xref: /openbmc/linux/fs/btrfs/file-item.c (revision a61127c2)
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 /*
417  * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
418  * @inode:	 Owner of the data inside the bio
419  * @bio:	 Contains the data to be checksummed
420  * @file_start:  offset in file this bio begins to describe
421  * @contig:	 Boolean. If true/1 means all bio vecs in this bio are
422  *		 contiguous and they begin at @file_start in the file. False/0
423  *		 means this bio can contains potentially discontigous bio vecs
424  *		 so the logical offset of each should be calculated separately.
425  */
426 blk_status_t btrfs_csum_one_bio(struct inode *inode, struct bio *bio,
427 		       u64 file_start, int contig)
428 {
429 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
430 	struct btrfs_ordered_sum *sums;
431 	struct btrfs_ordered_extent *ordered = NULL;
432 	char *data;
433 	struct bvec_iter iter;
434 	struct bio_vec bvec;
435 	int index;
436 	int nr_sectors;
437 	unsigned long total_bytes = 0;
438 	unsigned long this_sum_bytes = 0;
439 	int i;
440 	u64 offset;
441 	unsigned nofs_flag;
442 
443 	nofs_flag = memalloc_nofs_save();
444 	sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
445 		       GFP_KERNEL);
446 	memalloc_nofs_restore(nofs_flag);
447 
448 	if (!sums)
449 		return BLK_STS_RESOURCE;
450 
451 	sums->len = bio->bi_iter.bi_size;
452 	INIT_LIST_HEAD(&sums->list);
453 
454 	if (contig)
455 		offset = file_start;
456 	else
457 		offset = 0; /* shut up gcc */
458 
459 	sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
460 	index = 0;
461 
462 	bio_for_each_segment(bvec, bio, iter) {
463 		if (!contig)
464 			offset = page_offset(bvec.bv_page) + bvec.bv_offset;
465 
466 		if (!ordered) {
467 			ordered = btrfs_lookup_ordered_extent(inode, offset);
468 			BUG_ON(!ordered); /* Logic error */
469 		}
470 
471 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
472 						 bvec.bv_len + fs_info->sectorsize
473 						 - 1);
474 
475 		for (i = 0; i < nr_sectors; i++) {
476 			if (offset >= ordered->file_offset + ordered->len ||
477 				offset < ordered->file_offset) {
478 				unsigned long bytes_left;
479 
480 				sums->len = this_sum_bytes;
481 				this_sum_bytes = 0;
482 				btrfs_add_ordered_sum(ordered, sums);
483 				btrfs_put_ordered_extent(ordered);
484 
485 				bytes_left = bio->bi_iter.bi_size - total_bytes;
486 
487 				nofs_flag = memalloc_nofs_save();
488 				sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
489 						      bytes_left), GFP_KERNEL);
490 				memalloc_nofs_restore(nofs_flag);
491 				BUG_ON(!sums); /* -ENOMEM */
492 				sums->len = bytes_left;
493 				ordered = btrfs_lookup_ordered_extent(inode,
494 								offset);
495 				ASSERT(ordered); /* Logic error */
496 				sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
497 					+ total_bytes;
498 				index = 0;
499 			}
500 
501 			sums->sums[index] = ~(u32)0;
502 			data = kmap_atomic(bvec.bv_page);
503 			sums->sums[index]
504 				= btrfs_csum_data(data + bvec.bv_offset
505 						+ (i * fs_info->sectorsize),
506 						sums->sums[index],
507 						fs_info->sectorsize);
508 			kunmap_atomic(data);
509 			btrfs_csum_final(sums->sums[index],
510 					(char *)(sums->sums + index));
511 			index++;
512 			offset += fs_info->sectorsize;
513 			this_sum_bytes += fs_info->sectorsize;
514 			total_bytes += fs_info->sectorsize;
515 		}
516 
517 	}
518 	this_sum_bytes = 0;
519 	btrfs_add_ordered_sum(ordered, sums);
520 	btrfs_put_ordered_extent(ordered);
521 	return 0;
522 }
523 
524 /*
525  * helper function for csum removal, this expects the
526  * key to describe the csum pointed to by the path, and it expects
527  * the csum to overlap the range [bytenr, len]
528  *
529  * The csum should not be entirely contained in the range and the
530  * range should not be entirely contained in the csum.
531  *
532  * This calls btrfs_truncate_item with the correct args based on the
533  * overlap, and fixes up the key as required.
534  */
535 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
536 				       struct btrfs_path *path,
537 				       struct btrfs_key *key,
538 				       u64 bytenr, u64 len)
539 {
540 	struct extent_buffer *leaf;
541 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
542 	u64 csum_end;
543 	u64 end_byte = bytenr + len;
544 	u32 blocksize_bits = fs_info->sb->s_blocksize_bits;
545 
546 	leaf = path->nodes[0];
547 	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
548 	csum_end <<= fs_info->sb->s_blocksize_bits;
549 	csum_end += key->offset;
550 
551 	if (key->offset < bytenr && csum_end <= end_byte) {
552 		/*
553 		 *         [ bytenr - len ]
554 		 *         [   ]
555 		 *   [csum     ]
556 		 *   A simple truncate off the end of the item
557 		 */
558 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
559 		new_size *= csum_size;
560 		btrfs_truncate_item(path, new_size, 1);
561 	} else if (key->offset >= bytenr && csum_end > end_byte &&
562 		   end_byte > key->offset) {
563 		/*
564 		 *         [ bytenr - len ]
565 		 *                 [ ]
566 		 *                 [csum     ]
567 		 * we need to truncate from the beginning of the csum
568 		 */
569 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
570 		new_size *= csum_size;
571 
572 		btrfs_truncate_item(path, new_size, 0);
573 
574 		key->offset = end_byte;
575 		btrfs_set_item_key_safe(fs_info, path, key);
576 	} else {
577 		BUG();
578 	}
579 }
580 
581 /*
582  * deletes the csum items from the csum tree for a given
583  * range of bytes.
584  */
585 int btrfs_del_csums(struct btrfs_trans_handle *trans,
586 		    struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
587 {
588 	struct btrfs_root *root = fs_info->csum_root;
589 	struct btrfs_path *path;
590 	struct btrfs_key key;
591 	u64 end_byte = bytenr + len;
592 	u64 csum_end;
593 	struct extent_buffer *leaf;
594 	int ret;
595 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
596 	int blocksize_bits = fs_info->sb->s_blocksize_bits;
597 
598 	path = btrfs_alloc_path();
599 	if (!path)
600 		return -ENOMEM;
601 
602 	while (1) {
603 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
604 		key.offset = end_byte - 1;
605 		key.type = BTRFS_EXTENT_CSUM_KEY;
606 
607 		path->leave_spinning = 1;
608 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
609 		if (ret > 0) {
610 			if (path->slots[0] == 0)
611 				break;
612 			path->slots[0]--;
613 		} else if (ret < 0) {
614 			break;
615 		}
616 
617 		leaf = path->nodes[0];
618 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
619 
620 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
621 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
622 			break;
623 		}
624 
625 		if (key.offset >= end_byte)
626 			break;
627 
628 		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
629 		csum_end <<= blocksize_bits;
630 		csum_end += key.offset;
631 
632 		/* this csum ends before we start, we're done */
633 		if (csum_end <= bytenr)
634 			break;
635 
636 		/* delete the entire item, it is inside our range */
637 		if (key.offset >= bytenr && csum_end <= end_byte) {
638 			int del_nr = 1;
639 
640 			/*
641 			 * Check how many csum items preceding this one in this
642 			 * leaf correspond to our range and then delete them all
643 			 * at once.
644 			 */
645 			if (key.offset > bytenr && path->slots[0] > 0) {
646 				int slot = path->slots[0] - 1;
647 
648 				while (slot >= 0) {
649 					struct btrfs_key pk;
650 
651 					btrfs_item_key_to_cpu(leaf, &pk, slot);
652 					if (pk.offset < bytenr ||
653 					    pk.type != BTRFS_EXTENT_CSUM_KEY ||
654 					    pk.objectid !=
655 					    BTRFS_EXTENT_CSUM_OBJECTID)
656 						break;
657 					path->slots[0] = slot;
658 					del_nr++;
659 					key.offset = pk.offset;
660 					slot--;
661 				}
662 			}
663 			ret = btrfs_del_items(trans, root, path,
664 					      path->slots[0], del_nr);
665 			if (ret)
666 				goto out;
667 			if (key.offset == bytenr)
668 				break;
669 		} else if (key.offset < bytenr && csum_end > end_byte) {
670 			unsigned long offset;
671 			unsigned long shift_len;
672 			unsigned long item_offset;
673 			/*
674 			 *        [ bytenr - len ]
675 			 *     [csum                ]
676 			 *
677 			 * Our bytes are in the middle of the csum,
678 			 * we need to split this item and insert a new one.
679 			 *
680 			 * But we can't drop the path because the
681 			 * csum could change, get removed, extended etc.
682 			 *
683 			 * The trick here is the max size of a csum item leaves
684 			 * enough room in the tree block for a single
685 			 * item header.  So, we split the item in place,
686 			 * adding a new header pointing to the existing
687 			 * bytes.  Then we loop around again and we have
688 			 * a nicely formed csum item that we can neatly
689 			 * truncate.
690 			 */
691 			offset = (bytenr - key.offset) >> blocksize_bits;
692 			offset *= csum_size;
693 
694 			shift_len = (len >> blocksize_bits) * csum_size;
695 
696 			item_offset = btrfs_item_ptr_offset(leaf,
697 							    path->slots[0]);
698 
699 			memzero_extent_buffer(leaf, item_offset + offset,
700 					     shift_len);
701 			key.offset = bytenr;
702 
703 			/*
704 			 * btrfs_split_item returns -EAGAIN when the
705 			 * item changed size or key
706 			 */
707 			ret = btrfs_split_item(trans, root, path, &key, offset);
708 			if (ret && ret != -EAGAIN) {
709 				btrfs_abort_transaction(trans, ret);
710 				goto out;
711 			}
712 
713 			key.offset = end_byte - 1;
714 		} else {
715 			truncate_one_csum(fs_info, path, &key, bytenr, len);
716 			if (key.offset < bytenr)
717 				break;
718 		}
719 		btrfs_release_path(path);
720 	}
721 	ret = 0;
722 out:
723 	btrfs_free_path(path);
724 	return ret;
725 }
726 
727 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
728 			   struct btrfs_root *root,
729 			   struct btrfs_ordered_sum *sums)
730 {
731 	struct btrfs_fs_info *fs_info = root->fs_info;
732 	struct btrfs_key file_key;
733 	struct btrfs_key found_key;
734 	struct btrfs_path *path;
735 	struct btrfs_csum_item *item;
736 	struct btrfs_csum_item *item_end;
737 	struct extent_buffer *leaf = NULL;
738 	u64 next_offset;
739 	u64 total_bytes = 0;
740 	u64 csum_offset;
741 	u64 bytenr;
742 	u32 nritems;
743 	u32 ins_size;
744 	int index = 0;
745 	int found_next;
746 	int ret;
747 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
748 
749 	path = btrfs_alloc_path();
750 	if (!path)
751 		return -ENOMEM;
752 again:
753 	next_offset = (u64)-1;
754 	found_next = 0;
755 	bytenr = sums->bytenr + total_bytes;
756 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
757 	file_key.offset = bytenr;
758 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
759 
760 	item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
761 	if (!IS_ERR(item)) {
762 		ret = 0;
763 		leaf = path->nodes[0];
764 		item_end = btrfs_item_ptr(leaf, path->slots[0],
765 					  struct btrfs_csum_item);
766 		item_end = (struct btrfs_csum_item *)((char *)item_end +
767 			   btrfs_item_size_nr(leaf, path->slots[0]));
768 		goto found;
769 	}
770 	ret = PTR_ERR(item);
771 	if (ret != -EFBIG && ret != -ENOENT)
772 		goto fail_unlock;
773 
774 	if (ret == -EFBIG) {
775 		u32 item_size;
776 		/* we found one, but it isn't big enough yet */
777 		leaf = path->nodes[0];
778 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
779 		if ((item_size / csum_size) >=
780 		    MAX_CSUM_ITEMS(fs_info, csum_size)) {
781 			/* already at max size, make a new one */
782 			goto insert;
783 		}
784 	} else {
785 		int slot = path->slots[0] + 1;
786 		/* we didn't find a csum item, insert one */
787 		nritems = btrfs_header_nritems(path->nodes[0]);
788 		if (!nritems || (path->slots[0] >= nritems - 1)) {
789 			ret = btrfs_next_leaf(root, path);
790 			if (ret == 1)
791 				found_next = 1;
792 			if (ret != 0)
793 				goto insert;
794 			slot = path->slots[0];
795 		}
796 		btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
797 		if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
798 		    found_key.type != BTRFS_EXTENT_CSUM_KEY) {
799 			found_next = 1;
800 			goto insert;
801 		}
802 		next_offset = found_key.offset;
803 		found_next = 1;
804 		goto insert;
805 	}
806 
807 	/*
808 	 * at this point, we know the tree has an item, but it isn't big
809 	 * enough yet to put our csum in.  Grow it
810 	 */
811 	btrfs_release_path(path);
812 	ret = btrfs_search_slot(trans, root, &file_key, path,
813 				csum_size, 1);
814 	if (ret < 0)
815 		goto fail_unlock;
816 
817 	if (ret > 0) {
818 		if (path->slots[0] == 0)
819 			goto insert;
820 		path->slots[0]--;
821 	}
822 
823 	leaf = path->nodes[0];
824 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
825 	csum_offset = (bytenr - found_key.offset) >>
826 			fs_info->sb->s_blocksize_bits;
827 
828 	if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
829 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
830 	    csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
831 		goto insert;
832 	}
833 
834 	if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
835 	    csum_size) {
836 		int extend_nr;
837 		u64 tmp;
838 		u32 diff;
839 		u32 free_space;
840 
841 		if (btrfs_leaf_free_space(leaf) <
842 				 sizeof(struct btrfs_item) + csum_size * 2)
843 			goto insert;
844 
845 		free_space = btrfs_leaf_free_space(leaf) -
846 					 sizeof(struct btrfs_item) - csum_size;
847 		tmp = sums->len - total_bytes;
848 		tmp >>= fs_info->sb->s_blocksize_bits;
849 		WARN_ON(tmp < 1);
850 
851 		extend_nr = max_t(int, 1, (int)tmp);
852 		diff = (csum_offset + extend_nr) * csum_size;
853 		diff = min(diff,
854 			   MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
855 
856 		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
857 		diff = min(free_space, diff);
858 		diff /= csum_size;
859 		diff *= csum_size;
860 
861 		btrfs_extend_item(path, diff);
862 		ret = 0;
863 		goto csum;
864 	}
865 
866 insert:
867 	btrfs_release_path(path);
868 	csum_offset = 0;
869 	if (found_next) {
870 		u64 tmp;
871 
872 		tmp = sums->len - total_bytes;
873 		tmp >>= fs_info->sb->s_blocksize_bits;
874 		tmp = min(tmp, (next_offset - file_key.offset) >>
875 					 fs_info->sb->s_blocksize_bits);
876 
877 		tmp = max_t(u64, 1, tmp);
878 		tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
879 		ins_size = csum_size * tmp;
880 	} else {
881 		ins_size = csum_size;
882 	}
883 	path->leave_spinning = 1;
884 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
885 				      ins_size);
886 	path->leave_spinning = 0;
887 	if (ret < 0)
888 		goto fail_unlock;
889 	if (WARN_ON(ret != 0))
890 		goto fail_unlock;
891 	leaf = path->nodes[0];
892 csum:
893 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
894 	item_end = (struct btrfs_csum_item *)((unsigned char *)item +
895 				      btrfs_item_size_nr(leaf, path->slots[0]));
896 	item = (struct btrfs_csum_item *)((unsigned char *)item +
897 					  csum_offset * csum_size);
898 found:
899 	ins_size = (u32)(sums->len - total_bytes) >>
900 		   fs_info->sb->s_blocksize_bits;
901 	ins_size *= csum_size;
902 	ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
903 			      ins_size);
904 	write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
905 			    ins_size);
906 
907 	ins_size /= csum_size;
908 	total_bytes += ins_size * fs_info->sectorsize;
909 	index += ins_size;
910 
911 	btrfs_mark_buffer_dirty(path->nodes[0]);
912 	if (total_bytes < sums->len) {
913 		btrfs_release_path(path);
914 		cond_resched();
915 		goto again;
916 	}
917 out:
918 	btrfs_free_path(path);
919 	return ret;
920 
921 fail_unlock:
922 	goto out;
923 }
924 
925 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
926 				     const struct btrfs_path *path,
927 				     struct btrfs_file_extent_item *fi,
928 				     const bool new_inline,
929 				     struct extent_map *em)
930 {
931 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
932 	struct btrfs_root *root = inode->root;
933 	struct extent_buffer *leaf = path->nodes[0];
934 	const int slot = path->slots[0];
935 	struct btrfs_key key;
936 	u64 extent_start, extent_end;
937 	u64 bytenr;
938 	u8 type = btrfs_file_extent_type(leaf, fi);
939 	int compress_type = btrfs_file_extent_compression(leaf, fi);
940 
941 	em->bdev = fs_info->fs_devices->latest_bdev;
942 	btrfs_item_key_to_cpu(leaf, &key, slot);
943 	extent_start = key.offset;
944 
945 	if (type == BTRFS_FILE_EXTENT_REG ||
946 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
947 		extent_end = extent_start +
948 			btrfs_file_extent_num_bytes(leaf, fi);
949 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
950 		size_t size;
951 		size = btrfs_file_extent_ram_bytes(leaf, fi);
952 		extent_end = ALIGN(extent_start + size,
953 				   fs_info->sectorsize);
954 	}
955 
956 	em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
957 	if (type == BTRFS_FILE_EXTENT_REG ||
958 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
959 		em->start = extent_start;
960 		em->len = extent_end - extent_start;
961 		em->orig_start = extent_start -
962 			btrfs_file_extent_offset(leaf, fi);
963 		em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
964 		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
965 		if (bytenr == 0) {
966 			em->block_start = EXTENT_MAP_HOLE;
967 			return;
968 		}
969 		if (compress_type != BTRFS_COMPRESS_NONE) {
970 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
971 			em->compress_type = compress_type;
972 			em->block_start = bytenr;
973 			em->block_len = em->orig_block_len;
974 		} else {
975 			bytenr += btrfs_file_extent_offset(leaf, fi);
976 			em->block_start = bytenr;
977 			em->block_len = em->len;
978 			if (type == BTRFS_FILE_EXTENT_PREALLOC)
979 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
980 		}
981 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
982 		em->block_start = EXTENT_MAP_INLINE;
983 		em->start = extent_start;
984 		em->len = extent_end - extent_start;
985 		/*
986 		 * Initialize orig_start and block_len with the same values
987 		 * as in inode.c:btrfs_get_extent().
988 		 */
989 		em->orig_start = EXTENT_MAP_HOLE;
990 		em->block_len = (u64)-1;
991 		if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
992 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
993 			em->compress_type = compress_type;
994 		}
995 	} else {
996 		btrfs_err(fs_info,
997 			  "unknown file extent item type %d, inode %llu, offset %llu, "
998 			  "root %llu", type, btrfs_ino(inode), extent_start,
999 			  root->root_key.objectid);
1000 	}
1001 }
1002