xref: /linux/fs/btrfs/file-item.c (revision 1e525507)
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 "ctree.h"
14 #include "disk-io.h"
15 #include "transaction.h"
16 #include "bio.h"
17 #include "compression.h"
18 #include "fs.h"
19 #include "accessors.h"
20 #include "file-item.h"
21 
22 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
23 				   sizeof(struct btrfs_item) * 2) / \
24 				  size) - 1))
25 
26 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
27 				       PAGE_SIZE))
28 
29 /*
30  * Set inode's size according to filesystem options.
31  *
32  * @inode:      inode we want to update the disk_i_size for
33  * @new_i_size: i_size we want to set to, 0 if we use i_size
34  *
35  * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read()
36  * returns as it is perfectly fine with a file that has holes without hole file
37  * extent items.
38  *
39  * However without NO_HOLES we need to only return the area that is contiguous
40  * from the 0 offset of the file.  Otherwise we could end up adjust i_size up
41  * to an extent that has a gap in between.
42  *
43  * Finally new_i_size should only be set in the case of truncate where we're not
44  * ready to use i_size_read() as the limiter yet.
45  */
46 void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size)
47 {
48 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
49 	u64 start, end, i_size;
50 	int ret;
51 
52 	spin_lock(&inode->lock);
53 	i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
54 	if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
55 		inode->disk_i_size = i_size;
56 		goto out_unlock;
57 	}
58 
59 	ret = find_contiguous_extent_bit(inode->file_extent_tree, 0, &start,
60 					 &end, EXTENT_DIRTY);
61 	if (!ret && start == 0)
62 		i_size = min(i_size, end + 1);
63 	else
64 		i_size = 0;
65 	inode->disk_i_size = i_size;
66 out_unlock:
67 	spin_unlock(&inode->lock);
68 }
69 
70 /*
71  * Mark range within a file as having a new extent inserted.
72  *
73  * @inode: inode being modified
74  * @start: start file offset of the file extent we've inserted
75  * @len:   logical length of the file extent item
76  *
77  * Call when we are inserting a new file extent where there was none before.
78  * Does not need to call this in the case where we're replacing an existing file
79  * extent, however if not sure it's fine to call this multiple times.
80  *
81  * The start and len must match the file extent item, so thus must be sectorsize
82  * aligned.
83  */
84 int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
85 				      u64 len)
86 {
87 	if (len == 0)
88 		return 0;
89 
90 	ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
91 
92 	if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
93 		return 0;
94 	return set_extent_bit(inode->file_extent_tree, start, start + len - 1,
95 			      EXTENT_DIRTY, NULL);
96 }
97 
98 /*
99  * Mark an inode range as not having a backing extent.
100  *
101  * @inode: inode being modified
102  * @start: start file offset of the file extent we've inserted
103  * @len:   logical length of the file extent item
104  *
105  * Called when we drop a file extent, for example when we truncate.  Doesn't
106  * need to be called for cases where we're replacing a file extent, like when
107  * we've COWed a file extent.
108  *
109  * The start and len must match the file extent item, so thus must be sectorsize
110  * aligned.
111  */
112 int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
113 					u64 len)
114 {
115 	if (len == 0)
116 		return 0;
117 
118 	ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
119 	       len == (u64)-1);
120 
121 	if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
122 		return 0;
123 	return clear_extent_bit(inode->file_extent_tree, start,
124 				start + len - 1, EXTENT_DIRTY, NULL);
125 }
126 
127 static size_t bytes_to_csum_size(const struct btrfs_fs_info *fs_info, u32 bytes)
128 {
129 	ASSERT(IS_ALIGNED(bytes, fs_info->sectorsize));
130 
131 	return (bytes >> fs_info->sectorsize_bits) * fs_info->csum_size;
132 }
133 
134 static size_t csum_size_to_bytes(const struct btrfs_fs_info *fs_info, u32 csum_size)
135 {
136 	ASSERT(IS_ALIGNED(csum_size, fs_info->csum_size));
137 
138 	return (csum_size / fs_info->csum_size) << fs_info->sectorsize_bits;
139 }
140 
141 static inline u32 max_ordered_sum_bytes(const struct btrfs_fs_info *fs_info)
142 {
143 	u32 max_csum_size = round_down(PAGE_SIZE - sizeof(struct btrfs_ordered_sum),
144 				       fs_info->csum_size);
145 
146 	return csum_size_to_bytes(fs_info, max_csum_size);
147 }
148 
149 /*
150  * Calculate the total size needed to allocate for an ordered sum structure
151  * spanning @bytes in the file.
152  */
153 static int btrfs_ordered_sum_size(struct btrfs_fs_info *fs_info, unsigned long bytes)
154 {
155 	return sizeof(struct btrfs_ordered_sum) + bytes_to_csum_size(fs_info, bytes);
156 }
157 
158 int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans,
159 			     struct btrfs_root *root,
160 			     u64 objectid, u64 pos, u64 num_bytes)
161 {
162 	int ret = 0;
163 	struct btrfs_file_extent_item *item;
164 	struct btrfs_key file_key;
165 	struct btrfs_path *path;
166 	struct extent_buffer *leaf;
167 
168 	path = btrfs_alloc_path();
169 	if (!path)
170 		return -ENOMEM;
171 	file_key.objectid = objectid;
172 	file_key.offset = pos;
173 	file_key.type = BTRFS_EXTENT_DATA_KEY;
174 
175 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
176 				      sizeof(*item));
177 	if (ret < 0)
178 		goto out;
179 	leaf = path->nodes[0];
180 	item = btrfs_item_ptr(leaf, path->slots[0],
181 			      struct btrfs_file_extent_item);
182 	btrfs_set_file_extent_disk_bytenr(leaf, item, 0);
183 	btrfs_set_file_extent_disk_num_bytes(leaf, item, 0);
184 	btrfs_set_file_extent_offset(leaf, item, 0);
185 	btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
186 	btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes);
187 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
188 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
189 	btrfs_set_file_extent_compression(leaf, item, 0);
190 	btrfs_set_file_extent_encryption(leaf, item, 0);
191 	btrfs_set_file_extent_other_encoding(leaf, item, 0);
192 
193 	btrfs_mark_buffer_dirty(trans, leaf);
194 out:
195 	btrfs_free_path(path);
196 	return ret;
197 }
198 
199 static struct btrfs_csum_item *
200 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
201 		  struct btrfs_root *root,
202 		  struct btrfs_path *path,
203 		  u64 bytenr, int cow)
204 {
205 	struct btrfs_fs_info *fs_info = root->fs_info;
206 	int ret;
207 	struct btrfs_key file_key;
208 	struct btrfs_key found_key;
209 	struct btrfs_csum_item *item;
210 	struct extent_buffer *leaf;
211 	u64 csum_offset = 0;
212 	const u32 csum_size = fs_info->csum_size;
213 	int csums_in_item;
214 
215 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
216 	file_key.offset = bytenr;
217 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
218 	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
219 	if (ret < 0)
220 		goto fail;
221 	leaf = path->nodes[0];
222 	if (ret > 0) {
223 		ret = 1;
224 		if (path->slots[0] == 0)
225 			goto fail;
226 		path->slots[0]--;
227 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
228 		if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
229 			goto fail;
230 
231 		csum_offset = (bytenr - found_key.offset) >>
232 				fs_info->sectorsize_bits;
233 		csums_in_item = btrfs_item_size(leaf, path->slots[0]);
234 		csums_in_item /= csum_size;
235 
236 		if (csum_offset == csums_in_item) {
237 			ret = -EFBIG;
238 			goto fail;
239 		} else if (csum_offset > csums_in_item) {
240 			goto fail;
241 		}
242 	}
243 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
244 	item = (struct btrfs_csum_item *)((unsigned char *)item +
245 					  csum_offset * csum_size);
246 	return item;
247 fail:
248 	if (ret > 0)
249 		ret = -ENOENT;
250 	return ERR_PTR(ret);
251 }
252 
253 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
254 			     struct btrfs_root *root,
255 			     struct btrfs_path *path, u64 objectid,
256 			     u64 offset, int mod)
257 {
258 	struct btrfs_key file_key;
259 	int ins_len = mod < 0 ? -1 : 0;
260 	int cow = mod != 0;
261 
262 	file_key.objectid = objectid;
263 	file_key.offset = offset;
264 	file_key.type = BTRFS_EXTENT_DATA_KEY;
265 
266 	return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
267 }
268 
269 /*
270  * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
271  * store the result to @dst.
272  *
273  * Return >0 for the number of sectors we found.
274  * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
275  * for it. Caller may want to try next sector until one range is hit.
276  * Return <0 for fatal error.
277  */
278 static int search_csum_tree(struct btrfs_fs_info *fs_info,
279 			    struct btrfs_path *path, u64 disk_bytenr,
280 			    u64 len, u8 *dst)
281 {
282 	struct btrfs_root *csum_root;
283 	struct btrfs_csum_item *item = NULL;
284 	struct btrfs_key key;
285 	const u32 sectorsize = fs_info->sectorsize;
286 	const u32 csum_size = fs_info->csum_size;
287 	u32 itemsize;
288 	int ret;
289 	u64 csum_start;
290 	u64 csum_len;
291 
292 	ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
293 	       IS_ALIGNED(len, sectorsize));
294 
295 	/* Check if the current csum item covers disk_bytenr */
296 	if (path->nodes[0]) {
297 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
298 				      struct btrfs_csum_item);
299 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
300 		itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
301 
302 		csum_start = key.offset;
303 		csum_len = (itemsize / csum_size) * sectorsize;
304 
305 		if (in_range(disk_bytenr, csum_start, csum_len))
306 			goto found;
307 	}
308 
309 	/* Current item doesn't contain the desired range, search again */
310 	btrfs_release_path(path);
311 	csum_root = btrfs_csum_root(fs_info, disk_bytenr);
312 	item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0);
313 	if (IS_ERR(item)) {
314 		ret = PTR_ERR(item);
315 		goto out;
316 	}
317 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
318 	itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
319 
320 	csum_start = key.offset;
321 	csum_len = (itemsize / csum_size) * sectorsize;
322 	ASSERT(in_range(disk_bytenr, csum_start, csum_len));
323 
324 found:
325 	ret = (min(csum_start + csum_len, disk_bytenr + len) -
326 		   disk_bytenr) >> fs_info->sectorsize_bits;
327 	read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
328 			ret * csum_size);
329 out:
330 	if (ret == -ENOENT || ret == -EFBIG)
331 		ret = 0;
332 	return ret;
333 }
334 
335 /*
336  * Lookup the checksum for the read bio in csum tree.
337  *
338  * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
339  */
340 blk_status_t btrfs_lookup_bio_sums(struct btrfs_bio *bbio)
341 {
342 	struct btrfs_inode *inode = bbio->inode;
343 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
344 	struct bio *bio = &bbio->bio;
345 	struct btrfs_path *path;
346 	const u32 sectorsize = fs_info->sectorsize;
347 	const u32 csum_size = fs_info->csum_size;
348 	u32 orig_len = bio->bi_iter.bi_size;
349 	u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
350 	const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
351 	blk_status_t ret = BLK_STS_OK;
352 	u32 bio_offset = 0;
353 
354 	if ((inode->flags & BTRFS_INODE_NODATASUM) ||
355 	    test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state))
356 		return BLK_STS_OK;
357 
358 	/*
359 	 * This function is only called for read bio.
360 	 *
361 	 * This means two things:
362 	 * - All our csums should only be in csum tree
363 	 *   No ordered extents csums, as ordered extents are only for write
364 	 *   path.
365 	 * - No need to bother any other info from bvec
366 	 *   Since we're looking up csums, the only important info is the
367 	 *   disk_bytenr and the length, which can be extracted from bi_iter
368 	 *   directly.
369 	 */
370 	ASSERT(bio_op(bio) == REQ_OP_READ);
371 	path = btrfs_alloc_path();
372 	if (!path)
373 		return BLK_STS_RESOURCE;
374 
375 	if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
376 		bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS);
377 		if (!bbio->csum) {
378 			btrfs_free_path(path);
379 			return BLK_STS_RESOURCE;
380 		}
381 	} else {
382 		bbio->csum = bbio->csum_inline;
383 	}
384 
385 	/*
386 	 * If requested number of sectors is larger than one leaf can contain,
387 	 * kick the readahead for csum tree.
388 	 */
389 	if (nblocks > fs_info->csums_per_leaf)
390 		path->reada = READA_FORWARD;
391 
392 	/*
393 	 * the free space stuff is only read when it hasn't been
394 	 * updated in the current transaction.  So, we can safely
395 	 * read from the commit root and sidestep a nasty deadlock
396 	 * between reading the free space cache and updating the csum tree.
397 	 */
398 	if (btrfs_is_free_space_inode(inode)) {
399 		path->search_commit_root = 1;
400 		path->skip_locking = 1;
401 	}
402 
403 	while (bio_offset < orig_len) {
404 		int count;
405 		u64 cur_disk_bytenr = orig_disk_bytenr + bio_offset;
406 		u8 *csum_dst = bbio->csum +
407 			(bio_offset >> fs_info->sectorsize_bits) * csum_size;
408 
409 		count = search_csum_tree(fs_info, path, cur_disk_bytenr,
410 					 orig_len - bio_offset, csum_dst);
411 		if (count < 0) {
412 			ret = errno_to_blk_status(count);
413 			if (bbio->csum != bbio->csum_inline)
414 				kfree(bbio->csum);
415 			bbio->csum = NULL;
416 			break;
417 		}
418 
419 		/*
420 		 * We didn't find a csum for this range.  We need to make sure
421 		 * we complain loudly about this, because we are not NODATASUM.
422 		 *
423 		 * However for the DATA_RELOC inode we could potentially be
424 		 * relocating data extents for a NODATASUM inode, so the inode
425 		 * itself won't be marked with NODATASUM, but the extent we're
426 		 * copying is in fact NODATASUM.  If we don't find a csum we
427 		 * assume this is the case.
428 		 */
429 		if (count == 0) {
430 			memset(csum_dst, 0, csum_size);
431 			count = 1;
432 
433 			if (inode->root->root_key.objectid ==
434 			    BTRFS_DATA_RELOC_TREE_OBJECTID) {
435 				u64 file_offset = bbio->file_offset + bio_offset;
436 
437 				set_extent_bit(&inode->io_tree, file_offset,
438 					       file_offset + sectorsize - 1,
439 					       EXTENT_NODATASUM, NULL);
440 			} else {
441 				btrfs_warn_rl(fs_info,
442 			"csum hole found for disk bytenr range [%llu, %llu)",
443 				cur_disk_bytenr, cur_disk_bytenr + sectorsize);
444 			}
445 		}
446 		bio_offset += count * sectorsize;
447 	}
448 
449 	btrfs_free_path(path);
450 	return ret;
451 }
452 
453 int btrfs_lookup_csums_list(struct btrfs_root *root, u64 start, u64 end,
454 			    struct list_head *list, int search_commit,
455 			    bool nowait)
456 {
457 	struct btrfs_fs_info *fs_info = root->fs_info;
458 	struct btrfs_key key;
459 	struct btrfs_path *path;
460 	struct extent_buffer *leaf;
461 	struct btrfs_ordered_sum *sums;
462 	struct btrfs_csum_item *item;
463 	LIST_HEAD(tmplist);
464 	int ret;
465 
466 	ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
467 	       IS_ALIGNED(end + 1, fs_info->sectorsize));
468 
469 	path = btrfs_alloc_path();
470 	if (!path)
471 		return -ENOMEM;
472 
473 	path->nowait = nowait;
474 	if (search_commit) {
475 		path->skip_locking = 1;
476 		path->reada = READA_FORWARD;
477 		path->search_commit_root = 1;
478 	}
479 
480 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
481 	key.offset = start;
482 	key.type = BTRFS_EXTENT_CSUM_KEY;
483 
484 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
485 	if (ret < 0)
486 		goto fail;
487 	if (ret > 0 && path->slots[0] > 0) {
488 		leaf = path->nodes[0];
489 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
490 
491 		/*
492 		 * There are two cases we can hit here for the previous csum
493 		 * item:
494 		 *
495 		 *		|<- search range ->|
496 		 *	|<- csum item ->|
497 		 *
498 		 * Or
499 		 *				|<- search range ->|
500 		 *	|<- csum item ->|
501 		 *
502 		 * Check if the previous csum item covers the leading part of
503 		 * the search range.  If so we have to start from previous csum
504 		 * item.
505 		 */
506 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
507 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
508 			if (bytes_to_csum_size(fs_info, start - key.offset) <
509 			    btrfs_item_size(leaf, path->slots[0] - 1))
510 				path->slots[0]--;
511 		}
512 	}
513 
514 	while (start <= end) {
515 		u64 csum_end;
516 
517 		leaf = path->nodes[0];
518 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
519 			ret = btrfs_next_leaf(root, path);
520 			if (ret < 0)
521 				goto fail;
522 			if (ret > 0)
523 				break;
524 			leaf = path->nodes[0];
525 		}
526 
527 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
528 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
529 		    key.type != BTRFS_EXTENT_CSUM_KEY ||
530 		    key.offset > end)
531 			break;
532 
533 		if (key.offset > start)
534 			start = key.offset;
535 
536 		csum_end = key.offset + csum_size_to_bytes(fs_info,
537 					btrfs_item_size(leaf, path->slots[0]));
538 		if (csum_end <= start) {
539 			path->slots[0]++;
540 			continue;
541 		}
542 
543 		csum_end = min(csum_end, end + 1);
544 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
545 				      struct btrfs_csum_item);
546 		while (start < csum_end) {
547 			unsigned long offset;
548 			size_t size;
549 
550 			size = min_t(size_t, csum_end - start,
551 				     max_ordered_sum_bytes(fs_info));
552 			sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
553 				       GFP_NOFS);
554 			if (!sums) {
555 				ret = -ENOMEM;
556 				goto fail;
557 			}
558 
559 			sums->logical = start;
560 			sums->len = size;
561 
562 			offset = bytes_to_csum_size(fs_info, start - key.offset);
563 
564 			read_extent_buffer(path->nodes[0],
565 					   sums->sums,
566 					   ((unsigned long)item) + offset,
567 					   bytes_to_csum_size(fs_info, size));
568 
569 			start += size;
570 			list_add_tail(&sums->list, &tmplist);
571 		}
572 		path->slots[0]++;
573 	}
574 	ret = 0;
575 fail:
576 	while (ret < 0 && !list_empty(&tmplist)) {
577 		sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
578 		list_del(&sums->list);
579 		kfree(sums);
580 	}
581 	list_splice_tail(&tmplist, list);
582 
583 	btrfs_free_path(path);
584 	return ret;
585 }
586 
587 /*
588  * Do the same work as btrfs_lookup_csums_list(), the difference is in how
589  * we return the result.
590  *
591  * This version will set the corresponding bits in @csum_bitmap to represent
592  * that there is a csum found.
593  * Each bit represents a sector. Thus caller should ensure @csum_buf passed
594  * in is large enough to contain all csums.
595  */
596 int btrfs_lookup_csums_bitmap(struct btrfs_root *root, struct btrfs_path *path,
597 			      u64 start, u64 end, u8 *csum_buf,
598 			      unsigned long *csum_bitmap)
599 {
600 	struct btrfs_fs_info *fs_info = root->fs_info;
601 	struct btrfs_key key;
602 	struct extent_buffer *leaf;
603 	struct btrfs_csum_item *item;
604 	const u64 orig_start = start;
605 	bool free_path = false;
606 	int ret;
607 
608 	ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
609 	       IS_ALIGNED(end + 1, fs_info->sectorsize));
610 
611 	if (!path) {
612 		path = btrfs_alloc_path();
613 		if (!path)
614 			return -ENOMEM;
615 		free_path = true;
616 	}
617 
618 	/* Check if we can reuse the previous path. */
619 	if (path->nodes[0]) {
620 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
621 
622 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
623 		    key.type == BTRFS_EXTENT_CSUM_KEY &&
624 		    key.offset <= start)
625 			goto search_forward;
626 		btrfs_release_path(path);
627 	}
628 
629 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
630 	key.type = BTRFS_EXTENT_CSUM_KEY;
631 	key.offset = start;
632 
633 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
634 	if (ret < 0)
635 		goto fail;
636 	if (ret > 0 && path->slots[0] > 0) {
637 		leaf = path->nodes[0];
638 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
639 
640 		/*
641 		 * There are two cases we can hit here for the previous csum
642 		 * item:
643 		 *
644 		 *		|<- search range ->|
645 		 *	|<- csum item ->|
646 		 *
647 		 * Or
648 		 *				|<- search range ->|
649 		 *	|<- csum item ->|
650 		 *
651 		 * Check if the previous csum item covers the leading part of
652 		 * the search range.  If so we have to start from previous csum
653 		 * item.
654 		 */
655 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
656 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
657 			if (bytes_to_csum_size(fs_info, start - key.offset) <
658 			    btrfs_item_size(leaf, path->slots[0] - 1))
659 				path->slots[0]--;
660 		}
661 	}
662 
663 search_forward:
664 	while (start <= end) {
665 		u64 csum_end;
666 
667 		leaf = path->nodes[0];
668 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
669 			ret = btrfs_next_leaf(root, path);
670 			if (ret < 0)
671 				goto fail;
672 			if (ret > 0)
673 				break;
674 			leaf = path->nodes[0];
675 		}
676 
677 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
678 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
679 		    key.type != BTRFS_EXTENT_CSUM_KEY ||
680 		    key.offset > end)
681 			break;
682 
683 		if (key.offset > start)
684 			start = key.offset;
685 
686 		csum_end = key.offset + csum_size_to_bytes(fs_info,
687 					btrfs_item_size(leaf, path->slots[0]));
688 		if (csum_end <= start) {
689 			path->slots[0]++;
690 			continue;
691 		}
692 
693 		csum_end = min(csum_end, end + 1);
694 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
695 				      struct btrfs_csum_item);
696 		while (start < csum_end) {
697 			unsigned long offset;
698 			size_t size;
699 			u8 *csum_dest = csum_buf + bytes_to_csum_size(fs_info,
700 						start - orig_start);
701 
702 			size = min_t(size_t, csum_end - start, end + 1 - start);
703 
704 			offset = bytes_to_csum_size(fs_info, start - key.offset);
705 
706 			read_extent_buffer(path->nodes[0], csum_dest,
707 					   ((unsigned long)item) + offset,
708 					   bytes_to_csum_size(fs_info, size));
709 
710 			bitmap_set(csum_bitmap,
711 				(start - orig_start) >> fs_info->sectorsize_bits,
712 				size >> fs_info->sectorsize_bits);
713 
714 			start += size;
715 		}
716 		path->slots[0]++;
717 	}
718 	ret = 0;
719 fail:
720 	if (free_path)
721 		btrfs_free_path(path);
722 	return ret;
723 }
724 
725 /*
726  * Calculate checksums of the data contained inside a bio.
727  */
728 blk_status_t btrfs_csum_one_bio(struct btrfs_bio *bbio)
729 {
730 	struct btrfs_ordered_extent *ordered = bbio->ordered;
731 	struct btrfs_inode *inode = bbio->inode;
732 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
733 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
734 	struct bio *bio = &bbio->bio;
735 	struct btrfs_ordered_sum *sums;
736 	char *data;
737 	struct bvec_iter iter;
738 	struct bio_vec bvec;
739 	int index;
740 	unsigned int blockcount;
741 	int i;
742 	unsigned nofs_flag;
743 
744 	nofs_flag = memalloc_nofs_save();
745 	sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
746 		       GFP_KERNEL);
747 	memalloc_nofs_restore(nofs_flag);
748 
749 	if (!sums)
750 		return BLK_STS_RESOURCE;
751 
752 	sums->len = bio->bi_iter.bi_size;
753 	INIT_LIST_HEAD(&sums->list);
754 
755 	sums->logical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
756 	index = 0;
757 
758 	shash->tfm = fs_info->csum_shash;
759 
760 	bio_for_each_segment(bvec, bio, iter) {
761 		blockcount = BTRFS_BYTES_TO_BLKS(fs_info,
762 						 bvec.bv_len + fs_info->sectorsize
763 						 - 1);
764 
765 		for (i = 0; i < blockcount; i++) {
766 			data = bvec_kmap_local(&bvec);
767 			crypto_shash_digest(shash,
768 					    data + (i * fs_info->sectorsize),
769 					    fs_info->sectorsize,
770 					    sums->sums + index);
771 			kunmap_local(data);
772 			index += fs_info->csum_size;
773 		}
774 
775 	}
776 
777 	bbio->sums = sums;
778 	btrfs_add_ordered_sum(ordered, sums);
779 	return 0;
780 }
781 
782 /*
783  * Nodatasum I/O on zoned file systems still requires an btrfs_ordered_sum to
784  * record the updated logical address on Zone Append completion.
785  * Allocate just the structure with an empty sums array here for that case.
786  */
787 blk_status_t btrfs_alloc_dummy_sum(struct btrfs_bio *bbio)
788 {
789 	bbio->sums = kmalloc(sizeof(*bbio->sums), GFP_NOFS);
790 	if (!bbio->sums)
791 		return BLK_STS_RESOURCE;
792 	bbio->sums->len = bbio->bio.bi_iter.bi_size;
793 	bbio->sums->logical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
794 	btrfs_add_ordered_sum(bbio->ordered, bbio->sums);
795 	return 0;
796 }
797 
798 /*
799  * Remove one checksum overlapping a range.
800  *
801  * This expects the key to describe the csum pointed to by the path, and it
802  * expects the csum to overlap the range [bytenr, len]
803  *
804  * The csum should not be entirely contained in the range and the range should
805  * not be entirely contained in the csum.
806  *
807  * This calls btrfs_truncate_item with the correct args based on the overlap,
808  * and fixes up the key as required.
809  */
810 static noinline void truncate_one_csum(struct btrfs_trans_handle *trans,
811 				       struct btrfs_path *path,
812 				       struct btrfs_key *key,
813 				       u64 bytenr, u64 len)
814 {
815 	struct btrfs_fs_info *fs_info = trans->fs_info;
816 	struct extent_buffer *leaf;
817 	const u32 csum_size = fs_info->csum_size;
818 	u64 csum_end;
819 	u64 end_byte = bytenr + len;
820 	u32 blocksize_bits = fs_info->sectorsize_bits;
821 
822 	leaf = path->nodes[0];
823 	csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
824 	csum_end <<= blocksize_bits;
825 	csum_end += key->offset;
826 
827 	if (key->offset < bytenr && csum_end <= end_byte) {
828 		/*
829 		 *         [ bytenr - len ]
830 		 *         [   ]
831 		 *   [csum     ]
832 		 *   A simple truncate off the end of the item
833 		 */
834 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
835 		new_size *= csum_size;
836 		btrfs_truncate_item(trans, path, new_size, 1);
837 	} else if (key->offset >= bytenr && csum_end > end_byte &&
838 		   end_byte > key->offset) {
839 		/*
840 		 *         [ bytenr - len ]
841 		 *                 [ ]
842 		 *                 [csum     ]
843 		 * we need to truncate from the beginning of the csum
844 		 */
845 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
846 		new_size *= csum_size;
847 
848 		btrfs_truncate_item(trans, path, new_size, 0);
849 
850 		key->offset = end_byte;
851 		btrfs_set_item_key_safe(trans, path, key);
852 	} else {
853 		BUG();
854 	}
855 }
856 
857 /*
858  * Delete the csum items from the csum tree for a given range of bytes.
859  */
860 int btrfs_del_csums(struct btrfs_trans_handle *trans,
861 		    struct btrfs_root *root, u64 bytenr, u64 len)
862 {
863 	struct btrfs_fs_info *fs_info = trans->fs_info;
864 	struct btrfs_path *path;
865 	struct btrfs_key key;
866 	u64 end_byte = bytenr + len;
867 	u64 csum_end;
868 	struct extent_buffer *leaf;
869 	int ret = 0;
870 	const u32 csum_size = fs_info->csum_size;
871 	u32 blocksize_bits = fs_info->sectorsize_bits;
872 
873 	ASSERT(root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID ||
874 	       root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
875 
876 	path = btrfs_alloc_path();
877 	if (!path)
878 		return -ENOMEM;
879 
880 	while (1) {
881 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
882 		key.offset = end_byte - 1;
883 		key.type = BTRFS_EXTENT_CSUM_KEY;
884 
885 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
886 		if (ret > 0) {
887 			ret = 0;
888 			if (path->slots[0] == 0)
889 				break;
890 			path->slots[0]--;
891 		} else if (ret < 0) {
892 			break;
893 		}
894 
895 		leaf = path->nodes[0];
896 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
897 
898 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
899 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
900 			break;
901 		}
902 
903 		if (key.offset >= end_byte)
904 			break;
905 
906 		csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
907 		csum_end <<= blocksize_bits;
908 		csum_end += key.offset;
909 
910 		/* this csum ends before we start, we're done */
911 		if (csum_end <= bytenr)
912 			break;
913 
914 		/* delete the entire item, it is inside our range */
915 		if (key.offset >= bytenr && csum_end <= end_byte) {
916 			int del_nr = 1;
917 
918 			/*
919 			 * Check how many csum items preceding this one in this
920 			 * leaf correspond to our range and then delete them all
921 			 * at once.
922 			 */
923 			if (key.offset > bytenr && path->slots[0] > 0) {
924 				int slot = path->slots[0] - 1;
925 
926 				while (slot >= 0) {
927 					struct btrfs_key pk;
928 
929 					btrfs_item_key_to_cpu(leaf, &pk, slot);
930 					if (pk.offset < bytenr ||
931 					    pk.type != BTRFS_EXTENT_CSUM_KEY ||
932 					    pk.objectid !=
933 					    BTRFS_EXTENT_CSUM_OBJECTID)
934 						break;
935 					path->slots[0] = slot;
936 					del_nr++;
937 					key.offset = pk.offset;
938 					slot--;
939 				}
940 			}
941 			ret = btrfs_del_items(trans, root, path,
942 					      path->slots[0], del_nr);
943 			if (ret)
944 				break;
945 			if (key.offset == bytenr)
946 				break;
947 		} else if (key.offset < bytenr && csum_end > end_byte) {
948 			unsigned long offset;
949 			unsigned long shift_len;
950 			unsigned long item_offset;
951 			/*
952 			 *        [ bytenr - len ]
953 			 *     [csum                ]
954 			 *
955 			 * Our bytes are in the middle of the csum,
956 			 * we need to split this item and insert a new one.
957 			 *
958 			 * But we can't drop the path because the
959 			 * csum could change, get removed, extended etc.
960 			 *
961 			 * The trick here is the max size of a csum item leaves
962 			 * enough room in the tree block for a single
963 			 * item header.  So, we split the item in place,
964 			 * adding a new header pointing to the existing
965 			 * bytes.  Then we loop around again and we have
966 			 * a nicely formed csum item that we can neatly
967 			 * truncate.
968 			 */
969 			offset = (bytenr - key.offset) >> blocksize_bits;
970 			offset *= csum_size;
971 
972 			shift_len = (len >> blocksize_bits) * csum_size;
973 
974 			item_offset = btrfs_item_ptr_offset(leaf,
975 							    path->slots[0]);
976 
977 			memzero_extent_buffer(leaf, item_offset + offset,
978 					     shift_len);
979 			key.offset = bytenr;
980 
981 			/*
982 			 * btrfs_split_item returns -EAGAIN when the
983 			 * item changed size or key
984 			 */
985 			ret = btrfs_split_item(trans, root, path, &key, offset);
986 			if (ret && ret != -EAGAIN) {
987 				btrfs_abort_transaction(trans, ret);
988 				break;
989 			}
990 			ret = 0;
991 
992 			key.offset = end_byte - 1;
993 		} else {
994 			truncate_one_csum(trans, path, &key, bytenr, len);
995 			if (key.offset < bytenr)
996 				break;
997 		}
998 		btrfs_release_path(path);
999 	}
1000 	btrfs_free_path(path);
1001 	return ret;
1002 }
1003 
1004 static int find_next_csum_offset(struct btrfs_root *root,
1005 				 struct btrfs_path *path,
1006 				 u64 *next_offset)
1007 {
1008 	const u32 nritems = btrfs_header_nritems(path->nodes[0]);
1009 	struct btrfs_key found_key;
1010 	int slot = path->slots[0] + 1;
1011 	int ret;
1012 
1013 	if (nritems == 0 || slot >= nritems) {
1014 		ret = btrfs_next_leaf(root, path);
1015 		if (ret < 0) {
1016 			return ret;
1017 		} else if (ret > 0) {
1018 			*next_offset = (u64)-1;
1019 			return 0;
1020 		}
1021 		slot = path->slots[0];
1022 	}
1023 
1024 	btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
1025 
1026 	if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1027 	    found_key.type != BTRFS_EXTENT_CSUM_KEY)
1028 		*next_offset = (u64)-1;
1029 	else
1030 		*next_offset = found_key.offset;
1031 
1032 	return 0;
1033 }
1034 
1035 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
1036 			   struct btrfs_root *root,
1037 			   struct btrfs_ordered_sum *sums)
1038 {
1039 	struct btrfs_fs_info *fs_info = root->fs_info;
1040 	struct btrfs_key file_key;
1041 	struct btrfs_key found_key;
1042 	struct btrfs_path *path;
1043 	struct btrfs_csum_item *item;
1044 	struct btrfs_csum_item *item_end;
1045 	struct extent_buffer *leaf = NULL;
1046 	u64 next_offset;
1047 	u64 total_bytes = 0;
1048 	u64 csum_offset;
1049 	u64 bytenr;
1050 	u32 ins_size;
1051 	int index = 0;
1052 	int found_next;
1053 	int ret;
1054 	const u32 csum_size = fs_info->csum_size;
1055 
1056 	path = btrfs_alloc_path();
1057 	if (!path)
1058 		return -ENOMEM;
1059 again:
1060 	next_offset = (u64)-1;
1061 	found_next = 0;
1062 	bytenr = sums->logical + total_bytes;
1063 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1064 	file_key.offset = bytenr;
1065 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
1066 
1067 	item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1068 	if (!IS_ERR(item)) {
1069 		ret = 0;
1070 		leaf = path->nodes[0];
1071 		item_end = btrfs_item_ptr(leaf, path->slots[0],
1072 					  struct btrfs_csum_item);
1073 		item_end = (struct btrfs_csum_item *)((char *)item_end +
1074 			   btrfs_item_size(leaf, path->slots[0]));
1075 		goto found;
1076 	}
1077 	ret = PTR_ERR(item);
1078 	if (ret != -EFBIG && ret != -ENOENT)
1079 		goto out;
1080 
1081 	if (ret == -EFBIG) {
1082 		u32 item_size;
1083 		/* we found one, but it isn't big enough yet */
1084 		leaf = path->nodes[0];
1085 		item_size = btrfs_item_size(leaf, path->slots[0]);
1086 		if ((item_size / csum_size) >=
1087 		    MAX_CSUM_ITEMS(fs_info, csum_size)) {
1088 			/* already at max size, make a new one */
1089 			goto insert;
1090 		}
1091 	} else {
1092 		/* We didn't find a csum item, insert one. */
1093 		ret = find_next_csum_offset(root, path, &next_offset);
1094 		if (ret < 0)
1095 			goto out;
1096 		found_next = 1;
1097 		goto insert;
1098 	}
1099 
1100 	/*
1101 	 * At this point, we know the tree has a checksum item that ends at an
1102 	 * offset matching the start of the checksum range we want to insert.
1103 	 * We try to extend that item as much as possible and then add as many
1104 	 * checksums to it as they fit.
1105 	 *
1106 	 * First check if the leaf has enough free space for at least one
1107 	 * checksum. If it has go directly to the item extension code, otherwise
1108 	 * release the path and do a search for insertion before the extension.
1109 	 */
1110 	if (btrfs_leaf_free_space(leaf) >= csum_size) {
1111 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1112 		csum_offset = (bytenr - found_key.offset) >>
1113 			fs_info->sectorsize_bits;
1114 		goto extend_csum;
1115 	}
1116 
1117 	btrfs_release_path(path);
1118 	path->search_for_extension = 1;
1119 	ret = btrfs_search_slot(trans, root, &file_key, path,
1120 				csum_size, 1);
1121 	path->search_for_extension = 0;
1122 	if (ret < 0)
1123 		goto out;
1124 
1125 	if (ret > 0) {
1126 		if (path->slots[0] == 0)
1127 			goto insert;
1128 		path->slots[0]--;
1129 	}
1130 
1131 	leaf = path->nodes[0];
1132 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1133 	csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1134 
1135 	if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1136 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1137 	    csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1138 		goto insert;
1139 	}
1140 
1141 extend_csum:
1142 	if (csum_offset == btrfs_item_size(leaf, path->slots[0]) /
1143 	    csum_size) {
1144 		int extend_nr;
1145 		u64 tmp;
1146 		u32 diff;
1147 
1148 		tmp = sums->len - total_bytes;
1149 		tmp >>= fs_info->sectorsize_bits;
1150 		WARN_ON(tmp < 1);
1151 		extend_nr = max_t(int, 1, tmp);
1152 
1153 		/*
1154 		 * A log tree can already have checksum items with a subset of
1155 		 * the checksums we are trying to log. This can happen after
1156 		 * doing a sequence of partial writes into prealloc extents and
1157 		 * fsyncs in between, with a full fsync logging a larger subrange
1158 		 * of an extent for which a previous fast fsync logged a smaller
1159 		 * subrange. And this happens in particular due to merging file
1160 		 * extent items when we complete an ordered extent for a range
1161 		 * covered by a prealloc extent - this is done at
1162 		 * btrfs_mark_extent_written().
1163 		 *
1164 		 * So if we try to extend the previous checksum item, which has
1165 		 * a range that ends at the start of the range we want to insert,
1166 		 * make sure we don't extend beyond the start offset of the next
1167 		 * checksum item. If we are at the last item in the leaf, then
1168 		 * forget the optimization of extending and add a new checksum
1169 		 * item - it is not worth the complexity of releasing the path,
1170 		 * getting the first key for the next leaf, repeat the btree
1171 		 * search, etc, because log trees are temporary anyway and it
1172 		 * would only save a few bytes of leaf space.
1173 		 */
1174 		if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1175 			if (path->slots[0] + 1 >=
1176 			    btrfs_header_nritems(path->nodes[0])) {
1177 				ret = find_next_csum_offset(root, path, &next_offset);
1178 				if (ret < 0)
1179 					goto out;
1180 				found_next = 1;
1181 				goto insert;
1182 			}
1183 
1184 			ret = find_next_csum_offset(root, path, &next_offset);
1185 			if (ret < 0)
1186 				goto out;
1187 
1188 			tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1189 			if (tmp <= INT_MAX)
1190 				extend_nr = min_t(int, extend_nr, tmp);
1191 		}
1192 
1193 		diff = (csum_offset + extend_nr) * csum_size;
1194 		diff = min(diff,
1195 			   MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1196 
1197 		diff = diff - btrfs_item_size(leaf, path->slots[0]);
1198 		diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1199 		diff /= csum_size;
1200 		diff *= csum_size;
1201 
1202 		btrfs_extend_item(trans, path, diff);
1203 		ret = 0;
1204 		goto csum;
1205 	}
1206 
1207 insert:
1208 	btrfs_release_path(path);
1209 	csum_offset = 0;
1210 	if (found_next) {
1211 		u64 tmp;
1212 
1213 		tmp = sums->len - total_bytes;
1214 		tmp >>= fs_info->sectorsize_bits;
1215 		tmp = min(tmp, (next_offset - file_key.offset) >>
1216 					 fs_info->sectorsize_bits);
1217 
1218 		tmp = max_t(u64, 1, tmp);
1219 		tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1220 		ins_size = csum_size * tmp;
1221 	} else {
1222 		ins_size = csum_size;
1223 	}
1224 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1225 				      ins_size);
1226 	if (ret < 0)
1227 		goto out;
1228 	leaf = path->nodes[0];
1229 csum:
1230 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1231 	item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1232 				      btrfs_item_size(leaf, path->slots[0]));
1233 	item = (struct btrfs_csum_item *)((unsigned char *)item +
1234 					  csum_offset * csum_size);
1235 found:
1236 	ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1237 	ins_size *= csum_size;
1238 	ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1239 			      ins_size);
1240 	write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1241 			    ins_size);
1242 
1243 	index += ins_size;
1244 	ins_size /= csum_size;
1245 	total_bytes += ins_size * fs_info->sectorsize;
1246 
1247 	btrfs_mark_buffer_dirty(trans, path->nodes[0]);
1248 	if (total_bytes < sums->len) {
1249 		btrfs_release_path(path);
1250 		cond_resched();
1251 		goto again;
1252 	}
1253 out:
1254 	btrfs_free_path(path);
1255 	return ret;
1256 }
1257 
1258 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1259 				     const struct btrfs_path *path,
1260 				     struct btrfs_file_extent_item *fi,
1261 				     struct extent_map *em)
1262 {
1263 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1264 	struct btrfs_root *root = inode->root;
1265 	struct extent_buffer *leaf = path->nodes[0];
1266 	const int slot = path->slots[0];
1267 	struct btrfs_key key;
1268 	u64 extent_start, extent_end;
1269 	u64 bytenr;
1270 	u8 type = btrfs_file_extent_type(leaf, fi);
1271 	int compress_type = btrfs_file_extent_compression(leaf, fi);
1272 
1273 	btrfs_item_key_to_cpu(leaf, &key, slot);
1274 	extent_start = key.offset;
1275 	extent_end = btrfs_file_extent_end(path);
1276 	em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1277 	em->generation = btrfs_file_extent_generation(leaf, fi);
1278 	if (type == BTRFS_FILE_EXTENT_REG ||
1279 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
1280 		em->start = extent_start;
1281 		em->len = extent_end - extent_start;
1282 		em->orig_start = extent_start -
1283 			btrfs_file_extent_offset(leaf, fi);
1284 		em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1285 		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1286 		if (bytenr == 0) {
1287 			em->block_start = EXTENT_MAP_HOLE;
1288 			return;
1289 		}
1290 		if (compress_type != BTRFS_COMPRESS_NONE) {
1291 			extent_map_set_compression(em, compress_type);
1292 			em->block_start = bytenr;
1293 			em->block_len = em->orig_block_len;
1294 		} else {
1295 			bytenr += btrfs_file_extent_offset(leaf, fi);
1296 			em->block_start = bytenr;
1297 			em->block_len = em->len;
1298 			if (type == BTRFS_FILE_EXTENT_PREALLOC)
1299 				em->flags |= EXTENT_FLAG_PREALLOC;
1300 		}
1301 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
1302 		em->block_start = EXTENT_MAP_INLINE;
1303 		em->start = extent_start;
1304 		em->len = extent_end - extent_start;
1305 		/*
1306 		 * Initialize orig_start and block_len with the same values
1307 		 * as in inode.c:btrfs_get_extent().
1308 		 */
1309 		em->orig_start = EXTENT_MAP_HOLE;
1310 		em->block_len = (u64)-1;
1311 		extent_map_set_compression(em, compress_type);
1312 	} else {
1313 		btrfs_err(fs_info,
1314 			  "unknown file extent item type %d, inode %llu, offset %llu, "
1315 			  "root %llu", type, btrfs_ino(inode), extent_start,
1316 			  root->root_key.objectid);
1317 	}
1318 }
1319 
1320 /*
1321  * Returns the end offset (non inclusive) of the file extent item the given path
1322  * points to. If it points to an inline extent, the returned offset is rounded
1323  * up to the sector size.
1324  */
1325 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1326 {
1327 	const struct extent_buffer *leaf = path->nodes[0];
1328 	const int slot = path->slots[0];
1329 	struct btrfs_file_extent_item *fi;
1330 	struct btrfs_key key;
1331 	u64 end;
1332 
1333 	btrfs_item_key_to_cpu(leaf, &key, slot);
1334 	ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1335 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1336 
1337 	if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1338 		end = btrfs_file_extent_ram_bytes(leaf, fi);
1339 		end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1340 	} else {
1341 		end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1342 	}
1343 
1344 	return end;
1345 }
1346