xref: /openbmc/linux/fs/btrfs/file-item.c (revision 4bce6fce)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/pagemap.h>
22 #include <linux/highmem.h>
23 #include "ctree.h"
24 #include "disk-io.h"
25 #include "transaction.h"
26 #include "volumes.h"
27 #include "print-tree.h"
28 
29 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
30 				   sizeof(struct btrfs_item) * 2) / \
31 				  size) - 1))
32 
33 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
34 				       PAGE_CACHE_SIZE))
35 
36 #define MAX_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
37 				   sizeof(struct btrfs_ordered_sum)) / \
38 				   sizeof(u32) * (r)->sectorsize)
39 
40 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
41 			     struct btrfs_root *root,
42 			     u64 objectid, u64 pos,
43 			     u64 disk_offset, u64 disk_num_bytes,
44 			     u64 num_bytes, u64 offset, u64 ram_bytes,
45 			     u8 compression, u8 encryption, u16 other_encoding)
46 {
47 	int ret = 0;
48 	struct btrfs_file_extent_item *item;
49 	struct btrfs_key file_key;
50 	struct btrfs_path *path;
51 	struct extent_buffer *leaf;
52 
53 	path = btrfs_alloc_path();
54 	if (!path)
55 		return -ENOMEM;
56 	file_key.objectid = objectid;
57 	file_key.offset = pos;
58 	file_key.type = BTRFS_EXTENT_DATA_KEY;
59 
60 	path->leave_spinning = 1;
61 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
62 				      sizeof(*item));
63 	if (ret < 0)
64 		goto out;
65 	BUG_ON(ret); /* Can't happen */
66 	leaf = path->nodes[0];
67 	item = btrfs_item_ptr(leaf, path->slots[0],
68 			      struct btrfs_file_extent_item);
69 	btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
70 	btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
71 	btrfs_set_file_extent_offset(leaf, item, offset);
72 	btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
73 	btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
74 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
75 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
76 	btrfs_set_file_extent_compression(leaf, item, compression);
77 	btrfs_set_file_extent_encryption(leaf, item, encryption);
78 	btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
79 
80 	btrfs_mark_buffer_dirty(leaf);
81 out:
82 	btrfs_free_path(path);
83 	return ret;
84 }
85 
86 static struct btrfs_csum_item *
87 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
88 		  struct btrfs_root *root,
89 		  struct btrfs_path *path,
90 		  u64 bytenr, int cow)
91 {
92 	int ret;
93 	struct btrfs_key file_key;
94 	struct btrfs_key found_key;
95 	struct btrfs_csum_item *item;
96 	struct extent_buffer *leaf;
97 	u64 csum_offset = 0;
98 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
99 	int csums_in_item;
100 
101 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
102 	file_key.offset = bytenr;
103 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
104 	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
105 	if (ret < 0)
106 		goto fail;
107 	leaf = path->nodes[0];
108 	if (ret > 0) {
109 		ret = 1;
110 		if (path->slots[0] == 0)
111 			goto fail;
112 		path->slots[0]--;
113 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
114 		if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
115 			goto fail;
116 
117 		csum_offset = (bytenr - found_key.offset) >>
118 				root->fs_info->sb->s_blocksize_bits;
119 		csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
120 		csums_in_item /= csum_size;
121 
122 		if (csum_offset == csums_in_item) {
123 			ret = -EFBIG;
124 			goto fail;
125 		} else if (csum_offset > csums_in_item) {
126 			goto fail;
127 		}
128 	}
129 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
130 	item = (struct btrfs_csum_item *)((unsigned char *)item +
131 					  csum_offset * csum_size);
132 	return item;
133 fail:
134 	if (ret > 0)
135 		ret = -ENOENT;
136 	return ERR_PTR(ret);
137 }
138 
139 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
140 			     struct btrfs_root *root,
141 			     struct btrfs_path *path, u64 objectid,
142 			     u64 offset, int mod)
143 {
144 	int ret;
145 	struct btrfs_key file_key;
146 	int ins_len = mod < 0 ? -1 : 0;
147 	int cow = mod != 0;
148 
149 	file_key.objectid = objectid;
150 	file_key.offset = offset;
151 	file_key.type = BTRFS_EXTENT_DATA_KEY;
152 	ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
153 	return ret;
154 }
155 
156 static void btrfs_io_bio_endio_readpage(struct btrfs_io_bio *bio, int err)
157 {
158 	kfree(bio->csum_allocated);
159 }
160 
161 static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
162 				   struct inode *inode, struct bio *bio,
163 				   u64 logical_offset, u32 *dst, int dio)
164 {
165 	struct bio_vec *bvec = bio->bi_io_vec;
166 	struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
167 	struct btrfs_csum_item *item = NULL;
168 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
169 	struct btrfs_path *path;
170 	u8 *csum;
171 	u64 offset = 0;
172 	u64 item_start_offset = 0;
173 	u64 item_last_offset = 0;
174 	u64 disk_bytenr;
175 	u32 diff;
176 	int nblocks;
177 	int bio_index = 0;
178 	int count;
179 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
180 
181 	path = btrfs_alloc_path();
182 	if (!path)
183 		return -ENOMEM;
184 
185 	nblocks = bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
186 	if (!dst) {
187 		if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
188 			btrfs_bio->csum_allocated = kmalloc_array(nblocks,
189 					csum_size, GFP_NOFS);
190 			if (!btrfs_bio->csum_allocated) {
191 				btrfs_free_path(path);
192 				return -ENOMEM;
193 			}
194 			btrfs_bio->csum = btrfs_bio->csum_allocated;
195 			btrfs_bio->end_io = btrfs_io_bio_endio_readpage;
196 		} else {
197 			btrfs_bio->csum = btrfs_bio->csum_inline;
198 		}
199 		csum = btrfs_bio->csum;
200 	} else {
201 		csum = (u8 *)dst;
202 	}
203 
204 	if (bio->bi_iter.bi_size > PAGE_CACHE_SIZE * 8)
205 		path->reada = 2;
206 
207 	WARN_ON(bio->bi_vcnt <= 0);
208 
209 	/*
210 	 * the free space stuff is only read when it hasn't been
211 	 * updated in the current transaction.  So, we can safely
212 	 * read from the commit root and sidestep a nasty deadlock
213 	 * between reading the free space cache and updating the csum tree.
214 	 */
215 	if (btrfs_is_free_space_inode(inode)) {
216 		path->search_commit_root = 1;
217 		path->skip_locking = 1;
218 	}
219 
220 	disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
221 	if (dio)
222 		offset = logical_offset;
223 	while (bio_index < bio->bi_vcnt) {
224 		if (!dio)
225 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
226 		count = btrfs_find_ordered_sum(inode, offset, disk_bytenr,
227 					       (u32 *)csum, nblocks);
228 		if (count)
229 			goto found;
230 
231 		if (!item || disk_bytenr < item_start_offset ||
232 		    disk_bytenr >= item_last_offset) {
233 			struct btrfs_key found_key;
234 			u32 item_size;
235 
236 			if (item)
237 				btrfs_release_path(path);
238 			item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
239 						 path, disk_bytenr, 0);
240 			if (IS_ERR(item)) {
241 				count = 1;
242 				memset(csum, 0, csum_size);
243 				if (BTRFS_I(inode)->root->root_key.objectid ==
244 				    BTRFS_DATA_RELOC_TREE_OBJECTID) {
245 					set_extent_bits(io_tree, offset,
246 						offset + bvec->bv_len - 1,
247 						EXTENT_NODATASUM, GFP_NOFS);
248 				} else {
249 					btrfs_info(BTRFS_I(inode)->root->fs_info,
250 						   "no csum found for inode %llu start %llu",
251 					       btrfs_ino(inode), offset);
252 				}
253 				item = NULL;
254 				btrfs_release_path(path);
255 				goto found;
256 			}
257 			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
258 					      path->slots[0]);
259 
260 			item_start_offset = found_key.offset;
261 			item_size = btrfs_item_size_nr(path->nodes[0],
262 						       path->slots[0]);
263 			item_last_offset = item_start_offset +
264 				(item_size / csum_size) *
265 				root->sectorsize;
266 			item = btrfs_item_ptr(path->nodes[0], path->slots[0],
267 					      struct btrfs_csum_item);
268 		}
269 		/*
270 		 * this byte range must be able to fit inside
271 		 * a single leaf so it will also fit inside a u32
272 		 */
273 		diff = disk_bytenr - item_start_offset;
274 		diff = diff / root->sectorsize;
275 		diff = diff * csum_size;
276 		count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
277 					    inode->i_sb->s_blocksize_bits);
278 		read_extent_buffer(path->nodes[0], csum,
279 				   ((unsigned long)item) + diff,
280 				   csum_size * count);
281 found:
282 		csum += count * csum_size;
283 		nblocks -= count;
284 		bio_index += count;
285 		while (count--) {
286 			disk_bytenr += bvec->bv_len;
287 			offset += bvec->bv_len;
288 			bvec++;
289 		}
290 	}
291 	btrfs_free_path(path);
292 	return 0;
293 }
294 
295 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
296 			  struct bio *bio, u32 *dst)
297 {
298 	return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
299 }
300 
301 int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
302 			      struct bio *bio, u64 offset)
303 {
304 	return __btrfs_lookup_bio_sums(root, inode, bio, offset, NULL, 1);
305 }
306 
307 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
308 			     struct list_head *list, int search_commit)
309 {
310 	struct btrfs_key key;
311 	struct btrfs_path *path;
312 	struct extent_buffer *leaf;
313 	struct btrfs_ordered_sum *sums;
314 	struct btrfs_csum_item *item;
315 	LIST_HEAD(tmplist);
316 	unsigned long offset;
317 	int ret;
318 	size_t size;
319 	u64 csum_end;
320 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
321 
322 	ASSERT(IS_ALIGNED(start, root->sectorsize) &&
323 	       IS_ALIGNED(end + 1, root->sectorsize));
324 
325 	path = btrfs_alloc_path();
326 	if (!path)
327 		return -ENOMEM;
328 
329 	if (search_commit) {
330 		path->skip_locking = 1;
331 		path->reada = 2;
332 		path->search_commit_root = 1;
333 	}
334 
335 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
336 	key.offset = start;
337 	key.type = BTRFS_EXTENT_CSUM_KEY;
338 
339 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
340 	if (ret < 0)
341 		goto fail;
342 	if (ret > 0 && path->slots[0] > 0) {
343 		leaf = path->nodes[0];
344 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
345 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
346 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
347 			offset = (start - key.offset) >>
348 				 root->fs_info->sb->s_blocksize_bits;
349 			if (offset * csum_size <
350 			    btrfs_item_size_nr(leaf, path->slots[0] - 1))
351 				path->slots[0]--;
352 		}
353 	}
354 
355 	while (start <= end) {
356 		leaf = path->nodes[0];
357 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
358 			ret = btrfs_next_leaf(root, path);
359 			if (ret < 0)
360 				goto fail;
361 			if (ret > 0)
362 				break;
363 			leaf = path->nodes[0];
364 		}
365 
366 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
367 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
368 		    key.type != BTRFS_EXTENT_CSUM_KEY ||
369 		    key.offset > end)
370 			break;
371 
372 		if (key.offset > start)
373 			start = key.offset;
374 
375 		size = btrfs_item_size_nr(leaf, path->slots[0]);
376 		csum_end = key.offset + (size / csum_size) * root->sectorsize;
377 		if (csum_end <= start) {
378 			path->slots[0]++;
379 			continue;
380 		}
381 
382 		csum_end = min(csum_end, end + 1);
383 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
384 				      struct btrfs_csum_item);
385 		while (start < csum_end) {
386 			size = min_t(size_t, csum_end - start,
387 				     MAX_ORDERED_SUM_BYTES(root));
388 			sums = kzalloc(btrfs_ordered_sum_size(root, size),
389 				       GFP_NOFS);
390 			if (!sums) {
391 				ret = -ENOMEM;
392 				goto fail;
393 			}
394 
395 			sums->bytenr = start;
396 			sums->len = (int)size;
397 
398 			offset = (start - key.offset) >>
399 				root->fs_info->sb->s_blocksize_bits;
400 			offset *= csum_size;
401 			size >>= root->fs_info->sb->s_blocksize_bits;
402 
403 			read_extent_buffer(path->nodes[0],
404 					   sums->sums,
405 					   ((unsigned long)item) + offset,
406 					   csum_size * size);
407 
408 			start += root->sectorsize * size;
409 			list_add_tail(&sums->list, &tmplist);
410 		}
411 		path->slots[0]++;
412 	}
413 	ret = 0;
414 fail:
415 	while (ret < 0 && !list_empty(&tmplist)) {
416 		sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
417 		list_del(&sums->list);
418 		kfree(sums);
419 	}
420 	list_splice_tail(&tmplist, list);
421 
422 	btrfs_free_path(path);
423 	return ret;
424 }
425 
426 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
427 		       struct bio *bio, u64 file_start, int contig)
428 {
429 	struct btrfs_ordered_sum *sums;
430 	struct btrfs_ordered_extent *ordered;
431 	char *data;
432 	struct bio_vec *bvec = bio->bi_io_vec;
433 	int bio_index = 0;
434 	int index;
435 	unsigned long total_bytes = 0;
436 	unsigned long this_sum_bytes = 0;
437 	u64 offset;
438 
439 	WARN_ON(bio->bi_vcnt <= 0);
440 	sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_iter.bi_size),
441 		       GFP_NOFS);
442 	if (!sums)
443 		return -ENOMEM;
444 
445 	sums->len = bio->bi_iter.bi_size;
446 	INIT_LIST_HEAD(&sums->list);
447 
448 	if (contig)
449 		offset = file_start;
450 	else
451 		offset = page_offset(bvec->bv_page) + bvec->bv_offset;
452 
453 	ordered = btrfs_lookup_ordered_extent(inode, offset);
454 	BUG_ON(!ordered); /* Logic error */
455 	sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
456 	index = 0;
457 
458 	while (bio_index < bio->bi_vcnt) {
459 		if (!contig)
460 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
461 
462 		if (offset >= ordered->file_offset + ordered->len ||
463 		    offset < ordered->file_offset) {
464 			unsigned long bytes_left;
465 			sums->len = this_sum_bytes;
466 			this_sum_bytes = 0;
467 			btrfs_add_ordered_sum(inode, ordered, sums);
468 			btrfs_put_ordered_extent(ordered);
469 
470 			bytes_left = bio->bi_iter.bi_size - total_bytes;
471 
472 			sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
473 				       GFP_NOFS);
474 			BUG_ON(!sums); /* -ENOMEM */
475 			sums->len = bytes_left;
476 			ordered = btrfs_lookup_ordered_extent(inode, offset);
477 			BUG_ON(!ordered); /* Logic error */
478 			sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9) +
479 				       total_bytes;
480 			index = 0;
481 		}
482 
483 		data = kmap_atomic(bvec->bv_page);
484 		sums->sums[index] = ~(u32)0;
485 		sums->sums[index] = btrfs_csum_data(data + bvec->bv_offset,
486 						    sums->sums[index],
487 						    bvec->bv_len);
488 		kunmap_atomic(data);
489 		btrfs_csum_final(sums->sums[index],
490 				 (char *)(sums->sums + index));
491 
492 		bio_index++;
493 		index++;
494 		total_bytes += bvec->bv_len;
495 		this_sum_bytes += bvec->bv_len;
496 		offset += bvec->bv_len;
497 		bvec++;
498 	}
499 	this_sum_bytes = 0;
500 	btrfs_add_ordered_sum(inode, ordered, sums);
501 	btrfs_put_ordered_extent(ordered);
502 	return 0;
503 }
504 
505 /*
506  * helper function for csum removal, this expects the
507  * key to describe the csum pointed to by the path, and it expects
508  * the csum to overlap the range [bytenr, len]
509  *
510  * The csum should not be entirely contained in the range and the
511  * range should not be entirely contained in the csum.
512  *
513  * This calls btrfs_truncate_item with the correct args based on the
514  * overlap, and fixes up the key as required.
515  */
516 static noinline void truncate_one_csum(struct btrfs_root *root,
517 				       struct btrfs_path *path,
518 				       struct btrfs_key *key,
519 				       u64 bytenr, u64 len)
520 {
521 	struct extent_buffer *leaf;
522 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
523 	u64 csum_end;
524 	u64 end_byte = bytenr + len;
525 	u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
526 
527 	leaf = path->nodes[0];
528 	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
529 	csum_end <<= root->fs_info->sb->s_blocksize_bits;
530 	csum_end += key->offset;
531 
532 	if (key->offset < bytenr && csum_end <= end_byte) {
533 		/*
534 		 *         [ bytenr - len ]
535 		 *         [   ]
536 		 *   [csum     ]
537 		 *   A simple truncate off the end of the item
538 		 */
539 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
540 		new_size *= csum_size;
541 		btrfs_truncate_item(root, path, new_size, 1);
542 	} else if (key->offset >= bytenr && csum_end > end_byte &&
543 		   end_byte > key->offset) {
544 		/*
545 		 *         [ bytenr - len ]
546 		 *                 [ ]
547 		 *                 [csum     ]
548 		 * we need to truncate from the beginning of the csum
549 		 */
550 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
551 		new_size *= csum_size;
552 
553 		btrfs_truncate_item(root, path, new_size, 0);
554 
555 		key->offset = end_byte;
556 		btrfs_set_item_key_safe(root->fs_info, path, key);
557 	} else {
558 		BUG();
559 	}
560 }
561 
562 /*
563  * deletes the csum items from the csum tree for a given
564  * range of bytes.
565  */
566 int btrfs_del_csums(struct btrfs_trans_handle *trans,
567 		    struct btrfs_root *root, u64 bytenr, u64 len)
568 {
569 	struct btrfs_path *path;
570 	struct btrfs_key key;
571 	u64 end_byte = bytenr + len;
572 	u64 csum_end;
573 	struct extent_buffer *leaf;
574 	int ret;
575 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
576 	int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
577 
578 	root = root->fs_info->csum_root;
579 
580 	path = btrfs_alloc_path();
581 	if (!path)
582 		return -ENOMEM;
583 
584 	while (1) {
585 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
586 		key.offset = end_byte - 1;
587 		key.type = BTRFS_EXTENT_CSUM_KEY;
588 
589 		path->leave_spinning = 1;
590 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
591 		if (ret > 0) {
592 			if (path->slots[0] == 0)
593 				break;
594 			path->slots[0]--;
595 		} else if (ret < 0) {
596 			break;
597 		}
598 
599 		leaf = path->nodes[0];
600 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
601 
602 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
603 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
604 			break;
605 		}
606 
607 		if (key.offset >= end_byte)
608 			break;
609 
610 		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
611 		csum_end <<= blocksize_bits;
612 		csum_end += key.offset;
613 
614 		/* this csum ends before we start, we're done */
615 		if (csum_end <= bytenr)
616 			break;
617 
618 		/* delete the entire item, it is inside our range */
619 		if (key.offset >= bytenr && csum_end <= end_byte) {
620 			ret = btrfs_del_item(trans, root, path);
621 			if (ret)
622 				goto out;
623 			if (key.offset == bytenr)
624 				break;
625 		} else if (key.offset < bytenr && csum_end > end_byte) {
626 			unsigned long offset;
627 			unsigned long shift_len;
628 			unsigned long item_offset;
629 			/*
630 			 *        [ bytenr - len ]
631 			 *     [csum                ]
632 			 *
633 			 * Our bytes are in the middle of the csum,
634 			 * we need to split this item and insert a new one.
635 			 *
636 			 * But we can't drop the path because the
637 			 * csum could change, get removed, extended etc.
638 			 *
639 			 * The trick here is the max size of a csum item leaves
640 			 * enough room in the tree block for a single
641 			 * item header.  So, we split the item in place,
642 			 * adding a new header pointing to the existing
643 			 * bytes.  Then we loop around again and we have
644 			 * a nicely formed csum item that we can neatly
645 			 * truncate.
646 			 */
647 			offset = (bytenr - key.offset) >> blocksize_bits;
648 			offset *= csum_size;
649 
650 			shift_len = (len >> blocksize_bits) * csum_size;
651 
652 			item_offset = btrfs_item_ptr_offset(leaf,
653 							    path->slots[0]);
654 
655 			memset_extent_buffer(leaf, 0, item_offset + offset,
656 					     shift_len);
657 			key.offset = bytenr;
658 
659 			/*
660 			 * btrfs_split_item returns -EAGAIN when the
661 			 * item changed size or key
662 			 */
663 			ret = btrfs_split_item(trans, root, path, &key, offset);
664 			if (ret && ret != -EAGAIN) {
665 				btrfs_abort_transaction(trans, root, ret);
666 				goto out;
667 			}
668 
669 			key.offset = end_byte - 1;
670 		} else {
671 			truncate_one_csum(root, path, &key, bytenr, len);
672 			if (key.offset < bytenr)
673 				break;
674 		}
675 		btrfs_release_path(path);
676 	}
677 	ret = 0;
678 out:
679 	btrfs_free_path(path);
680 	return ret;
681 }
682 
683 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
684 			   struct btrfs_root *root,
685 			   struct btrfs_ordered_sum *sums)
686 {
687 	struct btrfs_key file_key;
688 	struct btrfs_key found_key;
689 	struct btrfs_path *path;
690 	struct btrfs_csum_item *item;
691 	struct btrfs_csum_item *item_end;
692 	struct extent_buffer *leaf = NULL;
693 	u64 next_offset;
694 	u64 total_bytes = 0;
695 	u64 csum_offset;
696 	u64 bytenr;
697 	u32 nritems;
698 	u32 ins_size;
699 	int index = 0;
700 	int found_next;
701 	int ret;
702 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
703 
704 	path = btrfs_alloc_path();
705 	if (!path)
706 		return -ENOMEM;
707 again:
708 	next_offset = (u64)-1;
709 	found_next = 0;
710 	bytenr = sums->bytenr + total_bytes;
711 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
712 	file_key.offset = bytenr;
713 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
714 
715 	item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
716 	if (!IS_ERR(item)) {
717 		ret = 0;
718 		leaf = path->nodes[0];
719 		item_end = btrfs_item_ptr(leaf, path->slots[0],
720 					  struct btrfs_csum_item);
721 		item_end = (struct btrfs_csum_item *)((char *)item_end +
722 			   btrfs_item_size_nr(leaf, path->slots[0]));
723 		goto found;
724 	}
725 	ret = PTR_ERR(item);
726 	if (ret != -EFBIG && ret != -ENOENT)
727 		goto fail_unlock;
728 
729 	if (ret == -EFBIG) {
730 		u32 item_size;
731 		/* we found one, but it isn't big enough yet */
732 		leaf = path->nodes[0];
733 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
734 		if ((item_size / csum_size) >=
735 		    MAX_CSUM_ITEMS(root, csum_size)) {
736 			/* already at max size, make a new one */
737 			goto insert;
738 		}
739 	} else {
740 		int slot = path->slots[0] + 1;
741 		/* we didn't find a csum item, insert one */
742 		nritems = btrfs_header_nritems(path->nodes[0]);
743 		if (!nritems || (path->slots[0] >= nritems - 1)) {
744 			ret = btrfs_next_leaf(root, path);
745 			if (ret == 1)
746 				found_next = 1;
747 			if (ret != 0)
748 				goto insert;
749 			slot = path->slots[0];
750 		}
751 		btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
752 		if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
753 		    found_key.type != BTRFS_EXTENT_CSUM_KEY) {
754 			found_next = 1;
755 			goto insert;
756 		}
757 		next_offset = found_key.offset;
758 		found_next = 1;
759 		goto insert;
760 	}
761 
762 	/*
763 	 * at this point, we know the tree has an item, but it isn't big
764 	 * enough yet to put our csum in.  Grow it
765 	 */
766 	btrfs_release_path(path);
767 	ret = btrfs_search_slot(trans, root, &file_key, path,
768 				csum_size, 1);
769 	if (ret < 0)
770 		goto fail_unlock;
771 
772 	if (ret > 0) {
773 		if (path->slots[0] == 0)
774 			goto insert;
775 		path->slots[0]--;
776 	}
777 
778 	leaf = path->nodes[0];
779 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
780 	csum_offset = (bytenr - found_key.offset) >>
781 			root->fs_info->sb->s_blocksize_bits;
782 
783 	if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
784 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
785 	    csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
786 		goto insert;
787 	}
788 
789 	if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
790 	    csum_size) {
791 		int extend_nr;
792 		u64 tmp;
793 		u32 diff;
794 		u32 free_space;
795 
796 		if (btrfs_leaf_free_space(root, leaf) <
797 				 sizeof(struct btrfs_item) + csum_size * 2)
798 			goto insert;
799 
800 		free_space = btrfs_leaf_free_space(root, leaf) -
801 					 sizeof(struct btrfs_item) - csum_size;
802 		tmp = sums->len - total_bytes;
803 		tmp >>= root->fs_info->sb->s_blocksize_bits;
804 		WARN_ON(tmp < 1);
805 
806 		extend_nr = max_t(int, 1, (int)tmp);
807 		diff = (csum_offset + extend_nr) * csum_size;
808 		diff = min(diff, MAX_CSUM_ITEMS(root, csum_size) * csum_size);
809 
810 		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
811 		diff = min(free_space, diff);
812 		diff /= csum_size;
813 		diff *= csum_size;
814 
815 		btrfs_extend_item(root, path, diff);
816 		ret = 0;
817 		goto csum;
818 	}
819 
820 insert:
821 	btrfs_release_path(path);
822 	csum_offset = 0;
823 	if (found_next) {
824 		u64 tmp;
825 
826 		tmp = sums->len - total_bytes;
827 		tmp >>= root->fs_info->sb->s_blocksize_bits;
828 		tmp = min(tmp, (next_offset - file_key.offset) >>
829 					 root->fs_info->sb->s_blocksize_bits);
830 
831 		tmp = max((u64)1, tmp);
832 		tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
833 		ins_size = csum_size * tmp;
834 	} else {
835 		ins_size = csum_size;
836 	}
837 	path->leave_spinning = 1;
838 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
839 				      ins_size);
840 	path->leave_spinning = 0;
841 	if (ret < 0)
842 		goto fail_unlock;
843 	if (WARN_ON(ret != 0))
844 		goto fail_unlock;
845 	leaf = path->nodes[0];
846 csum:
847 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
848 	item_end = (struct btrfs_csum_item *)((unsigned char *)item +
849 				      btrfs_item_size_nr(leaf, path->slots[0]));
850 	item = (struct btrfs_csum_item *)((unsigned char *)item +
851 					  csum_offset * csum_size);
852 found:
853 	ins_size = (u32)(sums->len - total_bytes) >>
854 		   root->fs_info->sb->s_blocksize_bits;
855 	ins_size *= csum_size;
856 	ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
857 			      ins_size);
858 	write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
859 			    ins_size);
860 
861 	ins_size /= csum_size;
862 	total_bytes += ins_size * root->sectorsize;
863 	index += ins_size;
864 
865 	btrfs_mark_buffer_dirty(path->nodes[0]);
866 	if (total_bytes < sums->len) {
867 		btrfs_release_path(path);
868 		cond_resched();
869 		goto again;
870 	}
871 out:
872 	btrfs_free_path(path);
873 	return ret;
874 
875 fail_unlock:
876 	goto out;
877 }
878 
879 void btrfs_extent_item_to_extent_map(struct inode *inode,
880 				     const struct btrfs_path *path,
881 				     struct btrfs_file_extent_item *fi,
882 				     const bool new_inline,
883 				     struct extent_map *em)
884 {
885 	struct btrfs_root *root = BTRFS_I(inode)->root;
886 	struct extent_buffer *leaf = path->nodes[0];
887 	const int slot = path->slots[0];
888 	struct btrfs_key key;
889 	u64 extent_start, extent_end;
890 	u64 bytenr;
891 	u8 type = btrfs_file_extent_type(leaf, fi);
892 	int compress_type = btrfs_file_extent_compression(leaf, fi);
893 
894 	em->bdev = root->fs_info->fs_devices->latest_bdev;
895 	btrfs_item_key_to_cpu(leaf, &key, slot);
896 	extent_start = key.offset;
897 
898 	if (type == BTRFS_FILE_EXTENT_REG ||
899 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
900 		extent_end = extent_start +
901 			btrfs_file_extent_num_bytes(leaf, fi);
902 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
903 		size_t size;
904 		size = btrfs_file_extent_inline_len(leaf, slot, fi);
905 		extent_end = ALIGN(extent_start + size, root->sectorsize);
906 	}
907 
908 	em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
909 	if (type == BTRFS_FILE_EXTENT_REG ||
910 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
911 		em->start = extent_start;
912 		em->len = extent_end - extent_start;
913 		em->orig_start = extent_start -
914 			btrfs_file_extent_offset(leaf, fi);
915 		em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
916 		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
917 		if (bytenr == 0) {
918 			em->block_start = EXTENT_MAP_HOLE;
919 			return;
920 		}
921 		if (compress_type != BTRFS_COMPRESS_NONE) {
922 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
923 			em->compress_type = compress_type;
924 			em->block_start = bytenr;
925 			em->block_len = em->orig_block_len;
926 		} else {
927 			bytenr += btrfs_file_extent_offset(leaf, fi);
928 			em->block_start = bytenr;
929 			em->block_len = em->len;
930 			if (type == BTRFS_FILE_EXTENT_PREALLOC)
931 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
932 		}
933 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
934 		em->block_start = EXTENT_MAP_INLINE;
935 		em->start = extent_start;
936 		em->len = extent_end - extent_start;
937 		/*
938 		 * Initialize orig_start and block_len with the same values
939 		 * as in inode.c:btrfs_get_extent().
940 		 */
941 		em->orig_start = EXTENT_MAP_HOLE;
942 		em->block_len = (u64)-1;
943 		if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
944 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
945 			em->compress_type = compress_type;
946 		}
947 	} else {
948 		btrfs_err(root->fs_info,
949 			  "unknown file extent item type %d, inode %llu, offset %llu, root %llu",
950 			  type, btrfs_ino(inode), extent_start,
951 			  root->root_key.objectid);
952 	}
953 }
954