xref: /openbmc/linux/fs/btrfs/file-item.c (revision a266ef69)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/bio.h>
7 #include <linux/slab.h>
8 #include <linux/pagemap.h>
9 #include <linux/highmem.h>
10 #include <linux/sched/mm.h>
11 #include <crypto/hash.h>
12 #include "messages.h"
13 #include "misc.h"
14 #include "ctree.h"
15 #include "disk-io.h"
16 #include "transaction.h"
17 #include "bio.h"
18 #include "print-tree.h"
19 #include "compression.h"
20 #include "fs.h"
21 #include "accessors.h"
22 #include "file-item.h"
23 #include "super.h"
24 
25 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
26 				   sizeof(struct btrfs_item) * 2) / \
27 				  size) - 1))
28 
29 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
30 				       PAGE_SIZE))
31 
32 /*
33  * Set inode's size according to filesystem options.
34  *
35  * @inode:      inode we want to update the disk_i_size for
36  * @new_i_size: i_size we want to set to, 0 if we use i_size
37  *
38  * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read()
39  * returns as it is perfectly fine with a file that has holes without hole file
40  * extent items.
41  *
42  * However without NO_HOLES we need to only return the area that is contiguous
43  * from the 0 offset of the file.  Otherwise we could end up adjust i_size up
44  * to an extent that has a gap in between.
45  *
46  * Finally new_i_size should only be set in the case of truncate where we're not
47  * ready to use i_size_read() as the limiter yet.
48  */
49 void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size)
50 {
51 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
52 	u64 start, end, i_size;
53 	int ret;
54 
55 	i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
56 	if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
57 		inode->disk_i_size = i_size;
58 		return;
59 	}
60 
61 	spin_lock(&inode->lock);
62 	ret = find_contiguous_extent_bit(&inode->file_extent_tree, 0, &start,
63 					 &end, EXTENT_DIRTY);
64 	if (!ret && start == 0)
65 		i_size = min(i_size, end + 1);
66 	else
67 		i_size = 0;
68 	inode->disk_i_size = i_size;
69 	spin_unlock(&inode->lock);
70 }
71 
72 /*
73  * Mark range within a file as having a new extent inserted.
74  *
75  * @inode: inode being modified
76  * @start: start file offset of the file extent we've inserted
77  * @len:   logical length of the file extent item
78  *
79  * Call when we are inserting a new file extent where there was none before.
80  * Does not need to call this in the case where we're replacing an existing file
81  * extent, however if not sure it's fine to call this multiple times.
82  *
83  * The start and len must match the file extent item, so thus must be sectorsize
84  * aligned.
85  */
86 int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
87 				      u64 len)
88 {
89 	if (len == 0)
90 		return 0;
91 
92 	ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
93 
94 	if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
95 		return 0;
96 	return set_extent_bits(&inode->file_extent_tree, start, start + len - 1,
97 			       EXTENT_DIRTY);
98 }
99 
100 /*
101  * Mark an inode range as not having a backing extent.
102  *
103  * @inode: inode being modified
104  * @start: start file offset of the file extent we've inserted
105  * @len:   logical length of the file extent item
106  *
107  * Called when we drop a file extent, for example when we truncate.  Doesn't
108  * need to be called for cases where we're replacing a file extent, like when
109  * we've COWed a file extent.
110  *
111  * The start and len must match the file extent item, so thus must be sectorsize
112  * aligned.
113  */
114 int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
115 					u64 len)
116 {
117 	if (len == 0)
118 		return 0;
119 
120 	ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
121 	       len == (u64)-1);
122 
123 	if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
124 		return 0;
125 	return clear_extent_bit(&inode->file_extent_tree, start,
126 				start + len - 1, EXTENT_DIRTY, NULL);
127 }
128 
129 static size_t bytes_to_csum_size(const struct btrfs_fs_info *fs_info, u32 bytes)
130 {
131 	ASSERT(IS_ALIGNED(bytes, fs_info->sectorsize));
132 
133 	return (bytes >> fs_info->sectorsize_bits) * fs_info->csum_size;
134 }
135 
136 static size_t csum_size_to_bytes(const struct btrfs_fs_info *fs_info, u32 csum_size)
137 {
138 	ASSERT(IS_ALIGNED(csum_size, fs_info->csum_size));
139 
140 	return (csum_size / fs_info->csum_size) << fs_info->sectorsize_bits;
141 }
142 
143 static inline u32 max_ordered_sum_bytes(const struct btrfs_fs_info *fs_info)
144 {
145 	u32 max_csum_size = round_down(PAGE_SIZE - sizeof(struct btrfs_ordered_sum),
146 				       fs_info->csum_size);
147 
148 	return csum_size_to_bytes(fs_info, max_csum_size);
149 }
150 
151 /*
152  * Calculate the total size needed to allocate for an ordered sum structure
153  * spanning @bytes in the file.
154  */
155 static int btrfs_ordered_sum_size(struct btrfs_fs_info *fs_info, unsigned long bytes)
156 {
157 	return sizeof(struct btrfs_ordered_sum) + bytes_to_csum_size(fs_info, bytes);
158 }
159 
160 int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans,
161 			     struct btrfs_root *root,
162 			     u64 objectid, u64 pos, u64 num_bytes)
163 {
164 	int ret = 0;
165 	struct btrfs_file_extent_item *item;
166 	struct btrfs_key file_key;
167 	struct btrfs_path *path;
168 	struct extent_buffer *leaf;
169 
170 	path = btrfs_alloc_path();
171 	if (!path)
172 		return -ENOMEM;
173 	file_key.objectid = objectid;
174 	file_key.offset = pos;
175 	file_key.type = BTRFS_EXTENT_DATA_KEY;
176 
177 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
178 				      sizeof(*item));
179 	if (ret < 0)
180 		goto out;
181 	BUG_ON(ret); /* Can't happen */
182 	leaf = path->nodes[0];
183 	item = btrfs_item_ptr(leaf, path->slots[0],
184 			      struct btrfs_file_extent_item);
185 	btrfs_set_file_extent_disk_bytenr(leaf, item, 0);
186 	btrfs_set_file_extent_disk_num_bytes(leaf, item, 0);
187 	btrfs_set_file_extent_offset(leaf, item, 0);
188 	btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
189 	btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes);
190 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
191 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
192 	btrfs_set_file_extent_compression(leaf, item, 0);
193 	btrfs_set_file_extent_encryption(leaf, item, 0);
194 	btrfs_set_file_extent_other_encoding(leaf, item, 0);
195 
196 	btrfs_mark_buffer_dirty(leaf);
197 out:
198 	btrfs_free_path(path);
199 	return ret;
200 }
201 
202 static struct btrfs_csum_item *
203 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
204 		  struct btrfs_root *root,
205 		  struct btrfs_path *path,
206 		  u64 bytenr, int cow)
207 {
208 	struct btrfs_fs_info *fs_info = root->fs_info;
209 	int ret;
210 	struct btrfs_key file_key;
211 	struct btrfs_key found_key;
212 	struct btrfs_csum_item *item;
213 	struct extent_buffer *leaf;
214 	u64 csum_offset = 0;
215 	const u32 csum_size = fs_info->csum_size;
216 	int csums_in_item;
217 
218 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
219 	file_key.offset = bytenr;
220 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
221 	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
222 	if (ret < 0)
223 		goto fail;
224 	leaf = path->nodes[0];
225 	if (ret > 0) {
226 		ret = 1;
227 		if (path->slots[0] == 0)
228 			goto fail;
229 		path->slots[0]--;
230 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
231 		if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
232 			goto fail;
233 
234 		csum_offset = (bytenr - found_key.offset) >>
235 				fs_info->sectorsize_bits;
236 		csums_in_item = btrfs_item_size(leaf, path->slots[0]);
237 		csums_in_item /= csum_size;
238 
239 		if (csum_offset == csums_in_item) {
240 			ret = -EFBIG;
241 			goto fail;
242 		} else if (csum_offset > csums_in_item) {
243 			goto fail;
244 		}
245 	}
246 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
247 	item = (struct btrfs_csum_item *)((unsigned char *)item +
248 					  csum_offset * csum_size);
249 	return item;
250 fail:
251 	if (ret > 0)
252 		ret = -ENOENT;
253 	return ERR_PTR(ret);
254 }
255 
256 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
257 			     struct btrfs_root *root,
258 			     struct btrfs_path *path, u64 objectid,
259 			     u64 offset, int mod)
260 {
261 	struct btrfs_key file_key;
262 	int ins_len = mod < 0 ? -1 : 0;
263 	int cow = mod != 0;
264 
265 	file_key.objectid = objectid;
266 	file_key.offset = offset;
267 	file_key.type = BTRFS_EXTENT_DATA_KEY;
268 
269 	return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
270 }
271 
272 /*
273  * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
274  * store the result to @dst.
275  *
276  * Return >0 for the number of sectors we found.
277  * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
278  * for it. Caller may want to try next sector until one range is hit.
279  * Return <0 for fatal error.
280  */
281 static int search_csum_tree(struct btrfs_fs_info *fs_info,
282 			    struct btrfs_path *path, u64 disk_bytenr,
283 			    u64 len, u8 *dst)
284 {
285 	struct btrfs_root *csum_root;
286 	struct btrfs_csum_item *item = NULL;
287 	struct btrfs_key key;
288 	const u32 sectorsize = fs_info->sectorsize;
289 	const u32 csum_size = fs_info->csum_size;
290 	u32 itemsize;
291 	int ret;
292 	u64 csum_start;
293 	u64 csum_len;
294 
295 	ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
296 	       IS_ALIGNED(len, sectorsize));
297 
298 	/* Check if the current csum item covers disk_bytenr */
299 	if (path->nodes[0]) {
300 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
301 				      struct btrfs_csum_item);
302 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
303 		itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
304 
305 		csum_start = key.offset;
306 		csum_len = (itemsize / csum_size) * sectorsize;
307 
308 		if (in_range(disk_bytenr, csum_start, csum_len))
309 			goto found;
310 	}
311 
312 	/* Current item doesn't contain the desired range, search again */
313 	btrfs_release_path(path);
314 	csum_root = btrfs_csum_root(fs_info, disk_bytenr);
315 	item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0);
316 	if (IS_ERR(item)) {
317 		ret = PTR_ERR(item);
318 		goto out;
319 	}
320 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
321 	itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
322 
323 	csum_start = key.offset;
324 	csum_len = (itemsize / csum_size) * sectorsize;
325 	ASSERT(in_range(disk_bytenr, csum_start, csum_len));
326 
327 found:
328 	ret = (min(csum_start + csum_len, disk_bytenr + len) -
329 		   disk_bytenr) >> fs_info->sectorsize_bits;
330 	read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
331 			ret * csum_size);
332 out:
333 	if (ret == -ENOENT || ret == -EFBIG)
334 		ret = 0;
335 	return ret;
336 }
337 
338 /*
339  * Locate the file_offset of @cur_disk_bytenr of a @bio.
340  *
341  * Bio of btrfs represents read range of
342  * [bi_sector << 9, bi_sector << 9 + bi_size).
343  * Knowing this, we can iterate through each bvec to locate the page belong to
344  * @cur_disk_bytenr and get the file offset.
345  *
346  * @inode is used to determine if the bvec page really belongs to @inode.
347  *
348  * Return 0 if we can't find the file offset
349  * Return >0 if we find the file offset and restore it to @file_offset_ret
350  */
351 static int search_file_offset_in_bio(struct bio *bio, struct inode *inode,
352 				     u64 disk_bytenr, u64 *file_offset_ret)
353 {
354 	struct bvec_iter iter;
355 	struct bio_vec bvec;
356 	u64 cur = bio->bi_iter.bi_sector << SECTOR_SHIFT;
357 	int ret = 0;
358 
359 	bio_for_each_segment(bvec, bio, iter) {
360 		struct page *page = bvec.bv_page;
361 
362 		if (cur > disk_bytenr)
363 			break;
364 		if (cur + bvec.bv_len <= disk_bytenr) {
365 			cur += bvec.bv_len;
366 			continue;
367 		}
368 		ASSERT(in_range(disk_bytenr, cur, bvec.bv_len));
369 		if (page->mapping && page->mapping->host &&
370 		    page->mapping->host == inode) {
371 			ret = 1;
372 			*file_offset_ret = page_offset(page) + bvec.bv_offset +
373 					   disk_bytenr - cur;
374 			break;
375 		}
376 	}
377 	return ret;
378 }
379 
380 /*
381  * Lookup the checksum for the read bio in csum tree.
382  *
383  * @inode:  inode that the bio is for.
384  * @bio:    bio to look up.
385  * @dst:    Buffer of size nblocks * btrfs_super_csum_size() used to return
386  *          checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If
387  *          NULL, the checksum buffer is allocated and returned in
388  *          btrfs_bio(bio)->csum instead.
389  *
390  * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
391  */
392 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst)
393 {
394 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
395 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
396 	struct btrfs_bio *bbio = NULL;
397 	struct btrfs_path *path;
398 	const u32 sectorsize = fs_info->sectorsize;
399 	const u32 csum_size = fs_info->csum_size;
400 	u32 orig_len = bio->bi_iter.bi_size;
401 	u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
402 	u64 cur_disk_bytenr;
403 	u8 *csum;
404 	const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
405 	int count = 0;
406 	blk_status_t ret = BLK_STS_OK;
407 
408 	if ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
409 	    test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state))
410 		return BLK_STS_OK;
411 
412 	/*
413 	 * This function is only called for read bio.
414 	 *
415 	 * This means two things:
416 	 * - All our csums should only be in csum tree
417 	 *   No ordered extents csums, as ordered extents are only for write
418 	 *   path.
419 	 * - No need to bother any other info from bvec
420 	 *   Since we're looking up csums, the only important info is the
421 	 *   disk_bytenr and the length, which can be extracted from bi_iter
422 	 *   directly.
423 	 */
424 	ASSERT(bio_op(bio) == REQ_OP_READ);
425 	path = btrfs_alloc_path();
426 	if (!path)
427 		return BLK_STS_RESOURCE;
428 
429 	if (!dst) {
430 		bbio = btrfs_bio(bio);
431 
432 		if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
433 			bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS);
434 			if (!bbio->csum) {
435 				btrfs_free_path(path);
436 				return BLK_STS_RESOURCE;
437 			}
438 		} else {
439 			bbio->csum = bbio->csum_inline;
440 		}
441 		csum = bbio->csum;
442 	} else {
443 		csum = dst;
444 	}
445 
446 	/*
447 	 * If requested number of sectors is larger than one leaf can contain,
448 	 * kick the readahead for csum tree.
449 	 */
450 	if (nblocks > fs_info->csums_per_leaf)
451 		path->reada = READA_FORWARD;
452 
453 	/*
454 	 * the free space stuff is only read when it hasn't been
455 	 * updated in the current transaction.  So, we can safely
456 	 * read from the commit root and sidestep a nasty deadlock
457 	 * between reading the free space cache and updating the csum tree.
458 	 */
459 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
460 		path->search_commit_root = 1;
461 		path->skip_locking = 1;
462 	}
463 
464 	for (cur_disk_bytenr = orig_disk_bytenr;
465 	     cur_disk_bytenr < orig_disk_bytenr + orig_len;
466 	     cur_disk_bytenr += (count * sectorsize)) {
467 		u64 search_len = orig_disk_bytenr + orig_len - cur_disk_bytenr;
468 		unsigned int sector_offset;
469 		u8 *csum_dst;
470 
471 		/*
472 		 * Although both cur_disk_bytenr and orig_disk_bytenr is u64,
473 		 * we're calculating the offset to the bio start.
474 		 *
475 		 * Bio size is limited to UINT_MAX, thus unsigned int is large
476 		 * enough to contain the raw result, not to mention the right
477 		 * shifted result.
478 		 */
479 		ASSERT(cur_disk_bytenr - orig_disk_bytenr < UINT_MAX);
480 		sector_offset = (cur_disk_bytenr - orig_disk_bytenr) >>
481 				fs_info->sectorsize_bits;
482 		csum_dst = csum + sector_offset * csum_size;
483 
484 		count = search_csum_tree(fs_info, path, cur_disk_bytenr,
485 					 search_len, csum_dst);
486 		if (count < 0) {
487 			ret = errno_to_blk_status(count);
488 			if (bbio)
489 				btrfs_bio_free_csum(bbio);
490 			break;
491 		}
492 
493 		/*
494 		 * We didn't find a csum for this range.  We need to make sure
495 		 * we complain loudly about this, because we are not NODATASUM.
496 		 *
497 		 * However for the DATA_RELOC inode we could potentially be
498 		 * relocating data extents for a NODATASUM inode, so the inode
499 		 * itself won't be marked with NODATASUM, but the extent we're
500 		 * copying is in fact NODATASUM.  If we don't find a csum we
501 		 * assume this is the case.
502 		 */
503 		if (count == 0) {
504 			memset(csum_dst, 0, csum_size);
505 			count = 1;
506 
507 			if (BTRFS_I(inode)->root->root_key.objectid ==
508 			    BTRFS_DATA_RELOC_TREE_OBJECTID) {
509 				u64 file_offset;
510 				int ret;
511 
512 				ret = search_file_offset_in_bio(bio, inode,
513 						cur_disk_bytenr, &file_offset);
514 				if (ret)
515 					set_extent_bits(io_tree, file_offset,
516 						file_offset + sectorsize - 1,
517 						EXTENT_NODATASUM);
518 			} else {
519 				btrfs_warn_rl(fs_info,
520 			"csum hole found for disk bytenr range [%llu, %llu)",
521 				cur_disk_bytenr, cur_disk_bytenr + sectorsize);
522 			}
523 		}
524 	}
525 
526 	btrfs_free_path(path);
527 	return ret;
528 }
529 
530 int btrfs_lookup_csums_list(struct btrfs_root *root, u64 start, u64 end,
531 			    struct list_head *list, int search_commit,
532 			    bool nowait)
533 {
534 	struct btrfs_fs_info *fs_info = root->fs_info;
535 	struct btrfs_key key;
536 	struct btrfs_path *path;
537 	struct extent_buffer *leaf;
538 	struct btrfs_ordered_sum *sums;
539 	struct btrfs_csum_item *item;
540 	LIST_HEAD(tmplist);
541 	int ret;
542 
543 	ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
544 	       IS_ALIGNED(end + 1, fs_info->sectorsize));
545 
546 	path = btrfs_alloc_path();
547 	if (!path)
548 		return -ENOMEM;
549 
550 	path->nowait = nowait;
551 	if (search_commit) {
552 		path->skip_locking = 1;
553 		path->reada = READA_FORWARD;
554 		path->search_commit_root = 1;
555 	}
556 
557 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
558 	key.offset = start;
559 	key.type = BTRFS_EXTENT_CSUM_KEY;
560 
561 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
562 	if (ret < 0)
563 		goto fail;
564 	if (ret > 0 && path->slots[0] > 0) {
565 		leaf = path->nodes[0];
566 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
567 
568 		/*
569 		 * There are two cases we can hit here for the previous csum
570 		 * item:
571 		 *
572 		 *		|<- search range ->|
573 		 *	|<- csum item ->|
574 		 *
575 		 * Or
576 		 *				|<- search range ->|
577 		 *	|<- csum item ->|
578 		 *
579 		 * Check if the previous csum item covers the leading part of
580 		 * the search range.  If so we have to start from previous csum
581 		 * item.
582 		 */
583 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
584 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
585 			if (bytes_to_csum_size(fs_info, start - key.offset) <
586 			    btrfs_item_size(leaf, path->slots[0] - 1))
587 				path->slots[0]--;
588 		}
589 	}
590 
591 	while (start <= end) {
592 		u64 csum_end;
593 
594 		leaf = path->nodes[0];
595 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
596 			ret = btrfs_next_leaf(root, path);
597 			if (ret < 0)
598 				goto fail;
599 			if (ret > 0)
600 				break;
601 			leaf = path->nodes[0];
602 		}
603 
604 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
605 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
606 		    key.type != BTRFS_EXTENT_CSUM_KEY ||
607 		    key.offset > end)
608 			break;
609 
610 		if (key.offset > start)
611 			start = key.offset;
612 
613 		csum_end = key.offset + csum_size_to_bytes(fs_info,
614 					btrfs_item_size(leaf, path->slots[0]));
615 		if (csum_end <= start) {
616 			path->slots[0]++;
617 			continue;
618 		}
619 
620 		csum_end = min(csum_end, end + 1);
621 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
622 				      struct btrfs_csum_item);
623 		while (start < csum_end) {
624 			unsigned long offset;
625 			size_t size;
626 
627 			size = min_t(size_t, csum_end - start,
628 				     max_ordered_sum_bytes(fs_info));
629 			sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
630 				       GFP_NOFS);
631 			if (!sums) {
632 				ret = -ENOMEM;
633 				goto fail;
634 			}
635 
636 			sums->bytenr = start;
637 			sums->len = (int)size;
638 
639 			offset = bytes_to_csum_size(fs_info, start - key.offset);
640 
641 			read_extent_buffer(path->nodes[0],
642 					   sums->sums,
643 					   ((unsigned long)item) + offset,
644 					   bytes_to_csum_size(fs_info, size));
645 
646 			start += size;
647 			list_add_tail(&sums->list, &tmplist);
648 		}
649 		path->slots[0]++;
650 	}
651 	ret = 0;
652 fail:
653 	while (ret < 0 && !list_empty(&tmplist)) {
654 		sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
655 		list_del(&sums->list);
656 		kfree(sums);
657 	}
658 	list_splice_tail(&tmplist, list);
659 
660 	btrfs_free_path(path);
661 	return ret;
662 }
663 
664 /*
665  * Do the same work as btrfs_lookup_csums_list(), the difference is in how
666  * we return the result.
667  *
668  * This version will set the corresponding bits in @csum_bitmap to represent
669  * that there is a csum found.
670  * Each bit represents a sector. Thus caller should ensure @csum_buf passed
671  * in is large enough to contain all csums.
672  */
673 int btrfs_lookup_csums_bitmap(struct btrfs_root *root, u64 start, u64 end,
674 			      u8 *csum_buf, unsigned long *csum_bitmap)
675 {
676 	struct btrfs_fs_info *fs_info = root->fs_info;
677 	struct btrfs_key key;
678 	struct btrfs_path *path;
679 	struct extent_buffer *leaf;
680 	struct btrfs_csum_item *item;
681 	const u64 orig_start = start;
682 	int ret;
683 
684 	ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
685 	       IS_ALIGNED(end + 1, fs_info->sectorsize));
686 
687 	path = btrfs_alloc_path();
688 	if (!path)
689 		return -ENOMEM;
690 
691 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
692 	key.type = BTRFS_EXTENT_CSUM_KEY;
693 	key.offset = start;
694 
695 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
696 	if (ret < 0)
697 		goto fail;
698 	if (ret > 0 && path->slots[0] > 0) {
699 		leaf = path->nodes[0];
700 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
701 
702 		/*
703 		 * There are two cases we can hit here for the previous csum
704 		 * item:
705 		 *
706 		 *		|<- search range ->|
707 		 *	|<- csum item ->|
708 		 *
709 		 * Or
710 		 *				|<- search range ->|
711 		 *	|<- csum item ->|
712 		 *
713 		 * Check if the previous csum item covers the leading part of
714 		 * the search range.  If so we have to start from previous csum
715 		 * item.
716 		 */
717 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
718 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
719 			if (bytes_to_csum_size(fs_info, start - key.offset) <
720 			    btrfs_item_size(leaf, path->slots[0] - 1))
721 				path->slots[0]--;
722 		}
723 	}
724 
725 	while (start <= end) {
726 		u64 csum_end;
727 
728 		leaf = path->nodes[0];
729 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
730 			ret = btrfs_next_leaf(root, path);
731 			if (ret < 0)
732 				goto fail;
733 			if (ret > 0)
734 				break;
735 			leaf = path->nodes[0];
736 		}
737 
738 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
739 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
740 		    key.type != BTRFS_EXTENT_CSUM_KEY ||
741 		    key.offset > end)
742 			break;
743 
744 		if (key.offset > start)
745 			start = key.offset;
746 
747 		csum_end = key.offset + csum_size_to_bytes(fs_info,
748 					btrfs_item_size(leaf, path->slots[0]));
749 		if (csum_end <= start) {
750 			path->slots[0]++;
751 			continue;
752 		}
753 
754 		csum_end = min(csum_end, end + 1);
755 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
756 				      struct btrfs_csum_item);
757 		while (start < csum_end) {
758 			unsigned long offset;
759 			size_t size;
760 			u8 *csum_dest = csum_buf + bytes_to_csum_size(fs_info,
761 						start - orig_start);
762 
763 			size = min_t(size_t, csum_end - start, end + 1 - start);
764 
765 			offset = bytes_to_csum_size(fs_info, start - key.offset);
766 
767 			read_extent_buffer(path->nodes[0], csum_dest,
768 					   ((unsigned long)item) + offset,
769 					   bytes_to_csum_size(fs_info, size));
770 
771 			bitmap_set(csum_bitmap,
772 				(start - orig_start) >> fs_info->sectorsize_bits,
773 				size >> fs_info->sectorsize_bits);
774 
775 			start += size;
776 		}
777 		path->slots[0]++;
778 	}
779 	ret = 0;
780 fail:
781 	btrfs_free_path(path);
782 	return ret;
783 }
784 
785 /*
786  * Calculate checksums of the data contained inside a bio.
787  *
788  * @inode:	 Owner of the data inside the bio
789  * @bio:	 Contains the data to be checksummed
790  * @offset:      If (u64)-1, @bio may contain discontiguous bio vecs, so the
791  *               file offsets are determined from the page offsets in the bio.
792  *               Otherwise, this is the starting file offset of the bio vecs in
793  *               @bio, which must be contiguous.
794  * @one_ordered: If true, @bio only refers to one ordered extent.
795  */
796 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
797 				u64 offset, bool one_ordered)
798 {
799 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
800 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
801 	struct btrfs_ordered_sum *sums;
802 	struct btrfs_ordered_extent *ordered = NULL;
803 	const bool use_page_offsets = (offset == (u64)-1);
804 	char *data;
805 	struct bvec_iter iter;
806 	struct bio_vec bvec;
807 	int index;
808 	unsigned int blockcount;
809 	unsigned long total_bytes = 0;
810 	unsigned long this_sum_bytes = 0;
811 	int i;
812 	unsigned nofs_flag;
813 
814 	nofs_flag = memalloc_nofs_save();
815 	sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
816 		       GFP_KERNEL);
817 	memalloc_nofs_restore(nofs_flag);
818 
819 	if (!sums)
820 		return BLK_STS_RESOURCE;
821 
822 	sums->len = bio->bi_iter.bi_size;
823 	INIT_LIST_HEAD(&sums->list);
824 
825 	sums->bytenr = bio->bi_iter.bi_sector << 9;
826 	index = 0;
827 
828 	shash->tfm = fs_info->csum_shash;
829 
830 	bio_for_each_segment(bvec, bio, iter) {
831 		if (use_page_offsets)
832 			offset = page_offset(bvec.bv_page) + bvec.bv_offset;
833 
834 		if (!ordered) {
835 			ordered = btrfs_lookup_ordered_extent(inode, offset);
836 			/*
837 			 * The bio range is not covered by any ordered extent,
838 			 * must be a code logic error.
839 			 */
840 			if (unlikely(!ordered)) {
841 				WARN(1, KERN_WARNING
842 			"no ordered extent for root %llu ino %llu offset %llu\n",
843 				     inode->root->root_key.objectid,
844 				     btrfs_ino(inode), offset);
845 				kvfree(sums);
846 				return BLK_STS_IOERR;
847 			}
848 		}
849 
850 		blockcount = BTRFS_BYTES_TO_BLKS(fs_info,
851 						 bvec.bv_len + fs_info->sectorsize
852 						 - 1);
853 
854 		for (i = 0; i < blockcount; i++) {
855 			if (!one_ordered &&
856 			    !in_range(offset, ordered->file_offset,
857 				      ordered->num_bytes)) {
858 				unsigned long bytes_left;
859 
860 				sums->len = this_sum_bytes;
861 				this_sum_bytes = 0;
862 				btrfs_add_ordered_sum(ordered, sums);
863 				btrfs_put_ordered_extent(ordered);
864 
865 				bytes_left = bio->bi_iter.bi_size - total_bytes;
866 
867 				nofs_flag = memalloc_nofs_save();
868 				sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
869 						      bytes_left), GFP_KERNEL);
870 				memalloc_nofs_restore(nofs_flag);
871 				BUG_ON(!sums); /* -ENOMEM */
872 				sums->len = bytes_left;
873 				ordered = btrfs_lookup_ordered_extent(inode,
874 								offset);
875 				ASSERT(ordered); /* Logic error */
876 				sums->bytenr = (bio->bi_iter.bi_sector << 9)
877 					+ total_bytes;
878 				index = 0;
879 			}
880 
881 			data = bvec_kmap_local(&bvec);
882 			crypto_shash_digest(shash,
883 					    data + (i * fs_info->sectorsize),
884 					    fs_info->sectorsize,
885 					    sums->sums + index);
886 			kunmap_local(data);
887 			index += fs_info->csum_size;
888 			offset += fs_info->sectorsize;
889 			this_sum_bytes += fs_info->sectorsize;
890 			total_bytes += fs_info->sectorsize;
891 		}
892 
893 	}
894 	this_sum_bytes = 0;
895 	btrfs_add_ordered_sum(ordered, sums);
896 	btrfs_put_ordered_extent(ordered);
897 	return 0;
898 }
899 
900 /*
901  * Remove one checksum overlapping a range.
902  *
903  * This expects the key to describe the csum pointed to by the path, and it
904  * expects the csum to overlap the range [bytenr, len]
905  *
906  * The csum should not be entirely contained in the range and the range should
907  * not be entirely contained in the csum.
908  *
909  * This calls btrfs_truncate_item with the correct args based on the overlap,
910  * and fixes up the key as required.
911  */
912 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
913 				       struct btrfs_path *path,
914 				       struct btrfs_key *key,
915 				       u64 bytenr, u64 len)
916 {
917 	struct extent_buffer *leaf;
918 	const u32 csum_size = fs_info->csum_size;
919 	u64 csum_end;
920 	u64 end_byte = bytenr + len;
921 	u32 blocksize_bits = fs_info->sectorsize_bits;
922 
923 	leaf = path->nodes[0];
924 	csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
925 	csum_end <<= blocksize_bits;
926 	csum_end += key->offset;
927 
928 	if (key->offset < bytenr && csum_end <= end_byte) {
929 		/*
930 		 *         [ bytenr - len ]
931 		 *         [   ]
932 		 *   [csum     ]
933 		 *   A simple truncate off the end of the item
934 		 */
935 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
936 		new_size *= csum_size;
937 		btrfs_truncate_item(path, new_size, 1);
938 	} else if (key->offset >= bytenr && csum_end > end_byte &&
939 		   end_byte > key->offset) {
940 		/*
941 		 *         [ bytenr - len ]
942 		 *                 [ ]
943 		 *                 [csum     ]
944 		 * we need to truncate from the beginning of the csum
945 		 */
946 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
947 		new_size *= csum_size;
948 
949 		btrfs_truncate_item(path, new_size, 0);
950 
951 		key->offset = end_byte;
952 		btrfs_set_item_key_safe(fs_info, path, key);
953 	} else {
954 		BUG();
955 	}
956 }
957 
958 /*
959  * Delete the csum items from the csum tree for a given range of bytes.
960  */
961 int btrfs_del_csums(struct btrfs_trans_handle *trans,
962 		    struct btrfs_root *root, u64 bytenr, u64 len)
963 {
964 	struct btrfs_fs_info *fs_info = trans->fs_info;
965 	struct btrfs_path *path;
966 	struct btrfs_key key;
967 	u64 end_byte = bytenr + len;
968 	u64 csum_end;
969 	struct extent_buffer *leaf;
970 	int ret = 0;
971 	const u32 csum_size = fs_info->csum_size;
972 	u32 blocksize_bits = fs_info->sectorsize_bits;
973 
974 	ASSERT(root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID ||
975 	       root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
976 
977 	path = btrfs_alloc_path();
978 	if (!path)
979 		return -ENOMEM;
980 
981 	while (1) {
982 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
983 		key.offset = end_byte - 1;
984 		key.type = BTRFS_EXTENT_CSUM_KEY;
985 
986 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
987 		if (ret > 0) {
988 			ret = 0;
989 			if (path->slots[0] == 0)
990 				break;
991 			path->slots[0]--;
992 		} else if (ret < 0) {
993 			break;
994 		}
995 
996 		leaf = path->nodes[0];
997 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
998 
999 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1000 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
1001 			break;
1002 		}
1003 
1004 		if (key.offset >= end_byte)
1005 			break;
1006 
1007 		csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
1008 		csum_end <<= blocksize_bits;
1009 		csum_end += key.offset;
1010 
1011 		/* this csum ends before we start, we're done */
1012 		if (csum_end <= bytenr)
1013 			break;
1014 
1015 		/* delete the entire item, it is inside our range */
1016 		if (key.offset >= bytenr && csum_end <= end_byte) {
1017 			int del_nr = 1;
1018 
1019 			/*
1020 			 * Check how many csum items preceding this one in this
1021 			 * leaf correspond to our range and then delete them all
1022 			 * at once.
1023 			 */
1024 			if (key.offset > bytenr && path->slots[0] > 0) {
1025 				int slot = path->slots[0] - 1;
1026 
1027 				while (slot >= 0) {
1028 					struct btrfs_key pk;
1029 
1030 					btrfs_item_key_to_cpu(leaf, &pk, slot);
1031 					if (pk.offset < bytenr ||
1032 					    pk.type != BTRFS_EXTENT_CSUM_KEY ||
1033 					    pk.objectid !=
1034 					    BTRFS_EXTENT_CSUM_OBJECTID)
1035 						break;
1036 					path->slots[0] = slot;
1037 					del_nr++;
1038 					key.offset = pk.offset;
1039 					slot--;
1040 				}
1041 			}
1042 			ret = btrfs_del_items(trans, root, path,
1043 					      path->slots[0], del_nr);
1044 			if (ret)
1045 				break;
1046 			if (key.offset == bytenr)
1047 				break;
1048 		} else if (key.offset < bytenr && csum_end > end_byte) {
1049 			unsigned long offset;
1050 			unsigned long shift_len;
1051 			unsigned long item_offset;
1052 			/*
1053 			 *        [ bytenr - len ]
1054 			 *     [csum                ]
1055 			 *
1056 			 * Our bytes are in the middle of the csum,
1057 			 * we need to split this item and insert a new one.
1058 			 *
1059 			 * But we can't drop the path because the
1060 			 * csum could change, get removed, extended etc.
1061 			 *
1062 			 * The trick here is the max size of a csum item leaves
1063 			 * enough room in the tree block for a single
1064 			 * item header.  So, we split the item in place,
1065 			 * adding a new header pointing to the existing
1066 			 * bytes.  Then we loop around again and we have
1067 			 * a nicely formed csum item that we can neatly
1068 			 * truncate.
1069 			 */
1070 			offset = (bytenr - key.offset) >> blocksize_bits;
1071 			offset *= csum_size;
1072 
1073 			shift_len = (len >> blocksize_bits) * csum_size;
1074 
1075 			item_offset = btrfs_item_ptr_offset(leaf,
1076 							    path->slots[0]);
1077 
1078 			memzero_extent_buffer(leaf, item_offset + offset,
1079 					     shift_len);
1080 			key.offset = bytenr;
1081 
1082 			/*
1083 			 * btrfs_split_item returns -EAGAIN when the
1084 			 * item changed size or key
1085 			 */
1086 			ret = btrfs_split_item(trans, root, path, &key, offset);
1087 			if (ret && ret != -EAGAIN) {
1088 				btrfs_abort_transaction(trans, ret);
1089 				break;
1090 			}
1091 			ret = 0;
1092 
1093 			key.offset = end_byte - 1;
1094 		} else {
1095 			truncate_one_csum(fs_info, path, &key, bytenr, len);
1096 			if (key.offset < bytenr)
1097 				break;
1098 		}
1099 		btrfs_release_path(path);
1100 	}
1101 	btrfs_free_path(path);
1102 	return ret;
1103 }
1104 
1105 static int find_next_csum_offset(struct btrfs_root *root,
1106 				 struct btrfs_path *path,
1107 				 u64 *next_offset)
1108 {
1109 	const u32 nritems = btrfs_header_nritems(path->nodes[0]);
1110 	struct btrfs_key found_key;
1111 	int slot = path->slots[0] + 1;
1112 	int ret;
1113 
1114 	if (nritems == 0 || slot >= nritems) {
1115 		ret = btrfs_next_leaf(root, path);
1116 		if (ret < 0) {
1117 			return ret;
1118 		} else if (ret > 0) {
1119 			*next_offset = (u64)-1;
1120 			return 0;
1121 		}
1122 		slot = path->slots[0];
1123 	}
1124 
1125 	btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
1126 
1127 	if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1128 	    found_key.type != BTRFS_EXTENT_CSUM_KEY)
1129 		*next_offset = (u64)-1;
1130 	else
1131 		*next_offset = found_key.offset;
1132 
1133 	return 0;
1134 }
1135 
1136 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
1137 			   struct btrfs_root *root,
1138 			   struct btrfs_ordered_sum *sums)
1139 {
1140 	struct btrfs_fs_info *fs_info = root->fs_info;
1141 	struct btrfs_key file_key;
1142 	struct btrfs_key found_key;
1143 	struct btrfs_path *path;
1144 	struct btrfs_csum_item *item;
1145 	struct btrfs_csum_item *item_end;
1146 	struct extent_buffer *leaf = NULL;
1147 	u64 next_offset;
1148 	u64 total_bytes = 0;
1149 	u64 csum_offset;
1150 	u64 bytenr;
1151 	u32 ins_size;
1152 	int index = 0;
1153 	int found_next;
1154 	int ret;
1155 	const u32 csum_size = fs_info->csum_size;
1156 
1157 	path = btrfs_alloc_path();
1158 	if (!path)
1159 		return -ENOMEM;
1160 again:
1161 	next_offset = (u64)-1;
1162 	found_next = 0;
1163 	bytenr = sums->bytenr + total_bytes;
1164 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1165 	file_key.offset = bytenr;
1166 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
1167 
1168 	item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1169 	if (!IS_ERR(item)) {
1170 		ret = 0;
1171 		leaf = path->nodes[0];
1172 		item_end = btrfs_item_ptr(leaf, path->slots[0],
1173 					  struct btrfs_csum_item);
1174 		item_end = (struct btrfs_csum_item *)((char *)item_end +
1175 			   btrfs_item_size(leaf, path->slots[0]));
1176 		goto found;
1177 	}
1178 	ret = PTR_ERR(item);
1179 	if (ret != -EFBIG && ret != -ENOENT)
1180 		goto out;
1181 
1182 	if (ret == -EFBIG) {
1183 		u32 item_size;
1184 		/* we found one, but it isn't big enough yet */
1185 		leaf = path->nodes[0];
1186 		item_size = btrfs_item_size(leaf, path->slots[0]);
1187 		if ((item_size / csum_size) >=
1188 		    MAX_CSUM_ITEMS(fs_info, csum_size)) {
1189 			/* already at max size, make a new one */
1190 			goto insert;
1191 		}
1192 	} else {
1193 		/* We didn't find a csum item, insert one. */
1194 		ret = find_next_csum_offset(root, path, &next_offset);
1195 		if (ret < 0)
1196 			goto out;
1197 		found_next = 1;
1198 		goto insert;
1199 	}
1200 
1201 	/*
1202 	 * At this point, we know the tree has a checksum item that ends at an
1203 	 * offset matching the start of the checksum range we want to insert.
1204 	 * We try to extend that item as much as possible and then add as many
1205 	 * checksums to it as they fit.
1206 	 *
1207 	 * First check if the leaf has enough free space for at least one
1208 	 * checksum. If it has go directly to the item extension code, otherwise
1209 	 * release the path and do a search for insertion before the extension.
1210 	 */
1211 	if (btrfs_leaf_free_space(leaf) >= csum_size) {
1212 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1213 		csum_offset = (bytenr - found_key.offset) >>
1214 			fs_info->sectorsize_bits;
1215 		goto extend_csum;
1216 	}
1217 
1218 	btrfs_release_path(path);
1219 	path->search_for_extension = 1;
1220 	ret = btrfs_search_slot(trans, root, &file_key, path,
1221 				csum_size, 1);
1222 	path->search_for_extension = 0;
1223 	if (ret < 0)
1224 		goto out;
1225 
1226 	if (ret > 0) {
1227 		if (path->slots[0] == 0)
1228 			goto insert;
1229 		path->slots[0]--;
1230 	}
1231 
1232 	leaf = path->nodes[0];
1233 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1234 	csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1235 
1236 	if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1237 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1238 	    csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1239 		goto insert;
1240 	}
1241 
1242 extend_csum:
1243 	if (csum_offset == btrfs_item_size(leaf, path->slots[0]) /
1244 	    csum_size) {
1245 		int extend_nr;
1246 		u64 tmp;
1247 		u32 diff;
1248 
1249 		tmp = sums->len - total_bytes;
1250 		tmp >>= fs_info->sectorsize_bits;
1251 		WARN_ON(tmp < 1);
1252 		extend_nr = max_t(int, 1, tmp);
1253 
1254 		/*
1255 		 * A log tree can already have checksum items with a subset of
1256 		 * the checksums we are trying to log. This can happen after
1257 		 * doing a sequence of partial writes into prealloc extents and
1258 		 * fsyncs in between, with a full fsync logging a larger subrange
1259 		 * of an extent for which a previous fast fsync logged a smaller
1260 		 * subrange. And this happens in particular due to merging file
1261 		 * extent items when we complete an ordered extent for a range
1262 		 * covered by a prealloc extent - this is done at
1263 		 * btrfs_mark_extent_written().
1264 		 *
1265 		 * So if we try to extend the previous checksum item, which has
1266 		 * a range that ends at the start of the range we want to insert,
1267 		 * make sure we don't extend beyond the start offset of the next
1268 		 * checksum item. If we are at the last item in the leaf, then
1269 		 * forget the optimization of extending and add a new checksum
1270 		 * item - it is not worth the complexity of releasing the path,
1271 		 * getting the first key for the next leaf, repeat the btree
1272 		 * search, etc, because log trees are temporary anyway and it
1273 		 * would only save a few bytes of leaf space.
1274 		 */
1275 		if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1276 			if (path->slots[0] + 1 >=
1277 			    btrfs_header_nritems(path->nodes[0])) {
1278 				ret = find_next_csum_offset(root, path, &next_offset);
1279 				if (ret < 0)
1280 					goto out;
1281 				found_next = 1;
1282 				goto insert;
1283 			}
1284 
1285 			ret = find_next_csum_offset(root, path, &next_offset);
1286 			if (ret < 0)
1287 				goto out;
1288 
1289 			tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1290 			if (tmp <= INT_MAX)
1291 				extend_nr = min_t(int, extend_nr, tmp);
1292 		}
1293 
1294 		diff = (csum_offset + extend_nr) * csum_size;
1295 		diff = min(diff,
1296 			   MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1297 
1298 		diff = diff - btrfs_item_size(leaf, path->slots[0]);
1299 		diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1300 		diff /= csum_size;
1301 		diff *= csum_size;
1302 
1303 		btrfs_extend_item(path, diff);
1304 		ret = 0;
1305 		goto csum;
1306 	}
1307 
1308 insert:
1309 	btrfs_release_path(path);
1310 	csum_offset = 0;
1311 	if (found_next) {
1312 		u64 tmp;
1313 
1314 		tmp = sums->len - total_bytes;
1315 		tmp >>= fs_info->sectorsize_bits;
1316 		tmp = min(tmp, (next_offset - file_key.offset) >>
1317 					 fs_info->sectorsize_bits);
1318 
1319 		tmp = max_t(u64, 1, tmp);
1320 		tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1321 		ins_size = csum_size * tmp;
1322 	} else {
1323 		ins_size = csum_size;
1324 	}
1325 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1326 				      ins_size);
1327 	if (ret < 0)
1328 		goto out;
1329 	if (WARN_ON(ret != 0))
1330 		goto out;
1331 	leaf = path->nodes[0];
1332 csum:
1333 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1334 	item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1335 				      btrfs_item_size(leaf, path->slots[0]));
1336 	item = (struct btrfs_csum_item *)((unsigned char *)item +
1337 					  csum_offset * csum_size);
1338 found:
1339 	ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1340 	ins_size *= csum_size;
1341 	ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1342 			      ins_size);
1343 	write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1344 			    ins_size);
1345 
1346 	index += ins_size;
1347 	ins_size /= csum_size;
1348 	total_bytes += ins_size * fs_info->sectorsize;
1349 
1350 	btrfs_mark_buffer_dirty(path->nodes[0]);
1351 	if (total_bytes < sums->len) {
1352 		btrfs_release_path(path);
1353 		cond_resched();
1354 		goto again;
1355 	}
1356 out:
1357 	btrfs_free_path(path);
1358 	return ret;
1359 }
1360 
1361 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1362 				     const struct btrfs_path *path,
1363 				     struct btrfs_file_extent_item *fi,
1364 				     struct extent_map *em)
1365 {
1366 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1367 	struct btrfs_root *root = inode->root;
1368 	struct extent_buffer *leaf = path->nodes[0];
1369 	const int slot = path->slots[0];
1370 	struct btrfs_key key;
1371 	u64 extent_start, extent_end;
1372 	u64 bytenr;
1373 	u8 type = btrfs_file_extent_type(leaf, fi);
1374 	int compress_type = btrfs_file_extent_compression(leaf, fi);
1375 
1376 	btrfs_item_key_to_cpu(leaf, &key, slot);
1377 	extent_start = key.offset;
1378 	extent_end = btrfs_file_extent_end(path);
1379 	em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1380 	em->generation = btrfs_file_extent_generation(leaf, fi);
1381 	if (type == BTRFS_FILE_EXTENT_REG ||
1382 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
1383 		em->start = extent_start;
1384 		em->len = extent_end - extent_start;
1385 		em->orig_start = extent_start -
1386 			btrfs_file_extent_offset(leaf, fi);
1387 		em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1388 		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1389 		if (bytenr == 0) {
1390 			em->block_start = EXTENT_MAP_HOLE;
1391 			return;
1392 		}
1393 		if (compress_type != BTRFS_COMPRESS_NONE) {
1394 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1395 			em->compress_type = compress_type;
1396 			em->block_start = bytenr;
1397 			em->block_len = em->orig_block_len;
1398 		} else {
1399 			bytenr += btrfs_file_extent_offset(leaf, fi);
1400 			em->block_start = bytenr;
1401 			em->block_len = em->len;
1402 			if (type == BTRFS_FILE_EXTENT_PREALLOC)
1403 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1404 		}
1405 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
1406 		em->block_start = EXTENT_MAP_INLINE;
1407 		em->start = extent_start;
1408 		em->len = extent_end - extent_start;
1409 		/*
1410 		 * Initialize orig_start and block_len with the same values
1411 		 * as in inode.c:btrfs_get_extent().
1412 		 */
1413 		em->orig_start = EXTENT_MAP_HOLE;
1414 		em->block_len = (u64)-1;
1415 		em->compress_type = compress_type;
1416 		if (compress_type != BTRFS_COMPRESS_NONE)
1417 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1418 	} else {
1419 		btrfs_err(fs_info,
1420 			  "unknown file extent item type %d, inode %llu, offset %llu, "
1421 			  "root %llu", type, btrfs_ino(inode), extent_start,
1422 			  root->root_key.objectid);
1423 	}
1424 }
1425 
1426 /*
1427  * Returns the end offset (non inclusive) of the file extent item the given path
1428  * points to. If it points to an inline extent, the returned offset is rounded
1429  * up to the sector size.
1430  */
1431 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1432 {
1433 	const struct extent_buffer *leaf = path->nodes[0];
1434 	const int slot = path->slots[0];
1435 	struct btrfs_file_extent_item *fi;
1436 	struct btrfs_key key;
1437 	u64 end;
1438 
1439 	btrfs_item_key_to_cpu(leaf, &key, slot);
1440 	ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1441 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1442 
1443 	if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1444 		end = btrfs_file_extent_ram_bytes(leaf, fi);
1445 		end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1446 	} else {
1447 		end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1448 	}
1449 
1450 	return end;
1451 }
1452