517  * the csum to overlap the range [bytenr, len]
518  *
519  * The csum should not be entirely contained in the range and the
520  * range should not be entirely contained in the csum.
521  *
522  * This calls btrfs_truncate_item with the correct args based on the
523  * overlap, and fixes up the key as required.
524  */
525 static noinline void truncate_one_csum(struct btrfs_root *root,
526 				       struct btrfs_path *path,
527 				       struct btrfs_key *key,
528 				       u64 bytenr, u64 len)
529 {
530 	struct extent_buffer *leaf;
531 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
532 	u64 csum_end;
533 	u64 end_byte = bytenr + len;
534 	u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
535 
536 	leaf = path->nodes[0];
537 	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
538 	csum_end <<= root->fs_info->sb->s_blocksize_bits;
539 	csum_end += key->offset;
540 
541 	if (key->offset < bytenr && csum_end <= end_byte) {
542 		/*
543 		 *         [ bytenr - len ]
544 		 *         [   ]
545 		 *   [csum     ]
546 		 *   A simple truncate off the end of the item
547 		 */
548 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
549 		new_size *= csum_size;
550 		btrfs_truncate_item(root, path, new_size, 1);
551 	} else if (key->offset >= bytenr && csum_end > end_byte &&
552 		   end_byte > key->offset) {
553 		/*
554 		 *         [ bytenr - len ]
555 		 *                 [ ]
556 		 *                 [csum     ]
557 		 * we need to truncate from the beginning of the csum
558 		 */
559 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
560 		new_size *= csum_size;
561 
562 		btrfs_truncate_item(root, path, new_size, 0);
563 
564 		key->offset = end_byte;
565 		btrfs_set_item_key_safe(root, path, key);
566 	} else {
567 		BUG();
568 	}
569 }
570 
571 /*
572  * deletes the csum items from the csum tree for a given
573  * range of bytes.
574  */
575 int btrfs_del_csums(struct btrfs_trans_handle *trans,
576 		    struct btrfs_root *root, u64 bytenr, u64 len)
577 {
578 	struct btrfs_path *path;
579 	struct btrfs_key key;
580 	u64 end_byte = bytenr + len;
581 	u64 csum_end;
582 	struct extent_buffer *leaf;
583 	int ret;
584 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
585 	int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
586 
587 	root = root->fs_info->csum_root;
588 
589 	path = btrfs_alloc_path();
590 	if (!path)
591 		return -ENOMEM;
592 
593 	while (1) {
594 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
595 		key.offset = end_byte - 1;
596 		key.type = BTRFS_EXTENT_CSUM_KEY;
597 
598 		path->leave_spinning = 1;
599 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
600 		if (ret > 0) {
601 			if (path->slots[0] == 0)
602 				break;
603 			path->slots[0]--;
604 		} else if (ret < 0) {
605 			break;
606 		}
607 
608 		leaf = path->nodes[0];
609 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
610 
611 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
612 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
613 			break;
614 		}
615 
616 		if (key.offset >= end_byte)
617 			break;
618 
619 		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
620 		csum_end <<= blocksize_bits;
621 		csum_end += key.offset;
622 
623 		/* this csum ends before we start, we're done */
624 		if (csum_end <= bytenr)
625 			break;
626 
627 		/* delete the entire item, it is inside our range */
628 		if (key.offset >= bytenr && csum_end <= end_byte) {
629 			ret = btrfs_del_item(trans, root, path);
630 			if (ret)
631 				goto out;
632 			if (key.offset == bytenr)
633 				break;
634 		} else if (key.offset < bytenr && csum_end > end_byte) {
635 			unsigned long offset;
636 			unsigned long shift_len;
637 			unsigned long item_offset;
638 			/*
639 			 *        [ bytenr - len ]
640 			 *     [csum                ]
641 			 *
642 			 * Our bytes are in the middle of the csum,
643 			 * we need to split this item and insert a new one.
644 			 *
645 			 * But we can't drop the path because the
646 			 * csum could change, get removed, extended etc.
647 			 *
648 			 * The trick here is the max size of a csum item leaves
649 			 * enough room in the tree block for a single
650 			 * item header.  So, we split the item in place,
651 			 * adding a new header pointing to the existing
652 			 * bytes.  Then we loop around again and we have
653 			 * a nicely formed csum item that we can neatly
654 			 * truncate.
655 			 */
656 			offset = (bytenr - key.offset) >> blocksize_bits;
657 			offset *= csum_size;
658 
659 			shift_len = (len >> blocksize_bits) * csum_size;
660 
661 			item_offset = btrfs_item_ptr_offset(leaf,
662 							    path->slots[0]);
663 
664 			memset_extent_buffer(leaf, 0, item_offset + offset,
665 					     shift_len);
666 			key.offset = bytenr;
667 
668 			/*
669 			 * btrfs_split_item returns -EAGAIN when the
670 			 * item changed size or key
671 			 */
672 			ret = btrfs_split_item(trans, root, path, &key, offset);
673 			if (ret && ret != -EAGAIN) {
674 				btrfs_abort_transaction(trans, root, ret);
675 				goto out;
676 			}
677 
678 			key.offset = end_byte - 1;
679 		} else {
680 			truncate_one_csum(root, path, &key, bytenr, len);
681 			if (key.offset < bytenr)
682 				break;
683 		}
684 		btrfs_release_path(path);
685 	}
686 	ret = 0;
687 out:
688 	btrfs_free_path(path);
689 	return ret;
690 }
691 
692 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
693 			   struct btrfs_root *root,
694 			   struct btrfs_ordered_sum *sums)
695 {
696 	struct btrfs_key file_key;
697 	struct btrfs_key found_key;
698 	struct btrfs_path *path;
699 	struct btrfs_csum_item *item;
700 	struct btrfs_csum_item *item_end;
701 	struct extent_buffer *leaf = NULL;
702 	u64 next_offset;
703 	u64 total_bytes = 0;
704 	u64 csum_offset;
705 	u64 bytenr;
706 	u32 nritems;
707 	u32 ins_size;
708 	int index = 0;
709 	int found_next;
710 	int ret;
711 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
712 
713 	path = btrfs_alloc_path();
714 	if (!path)
715 		return -ENOMEM;
716 again:
717 	next_offset = (u64)-1;
718 	found_next = 0;
719 	bytenr = sums->bytenr + total_bytes;
720 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
721 	file_key.offset = bytenr;
722 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
723 
724 	item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
725 	if (!IS_ERR(item)) {
726 		ret = 0;
727 		leaf = path->nodes[0];
728 		item_end = btrfs_item_ptr(leaf, path->slots[0],
729 					  struct btrfs_csum_item);
730 		item_end = (struct btrfs_csum_item *)((char *)item_end +
731 			   btrfs_item_size_nr(leaf, path->slots[0]));
732 		goto found;
733 	}
734 	ret = PTR_ERR(item);
735 	if (ret != -EFBIG && ret != -ENOENT)
736 		goto fail_unlock;
737 
738 	if (ret == -EFBIG) {
739 		u32 item_size;
740 		/* we found one, but it isn't big enough yet */
741 		leaf = path->nodes[0];
742 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
743 		if ((item_size / csum_size) >=
744 		    MAX_CSUM_ITEMS(root, csum_size)) {
745 			/* already at max size, make a new one */
746 			goto insert;
747 		}
748 	} else {
749 		int slot = path->slots[0] + 1;
750 		/* we didn't find a csum item, insert one */
751 		nritems = btrfs_header_nritems(path->nodes[0]);
752 		if (path->slots[0] >= nritems - 1) {
753 			ret = btrfs_next_leaf(root, path);
754 			if (ret == 1)
755 				found_next = 1;
756 			if (ret != 0)
757 				goto insert;
758 			slot = 0;
759 		}
760 		btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
761 		if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
762 		    found_key.type != BTRFS_EXTENT_CSUM_KEY) {
763 			found_next = 1;
764 			goto insert;
765 		}
766 		next_offset = found_key.offset;
767 		found_next = 1;
768 		goto insert;
769 	}
770 
771 	/*
772 	 * at this point, we know the tree has an item, but it isn't big
773 	 * enough yet to put our csum in.  Grow it
774 	 */
775 	btrfs_release_path(path);
776 	ret = btrfs_search_slot(trans, root, &file_key, path,
777 				csum_size, 1);
778 	if (ret < 0)
779 		goto fail_unlock;
780 
781 	if (ret > 0) {
782 		if (path->slots[0] == 0)
783 			goto insert;
784 		path->slots[0]--;
785 	}
786 
787 	leaf = path->nodes[0];
788 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
789 	csum_offset = (bytenr - found_key.offset) >>
790 			root->fs_info->sb->s_blocksize_bits;
791 
792 	if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
793 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
794 	    csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
795 		goto insert;
796 	}
797 
798 	if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
799 	    csum_size) {
800 		int extend_nr;
801 		u64 tmp;
802 		u32 diff;
803 		u32 free_space;
804 
805 		if (btrfs_leaf_free_space(root, leaf) <
806 				 sizeof(struct btrfs_item) + csum_size * 2)
807 			goto insert;
808 
809 		free_space = btrfs_leaf_free_space(root, leaf) -
810 					 sizeof(struct btrfs_item) - csum_size;
811 		tmp = sums->len - total_bytes;
812 		tmp >>= root->fs_info->sb->s_blocksize_bits;
813 		WARN_ON(tmp < 1);
814 
815 		extend_nr = max_t(int, 1, (int)tmp);
816 		diff = (csum_offset + extend_nr) * csum_size;
817 		diff = min(diff, MAX_CSUM_ITEMS(root, csum_size) * csum_size);
818 
819 		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
820 		diff = min(free_space, diff);
821 		diff /= csum_size;
822 		diff *= csum_size;
823 
824 		btrfs_extend_item(root, path, diff);
825 		ret = 0;
826 		goto csum;
827 	}
828 
829 insert:
830 	btrfs_release_path(path);
831 	csum_offset = 0;
832 	if (found_next) {
833 		u64 tmp;
834 
835 		tmp = sums->len - total_bytes;
836 		tmp >>= root->fs_info->sb->s_blocksize_bits;
837 		tmp = min(tmp, (next_offset - file_key.offset) >>
838 					 root->fs_info->sb->s_blocksize_bits);
839 
840 		tmp = max((u64)1, tmp);
841 		tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
842 		ins_size = csum_size * tmp;
843 	} else {
844 		ins_size = csum_size;
845 	}
846 	path->leave_spinning = 1;
847 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
848 				      ins_size);
849 	path->leave_spinning = 0;
850 	if (ret < 0)
851 		goto fail_unlock;
852 	if (WARN_ON(ret != 0))
853 		goto fail_unlock;
854 	leaf = path->nodes[0];
855 csum:
856 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
857 	item_end = (struct btrfs_csum_item *)((unsigned char *)item +
858 				      btrfs_item_size_nr(leaf, path->slots[0]));
859 	item = (struct btrfs_csum_item *)((unsigned char *)item +
860 					  csum_offset * csum_size);
861 found:
862 	ins_size = (u32)(sums->len - total_bytes) >>
863 		   root->fs_info->sb->s_blocksize_bits;
864 	ins_size *= csum_size;
865 	ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
866 			      ins_size);
867 	write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
868 			    ins_size);
869 
870 	ins_size /= csum_size;
871 	total_bytes += ins_size * root->sectorsize;
872 	index += ins_size;
873 
874 	btrfs_mark_buffer_dirty(path->nodes[0]);
875 	if (total_bytes < sums->len) {
876 		btrfs_release_path(path);
877 		cond_resched();
878 		goto again;
879 	}
880 out:
881 	btrfs_free_path(path);
882 	return ret;
883 
884 fail_unlock:
885 	goto out;
886 }
887