xref: /linux/fs/btrfs/file-item.c (revision 42466b9f29b415c254dc4c2f4618e2a96951a406)
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 "ctree.h"
13 #include "disk-io.h"
14 #include "transaction.h"
15 #include "volumes.h"
16 #include "print-tree.h"
17 #include "compression.h"
18 
19 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
20 				   sizeof(struct btrfs_item) * 2) / \
21 				  size) - 1))
22 
23 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
24 				       PAGE_SIZE))
25 
26 static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info,
27 					u16 csum_size)
28 {
29 	u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size;
30 
31 	return ncsums * fs_info->sectorsize;
32 }
33 
34 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
35 			     struct btrfs_root *root,
36 			     u64 objectid, u64 pos,
37 			     u64 disk_offset, u64 disk_num_bytes,
38 			     u64 num_bytes, u64 offset, u64 ram_bytes,
39 			     u8 compression, u8 encryption, u16 other_encoding)
40 {
41 	int ret = 0;
42 	struct btrfs_file_extent_item *item;
43 	struct btrfs_key file_key;
44 	struct btrfs_path *path;
45 	struct extent_buffer *leaf;
46 
47 	path = btrfs_alloc_path();
48 	if (!path)
49 		return -ENOMEM;
50 	file_key.objectid = objectid;
51 	file_key.offset = pos;
52 	file_key.type = BTRFS_EXTENT_DATA_KEY;
53 
54 	path->leave_spinning = 1;
55 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
56 				      sizeof(*item));
57 	if (ret < 0)
58 		goto out;
59 	BUG_ON(ret); /* Can't happen */
60 	leaf = path->nodes[0];
61 	item = btrfs_item_ptr(leaf, path->slots[0],
62 			      struct btrfs_file_extent_item);
63 	btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
64 	btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
65 	btrfs_set_file_extent_offset(leaf, item, offset);
66 	btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
67 	btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
68 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
69 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
70 	btrfs_set_file_extent_compression(leaf, item, compression);
71 	btrfs_set_file_extent_encryption(leaf, item, encryption);
72 	btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
73 
74 	btrfs_mark_buffer_dirty(leaf);
75 out:
76 	btrfs_free_path(path);
77 	return ret;
78 }
79 
80 static struct btrfs_csum_item *
81 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
82 		  struct btrfs_root *root,
83 		  struct btrfs_path *path,
84 		  u64 bytenr, int cow)
85 {
86 	struct btrfs_fs_info *fs_info = root->fs_info;
87 	int ret;
88 	struct btrfs_key file_key;
89 	struct btrfs_key found_key;
90 	struct btrfs_csum_item *item;
91 	struct extent_buffer *leaf;
92 	u64 csum_offset = 0;
93 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
94 	int csums_in_item;
95 
96 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
97 	file_key.offset = bytenr;
98 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
99 	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
100 	if (ret < 0)
101 		goto fail;
102 	leaf = path->nodes[0];
103 	if (ret > 0) {
104 		ret = 1;
105 		if (path->slots[0] == 0)
106 			goto fail;
107 		path->slots[0]--;
108 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
109 		if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
110 			goto fail;
111 
112 		csum_offset = (bytenr - found_key.offset) >>
113 				fs_info->sb->s_blocksize_bits;
114 		csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
115 		csums_in_item /= csum_size;
116 
117 		if (csum_offset == csums_in_item) {
118 			ret = -EFBIG;
119 			goto fail;
120 		} else if (csum_offset > csums_in_item) {
121 			goto fail;
122 		}
123 	}
124 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
125 	item = (struct btrfs_csum_item *)((unsigned char *)item +
126 					  csum_offset * csum_size);
127 	return item;
128 fail:
129 	if (ret > 0)
130 		ret = -ENOENT;
131 	return ERR_PTR(ret);
132 }
133 
134 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
135 			     struct btrfs_root *root,
136 			     struct btrfs_path *path, u64 objectid,
137 			     u64 offset, int mod)
138 {
139 	int ret;
140 	struct btrfs_key file_key;
141 	int ins_len = mod < 0 ? -1 : 0;
142 	int cow = mod != 0;
143 
144 	file_key.objectid = objectid;
145 	file_key.offset = offset;
146 	file_key.type = BTRFS_EXTENT_DATA_KEY;
147 	ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
148 	return ret;
149 }
150 
151 static blk_status_t __btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,
152 				   u64 logical_offset, u8 *dst, int dio)
153 {
154 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
155 	struct bio_vec bvec;
156 	struct bvec_iter iter;
157 	struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
158 	struct btrfs_csum_item *item = NULL;
159 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
160 	struct btrfs_path *path;
161 	u8 *csum;
162 	u64 offset = 0;
163 	u64 item_start_offset = 0;
164 	u64 item_last_offset = 0;
165 	u64 disk_bytenr;
166 	u64 page_bytes_left;
167 	u32 diff;
168 	int nblocks;
169 	int count = 0;
170 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
171 
172 	path = btrfs_alloc_path();
173 	if (!path)
174 		return BLK_STS_RESOURCE;
175 
176 	nblocks = bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
177 	if (!dst) {
178 		if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
179 			btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
180 							GFP_NOFS);
181 			if (!btrfs_bio->csum) {
182 				btrfs_free_path(path);
183 				return BLK_STS_RESOURCE;
184 			}
185 		} else {
186 			btrfs_bio->csum = btrfs_bio->csum_inline;
187 		}
188 		csum = btrfs_bio->csum;
189 	} else {
190 		csum = dst;
191 	}
192 
193 	if (bio->bi_iter.bi_size > PAGE_SIZE * 8)
194 		path->reada = READA_FORWARD;
195 
196 	/*
197 	 * the free space stuff is only read when it hasn't been
198 	 * updated in the current transaction.  So, we can safely
199 	 * read from the commit root and sidestep a nasty deadlock
200 	 * between reading the free space cache and updating the csum tree.
201 	 */
202 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
203 		path->search_commit_root = 1;
204 		path->skip_locking = 1;
205 	}
206 
207 	disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
208 	if (dio)
209 		offset = logical_offset;
210 
211 	bio_for_each_segment(bvec, bio, iter) {
212 		page_bytes_left = bvec.bv_len;
213 		if (count)
214 			goto next;
215 
216 		if (!dio)
217 			offset = page_offset(bvec.bv_page) + bvec.bv_offset;
218 		count = btrfs_find_ordered_sum(inode, offset, disk_bytenr,
219 					       csum, nblocks);
220 		if (count)
221 			goto found;
222 
223 		if (!item || disk_bytenr < item_start_offset ||
224 		    disk_bytenr >= item_last_offset) {
225 			struct btrfs_key found_key;
226 			u32 item_size;
227 
228 			if (item)
229 				btrfs_release_path(path);
230 			item = btrfs_lookup_csum(NULL, fs_info->csum_root,
231 						 path, disk_bytenr, 0);
232 			if (IS_ERR(item)) {
233 				count = 1;
234 				memset(csum, 0, csum_size);
235 				if (BTRFS_I(inode)->root->root_key.objectid ==
236 				    BTRFS_DATA_RELOC_TREE_OBJECTID) {
237 					set_extent_bits(io_tree, offset,
238 						offset + fs_info->sectorsize - 1,
239 						EXTENT_NODATASUM);
240 				} else {
241 					btrfs_info_rl(fs_info,
242 						   "no csum found for inode %llu start %llu",
243 					       btrfs_ino(BTRFS_I(inode)), offset);
244 				}
245 				item = NULL;
246 				btrfs_release_path(path);
247 				goto found;
248 			}
249 			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
250 					      path->slots[0]);
251 
252 			item_start_offset = found_key.offset;
253 			item_size = btrfs_item_size_nr(path->nodes[0],
254 						       path->slots[0]);
255 			item_last_offset = item_start_offset +
256 				(item_size / csum_size) *
257 				fs_info->sectorsize;
258 			item = btrfs_item_ptr(path->nodes[0], path->slots[0],
259 					      struct btrfs_csum_item);
260 		}
261 		/*
262 		 * this byte range must be able to fit inside
263 		 * a single leaf so it will also fit inside a u32
264 		 */
265 		diff = disk_bytenr - item_start_offset;
266 		diff = diff / fs_info->sectorsize;
267 		diff = diff * csum_size;
268 		count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
269 					    inode->i_sb->s_blocksize_bits);
270 		read_extent_buffer(path->nodes[0], csum,
271 				   ((unsigned long)item) + diff,
272 				   csum_size * count);
273 found:
274 		csum += count * csum_size;
275 		nblocks -= count;
276 next:
277 		while (count--) {
278 			disk_bytenr += fs_info->sectorsize;
279 			offset += fs_info->sectorsize;
280 			page_bytes_left -= fs_info->sectorsize;
281 			if (!page_bytes_left)
282 				break; /* move to next bio */
283 		}
284 	}
285 
286 	WARN_ON_ONCE(count);
287 	btrfs_free_path(path);
288 	return 0;
289 }
290 
291 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,
292 				   u8 *dst)
293 {
294 	return __btrfs_lookup_bio_sums(inode, bio, 0, dst, 0);
295 }
296 
297 blk_status_t btrfs_lookup_bio_sums_dio(struct inode *inode, struct bio *bio, u64 offset)
298 {
299 	return __btrfs_lookup_bio_sums(inode, bio, offset, NULL, 1);
300 }
301 
302 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
303 			     struct list_head *list, int search_commit)
304 {
305 	struct btrfs_fs_info *fs_info = root->fs_info;
306 	struct btrfs_key key;
307 	struct btrfs_path *path;
308 	struct extent_buffer *leaf;
309 	struct btrfs_ordered_sum *sums;
310 	struct btrfs_csum_item *item;
311 	LIST_HEAD(tmplist);
312 	unsigned long offset;
313 	int ret;
314 	size_t size;
315 	u64 csum_end;
316 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
317 
318 	ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
319 	       IS_ALIGNED(end + 1, fs_info->sectorsize));
320 
321 	path = btrfs_alloc_path();
322 	if (!path)
323 		return -ENOMEM;
324 
325 	if (search_commit) {
326 		path->skip_locking = 1;
327 		path->reada = READA_FORWARD;
328 		path->search_commit_root = 1;
329 	}
330 
331 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
332 	key.offset = start;
333 	key.type = BTRFS_EXTENT_CSUM_KEY;
334 
335 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
336 	if (ret < 0)
337 		goto fail;
338 	if (ret > 0 && path->slots[0] > 0) {
339 		leaf = path->nodes[0];
340 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
341 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
342 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
343 			offset = (start - key.offset) >>
344 				 fs_info->sb->s_blocksize_bits;
345 			if (offset * csum_size <
346 			    btrfs_item_size_nr(leaf, path->slots[0] - 1))
347 				path->slots[0]--;
348 		}
349 	}
350 
351 	while (start <= end) {
352 		leaf = path->nodes[0];
353 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
354 			ret = btrfs_next_leaf(root, path);
355 			if (ret < 0)
356 				goto fail;
357 			if (ret > 0)
358 				break;
359 			leaf = path->nodes[0];
360 		}
361 
362 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
363 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
364 		    key.type != BTRFS_EXTENT_CSUM_KEY ||
365 		    key.offset > end)
366 			break;
367 
368 		if (key.offset > start)
369 			start = key.offset;
370 
371 		size = btrfs_item_size_nr(leaf, path->slots[0]);
372 		csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
373 		if (csum_end <= start) {
374 			path->slots[0]++;
375 			continue;
376 		}
377 
378 		csum_end = min(csum_end, end + 1);
379 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
380 				      struct btrfs_csum_item);
381 		while (start < csum_end) {
382 			size = min_t(size_t, csum_end - start,
383 				     max_ordered_sum_bytes(fs_info, csum_size));
384 			sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
385 				       GFP_NOFS);
386 			if (!sums) {
387 				ret = -ENOMEM;
388 				goto fail;
389 			}
390 
391 			sums->bytenr = start;
392 			sums->len = (int)size;
393 
394 			offset = (start - key.offset) >>
395 				fs_info->sb->s_blocksize_bits;
396 			offset *= csum_size;
397 			size >>= fs_info->sb->s_blocksize_bits;
398 
399 			read_extent_buffer(path->nodes[0],
400 					   sums->sums,
401 					   ((unsigned long)item) + offset,
402 					   csum_size * size);
403 
404 			start += fs_info->sectorsize * size;
405 			list_add_tail(&sums->list, &tmplist);
406 		}
407 		path->slots[0]++;
408 	}
409 	ret = 0;
410 fail:
411 	while (ret < 0 && !list_empty(&tmplist)) {
412 		sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
413 		list_del(&sums->list);
414 		kfree(sums);
415 	}
416 	list_splice_tail(&tmplist, list);
417 
418 	btrfs_free_path(path);
419 	return ret;
420 }
421 
422 /*
423  * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
424  * @inode:	 Owner of the data inside the bio
425  * @bio:	 Contains the data to be checksummed
426  * @file_start:  offset in file this bio begins to describe
427  * @contig:	 Boolean. If true/1 means all bio vecs in this bio are
428  *		 contiguous and they begin at @file_start in the file. False/0
429  *		 means this bio can contains potentially discontigous bio vecs
430  *		 so the logical offset of each should be calculated separately.
431  */
432 blk_status_t btrfs_csum_one_bio(struct inode *inode, struct bio *bio,
433 		       u64 file_start, int contig)
434 {
435 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
436 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
437 	struct btrfs_ordered_sum *sums;
438 	struct btrfs_ordered_extent *ordered = NULL;
439 	char *data;
440 	struct bvec_iter iter;
441 	struct bio_vec bvec;
442 	int index;
443 	int nr_sectors;
444 	unsigned long total_bytes = 0;
445 	unsigned long this_sum_bytes = 0;
446 	int i;
447 	u64 offset;
448 	unsigned nofs_flag;
449 	const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
450 
451 	nofs_flag = memalloc_nofs_save();
452 	sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
453 		       GFP_KERNEL);
454 	memalloc_nofs_restore(nofs_flag);
455 
456 	if (!sums)
457 		return BLK_STS_RESOURCE;
458 
459 	sums->len = bio->bi_iter.bi_size;
460 	INIT_LIST_HEAD(&sums->list);
461 
462 	if (contig)
463 		offset = file_start;
464 	else
465 		offset = 0; /* shut up gcc */
466 
467 	sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
468 	index = 0;
469 
470 	shash->tfm = fs_info->csum_shash;
471 
472 	bio_for_each_segment(bvec, bio, iter) {
473 		if (!contig)
474 			offset = page_offset(bvec.bv_page) + bvec.bv_offset;
475 
476 		if (!ordered) {
477 			ordered = btrfs_lookup_ordered_extent(inode, offset);
478 			BUG_ON(!ordered); /* Logic error */
479 		}
480 
481 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
482 						 bvec.bv_len + fs_info->sectorsize
483 						 - 1);
484 
485 		for (i = 0; i < nr_sectors; i++) {
486 			if (offset >= ordered->file_offset + ordered->len ||
487 				offset < ordered->file_offset) {
488 				unsigned long bytes_left;
489 
490 				sums->len = this_sum_bytes;
491 				this_sum_bytes = 0;
492 				btrfs_add_ordered_sum(ordered, sums);
493 				btrfs_put_ordered_extent(ordered);
494 
495 				bytes_left = bio->bi_iter.bi_size - total_bytes;
496 
497 				nofs_flag = memalloc_nofs_save();
498 				sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
499 						      bytes_left), GFP_KERNEL);
500 				memalloc_nofs_restore(nofs_flag);
501 				BUG_ON(!sums); /* -ENOMEM */
502 				sums->len = bytes_left;
503 				ordered = btrfs_lookup_ordered_extent(inode,
504 								offset);
505 				ASSERT(ordered); /* Logic error */
506 				sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
507 					+ total_bytes;
508 				index = 0;
509 			}
510 
511 			crypto_shash_init(shash);
512 			data = kmap_atomic(bvec.bv_page);
513 			crypto_shash_update(shash, data + bvec.bv_offset
514 					    + (i * fs_info->sectorsize),
515 					    fs_info->sectorsize);
516 			kunmap_atomic(data);
517 			crypto_shash_final(shash, (char *)(sums->sums + index));
518 			index += csum_size;
519 			offset += fs_info->sectorsize;
520 			this_sum_bytes += fs_info->sectorsize;
521 			total_bytes += fs_info->sectorsize;
522 		}
523 
524 	}
525 	this_sum_bytes = 0;
526 	btrfs_add_ordered_sum(ordered, sums);
527 	btrfs_put_ordered_extent(ordered);
528 	return 0;
529 }
530 
531 /*
532  * helper function for csum removal, this expects the
533  * key to describe the csum pointed to by the path, and it expects
534  * the csum to overlap the range [bytenr, len]
535  *
536  * The csum should not be entirely contained in the range and the
537  * range should not be entirely contained in the csum.
538  *
539  * This calls btrfs_truncate_item with the correct args based on the
540  * overlap, and fixes up the key as required.
541  */
542 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
543 				       struct btrfs_path *path,
544 				       struct btrfs_key *key,
545 				       u64 bytenr, u64 len)
546 {
547 	struct extent_buffer *leaf;
548 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
549 	u64 csum_end;
550 	u64 end_byte = bytenr + len;
551 	u32 blocksize_bits = fs_info->sb->s_blocksize_bits;
552 
553 	leaf = path->nodes[0];
554 	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
555 	csum_end <<= fs_info->sb->s_blocksize_bits;
556 	csum_end += key->offset;
557 
558 	if (key->offset < bytenr && csum_end <= end_byte) {
559 		/*
560 		 *         [ bytenr - len ]
561 		 *         [   ]
562 		 *   [csum     ]
563 		 *   A simple truncate off the end of the item
564 		 */
565 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
566 		new_size *= csum_size;
567 		btrfs_truncate_item(path, new_size, 1);
568 	} else if (key->offset >= bytenr && csum_end > end_byte &&
569 		   end_byte > key->offset) {
570 		/*
571 		 *         [ bytenr - len ]
572 		 *                 [ ]
573 		 *                 [csum     ]
574 		 * we need to truncate from the beginning of the csum
575 		 */
576 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
577 		new_size *= csum_size;
578 
579 		btrfs_truncate_item(path, new_size, 0);
580 
581 		key->offset = end_byte;
582 		btrfs_set_item_key_safe(fs_info, path, key);
583 	} else {
584 		BUG();
585 	}
586 }
587 
588 /*
589  * deletes the csum items from the csum tree for a given
590  * range of bytes.
591  */
592 int btrfs_del_csums(struct btrfs_trans_handle *trans,
593 		    struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
594 {
595 	struct btrfs_root *root = fs_info->csum_root;
596 	struct btrfs_path *path;
597 	struct btrfs_key key;
598 	u64 end_byte = bytenr + len;
599 	u64 csum_end;
600 	struct extent_buffer *leaf;
601 	int ret;
602 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
603 	int blocksize_bits = fs_info->sb->s_blocksize_bits;
604 
605 	path = btrfs_alloc_path();
606 	if (!path)
607 		return -ENOMEM;
608 
609 	while (1) {
610 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
611 		key.offset = end_byte - 1;
612 		key.type = BTRFS_EXTENT_CSUM_KEY;
613 
614 		path->leave_spinning = 1;
615 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
616 		if (ret > 0) {
617 			if (path->slots[0] == 0)
618 				break;
619 			path->slots[0]--;
620 		} else if (ret < 0) {
621 			break;
622 		}
623 
624 		leaf = path->nodes[0];
625 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
626 
627 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
628 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
629 			break;
630 		}
631 
632 		if (key.offset >= end_byte)
633 			break;
634 
635 		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
636 		csum_end <<= blocksize_bits;
637 		csum_end += key.offset;
638 
639 		/* this csum ends before we start, we're done */
640 		if (csum_end <= bytenr)
641 			break;
642 
643 		/* delete the entire item, it is inside our range */
644 		if (key.offset >= bytenr && csum_end <= end_byte) {
645 			int del_nr = 1;
646 
647 			/*
648 			 * Check how many csum items preceding this one in this
649 			 * leaf correspond to our range and then delete them all
650 			 * at once.
651 			 */
652 			if (key.offset > bytenr && path->slots[0] > 0) {
653 				int slot = path->slots[0] - 1;
654 
655 				while (slot >= 0) {
656 					struct btrfs_key pk;
657 
658 					btrfs_item_key_to_cpu(leaf, &pk, slot);
659 					if (pk.offset < bytenr ||
660 					    pk.type != BTRFS_EXTENT_CSUM_KEY ||
661 					    pk.objectid !=
662 					    BTRFS_EXTENT_CSUM_OBJECTID)
663 						break;
664 					path->slots[0] = slot;
665 					del_nr++;
666 					key.offset = pk.offset;
667 					slot--;
668 				}
669 			}
670 			ret = btrfs_del_items(trans, root, path,
671 					      path->slots[0], del_nr);
672 			if (ret)
673 				goto out;
674 			if (key.offset == bytenr)
675 				break;
676 		} else if (key.offset < bytenr && csum_end > end_byte) {
677 			unsigned long offset;
678 			unsigned long shift_len;
679 			unsigned long item_offset;
680 			/*
681 			 *        [ bytenr - len ]
682 			 *     [csum                ]
683 			 *
684 			 * Our bytes are in the middle of the csum,
685 			 * we need to split this item and insert a new one.
686 			 *
687 			 * But we can't drop the path because the
688 			 * csum could change, get removed, extended etc.
689 			 *
690 			 * The trick here is the max size of a csum item leaves
691 			 * enough room in the tree block for a single
692 			 * item header.  So, we split the item in place,
693 			 * adding a new header pointing to the existing
694 			 * bytes.  Then we loop around again and we have
695 			 * a nicely formed csum item that we can neatly
696 			 * truncate.
697 			 */
698 			offset = (bytenr - key.offset) >> blocksize_bits;
699 			offset *= csum_size;
700 
701 			shift_len = (len >> blocksize_bits) * csum_size;
702 
703 			item_offset = btrfs_item_ptr_offset(leaf,
704 							    path->slots[0]);
705 
706 			memzero_extent_buffer(leaf, item_offset + offset,
707 					     shift_len);
708 			key.offset = bytenr;
709 
710 			/*
711 			 * btrfs_split_item returns -EAGAIN when the
712 			 * item changed size or key
713 			 */
714 			ret = btrfs_split_item(trans, root, path, &key, offset);
715 			if (ret && ret != -EAGAIN) {
716 				btrfs_abort_transaction(trans, ret);
717 				goto out;
718 			}
719 
720 			key.offset = end_byte - 1;
721 		} else {
722 			truncate_one_csum(fs_info, path, &key, bytenr, len);
723 			if (key.offset < bytenr)
724 				break;
725 		}
726 		btrfs_release_path(path);
727 	}
728 	ret = 0;
729 out:
730 	btrfs_free_path(path);
731 	return ret;
732 }
733 
734 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
735 			   struct btrfs_root *root,
736 			   struct btrfs_ordered_sum *sums)
737 {
738 	struct btrfs_fs_info *fs_info = root->fs_info;
739 	struct btrfs_key file_key;
740 	struct btrfs_key found_key;
741 	struct btrfs_path *path;
742 	struct btrfs_csum_item *item;
743 	struct btrfs_csum_item *item_end;
744 	struct extent_buffer *leaf = NULL;
745 	u64 next_offset;
746 	u64 total_bytes = 0;
747 	u64 csum_offset;
748 	u64 bytenr;
749 	u32 nritems;
750 	u32 ins_size;
751 	int index = 0;
752 	int found_next;
753 	int ret;
754 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
755 
756 	path = btrfs_alloc_path();
757 	if (!path)
758 		return -ENOMEM;
759 again:
760 	next_offset = (u64)-1;
761 	found_next = 0;
762 	bytenr = sums->bytenr + total_bytes;
763 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
764 	file_key.offset = bytenr;
765 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
766 
767 	item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
768 	if (!IS_ERR(item)) {
769 		ret = 0;
770 		leaf = path->nodes[0];
771 		item_end = btrfs_item_ptr(leaf, path->slots[0],
772 					  struct btrfs_csum_item);
773 		item_end = (struct btrfs_csum_item *)((char *)item_end +
774 			   btrfs_item_size_nr(leaf, path->slots[0]));
775 		goto found;
776 	}
777 	ret = PTR_ERR(item);
778 	if (ret != -EFBIG && ret != -ENOENT)
779 		goto fail_unlock;
780 
781 	if (ret == -EFBIG) {
782 		u32 item_size;
783 		/* we found one, but it isn't big enough yet */
784 		leaf = path->nodes[0];
785 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
786 		if ((item_size / csum_size) >=
787 		    MAX_CSUM_ITEMS(fs_info, csum_size)) {
788 			/* already at max size, make a new one */
789 			goto insert;
790 		}
791 	} else {
792 		int slot = path->slots[0] + 1;
793 		/* we didn't find a csum item, insert one */
794 		nritems = btrfs_header_nritems(path->nodes[0]);
795 		if (!nritems || (path->slots[0] >= nritems - 1)) {
796 			ret = btrfs_next_leaf(root, path);
797 			if (ret == 1)
798 				found_next = 1;
799 			if (ret != 0)
800 				goto insert;
801 			slot = path->slots[0];
802 		}
803 		btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
804 		if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
805 		    found_key.type != BTRFS_EXTENT_CSUM_KEY) {
806 			found_next = 1;
807 			goto insert;
808 		}
809 		next_offset = found_key.offset;
810 		found_next = 1;
811 		goto insert;
812 	}
813 
814 	/*
815 	 * at this point, we know the tree has an item, but it isn't big
816 	 * enough yet to put our csum in.  Grow it
817 	 */
818 	btrfs_release_path(path);
819 	ret = btrfs_search_slot(trans, root, &file_key, path,
820 				csum_size, 1);
821 	if (ret < 0)
822 		goto fail_unlock;
823 
824 	if (ret > 0) {
825 		if (path->slots[0] == 0)
826 			goto insert;
827 		path->slots[0]--;
828 	}
829 
830 	leaf = path->nodes[0];
831 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
832 	csum_offset = (bytenr - found_key.offset) >>
833 			fs_info->sb->s_blocksize_bits;
834 
835 	if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
836 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
837 	    csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
838 		goto insert;
839 	}
840 
841 	if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
842 	    csum_size) {
843 		int extend_nr;
844 		u64 tmp;
845 		u32 diff;
846 		u32 free_space;
847 
848 		if (btrfs_leaf_free_space(leaf) <
849 				 sizeof(struct btrfs_item) + csum_size * 2)
850 			goto insert;
851 
852 		free_space = btrfs_leaf_free_space(leaf) -
853 					 sizeof(struct btrfs_item) - csum_size;
854 		tmp = sums->len - total_bytes;
855 		tmp >>= fs_info->sb->s_blocksize_bits;
856 		WARN_ON(tmp < 1);
857 
858 		extend_nr = max_t(int, 1, (int)tmp);
859 		diff = (csum_offset + extend_nr) * csum_size;
860 		diff = min(diff,
861 			   MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
862 
863 		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
864 		diff = min(free_space, diff);
865 		diff /= csum_size;
866 		diff *= csum_size;
867 
868 		btrfs_extend_item(path, diff);
869 		ret = 0;
870 		goto csum;
871 	}
872 
873 insert:
874 	btrfs_release_path(path);
875 	csum_offset = 0;
876 	if (found_next) {
877 		u64 tmp;
878 
879 		tmp = sums->len - total_bytes;
880 		tmp >>= fs_info->sb->s_blocksize_bits;
881 		tmp = min(tmp, (next_offset - file_key.offset) >>
882 					 fs_info->sb->s_blocksize_bits);
883 
884 		tmp = max_t(u64, 1, tmp);
885 		tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
886 		ins_size = csum_size * tmp;
887 	} else {
888 		ins_size = csum_size;
889 	}
890 	path->leave_spinning = 1;
891 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
892 				      ins_size);
893 	path->leave_spinning = 0;
894 	if (ret < 0)
895 		goto fail_unlock;
896 	if (WARN_ON(ret != 0))
897 		goto fail_unlock;
898 	leaf = path->nodes[0];
899 csum:
900 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
901 	item_end = (struct btrfs_csum_item *)((unsigned char *)item +
902 				      btrfs_item_size_nr(leaf, path->slots[0]));
903 	item = (struct btrfs_csum_item *)((unsigned char *)item +
904 					  csum_offset * csum_size);
905 found:
906 	ins_size = (u32)(sums->len - total_bytes) >>
907 		   fs_info->sb->s_blocksize_bits;
908 	ins_size *= csum_size;
909 	ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
910 			      ins_size);
911 	write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
912 			    ins_size);
913 
914 	index += ins_size;
915 	ins_size /= csum_size;
916 	total_bytes += ins_size * fs_info->sectorsize;
917 
918 	btrfs_mark_buffer_dirty(path->nodes[0]);
919 	if (total_bytes < sums->len) {
920 		btrfs_release_path(path);
921 		cond_resched();
922 		goto again;
923 	}
924 out:
925 	btrfs_free_path(path);
926 	return ret;
927 
928 fail_unlock:
929 	goto out;
930 }
931 
932 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
933 				     const struct btrfs_path *path,
934 				     struct btrfs_file_extent_item *fi,
935 				     const bool new_inline,
936 				     struct extent_map *em)
937 {
938 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
939 	struct btrfs_root *root = inode->root;
940 	struct extent_buffer *leaf = path->nodes[0];
941 	const int slot = path->slots[0];
942 	struct btrfs_key key;
943 	u64 extent_start, extent_end;
944 	u64 bytenr;
945 	u8 type = btrfs_file_extent_type(leaf, fi);
946 	int compress_type = btrfs_file_extent_compression(leaf, fi);
947 
948 	em->bdev = fs_info->fs_devices->latest_bdev;
949 	btrfs_item_key_to_cpu(leaf, &key, slot);
950 	extent_start = key.offset;
951 
952 	if (type == BTRFS_FILE_EXTENT_REG ||
953 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
954 		extent_end = extent_start +
955 			btrfs_file_extent_num_bytes(leaf, fi);
956 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
957 		size_t size;
958 		size = btrfs_file_extent_ram_bytes(leaf, fi);
959 		extent_end = ALIGN(extent_start + size,
960 				   fs_info->sectorsize);
961 	}
962 
963 	em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
964 	if (type == BTRFS_FILE_EXTENT_REG ||
965 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
966 		em->start = extent_start;
967 		em->len = extent_end - extent_start;
968 		em->orig_start = extent_start -
969 			btrfs_file_extent_offset(leaf, fi);
970 		em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
971 		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
972 		if (bytenr == 0) {
973 			em->block_start = EXTENT_MAP_HOLE;
974 			return;
975 		}
976 		if (compress_type != BTRFS_COMPRESS_NONE) {
977 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
978 			em->compress_type = compress_type;
979 			em->block_start = bytenr;
980 			em->block_len = em->orig_block_len;
981 		} else {
982 			bytenr += btrfs_file_extent_offset(leaf, fi);
983 			em->block_start = bytenr;
984 			em->block_len = em->len;
985 			if (type == BTRFS_FILE_EXTENT_PREALLOC)
986 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
987 		}
988 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
989 		em->block_start = EXTENT_MAP_INLINE;
990 		em->start = extent_start;
991 		em->len = extent_end - extent_start;
992 		/*
993 		 * Initialize orig_start and block_len with the same values
994 		 * as in inode.c:btrfs_get_extent().
995 		 */
996 		em->orig_start = EXTENT_MAP_HOLE;
997 		em->block_len = (u64)-1;
998 		if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
999 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1000 			em->compress_type = compress_type;
1001 		}
1002 	} else {
1003 		btrfs_err(fs_info,
1004 			  "unknown file extent item type %d, inode %llu, offset %llu, "
1005 			  "root %llu", type, btrfs_ino(inode), extent_start,
1006 			  root->root_key.objectid);
1007 	}
1008 }
1009