xref: /linux/fs/btrfs/file-item.c (revision fae4973c9a58858bc63a842cdaf7810955503399)
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 "ctree.h"
12 #include "disk-io.h"
13 #include "transaction.h"
14 #include "volumes.h"
15 #include "print-tree.h"
16 #include "compression.h"
17 
18 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
19 				   sizeof(struct btrfs_item) * 2) / \
20 				  size) - 1))
21 
22 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
23 				       PAGE_SIZE))
24 
25 #define MAX_ORDERED_SUM_BYTES(fs_info) ((PAGE_SIZE - \
26 				   sizeof(struct btrfs_ordered_sum)) / \
27 				   sizeof(u32) * (fs_info)->sectorsize)
28 
29 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
30 			     struct btrfs_root *root,
31 			     u64 objectid, u64 pos,
32 			     u64 disk_offset, u64 disk_num_bytes,
33 			     u64 num_bytes, u64 offset, u64 ram_bytes,
34 			     u8 compression, u8 encryption, u16 other_encoding)
35 {
36 	int ret = 0;
37 	struct btrfs_file_extent_item *item;
38 	struct btrfs_key file_key;
39 	struct btrfs_path *path;
40 	struct extent_buffer *leaf;
41 
42 	path = btrfs_alloc_path();
43 	if (!path)
44 		return -ENOMEM;
45 	file_key.objectid = objectid;
46 	file_key.offset = pos;
47 	file_key.type = BTRFS_EXTENT_DATA_KEY;
48 
49 	path->leave_spinning = 1;
50 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
51 				      sizeof(*item));
52 	if (ret < 0)
53 		goto out;
54 	BUG_ON(ret); /* Can't happen */
55 	leaf = path->nodes[0];
56 	item = btrfs_item_ptr(leaf, path->slots[0],
57 			      struct btrfs_file_extent_item);
58 	btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
59 	btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
60 	btrfs_set_file_extent_offset(leaf, item, offset);
61 	btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
62 	btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
63 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
64 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
65 	btrfs_set_file_extent_compression(leaf, item, compression);
66 	btrfs_set_file_extent_encryption(leaf, item, encryption);
67 	btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
68 
69 	btrfs_mark_buffer_dirty(leaf);
70 out:
71 	btrfs_free_path(path);
72 	return ret;
73 }
74 
75 static struct btrfs_csum_item *
76 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
77 		  struct btrfs_root *root,
78 		  struct btrfs_path *path,
79 		  u64 bytenr, int cow)
80 {
81 	struct btrfs_fs_info *fs_info = root->fs_info;
82 	int ret;
83 	struct btrfs_key file_key;
84 	struct btrfs_key found_key;
85 	struct btrfs_csum_item *item;
86 	struct extent_buffer *leaf;
87 	u64 csum_offset = 0;
88 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
89 	int csums_in_item;
90 
91 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
92 	file_key.offset = bytenr;
93 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
94 	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
95 	if (ret < 0)
96 		goto fail;
97 	leaf = path->nodes[0];
98 	if (ret > 0) {
99 		ret = 1;
100 		if (path->slots[0] == 0)
101 			goto fail;
102 		path->slots[0]--;
103 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
104 		if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
105 			goto fail;
106 
107 		csum_offset = (bytenr - found_key.offset) >>
108 				fs_info->sb->s_blocksize_bits;
109 		csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
110 		csums_in_item /= csum_size;
111 
112 		if (csum_offset == csums_in_item) {
113 			ret = -EFBIG;
114 			goto fail;
115 		} else if (csum_offset > csums_in_item) {
116 			goto fail;
117 		}
118 	}
119 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
120 	item = (struct btrfs_csum_item *)((unsigned char *)item +
121 					  csum_offset * csum_size);
122 	return item;
123 fail:
124 	if (ret > 0)
125 		ret = -ENOENT;
126 	return ERR_PTR(ret);
127 }
128 
129 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
130 			     struct btrfs_root *root,
131 			     struct btrfs_path *path, u64 objectid,
132 			     u64 offset, int mod)
133 {
134 	int ret;
135 	struct btrfs_key file_key;
136 	int ins_len = mod < 0 ? -1 : 0;
137 	int cow = mod != 0;
138 
139 	file_key.objectid = objectid;
140 	file_key.offset = offset;
141 	file_key.type = BTRFS_EXTENT_DATA_KEY;
142 	ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
143 	return ret;
144 }
145 
146 static blk_status_t __btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,
147 				   u64 logical_offset, u32 *dst, int dio)
148 {
149 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
150 	struct bio_vec bvec;
151 	struct bvec_iter iter;
152 	struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
153 	struct btrfs_csum_item *item = NULL;
154 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
155 	struct btrfs_path *path;
156 	u8 *csum;
157 	u64 offset = 0;
158 	u64 item_start_offset = 0;
159 	u64 item_last_offset = 0;
160 	u64 disk_bytenr;
161 	u64 page_bytes_left;
162 	u32 diff;
163 	int nblocks;
164 	int count = 0;
165 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
166 
167 	path = btrfs_alloc_path();
168 	if (!path)
169 		return BLK_STS_RESOURCE;
170 
171 	nblocks = bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
172 	if (!dst) {
173 		if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
174 			btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
175 							GFP_NOFS);
176 			if (!btrfs_bio->csum) {
177 				btrfs_free_path(path);
178 				return BLK_STS_RESOURCE;
179 			}
180 		} else {
181 			btrfs_bio->csum = btrfs_bio->csum_inline;
182 		}
183 		csum = btrfs_bio->csum;
184 	} else {
185 		csum = (u8 *)dst;
186 	}
187 
188 	if (bio->bi_iter.bi_size > PAGE_SIZE * 8)
189 		path->reada = READA_FORWARD;
190 
191 	/*
192 	 * the free space stuff is only read when it hasn't been
193 	 * updated in the current transaction.  So, we can safely
194 	 * read from the commit root and sidestep a nasty deadlock
195 	 * between reading the free space cache and updating the csum tree.
196 	 */
197 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
198 		path->search_commit_root = 1;
199 		path->skip_locking = 1;
200 	}
201 
202 	disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
203 	if (dio)
204 		offset = logical_offset;
205 
206 	bio_for_each_segment(bvec, bio, iter) {
207 		page_bytes_left = bvec.bv_len;
208 		if (count)
209 			goto next;
210 
211 		if (!dio)
212 			offset = page_offset(bvec.bv_page) + bvec.bv_offset;
213 		count = btrfs_find_ordered_sum(inode, offset, disk_bytenr,
214 					       (u32 *)csum, nblocks);
215 		if (count)
216 			goto found;
217 
218 		if (!item || disk_bytenr < item_start_offset ||
219 		    disk_bytenr >= item_last_offset) {
220 			struct btrfs_key found_key;
221 			u32 item_size;
222 
223 			if (item)
224 				btrfs_release_path(path);
225 			item = btrfs_lookup_csum(NULL, fs_info->csum_root,
226 						 path, disk_bytenr, 0);
227 			if (IS_ERR(item)) {
228 				count = 1;
229 				memset(csum, 0, csum_size);
230 				if (BTRFS_I(inode)->root->root_key.objectid ==
231 				    BTRFS_DATA_RELOC_TREE_OBJECTID) {
232 					set_extent_bits(io_tree, offset,
233 						offset + fs_info->sectorsize - 1,
234 						EXTENT_NODATASUM);
235 				} else {
236 					btrfs_info_rl(fs_info,
237 						   "no csum found for inode %llu start %llu",
238 					       btrfs_ino(BTRFS_I(inode)), offset);
239 				}
240 				item = NULL;
241 				btrfs_release_path(path);
242 				goto found;
243 			}
244 			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
245 					      path->slots[0]);
246 
247 			item_start_offset = found_key.offset;
248 			item_size = btrfs_item_size_nr(path->nodes[0],
249 						       path->slots[0]);
250 			item_last_offset = item_start_offset +
251 				(item_size / csum_size) *
252 				fs_info->sectorsize;
253 			item = btrfs_item_ptr(path->nodes[0], path->slots[0],
254 					      struct btrfs_csum_item);
255 		}
256 		/*
257 		 * this byte range must be able to fit inside
258 		 * a single leaf so it will also fit inside a u32
259 		 */
260 		diff = disk_bytenr - item_start_offset;
261 		diff = diff / fs_info->sectorsize;
262 		diff = diff * csum_size;
263 		count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
264 					    inode->i_sb->s_blocksize_bits);
265 		read_extent_buffer(path->nodes[0], csum,
266 				   ((unsigned long)item) + diff,
267 				   csum_size * count);
268 found:
269 		csum += count * csum_size;
270 		nblocks -= count;
271 next:
272 		while (count--) {
273 			disk_bytenr += fs_info->sectorsize;
274 			offset += fs_info->sectorsize;
275 			page_bytes_left -= fs_info->sectorsize;
276 			if (!page_bytes_left)
277 				break; /* move to next bio */
278 		}
279 	}
280 
281 	WARN_ON_ONCE(count);
282 	btrfs_free_path(path);
283 	return 0;
284 }
285 
286 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u32 *dst)
287 {
288 	return __btrfs_lookup_bio_sums(inode, bio, 0, dst, 0);
289 }
290 
291 blk_status_t btrfs_lookup_bio_sums_dio(struct inode *inode, struct bio *bio, u64 offset)
292 {
293 	return __btrfs_lookup_bio_sums(inode, bio, offset, NULL, 1);
294 }
295 
296 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
297 			     struct list_head *list, int search_commit)
298 {
299 	struct btrfs_fs_info *fs_info = root->fs_info;
300 	struct btrfs_key key;
301 	struct btrfs_path *path;
302 	struct extent_buffer *leaf;
303 	struct btrfs_ordered_sum *sums;
304 	struct btrfs_csum_item *item;
305 	LIST_HEAD(tmplist);
306 	unsigned long offset;
307 	int ret;
308 	size_t size;
309 	u64 csum_end;
310 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
311 
312 	ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
313 	       IS_ALIGNED(end + 1, fs_info->sectorsize));
314 
315 	path = btrfs_alloc_path();
316 	if (!path)
317 		return -ENOMEM;
318 
319 	if (search_commit) {
320 		path->skip_locking = 1;
321 		path->reada = READA_FORWARD;
322 		path->search_commit_root = 1;
323 	}
324 
325 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
326 	key.offset = start;
327 	key.type = BTRFS_EXTENT_CSUM_KEY;
328 
329 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
330 	if (ret < 0)
331 		goto fail;
332 	if (ret > 0 && path->slots[0] > 0) {
333 		leaf = path->nodes[0];
334 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
335 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
336 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
337 			offset = (start - key.offset) >>
338 				 fs_info->sb->s_blocksize_bits;
339 			if (offset * csum_size <
340 			    btrfs_item_size_nr(leaf, path->slots[0] - 1))
341 				path->slots[0]--;
342 		}
343 	}
344 
345 	while (start <= end) {
346 		leaf = path->nodes[0];
347 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
348 			ret = btrfs_next_leaf(root, path);
349 			if (ret < 0)
350 				goto fail;
351 			if (ret > 0)
352 				break;
353 			leaf = path->nodes[0];
354 		}
355 
356 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
357 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
358 		    key.type != BTRFS_EXTENT_CSUM_KEY ||
359 		    key.offset > end)
360 			break;
361 
362 		if (key.offset > start)
363 			start = key.offset;
364 
365 		size = btrfs_item_size_nr(leaf, path->slots[0]);
366 		csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
367 		if (csum_end <= start) {
368 			path->slots[0]++;
369 			continue;
370 		}
371 
372 		csum_end = min(csum_end, end + 1);
373 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
374 				      struct btrfs_csum_item);
375 		while (start < csum_end) {
376 			size = min_t(size_t, csum_end - start,
377 				     MAX_ORDERED_SUM_BYTES(fs_info));
378 			sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
379 				       GFP_NOFS);
380 			if (!sums) {
381 				ret = -ENOMEM;
382 				goto fail;
383 			}
384 
385 			sums->bytenr = start;
386 			sums->len = (int)size;
387 
388 			offset = (start - key.offset) >>
389 				fs_info->sb->s_blocksize_bits;
390 			offset *= csum_size;
391 			size >>= fs_info->sb->s_blocksize_bits;
392 
393 			read_extent_buffer(path->nodes[0],
394 					   sums->sums,
395 					   ((unsigned long)item) + offset,
396 					   csum_size * size);
397 
398 			start += fs_info->sectorsize * size;
399 			list_add_tail(&sums->list, &tmplist);
400 		}
401 		path->slots[0]++;
402 	}
403 	ret = 0;
404 fail:
405 	while (ret < 0 && !list_empty(&tmplist)) {
406 		sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
407 		list_del(&sums->list);
408 		kfree(sums);
409 	}
410 	list_splice_tail(&tmplist, list);
411 
412 	btrfs_free_path(path);
413 	return ret;
414 }
415 
416 blk_status_t btrfs_csum_one_bio(struct inode *inode, struct bio *bio,
417 		       u64 file_start, int contig)
418 {
419 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
420 	struct btrfs_ordered_sum *sums;
421 	struct btrfs_ordered_extent *ordered = NULL;
422 	char *data;
423 	struct bvec_iter iter;
424 	struct bio_vec bvec;
425 	int index;
426 	int nr_sectors;
427 	unsigned long total_bytes = 0;
428 	unsigned long this_sum_bytes = 0;
429 	int i;
430 	u64 offset;
431 	unsigned nofs_flag;
432 
433 	nofs_flag = memalloc_nofs_save();
434 	sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
435 		       GFP_KERNEL);
436 	memalloc_nofs_restore(nofs_flag);
437 
438 	if (!sums)
439 		return BLK_STS_RESOURCE;
440 
441 	sums->len = bio->bi_iter.bi_size;
442 	INIT_LIST_HEAD(&sums->list);
443 
444 	if (contig)
445 		offset = file_start;
446 	else
447 		offset = 0; /* shut up gcc */
448 
449 	sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
450 	index = 0;
451 
452 	bio_for_each_segment(bvec, bio, iter) {
453 		if (!contig)
454 			offset = page_offset(bvec.bv_page) + bvec.bv_offset;
455 
456 		if (!ordered) {
457 			ordered = btrfs_lookup_ordered_extent(inode, offset);
458 			BUG_ON(!ordered); /* Logic error */
459 		}
460 
461 		data = kmap_atomic(bvec.bv_page);
462 
463 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
464 						 bvec.bv_len + fs_info->sectorsize
465 						 - 1);
466 
467 		for (i = 0; i < nr_sectors; i++) {
468 			if (offset >= ordered->file_offset + ordered->len ||
469 				offset < ordered->file_offset) {
470 				unsigned long bytes_left;
471 
472 				kunmap_atomic(data);
473 				sums->len = this_sum_bytes;
474 				this_sum_bytes = 0;
475 				btrfs_add_ordered_sum(inode, ordered, sums);
476 				btrfs_put_ordered_extent(ordered);
477 
478 				bytes_left = bio->bi_iter.bi_size - total_bytes;
479 
480 				nofs_flag = memalloc_nofs_save();
481 				sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
482 						      bytes_left), GFP_KERNEL);
483 				memalloc_nofs_restore(nofs_flag);
484 				BUG_ON(!sums); /* -ENOMEM */
485 				sums->len = bytes_left;
486 				ordered = btrfs_lookup_ordered_extent(inode,
487 								offset);
488 				ASSERT(ordered); /* Logic error */
489 				sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
490 					+ total_bytes;
491 				index = 0;
492 
493 				data = kmap_atomic(bvec.bv_page);
494 			}
495 
496 			sums->sums[index] = ~(u32)0;
497 			sums->sums[index]
498 				= btrfs_csum_data(data + bvec.bv_offset
499 						+ (i * fs_info->sectorsize),
500 						sums->sums[index],
501 						fs_info->sectorsize);
502 			btrfs_csum_final(sums->sums[index],
503 					(char *)(sums->sums + index));
504 			index++;
505 			offset += fs_info->sectorsize;
506 			this_sum_bytes += fs_info->sectorsize;
507 			total_bytes += fs_info->sectorsize;
508 		}
509 
510 		kunmap_atomic(data);
511 	}
512 	this_sum_bytes = 0;
513 	btrfs_add_ordered_sum(inode, ordered, sums);
514 	btrfs_put_ordered_extent(ordered);
515 	return 0;
516 }
517 
518 /*
519  * helper function for csum removal, this expects the
520  * key to describe the csum pointed to by the path, and it expects
521  * the csum to overlap the range [bytenr, len]
522  *
523  * The csum should not be entirely contained in the range and the
524  * range should not be entirely contained in the csum.
525  *
526  * This calls btrfs_truncate_item with the correct args based on the
527  * overlap, and fixes up the key as required.
528  */
529 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
530 				       struct btrfs_path *path,
531 				       struct btrfs_key *key,
532 				       u64 bytenr, u64 len)
533 {
534 	struct extent_buffer *leaf;
535 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
536 	u64 csum_end;
537 	u64 end_byte = bytenr + len;
538 	u32 blocksize_bits = fs_info->sb->s_blocksize_bits;
539 
540 	leaf = path->nodes[0];
541 	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
542 	csum_end <<= fs_info->sb->s_blocksize_bits;
543 	csum_end += key->offset;
544 
545 	if (key->offset < bytenr && csum_end <= end_byte) {
546 		/*
547 		 *         [ bytenr - len ]
548 		 *         [   ]
549 		 *   [csum     ]
550 		 *   A simple truncate off the end of the item
551 		 */
552 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
553 		new_size *= csum_size;
554 		btrfs_truncate_item(fs_info, path, new_size, 1);
555 	} else if (key->offset >= bytenr && csum_end > end_byte &&
556 		   end_byte > key->offset) {
557 		/*
558 		 *         [ bytenr - len ]
559 		 *                 [ ]
560 		 *                 [csum     ]
561 		 * we need to truncate from the beginning of the csum
562 		 */
563 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
564 		new_size *= csum_size;
565 
566 		btrfs_truncate_item(fs_info, path, new_size, 0);
567 
568 		key->offset = end_byte;
569 		btrfs_set_item_key_safe(fs_info, path, key);
570 	} else {
571 		BUG();
572 	}
573 }
574 
575 /*
576  * deletes the csum items from the csum tree for a given
577  * range of bytes.
578  */
579 int btrfs_del_csums(struct btrfs_trans_handle *trans,
580 		    struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
581 {
582 	struct btrfs_root *root = fs_info->csum_root;
583 	struct btrfs_path *path;
584 	struct btrfs_key key;
585 	u64 end_byte = bytenr + len;
586 	u64 csum_end;
587 	struct extent_buffer *leaf;
588 	int ret;
589 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
590 	int blocksize_bits = fs_info->sb->s_blocksize_bits;
591 
592 	path = btrfs_alloc_path();
593 	if (!path)
594 		return -ENOMEM;
595 
596 	while (1) {
597 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
598 		key.offset = end_byte - 1;
599 		key.type = BTRFS_EXTENT_CSUM_KEY;
600 
601 		path->leave_spinning = 1;
602 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
603 		if (ret > 0) {
604 			if (path->slots[0] == 0)
605 				break;
606 			path->slots[0]--;
607 		} else if (ret < 0) {
608 			break;
609 		}
610 
611 		leaf = path->nodes[0];
612 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
613 
614 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
615 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
616 			break;
617 		}
618 
619 		if (key.offset >= end_byte)
620 			break;
621 
622 		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
623 		csum_end <<= blocksize_bits;
624 		csum_end += key.offset;
625 
626 		/* this csum ends before we start, we're done */
627 		if (csum_end <= bytenr)
628 			break;
629 
630 		/* delete the entire item, it is inside our range */
631 		if (key.offset >= bytenr && csum_end <= end_byte) {
632 			int del_nr = 1;
633 
634 			/*
635 			 * Check how many csum items preceding this one in this
636 			 * leaf correspond to our range and then delete them all
637 			 * at once.
638 			 */
639 			if (key.offset > bytenr && path->slots[0] > 0) {
640 				int slot = path->slots[0] - 1;
641 
642 				while (slot >= 0) {
643 					struct btrfs_key pk;
644 
645 					btrfs_item_key_to_cpu(leaf, &pk, slot);
646 					if (pk.offset < bytenr ||
647 					    pk.type != BTRFS_EXTENT_CSUM_KEY ||
648 					    pk.objectid !=
649 					    BTRFS_EXTENT_CSUM_OBJECTID)
650 						break;
651 					path->slots[0] = slot;
652 					del_nr++;
653 					key.offset = pk.offset;
654 					slot--;
655 				}
656 			}
657 			ret = btrfs_del_items(trans, root, path,
658 					      path->slots[0], del_nr);
659 			if (ret)
660 				goto out;
661 			if (key.offset == bytenr)
662 				break;
663 		} else if (key.offset < bytenr && csum_end > end_byte) {
664 			unsigned long offset;
665 			unsigned long shift_len;
666 			unsigned long item_offset;
667 			/*
668 			 *        [ bytenr - len ]
669 			 *     [csum                ]
670 			 *
671 			 * Our bytes are in the middle of the csum,
672 			 * we need to split this item and insert a new one.
673 			 *
674 			 * But we can't drop the path because the
675 			 * csum could change, get removed, extended etc.
676 			 *
677 			 * The trick here is the max size of a csum item leaves
678 			 * enough room in the tree block for a single
679 			 * item header.  So, we split the item in place,
680 			 * adding a new header pointing to the existing
681 			 * bytes.  Then we loop around again and we have
682 			 * a nicely formed csum item that we can neatly
683 			 * truncate.
684 			 */
685 			offset = (bytenr - key.offset) >> blocksize_bits;
686 			offset *= csum_size;
687 
688 			shift_len = (len >> blocksize_bits) * csum_size;
689 
690 			item_offset = btrfs_item_ptr_offset(leaf,
691 							    path->slots[0]);
692 
693 			memzero_extent_buffer(leaf, item_offset + offset,
694 					     shift_len);
695 			key.offset = bytenr;
696 
697 			/*
698 			 * btrfs_split_item returns -EAGAIN when the
699 			 * item changed size or key
700 			 */
701 			ret = btrfs_split_item(trans, root, path, &key, offset);
702 			if (ret && ret != -EAGAIN) {
703 				btrfs_abort_transaction(trans, ret);
704 				goto out;
705 			}
706 
707 			key.offset = end_byte - 1;
708 		} else {
709 			truncate_one_csum(fs_info, path, &key, bytenr, len);
710 			if (key.offset < bytenr)
711 				break;
712 		}
713 		btrfs_release_path(path);
714 	}
715 	ret = 0;
716 out:
717 	btrfs_free_path(path);
718 	return ret;
719 }
720 
721 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
722 			   struct btrfs_root *root,
723 			   struct btrfs_ordered_sum *sums)
724 {
725 	struct btrfs_fs_info *fs_info = root->fs_info;
726 	struct btrfs_key file_key;
727 	struct btrfs_key found_key;
728 	struct btrfs_path *path;
729 	struct btrfs_csum_item *item;
730 	struct btrfs_csum_item *item_end;
731 	struct extent_buffer *leaf = NULL;
732 	u64 next_offset;
733 	u64 total_bytes = 0;
734 	u64 csum_offset;
735 	u64 bytenr;
736 	u32 nritems;
737 	u32 ins_size;
738 	int index = 0;
739 	int found_next;
740 	int ret;
741 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
742 
743 	path = btrfs_alloc_path();
744 	if (!path)
745 		return -ENOMEM;
746 again:
747 	next_offset = (u64)-1;
748 	found_next = 0;
749 	bytenr = sums->bytenr + total_bytes;
750 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
751 	file_key.offset = bytenr;
752 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
753 
754 	item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
755 	if (!IS_ERR(item)) {
756 		ret = 0;
757 		leaf = path->nodes[0];
758 		item_end = btrfs_item_ptr(leaf, path->slots[0],
759 					  struct btrfs_csum_item);
760 		item_end = (struct btrfs_csum_item *)((char *)item_end +
761 			   btrfs_item_size_nr(leaf, path->slots[0]));
762 		goto found;
763 	}
764 	ret = PTR_ERR(item);
765 	if (ret != -EFBIG && ret != -ENOENT)
766 		goto fail_unlock;
767 
768 	if (ret == -EFBIG) {
769 		u32 item_size;
770 		/* we found one, but it isn't big enough yet */
771 		leaf = path->nodes[0];
772 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
773 		if ((item_size / csum_size) >=
774 		    MAX_CSUM_ITEMS(fs_info, csum_size)) {
775 			/* already at max size, make a new one */
776 			goto insert;
777 		}
778 	} else {
779 		int slot = path->slots[0] + 1;
780 		/* we didn't find a csum item, insert one */
781 		nritems = btrfs_header_nritems(path->nodes[0]);
782 		if (!nritems || (path->slots[0] >= nritems - 1)) {
783 			ret = btrfs_next_leaf(root, path);
784 			if (ret == 1)
785 				found_next = 1;
786 			if (ret != 0)
787 				goto insert;
788 			slot = path->slots[0];
789 		}
790 		btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
791 		if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
792 		    found_key.type != BTRFS_EXTENT_CSUM_KEY) {
793 			found_next = 1;
794 			goto insert;
795 		}
796 		next_offset = found_key.offset;
797 		found_next = 1;
798 		goto insert;
799 	}
800 
801 	/*
802 	 * at this point, we know the tree has an item, but it isn't big
803 	 * enough yet to put our csum in.  Grow it
804 	 */
805 	btrfs_release_path(path);
806 	ret = btrfs_search_slot(trans, root, &file_key, path,
807 				csum_size, 1);
808 	if (ret < 0)
809 		goto fail_unlock;
810 
811 	if (ret > 0) {
812 		if (path->slots[0] == 0)
813 			goto insert;
814 		path->slots[0]--;
815 	}
816 
817 	leaf = path->nodes[0];
818 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
819 	csum_offset = (bytenr - found_key.offset) >>
820 			fs_info->sb->s_blocksize_bits;
821 
822 	if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
823 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
824 	    csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
825 		goto insert;
826 	}
827 
828 	if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
829 	    csum_size) {
830 		int extend_nr;
831 		u64 tmp;
832 		u32 diff;
833 		u32 free_space;
834 
835 		if (btrfs_leaf_free_space(fs_info, leaf) <
836 				 sizeof(struct btrfs_item) + csum_size * 2)
837 			goto insert;
838 
839 		free_space = btrfs_leaf_free_space(fs_info, leaf) -
840 					 sizeof(struct btrfs_item) - csum_size;
841 		tmp = sums->len - total_bytes;
842 		tmp >>= fs_info->sb->s_blocksize_bits;
843 		WARN_ON(tmp < 1);
844 
845 		extend_nr = max_t(int, 1, (int)tmp);
846 		diff = (csum_offset + extend_nr) * csum_size;
847 		diff = min(diff,
848 			   MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
849 
850 		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
851 		diff = min(free_space, diff);
852 		diff /= csum_size;
853 		diff *= csum_size;
854 
855 		btrfs_extend_item(fs_info, path, diff);
856 		ret = 0;
857 		goto csum;
858 	}
859 
860 insert:
861 	btrfs_release_path(path);
862 	csum_offset = 0;
863 	if (found_next) {
864 		u64 tmp;
865 
866 		tmp = sums->len - total_bytes;
867 		tmp >>= fs_info->sb->s_blocksize_bits;
868 		tmp = min(tmp, (next_offset - file_key.offset) >>
869 					 fs_info->sb->s_blocksize_bits);
870 
871 		tmp = max_t(u64, 1, tmp);
872 		tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
873 		ins_size = csum_size * tmp;
874 	} else {
875 		ins_size = csum_size;
876 	}
877 	path->leave_spinning = 1;
878 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
879 				      ins_size);
880 	path->leave_spinning = 0;
881 	if (ret < 0)
882 		goto fail_unlock;
883 	if (WARN_ON(ret != 0))
884 		goto fail_unlock;
885 	leaf = path->nodes[0];
886 csum:
887 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
888 	item_end = (struct btrfs_csum_item *)((unsigned char *)item +
889 				      btrfs_item_size_nr(leaf, path->slots[0]));
890 	item = (struct btrfs_csum_item *)((unsigned char *)item +
891 					  csum_offset * csum_size);
892 found:
893 	ins_size = (u32)(sums->len - total_bytes) >>
894 		   fs_info->sb->s_blocksize_bits;
895 	ins_size *= csum_size;
896 	ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
897 			      ins_size);
898 	write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
899 			    ins_size);
900 
901 	ins_size /= csum_size;
902 	total_bytes += ins_size * fs_info->sectorsize;
903 	index += ins_size;
904 
905 	btrfs_mark_buffer_dirty(path->nodes[0]);
906 	if (total_bytes < sums->len) {
907 		btrfs_release_path(path);
908 		cond_resched();
909 		goto again;
910 	}
911 out:
912 	btrfs_free_path(path);
913 	return ret;
914 
915 fail_unlock:
916 	goto out;
917 }
918 
919 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
920 				     const struct btrfs_path *path,
921 				     struct btrfs_file_extent_item *fi,
922 				     const bool new_inline,
923 				     struct extent_map *em)
924 {
925 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
926 	struct btrfs_root *root = inode->root;
927 	struct extent_buffer *leaf = path->nodes[0];
928 	const int slot = path->slots[0];
929 	struct btrfs_key key;
930 	u64 extent_start, extent_end;
931 	u64 bytenr;
932 	u8 type = btrfs_file_extent_type(leaf, fi);
933 	int compress_type = btrfs_file_extent_compression(leaf, fi);
934 
935 	em->bdev = fs_info->fs_devices->latest_bdev;
936 	btrfs_item_key_to_cpu(leaf, &key, slot);
937 	extent_start = key.offset;
938 
939 	if (type == BTRFS_FILE_EXTENT_REG ||
940 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
941 		extent_end = extent_start +
942 			btrfs_file_extent_num_bytes(leaf, fi);
943 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
944 		size_t size;
945 		size = btrfs_file_extent_ram_bytes(leaf, fi);
946 		extent_end = ALIGN(extent_start + size,
947 				   fs_info->sectorsize);
948 	}
949 
950 	em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
951 	if (type == BTRFS_FILE_EXTENT_REG ||
952 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
953 		em->start = extent_start;
954 		em->len = extent_end - extent_start;
955 		em->orig_start = extent_start -
956 			btrfs_file_extent_offset(leaf, fi);
957 		em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
958 		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
959 		if (bytenr == 0) {
960 			em->block_start = EXTENT_MAP_HOLE;
961 			return;
962 		}
963 		if (compress_type != BTRFS_COMPRESS_NONE) {
964 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
965 			em->compress_type = compress_type;
966 			em->block_start = bytenr;
967 			em->block_len = em->orig_block_len;
968 		} else {
969 			bytenr += btrfs_file_extent_offset(leaf, fi);
970 			em->block_start = bytenr;
971 			em->block_len = em->len;
972 			if (type == BTRFS_FILE_EXTENT_PREALLOC)
973 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
974 		}
975 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
976 		em->block_start = EXTENT_MAP_INLINE;
977 		em->start = extent_start;
978 		em->len = extent_end - extent_start;
979 		/*
980 		 * Initialize orig_start and block_len with the same values
981 		 * as in inode.c:btrfs_get_extent().
982 		 */
983 		em->orig_start = EXTENT_MAP_HOLE;
984 		em->block_len = (u64)-1;
985 		if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
986 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
987 			em->compress_type = compress_type;
988 		}
989 	} else {
990 		btrfs_err(fs_info,
991 			  "unknown file extent item type %d, inode %llu, offset %llu, "
992 			  "root %llu", type, btrfs_ino(inode), extent_start,
993 			  root->root_key.objectid);
994 	}
995 }
996