• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 "misc.h"
13 #include "ctree.h"
14 #include "disk-io.h"
15 #include "transaction.h"
16 #include "volumes.h"
17 #include "print-tree.h"
18 #include "compression.h"
19 
20 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
21 				   sizeof(struct btrfs_item) * 2) / \
22 				  size) - 1))
23 
24 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
25 				       PAGE_SIZE))
26 
27 /**
28  * Set inode's size according to filesystem options
29  *
30  * @inode:      inode we want to update the disk_i_size for
31  * @new_i_size: i_size we want to set to, 0 if we use i_size
32  *
33  * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read()
34  * returns as it is perfectly fine with a file that has holes without hole file
35  * extent items.
36  *
37  * However without NO_HOLES we need to only return the area that is contiguous
38  * from the 0 offset of the file.  Otherwise we could end up adjust i_size up
39  * to an extent that has a gap in between.
40  *
41  * Finally new_i_size should only be set in the case of truncate where we're not
42  * ready to use i_size_read() as the limiter yet.
43  */
btrfs_inode_safe_disk_i_size_write(struct btrfs_inode * inode,u64 new_i_size)44 void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size)
45 {
46 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
47 	u64 start, end, i_size;
48 	int ret;
49 
50 	spin_lock(&inode->lock);
51 	i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
52 	if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
53 		inode->disk_i_size = i_size;
54 		goto out_unlock;
55 	}
56 
57 	ret = find_contiguous_extent_bit(&inode->file_extent_tree, 0, &start,
58 					 &end, EXTENT_DIRTY);
59 	if (!ret && start == 0)
60 		i_size = min(i_size, end + 1);
61 	else
62 		i_size = 0;
63 	inode->disk_i_size = i_size;
64 out_unlock:
65 	spin_unlock(&inode->lock);
66 }
67 
68 /**
69  * Mark range within a file as having a new extent inserted
70  *
71  * @inode: inode being modified
72  * @start: start file offset of the file extent we've inserted
73  * @len:   logical length of the file extent item
74  *
75  * Call when we are inserting a new file extent where there was none before.
76  * Does not need to call this in the case where we're replacing an existing file
77  * extent, however if not sure it's fine to call this multiple times.
78  *
79  * The start and len must match the file extent item, so thus must be sectorsize
80  * aligned.
81  */
btrfs_inode_set_file_extent_range(struct btrfs_inode * inode,u64 start,u64 len)82 int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
83 				      u64 len)
84 {
85 	if (len == 0)
86 		return 0;
87 
88 	ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
89 
90 	if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
91 		return 0;
92 	return set_extent_bits(&inode->file_extent_tree, start, start + len - 1,
93 			       EXTENT_DIRTY);
94 }
95 
96 /**
97  * Marks an inode range as not having a backing extent
98  *
99  * @inode: inode being modified
100  * @start: start file offset of the file extent we've inserted
101  * @len:   logical length of the file extent item
102  *
103  * Called when we drop a file extent, for example when we truncate.  Doesn't
104  * need to be called for cases where we're replacing a file extent, like when
105  * we've COWed a file extent.
106  *
107  * The start and len must match the file extent item, so thus must be sectorsize
108  * aligned.
109  */
btrfs_inode_clear_file_extent_range(struct btrfs_inode * inode,u64 start,u64 len)110 int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
111 					u64 len)
112 {
113 	if (len == 0)
114 		return 0;
115 
116 	ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
117 	       len == (u64)-1);
118 
119 	if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
120 		return 0;
121 	return clear_extent_bit(&inode->file_extent_tree, start,
122 				start + len - 1, EXTENT_DIRTY, 0, 0, NULL);
123 }
124 
max_ordered_sum_bytes(struct btrfs_fs_info * fs_info,u16 csum_size)125 static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info,
126 					u16 csum_size)
127 {
128 	u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size;
129 
130 	return ncsums * fs_info->sectorsize;
131 }
132 
btrfs_insert_file_extent(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 objectid,u64 pos,u64 disk_offset,u64 disk_num_bytes,u64 num_bytes,u64 offset,u64 ram_bytes,u8 compression,u8 encryption,u16 other_encoding)133 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
134 			     struct btrfs_root *root,
135 			     u64 objectid, u64 pos,
136 			     u64 disk_offset, u64 disk_num_bytes,
137 			     u64 num_bytes, u64 offset, u64 ram_bytes,
138 			     u8 compression, u8 encryption, u16 other_encoding)
139 {
140 	int ret = 0;
141 	struct btrfs_file_extent_item *item;
142 	struct btrfs_key file_key;
143 	struct btrfs_path *path;
144 	struct extent_buffer *leaf;
145 
146 	path = btrfs_alloc_path();
147 	if (!path)
148 		return -ENOMEM;
149 	file_key.objectid = objectid;
150 	file_key.offset = pos;
151 	file_key.type = BTRFS_EXTENT_DATA_KEY;
152 
153 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
154 				      sizeof(*item));
155 	if (ret < 0)
156 		goto out;
157 	BUG_ON(ret); /* Can't happen */
158 	leaf = path->nodes[0];
159 	item = btrfs_item_ptr(leaf, path->slots[0],
160 			      struct btrfs_file_extent_item);
161 	btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
162 	btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
163 	btrfs_set_file_extent_offset(leaf, item, offset);
164 	btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
165 	btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
166 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
167 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
168 	btrfs_set_file_extent_compression(leaf, item, compression);
169 	btrfs_set_file_extent_encryption(leaf, item, encryption);
170 	btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
171 
172 	btrfs_mark_buffer_dirty(leaf);
173 out:
174 	btrfs_free_path(path);
175 	return ret;
176 }
177 
178 static struct btrfs_csum_item *
btrfs_lookup_csum(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,u64 bytenr,int cow)179 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
180 		  struct btrfs_root *root,
181 		  struct btrfs_path *path,
182 		  u64 bytenr, int cow)
183 {
184 	struct btrfs_fs_info *fs_info = root->fs_info;
185 	int ret;
186 	struct btrfs_key file_key;
187 	struct btrfs_key found_key;
188 	struct btrfs_csum_item *item;
189 	struct extent_buffer *leaf;
190 	u64 csum_offset = 0;
191 	const u32 csum_size = fs_info->csum_size;
192 	int csums_in_item;
193 
194 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
195 	file_key.offset = bytenr;
196 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
197 	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
198 	if (ret < 0)
199 		goto fail;
200 	leaf = path->nodes[0];
201 	if (ret > 0) {
202 		ret = 1;
203 		if (path->slots[0] == 0)
204 			goto fail;
205 		path->slots[0]--;
206 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
207 		if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
208 			goto fail;
209 
210 		csum_offset = (bytenr - found_key.offset) >>
211 				fs_info->sectorsize_bits;
212 		csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
213 		csums_in_item /= csum_size;
214 
215 		if (csum_offset == csums_in_item) {
216 			ret = -EFBIG;
217 			goto fail;
218 		} else if (csum_offset > csums_in_item) {
219 			goto fail;
220 		}
221 	}
222 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
223 	item = (struct btrfs_csum_item *)((unsigned char *)item +
224 					  csum_offset * csum_size);
225 	return item;
226 fail:
227 	if (ret > 0)
228 		ret = -ENOENT;
229 	return ERR_PTR(ret);
230 }
231 
btrfs_lookup_file_extent(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,u64 objectid,u64 offset,int mod)232 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
233 			     struct btrfs_root *root,
234 			     struct btrfs_path *path, u64 objectid,
235 			     u64 offset, int mod)
236 {
237 	struct btrfs_key file_key;
238 	int ins_len = mod < 0 ? -1 : 0;
239 	int cow = mod != 0;
240 
241 	file_key.objectid = objectid;
242 	file_key.offset = offset;
243 	file_key.type = BTRFS_EXTENT_DATA_KEY;
244 
245 	return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
246 }
247 
248 /*
249  * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
250  * estore the result to @dst.
251  *
252  * Return >0 for the number of sectors we found.
253  * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
254  * for it. Caller may want to try next sector until one range is hit.
255  * Return <0 for fatal error.
256  */
search_csum_tree(struct btrfs_fs_info * fs_info,struct btrfs_path * path,u64 disk_bytenr,u64 len,u8 * dst)257 static int search_csum_tree(struct btrfs_fs_info *fs_info,
258 			    struct btrfs_path *path, u64 disk_bytenr,
259 			    u64 len, u8 *dst)
260 {
261 	struct btrfs_csum_item *item = NULL;
262 	struct btrfs_key key;
263 	const u32 sectorsize = fs_info->sectorsize;
264 	const u32 csum_size = fs_info->csum_size;
265 	u32 itemsize;
266 	int ret;
267 	u64 csum_start;
268 	u64 csum_len;
269 
270 	ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
271 	       IS_ALIGNED(len, sectorsize));
272 
273 	/* Check if the current csum item covers disk_bytenr */
274 	if (path->nodes[0]) {
275 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
276 				      struct btrfs_csum_item);
277 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
278 		itemsize = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
279 
280 		csum_start = key.offset;
281 		csum_len = (itemsize / csum_size) * sectorsize;
282 
283 		if (in_range(disk_bytenr, csum_start, csum_len))
284 			goto found;
285 	}
286 
287 	/* Current item doesn't contain the desired range, search again */
288 	btrfs_release_path(path);
289 	item = btrfs_lookup_csum(NULL, fs_info->csum_root, path, disk_bytenr, 0);
290 	if (IS_ERR(item)) {
291 		ret = PTR_ERR(item);
292 		goto out;
293 	}
294 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
295 	itemsize = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
296 
297 	csum_start = key.offset;
298 	csum_len = (itemsize / csum_size) * sectorsize;
299 	ASSERT(in_range(disk_bytenr, csum_start, csum_len));
300 
301 found:
302 	ret = (min(csum_start + csum_len, disk_bytenr + len) -
303 		   disk_bytenr) >> fs_info->sectorsize_bits;
304 	read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
305 			ret * csum_size);
306 out:
307 	if (ret == -ENOENT || ret == -EFBIG)
308 		ret = 0;
309 	return ret;
310 }
311 
312 /*
313  * Locate the file_offset of @cur_disk_bytenr of a @bio.
314  *
315  * Bio of btrfs represents read range of
316  * [bi_sector << 9, bi_sector << 9 + bi_size).
317  * Knowing this, we can iterate through each bvec to locate the page belong to
318  * @cur_disk_bytenr and get the file offset.
319  *
320  * @inode is used to determine if the bvec page really belongs to @inode.
321  *
322  * Return 0 if we can't find the file offset
323  * Return >0 if we find the file offset and restore it to @file_offset_ret
324  */
search_file_offset_in_bio(struct bio * bio,struct inode * inode,u64 disk_bytenr,u64 * file_offset_ret)325 static int search_file_offset_in_bio(struct bio *bio, struct inode *inode,
326 				     u64 disk_bytenr, u64 *file_offset_ret)
327 {
328 	struct bvec_iter iter;
329 	struct bio_vec bvec;
330 	u64 cur = bio->bi_iter.bi_sector << SECTOR_SHIFT;
331 	int ret = 0;
332 
333 	bio_for_each_segment(bvec, bio, iter) {
334 		struct page *page = bvec.bv_page;
335 
336 		if (cur > disk_bytenr)
337 			break;
338 		if (cur + bvec.bv_len <= disk_bytenr) {
339 			cur += bvec.bv_len;
340 			continue;
341 		}
342 		ASSERT(in_range(disk_bytenr, cur, bvec.bv_len));
343 		if (page->mapping && page->mapping->host &&
344 		    page->mapping->host == inode) {
345 			ret = 1;
346 			*file_offset_ret = page_offset(page) + bvec.bv_offset +
347 					   disk_bytenr - cur;
348 			break;
349 		}
350 	}
351 	return ret;
352 }
353 
354 /**
355  * Lookup the checksum for the read bio in csum tree.
356  *
357  * @inode: inode that the bio is for.
358  * @bio: bio to look up.
359  * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return
360  *       checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If
361  *       NULL, the checksum buffer is allocated and returned in
362  *       btrfs_io_bio(bio)->csum instead.
363  *
364  * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
365  */
btrfs_lookup_bio_sums(struct inode * inode,struct bio * bio,u8 * dst)366 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst)
367 {
368 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
369 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
370 	struct btrfs_path *path;
371 	const u32 sectorsize = fs_info->sectorsize;
372 	const u32 csum_size = fs_info->csum_size;
373 	u32 orig_len = bio->bi_iter.bi_size;
374 	u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
375 	u64 cur_disk_bytenr;
376 	u8 *csum;
377 	const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
378 	int count = 0;
379 
380 	if (!fs_info->csum_root || (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
381 		return BLK_STS_OK;
382 
383 	/*
384 	 * This function is only called for read bio.
385 	 *
386 	 * This means two things:
387 	 * - All our csums should only be in csum tree
388 	 *   No ordered extents csums, as ordered extents are only for write
389 	 *   path.
390 	 * - No need to bother any other info from bvec
391 	 *   Since we're looking up csums, the only important info is the
392 	 *   disk_bytenr and the length, which can be extracted from bi_iter
393 	 *   directly.
394 	 */
395 	ASSERT(bio_op(bio) == REQ_OP_READ);
396 	path = btrfs_alloc_path();
397 	if (!path)
398 		return BLK_STS_RESOURCE;
399 
400 	if (!dst) {
401 		struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
402 
403 		if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
404 			btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
405 							GFP_NOFS);
406 			if (!btrfs_bio->csum) {
407 				btrfs_free_path(path);
408 				return BLK_STS_RESOURCE;
409 			}
410 		} else {
411 			btrfs_bio->csum = btrfs_bio->csum_inline;
412 		}
413 		csum = btrfs_bio->csum;
414 	} else {
415 		csum = dst;
416 	}
417 
418 	/*
419 	 * If requested number of sectors is larger than one leaf can contain,
420 	 * kick the readahead for csum tree.
421 	 */
422 	if (nblocks > fs_info->csums_per_leaf)
423 		path->reada = READA_FORWARD;
424 
425 	/*
426 	 * the free space stuff is only read when it hasn't been
427 	 * updated in the current transaction.  So, we can safely
428 	 * read from the commit root and sidestep a nasty deadlock
429 	 * between reading the free space cache and updating the csum tree.
430 	 */
431 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
432 		path->search_commit_root = 1;
433 		path->skip_locking = 1;
434 	}
435 
436 	for (cur_disk_bytenr = orig_disk_bytenr;
437 	     cur_disk_bytenr < orig_disk_bytenr + orig_len;
438 	     cur_disk_bytenr += (count * sectorsize)) {
439 		u64 search_len = orig_disk_bytenr + orig_len - cur_disk_bytenr;
440 		unsigned int sector_offset;
441 		u8 *csum_dst;
442 
443 		/*
444 		 * Although both cur_disk_bytenr and orig_disk_bytenr is u64,
445 		 * we're calculating the offset to the bio start.
446 		 *
447 		 * Bio size is limited to UINT_MAX, thus unsigned int is large
448 		 * enough to contain the raw result, not to mention the right
449 		 * shifted result.
450 		 */
451 		ASSERT(cur_disk_bytenr - orig_disk_bytenr < UINT_MAX);
452 		sector_offset = (cur_disk_bytenr - orig_disk_bytenr) >>
453 				fs_info->sectorsize_bits;
454 		csum_dst = csum + sector_offset * csum_size;
455 
456 		count = search_csum_tree(fs_info, path, cur_disk_bytenr,
457 					 search_len, csum_dst);
458 		if (count <= 0) {
459 			/*
460 			 * Either we hit a critical error or we didn't find
461 			 * the csum.
462 			 * Either way, we put zero into the csums dst, and skip
463 			 * to the next sector.
464 			 */
465 			memset(csum_dst, 0, csum_size);
466 			count = 1;
467 
468 			/*
469 			 * For data reloc inode, we need to mark the range
470 			 * NODATASUM so that balance won't report false csum
471 			 * error.
472 			 */
473 			if (BTRFS_I(inode)->root->root_key.objectid ==
474 			    BTRFS_DATA_RELOC_TREE_OBJECTID) {
475 				u64 file_offset;
476 				int ret;
477 
478 				ret = search_file_offset_in_bio(bio, inode,
479 						cur_disk_bytenr, &file_offset);
480 				if (ret)
481 					set_extent_bits(io_tree, file_offset,
482 						file_offset + sectorsize - 1,
483 						EXTENT_NODATASUM);
484 			} else {
485 				btrfs_warn_rl(fs_info,
486 			"csum hole found for disk bytenr range [%llu, %llu)",
487 				cur_disk_bytenr, cur_disk_bytenr + sectorsize);
488 			}
489 		}
490 	}
491 
492 	btrfs_free_path(path);
493 	return BLK_STS_OK;
494 }
495 
btrfs_lookup_csums_range(struct btrfs_root * root,u64 start,u64 end,struct list_head * list,int search_commit)496 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
497 			     struct list_head *list, int search_commit)
498 {
499 	struct btrfs_fs_info *fs_info = root->fs_info;
500 	struct btrfs_key key;
501 	struct btrfs_path *path;
502 	struct extent_buffer *leaf;
503 	struct btrfs_ordered_sum *sums;
504 	struct btrfs_csum_item *item;
505 	LIST_HEAD(tmplist);
506 	unsigned long offset;
507 	int ret;
508 	size_t size;
509 	u64 csum_end;
510 	const u32 csum_size = fs_info->csum_size;
511 
512 	ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
513 	       IS_ALIGNED(end + 1, fs_info->sectorsize));
514 
515 	path = btrfs_alloc_path();
516 	if (!path)
517 		return -ENOMEM;
518 
519 	if (search_commit) {
520 		path->skip_locking = 1;
521 		path->reada = READA_FORWARD;
522 		path->search_commit_root = 1;
523 	}
524 
525 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
526 	key.offset = start;
527 	key.type = BTRFS_EXTENT_CSUM_KEY;
528 
529 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
530 	if (ret < 0)
531 		goto fail;
532 	if (ret > 0 && path->slots[0] > 0) {
533 		leaf = path->nodes[0];
534 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
535 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
536 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
537 			offset = (start - key.offset) >> fs_info->sectorsize_bits;
538 			if (offset * csum_size <
539 			    btrfs_item_size_nr(leaf, path->slots[0] - 1))
540 				path->slots[0]--;
541 		}
542 	}
543 
544 	while (start <= end) {
545 		leaf = path->nodes[0];
546 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
547 			ret = btrfs_next_leaf(root, path);
548 			if (ret < 0)
549 				goto fail;
550 			if (ret > 0)
551 				break;
552 			leaf = path->nodes[0];
553 		}
554 
555 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
556 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
557 		    key.type != BTRFS_EXTENT_CSUM_KEY ||
558 		    key.offset > end)
559 			break;
560 
561 		if (key.offset > start)
562 			start = key.offset;
563 
564 		size = btrfs_item_size_nr(leaf, path->slots[0]);
565 		csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
566 		if (csum_end <= start) {
567 			path->slots[0]++;
568 			continue;
569 		}
570 
571 		csum_end = min(csum_end, end + 1);
572 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
573 				      struct btrfs_csum_item);
574 		while (start < csum_end) {
575 			size = min_t(size_t, csum_end - start,
576 				     max_ordered_sum_bytes(fs_info, csum_size));
577 			sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
578 				       GFP_NOFS);
579 			if (!sums) {
580 				ret = -ENOMEM;
581 				goto fail;
582 			}
583 
584 			sums->bytenr = start;
585 			sums->len = (int)size;
586 
587 			offset = (start - key.offset) >> fs_info->sectorsize_bits;
588 			offset *= csum_size;
589 			size >>= fs_info->sectorsize_bits;
590 
591 			read_extent_buffer(path->nodes[0],
592 					   sums->sums,
593 					   ((unsigned long)item) + offset,
594 					   csum_size * size);
595 
596 			start += fs_info->sectorsize * size;
597 			list_add_tail(&sums->list, &tmplist);
598 		}
599 		path->slots[0]++;
600 	}
601 	ret = 0;
602 fail:
603 	while (ret < 0 && !list_empty(&tmplist)) {
604 		sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
605 		list_del(&sums->list);
606 		kfree(sums);
607 	}
608 	list_splice_tail(&tmplist, list);
609 
610 	btrfs_free_path(path);
611 	return ret;
612 }
613 
614 /*
615  * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
616  * @inode:	 Owner of the data inside the bio
617  * @bio:	 Contains the data to be checksummed
618  * @file_start:  offset in file this bio begins to describe
619  * @contig:	 Boolean. If true/1 means all bio vecs in this bio are
620  *		 contiguous and they begin at @file_start in the file. False/0
621  *		 means this bio can contain potentially discontiguous bio vecs
622  *		 so the logical offset of each should be calculated separately.
623  */
btrfs_csum_one_bio(struct btrfs_inode * inode,struct bio * bio,u64 file_start,int contig)624 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
625 		       u64 file_start, int contig)
626 {
627 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
628 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
629 	struct btrfs_ordered_sum *sums;
630 	struct btrfs_ordered_extent *ordered = NULL;
631 	char *data;
632 	struct bvec_iter iter;
633 	struct bio_vec bvec;
634 	int index;
635 	int nr_sectors;
636 	unsigned long total_bytes = 0;
637 	unsigned long this_sum_bytes = 0;
638 	int i;
639 	u64 offset;
640 	unsigned nofs_flag;
641 
642 	nofs_flag = memalloc_nofs_save();
643 	sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
644 		       GFP_KERNEL);
645 	memalloc_nofs_restore(nofs_flag);
646 
647 	if (!sums)
648 		return BLK_STS_RESOURCE;
649 
650 	sums->len = bio->bi_iter.bi_size;
651 	INIT_LIST_HEAD(&sums->list);
652 
653 	if (contig)
654 		offset = file_start;
655 	else
656 		offset = 0; /* shut up gcc */
657 
658 	sums->bytenr = bio->bi_iter.bi_sector << 9;
659 	index = 0;
660 
661 	shash->tfm = fs_info->csum_shash;
662 
663 	bio_for_each_segment(bvec, bio, iter) {
664 		if (!contig)
665 			offset = page_offset(bvec.bv_page) + bvec.bv_offset;
666 
667 		if (!ordered) {
668 			ordered = btrfs_lookup_ordered_extent(inode, offset);
669 			/*
670 			 * The bio range is not covered by any ordered extent,
671 			 * must be a code logic error.
672 			 */
673 			if (unlikely(!ordered)) {
674 				WARN(1, KERN_WARNING
675 			"no ordered extent for root %llu ino %llu offset %llu\n",
676 				     inode->root->root_key.objectid,
677 				     btrfs_ino(inode), offset);
678 				kvfree(sums);
679 				return BLK_STS_IOERR;
680 			}
681 		}
682 
683 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
684 						 bvec.bv_len + fs_info->sectorsize
685 						 - 1);
686 
687 		for (i = 0; i < nr_sectors; i++) {
688 			if (offset >= ordered->file_offset + ordered->num_bytes ||
689 			    offset < ordered->file_offset) {
690 				unsigned long bytes_left;
691 
692 				sums->len = this_sum_bytes;
693 				this_sum_bytes = 0;
694 				btrfs_add_ordered_sum(ordered, sums);
695 				btrfs_put_ordered_extent(ordered);
696 
697 				bytes_left = bio->bi_iter.bi_size - total_bytes;
698 
699 				nofs_flag = memalloc_nofs_save();
700 				sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
701 						      bytes_left), GFP_KERNEL);
702 				memalloc_nofs_restore(nofs_flag);
703 				if (!sums)
704 					return BLK_STS_RESOURCE;
705 
706 				sums->len = bytes_left;
707 				ordered = btrfs_lookup_ordered_extent(inode,
708 								offset);
709 				ASSERT(ordered); /* Logic error */
710 				sums->bytenr = (bio->bi_iter.bi_sector << 9)
711 					+ total_bytes;
712 				index = 0;
713 			}
714 
715 			data = kmap_atomic(bvec.bv_page);
716 			crypto_shash_digest(shash, data + bvec.bv_offset
717 					    + (i * fs_info->sectorsize),
718 					    fs_info->sectorsize,
719 					    sums->sums + index);
720 			kunmap_atomic(data);
721 			index += fs_info->csum_size;
722 			offset += fs_info->sectorsize;
723 			this_sum_bytes += fs_info->sectorsize;
724 			total_bytes += fs_info->sectorsize;
725 		}
726 
727 	}
728 	this_sum_bytes = 0;
729 	btrfs_add_ordered_sum(ordered, sums);
730 	btrfs_put_ordered_extent(ordered);
731 	return 0;
732 }
733 
734 /*
735  * helper function for csum removal, this expects the
736  * key to describe the csum pointed to by the path, and it expects
737  * the csum to overlap the range [bytenr, len]
738  *
739  * The csum should not be entirely contained in the range and the
740  * range should not be entirely contained in the csum.
741  *
742  * This calls btrfs_truncate_item with the correct args based on the
743  * overlap, and fixes up the key as required.
744  */
truncate_one_csum(struct btrfs_fs_info * fs_info,struct btrfs_path * path,struct btrfs_key * key,u64 bytenr,u64 len)745 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
746 				       struct btrfs_path *path,
747 				       struct btrfs_key *key,
748 				       u64 bytenr, u64 len)
749 {
750 	struct extent_buffer *leaf;
751 	const u32 csum_size = fs_info->csum_size;
752 	u64 csum_end;
753 	u64 end_byte = bytenr + len;
754 	u32 blocksize_bits = fs_info->sectorsize_bits;
755 
756 	leaf = path->nodes[0];
757 	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
758 	csum_end <<= blocksize_bits;
759 	csum_end += key->offset;
760 
761 	if (key->offset < bytenr && csum_end <= end_byte) {
762 		/*
763 		 *         [ bytenr - len ]
764 		 *         [   ]
765 		 *   [csum     ]
766 		 *   A simple truncate off the end of the item
767 		 */
768 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
769 		new_size *= csum_size;
770 		btrfs_truncate_item(path, new_size, 1);
771 	} else if (key->offset >= bytenr && csum_end > end_byte &&
772 		   end_byte > key->offset) {
773 		/*
774 		 *         [ bytenr - len ]
775 		 *                 [ ]
776 		 *                 [csum     ]
777 		 * we need to truncate from the beginning of the csum
778 		 */
779 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
780 		new_size *= csum_size;
781 
782 		btrfs_truncate_item(path, new_size, 0);
783 
784 		key->offset = end_byte;
785 		btrfs_set_item_key_safe(fs_info, path, key);
786 	} else {
787 		BUG();
788 	}
789 }
790 
791 /*
792  * deletes the csum items from the csum tree for a given
793  * range of bytes.
794  */
btrfs_del_csums(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 bytenr,u64 len)795 int btrfs_del_csums(struct btrfs_trans_handle *trans,
796 		    struct btrfs_root *root, u64 bytenr, u64 len)
797 {
798 	struct btrfs_fs_info *fs_info = trans->fs_info;
799 	struct btrfs_path *path;
800 	struct btrfs_key key;
801 	u64 end_byte = bytenr + len;
802 	u64 csum_end;
803 	struct extent_buffer *leaf;
804 	int ret = 0;
805 	const u32 csum_size = fs_info->csum_size;
806 	u32 blocksize_bits = fs_info->sectorsize_bits;
807 
808 	ASSERT(root == fs_info->csum_root ||
809 	       root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
810 
811 	path = btrfs_alloc_path();
812 	if (!path)
813 		return -ENOMEM;
814 
815 	while (1) {
816 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
817 		key.offset = end_byte - 1;
818 		key.type = BTRFS_EXTENT_CSUM_KEY;
819 
820 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
821 		if (ret > 0) {
822 			ret = 0;
823 			if (path->slots[0] == 0)
824 				break;
825 			path->slots[0]--;
826 		} else if (ret < 0) {
827 			break;
828 		}
829 
830 		leaf = path->nodes[0];
831 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
832 
833 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
834 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
835 			break;
836 		}
837 
838 		if (key.offset >= end_byte)
839 			break;
840 
841 		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
842 		csum_end <<= blocksize_bits;
843 		csum_end += key.offset;
844 
845 		/* this csum ends before we start, we're done */
846 		if (csum_end <= bytenr)
847 			break;
848 
849 		/* delete the entire item, it is inside our range */
850 		if (key.offset >= bytenr && csum_end <= end_byte) {
851 			int del_nr = 1;
852 
853 			/*
854 			 * Check how many csum items preceding this one in this
855 			 * leaf correspond to our range and then delete them all
856 			 * at once.
857 			 */
858 			if (key.offset > bytenr && path->slots[0] > 0) {
859 				int slot = path->slots[0] - 1;
860 
861 				while (slot >= 0) {
862 					struct btrfs_key pk;
863 
864 					btrfs_item_key_to_cpu(leaf, &pk, slot);
865 					if (pk.offset < bytenr ||
866 					    pk.type != BTRFS_EXTENT_CSUM_KEY ||
867 					    pk.objectid !=
868 					    BTRFS_EXTENT_CSUM_OBJECTID)
869 						break;
870 					path->slots[0] = slot;
871 					del_nr++;
872 					key.offset = pk.offset;
873 					slot--;
874 				}
875 			}
876 			ret = btrfs_del_items(trans, root, path,
877 					      path->slots[0], del_nr);
878 			if (ret)
879 				break;
880 			if (key.offset == bytenr)
881 				break;
882 		} else if (key.offset < bytenr && csum_end > end_byte) {
883 			unsigned long offset;
884 			unsigned long shift_len;
885 			unsigned long item_offset;
886 			/*
887 			 *        [ bytenr - len ]
888 			 *     [csum                ]
889 			 *
890 			 * Our bytes are in the middle of the csum,
891 			 * we need to split this item and insert a new one.
892 			 *
893 			 * But we can't drop the path because the
894 			 * csum could change, get removed, extended etc.
895 			 *
896 			 * The trick here is the max size of a csum item leaves
897 			 * enough room in the tree block for a single
898 			 * item header.  So, we split the item in place,
899 			 * adding a new header pointing to the existing
900 			 * bytes.  Then we loop around again and we have
901 			 * a nicely formed csum item that we can neatly
902 			 * truncate.
903 			 */
904 			offset = (bytenr - key.offset) >> blocksize_bits;
905 			offset *= csum_size;
906 
907 			shift_len = (len >> blocksize_bits) * csum_size;
908 
909 			item_offset = btrfs_item_ptr_offset(leaf,
910 							    path->slots[0]);
911 
912 			memzero_extent_buffer(leaf, item_offset + offset,
913 					     shift_len);
914 			key.offset = bytenr;
915 
916 			/*
917 			 * btrfs_split_item returns -EAGAIN when the
918 			 * item changed size or key
919 			 */
920 			ret = btrfs_split_item(trans, root, path, &key, offset);
921 			if (ret && ret != -EAGAIN) {
922 				btrfs_abort_transaction(trans, ret);
923 				break;
924 			}
925 			ret = 0;
926 
927 			key.offset = end_byte - 1;
928 		} else {
929 			truncate_one_csum(fs_info, path, &key, bytenr, len);
930 			if (key.offset < bytenr)
931 				break;
932 		}
933 		btrfs_release_path(path);
934 	}
935 	btrfs_free_path(path);
936 	return ret;
937 }
938 
find_next_csum_offset(struct btrfs_root * root,struct btrfs_path * path,u64 * next_offset)939 static int find_next_csum_offset(struct btrfs_root *root,
940 				 struct btrfs_path *path,
941 				 u64 *next_offset)
942 {
943 	const u32 nritems = btrfs_header_nritems(path->nodes[0]);
944 	struct btrfs_key found_key;
945 	int slot = path->slots[0] + 1;
946 	int ret;
947 
948 	if (nritems == 0 || slot >= nritems) {
949 		ret = btrfs_next_leaf(root, path);
950 		if (ret < 0) {
951 			return ret;
952 		} else if (ret > 0) {
953 			*next_offset = (u64)-1;
954 			return 0;
955 		}
956 		slot = path->slots[0];
957 	}
958 
959 	btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
960 
961 	if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
962 	    found_key.type != BTRFS_EXTENT_CSUM_KEY)
963 		*next_offset = (u64)-1;
964 	else
965 		*next_offset = found_key.offset;
966 
967 	return 0;
968 }
969 
btrfs_csum_file_blocks(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_ordered_sum * sums)970 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
971 			   struct btrfs_root *root,
972 			   struct btrfs_ordered_sum *sums)
973 {
974 	struct btrfs_fs_info *fs_info = root->fs_info;
975 	struct btrfs_key file_key;
976 	struct btrfs_key found_key;
977 	struct btrfs_path *path;
978 	struct btrfs_csum_item *item;
979 	struct btrfs_csum_item *item_end;
980 	struct extent_buffer *leaf = NULL;
981 	u64 next_offset;
982 	u64 total_bytes = 0;
983 	u64 csum_offset;
984 	u64 bytenr;
985 	u32 ins_size;
986 	int index = 0;
987 	int found_next;
988 	int ret;
989 	const u32 csum_size = fs_info->csum_size;
990 
991 	path = btrfs_alloc_path();
992 	if (!path)
993 		return -ENOMEM;
994 again:
995 	next_offset = (u64)-1;
996 	found_next = 0;
997 	bytenr = sums->bytenr + total_bytes;
998 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
999 	file_key.offset = bytenr;
1000 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
1001 
1002 	item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1003 	if (!IS_ERR(item)) {
1004 		ret = 0;
1005 		leaf = path->nodes[0];
1006 		item_end = btrfs_item_ptr(leaf, path->slots[0],
1007 					  struct btrfs_csum_item);
1008 		item_end = (struct btrfs_csum_item *)((char *)item_end +
1009 			   btrfs_item_size_nr(leaf, path->slots[0]));
1010 		goto found;
1011 	}
1012 	ret = PTR_ERR(item);
1013 	if (ret != -EFBIG && ret != -ENOENT)
1014 		goto out;
1015 
1016 	if (ret == -EFBIG) {
1017 		u32 item_size;
1018 		/* we found one, but it isn't big enough yet */
1019 		leaf = path->nodes[0];
1020 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1021 		if ((item_size / csum_size) >=
1022 		    MAX_CSUM_ITEMS(fs_info, csum_size)) {
1023 			/* already at max size, make a new one */
1024 			goto insert;
1025 		}
1026 	} else {
1027 		/* We didn't find a csum item, insert one. */
1028 		ret = find_next_csum_offset(root, path, &next_offset);
1029 		if (ret < 0)
1030 			goto out;
1031 		found_next = 1;
1032 		goto insert;
1033 	}
1034 
1035 	/*
1036 	 * At this point, we know the tree has a checksum item that ends at an
1037 	 * offset matching the start of the checksum range we want to insert.
1038 	 * We try to extend that item as much as possible and then add as many
1039 	 * checksums to it as they fit.
1040 	 *
1041 	 * First check if the leaf has enough free space for at least one
1042 	 * checksum. If it has go directly to the item extension code, otherwise
1043 	 * release the path and do a search for insertion before the extension.
1044 	 */
1045 	if (btrfs_leaf_free_space(leaf) >= csum_size) {
1046 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1047 		csum_offset = (bytenr - found_key.offset) >>
1048 			fs_info->sectorsize_bits;
1049 		goto extend_csum;
1050 	}
1051 
1052 	btrfs_release_path(path);
1053 	path->search_for_extension = 1;
1054 	ret = btrfs_search_slot(trans, root, &file_key, path,
1055 				csum_size, 1);
1056 	path->search_for_extension = 0;
1057 	if (ret < 0)
1058 		goto out;
1059 
1060 	if (ret > 0) {
1061 		if (path->slots[0] == 0)
1062 			goto insert;
1063 		path->slots[0]--;
1064 	}
1065 
1066 	leaf = path->nodes[0];
1067 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1068 	csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1069 
1070 	if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1071 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1072 	    csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1073 		goto insert;
1074 	}
1075 
1076 extend_csum:
1077 	if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
1078 	    csum_size) {
1079 		int extend_nr;
1080 		u64 tmp;
1081 		u32 diff;
1082 
1083 		tmp = sums->len - total_bytes;
1084 		tmp >>= fs_info->sectorsize_bits;
1085 		WARN_ON(tmp < 1);
1086 		extend_nr = max_t(int, 1, tmp);
1087 
1088 		/*
1089 		 * A log tree can already have checksum items with a subset of
1090 		 * the checksums we are trying to log. This can happen after
1091 		 * doing a sequence of partial writes into prealloc extents and
1092 		 * fsyncs in between, with a full fsync logging a larger subrange
1093 		 * of an extent for which a previous fast fsync logged a smaller
1094 		 * subrange. And this happens in particular due to merging file
1095 		 * extent items when we complete an ordered extent for a range
1096 		 * covered by a prealloc extent - this is done at
1097 		 * btrfs_mark_extent_written().
1098 		 *
1099 		 * So if we try to extend the previous checksum item, which has
1100 		 * a range that ends at the start of the range we want to insert,
1101 		 * make sure we don't extend beyond the start offset of the next
1102 		 * checksum item. If we are at the last item in the leaf, then
1103 		 * forget the optimization of extending and add a new checksum
1104 		 * item - it is not worth the complexity of releasing the path,
1105 		 * getting the first key for the next leaf, repeat the btree
1106 		 * search, etc, because log trees are temporary anyway and it
1107 		 * would only save a few bytes of leaf space.
1108 		 */
1109 		if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1110 			if (path->slots[0] + 1 >=
1111 			    btrfs_header_nritems(path->nodes[0])) {
1112 				ret = find_next_csum_offset(root, path, &next_offset);
1113 				if (ret < 0)
1114 					goto out;
1115 				found_next = 1;
1116 				goto insert;
1117 			}
1118 
1119 			ret = find_next_csum_offset(root, path, &next_offset);
1120 			if (ret < 0)
1121 				goto out;
1122 
1123 			tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1124 			if (tmp <= INT_MAX)
1125 				extend_nr = min_t(int, extend_nr, tmp);
1126 		}
1127 
1128 		diff = (csum_offset + extend_nr) * csum_size;
1129 		diff = min(diff,
1130 			   MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1131 
1132 		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
1133 		diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1134 		diff /= csum_size;
1135 		diff *= csum_size;
1136 
1137 		btrfs_extend_item(path, diff);
1138 		ret = 0;
1139 		goto csum;
1140 	}
1141 
1142 insert:
1143 	btrfs_release_path(path);
1144 	csum_offset = 0;
1145 	if (found_next) {
1146 		u64 tmp;
1147 
1148 		tmp = sums->len - total_bytes;
1149 		tmp >>= fs_info->sectorsize_bits;
1150 		tmp = min(tmp, (next_offset - file_key.offset) >>
1151 					 fs_info->sectorsize_bits);
1152 
1153 		tmp = max_t(u64, 1, tmp);
1154 		tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1155 		ins_size = csum_size * tmp;
1156 	} else {
1157 		ins_size = csum_size;
1158 	}
1159 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1160 				      ins_size);
1161 	if (ret < 0)
1162 		goto out;
1163 	if (WARN_ON(ret != 0))
1164 		goto out;
1165 	leaf = path->nodes[0];
1166 csum:
1167 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1168 	item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1169 				      btrfs_item_size_nr(leaf, path->slots[0]));
1170 	item = (struct btrfs_csum_item *)((unsigned char *)item +
1171 					  csum_offset * csum_size);
1172 found:
1173 	ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1174 	ins_size *= csum_size;
1175 	ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1176 			      ins_size);
1177 	write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1178 			    ins_size);
1179 
1180 	index += ins_size;
1181 	ins_size /= csum_size;
1182 	total_bytes += ins_size * fs_info->sectorsize;
1183 
1184 	btrfs_mark_buffer_dirty(path->nodes[0]);
1185 	if (total_bytes < sums->len) {
1186 		btrfs_release_path(path);
1187 		cond_resched();
1188 		goto again;
1189 	}
1190 out:
1191 	btrfs_free_path(path);
1192 	return ret;
1193 }
1194 
btrfs_extent_item_to_extent_map(struct btrfs_inode * inode,const struct btrfs_path * path,struct btrfs_file_extent_item * fi,const bool new_inline,struct extent_map * em)1195 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1196 				     const struct btrfs_path *path,
1197 				     struct btrfs_file_extent_item *fi,
1198 				     const bool new_inline,
1199 				     struct extent_map *em)
1200 {
1201 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1202 	struct btrfs_root *root = inode->root;
1203 	struct extent_buffer *leaf = path->nodes[0];
1204 	const int slot = path->slots[0];
1205 	struct btrfs_key key;
1206 	u64 extent_start, extent_end;
1207 	u64 bytenr;
1208 	u8 type = btrfs_file_extent_type(leaf, fi);
1209 	int compress_type = btrfs_file_extent_compression(leaf, fi);
1210 
1211 	btrfs_item_key_to_cpu(leaf, &key, slot);
1212 	extent_start = key.offset;
1213 	extent_end = btrfs_file_extent_end(path);
1214 	em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1215 	if (type == BTRFS_FILE_EXTENT_REG ||
1216 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
1217 		em->start = extent_start;
1218 		em->len = extent_end - extent_start;
1219 		em->orig_start = extent_start -
1220 			btrfs_file_extent_offset(leaf, fi);
1221 		em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1222 		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1223 		if (bytenr == 0) {
1224 			em->block_start = EXTENT_MAP_HOLE;
1225 			return;
1226 		}
1227 		if (compress_type != BTRFS_COMPRESS_NONE) {
1228 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1229 			em->compress_type = compress_type;
1230 			em->block_start = bytenr;
1231 			em->block_len = em->orig_block_len;
1232 		} else {
1233 			bytenr += btrfs_file_extent_offset(leaf, fi);
1234 			em->block_start = bytenr;
1235 			em->block_len = em->len;
1236 			if (type == BTRFS_FILE_EXTENT_PREALLOC)
1237 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1238 		}
1239 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
1240 		em->block_start = EXTENT_MAP_INLINE;
1241 		em->start = extent_start;
1242 		em->len = extent_end - extent_start;
1243 		/*
1244 		 * Initialize orig_start and block_len with the same values
1245 		 * as in inode.c:btrfs_get_extent().
1246 		 */
1247 		em->orig_start = EXTENT_MAP_HOLE;
1248 		em->block_len = (u64)-1;
1249 		if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1250 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1251 			em->compress_type = compress_type;
1252 		}
1253 	} else {
1254 		btrfs_err(fs_info,
1255 			  "unknown file extent item type %d, inode %llu, offset %llu, "
1256 			  "root %llu", type, btrfs_ino(inode), extent_start,
1257 			  root->root_key.objectid);
1258 	}
1259 }
1260 
1261 /*
1262  * Returns the end offset (non inclusive) of the file extent item the given path
1263  * points to. If it points to an inline extent, the returned offset is rounded
1264  * up to the sector size.
1265  */
btrfs_file_extent_end(const struct btrfs_path * path)1266 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1267 {
1268 	const struct extent_buffer *leaf = path->nodes[0];
1269 	const int slot = path->slots[0];
1270 	struct btrfs_file_extent_item *fi;
1271 	struct btrfs_key key;
1272 	u64 end;
1273 
1274 	btrfs_item_key_to_cpu(leaf, &key, slot);
1275 	ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1276 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1277 
1278 	if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1279 		end = btrfs_file_extent_ram_bytes(leaf, fi);
1280 		end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1281 	} else {
1282 		end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1283 	}
1284 
1285 	return end;
1286 }
1287