xref: /openbmc/linux/fs/btrfs/file-item.c (revision db181ce0)
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 	btrfs_set_key_type(&file_key, 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 	btrfs_set_key_type(&file_key, 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 (btrfs_key_type(&found_key) != 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 	btrfs_set_key_type(&file_key, 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(nblocks * csum_size,
189 							    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 btrfs_dio_private *dip, struct bio *bio,
303 			      u64 offset)
304 {
305 	int len = (bio->bi_iter.bi_sector << 9) - dip->disk_bytenr;
306 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
307 	int ret;
308 
309 	len >>= inode->i_sb->s_blocksize_bits;
310 	len *= csum_size;
311 
312 	ret = __btrfs_lookup_bio_sums(root, inode, bio, offset,
313 				      (u32 *)(dip->csum + len), 1);
314 	return ret;
315 }
316 
317 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
318 			     struct list_head *list, int search_commit)
319 {
320 	struct btrfs_key key;
321 	struct btrfs_path *path;
322 	struct extent_buffer *leaf;
323 	struct btrfs_ordered_sum *sums;
324 	struct btrfs_csum_item *item;
325 	LIST_HEAD(tmplist);
326 	unsigned long offset;
327 	int ret;
328 	size_t size;
329 	u64 csum_end;
330 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
331 
332 	ASSERT(start == ALIGN(start, root->sectorsize) &&
333 	       (end + 1) == ALIGN(end + 1, root->sectorsize));
334 
335 	path = btrfs_alloc_path();
336 	if (!path)
337 		return -ENOMEM;
338 
339 	if (search_commit) {
340 		path->skip_locking = 1;
341 		path->reada = 2;
342 		path->search_commit_root = 1;
343 	}
344 
345 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
346 	key.offset = start;
347 	key.type = BTRFS_EXTENT_CSUM_KEY;
348 
349 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
350 	if (ret < 0)
351 		goto fail;
352 	if (ret > 0 && path->slots[0] > 0) {
353 		leaf = path->nodes[0];
354 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
355 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
356 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
357 			offset = (start - key.offset) >>
358 				 root->fs_info->sb->s_blocksize_bits;
359 			if (offset * csum_size <
360 			    btrfs_item_size_nr(leaf, path->slots[0] - 1))
361 				path->slots[0]--;
362 		}
363 	}
364 
365 	while (start <= end) {
366 		leaf = path->nodes[0];
367 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
368 			ret = btrfs_next_leaf(root, path);
369 			if (ret < 0)
370 				goto fail;
371 			if (ret > 0)
372 				break;
373 			leaf = path->nodes[0];
374 		}
375 
376 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
377 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
378 		    key.type != BTRFS_EXTENT_CSUM_KEY ||
379 		    key.offset > end)
380 			break;
381 
382 		if (key.offset > start)
383 			start = key.offset;
384 
385 		size = btrfs_item_size_nr(leaf, path->slots[0]);
386 		csum_end = key.offset + (size / csum_size) * root->sectorsize;
387 		if (csum_end <= start) {
388 			path->slots[0]++;
389 			continue;
390 		}
391 
392 		csum_end = min(csum_end, end + 1);
393 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
394 				      struct btrfs_csum_item);
395 		while (start < csum_end) {
396 			size = min_t(size_t, csum_end - start,
397 				     MAX_ORDERED_SUM_BYTES(root));
398 			sums = kzalloc(btrfs_ordered_sum_size(root, size),
399 				       GFP_NOFS);
400 			if (!sums) {
401 				ret = -ENOMEM;
402 				goto fail;
403 			}
404 
405 			sums->bytenr = start;
406 			sums->len = (int)size;
407 
408 			offset = (start - key.offset) >>
409 				root->fs_info->sb->s_blocksize_bits;
410 			offset *= csum_size;
411 			size >>= root->fs_info->sb->s_blocksize_bits;
412 
413 			read_extent_buffer(path->nodes[0],
414 					   sums->sums,
415 					   ((unsigned long)item) + offset,
416 					   csum_size * size);
417 
418 			start += root->sectorsize * size;
419 			list_add_tail(&sums->list, &tmplist);
420 		}
421 		path->slots[0]++;
422 	}
423 	ret = 0;
424 fail:
425 	while (ret < 0 && !list_empty(&tmplist)) {
426 		sums = list_entry(&tmplist, struct btrfs_ordered_sum, list);
427 		list_del(&sums->list);
428 		kfree(sums);
429 	}
430 	list_splice_tail(&tmplist, list);
431 
432 	btrfs_free_path(path);
433 	return ret;
434 }
435 
436 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
437 		       struct bio *bio, u64 file_start, int contig)
438 {
439 	struct btrfs_ordered_sum *sums;
440 	struct btrfs_ordered_extent *ordered;
441 	char *data;
442 	struct bio_vec *bvec = bio->bi_io_vec;
443 	int bio_index = 0;
444 	int index;
445 	unsigned long total_bytes = 0;
446 	unsigned long this_sum_bytes = 0;
447 	u64 offset;
448 
449 	WARN_ON(bio->bi_vcnt <= 0);
450 	sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_iter.bi_size),
451 		       GFP_NOFS);
452 	if (!sums)
453 		return -ENOMEM;
454 
455 	sums->len = bio->bi_iter.bi_size;
456 	INIT_LIST_HEAD(&sums->list);
457 
458 	if (contig)
459 		offset = file_start;
460 	else
461 		offset = page_offset(bvec->bv_page) + bvec->bv_offset;
462 
463 	ordered = btrfs_lookup_ordered_extent(inode, offset);
464 	BUG_ON(!ordered); /* Logic error */
465 	sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
466 	index = 0;
467 
468 	while (bio_index < bio->bi_vcnt) {
469 		if (!contig)
470 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
471 
472 		if (offset >= ordered->file_offset + ordered->len ||
473 		    offset < ordered->file_offset) {
474 			unsigned long bytes_left;
475 			sums->len = this_sum_bytes;
476 			this_sum_bytes = 0;
477 			btrfs_add_ordered_sum(inode, ordered, sums);
478 			btrfs_put_ordered_extent(ordered);
479 
480 			bytes_left = bio->bi_iter.bi_size - total_bytes;
481 
482 			sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
483 				       GFP_NOFS);
484 			BUG_ON(!sums); /* -ENOMEM */
485 			sums->len = bytes_left;
486 			ordered = btrfs_lookup_ordered_extent(inode, offset);
487 			BUG_ON(!ordered); /* Logic error */
488 			sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9) +
489 				       total_bytes;
490 			index = 0;
491 		}
492 
493 		data = kmap_atomic(bvec->bv_page);
494 		sums->sums[index] = ~(u32)0;
495 		sums->sums[index] = btrfs_csum_data(data + bvec->bv_offset,
496 						    sums->sums[index],
497 						    bvec->bv_len);
498 		kunmap_atomic(data);
499 		btrfs_csum_final(sums->sums[index],
500 				 (char *)(sums->sums + index));
501 
502 		bio_index++;
503 		index++;
504 		total_bytes += bvec->bv_len;
505 		this_sum_bytes += bvec->bv_len;
506 		offset += bvec->bv_len;
507 		bvec++;
508 	}
509 	this_sum_bytes = 0;
510 	btrfs_add_ordered_sum(inode, ordered, sums);
511 	btrfs_put_ordered_extent(ordered);
512 	return 0;
513 }
514 
515 /*
516  * helper function for csum removal, this expects the
517  * key to describe the csum pointed to by the path, and it expects
518  * the csum to overlap the range [bytenr, len]
519  *
520  * The csum should not be entirely contained in the range and the
521  * range should not be entirely contained in the csum.
522  *
523  * This calls btrfs_truncate_item with the correct args based on the
524  * overlap, and fixes up the key as required.
525  */
526 static noinline void truncate_one_csum(struct btrfs_root *root,
527 				       struct btrfs_path *path,
528 				       struct btrfs_key *key,
529 				       u64 bytenr, u64 len)
530 {
531 	struct extent_buffer *leaf;
532 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
533 	u64 csum_end;
534 	u64 end_byte = bytenr + len;
535 	u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
536 
537 	leaf = path->nodes[0];
538 	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
539 	csum_end <<= root->fs_info->sb->s_blocksize_bits;
540 	csum_end += key->offset;
541 
542 	if (key->offset < bytenr && csum_end <= end_byte) {
543 		/*
544 		 *         [ bytenr - len ]
545 		 *         [   ]
546 		 *   [csum     ]
547 		 *   A simple truncate off the end of the item
548 		 */
549 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
550 		new_size *= csum_size;
551 		btrfs_truncate_item(root, path, new_size, 1);
552 	} else if (key->offset >= bytenr && csum_end > end_byte &&
553 		   end_byte > key->offset) {
554 		/*
555 		 *         [ bytenr - len ]
556 		 *                 [ ]
557 		 *                 [csum     ]
558 		 * we need to truncate from the beginning of the csum
559 		 */
560 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
561 		new_size *= csum_size;
562 
563 		btrfs_truncate_item(root, path, new_size, 0);
564 
565 		key->offset = end_byte;
566 		btrfs_set_item_key_safe(root, path, key);
567 	} else {
568 		BUG();
569 	}
570 }
571 
572 /*
573  * deletes the csum items from the csum tree for a given
574  * range of bytes.
575  */
576 int btrfs_del_csums(struct btrfs_trans_handle *trans,
577 		    struct btrfs_root *root, u64 bytenr, u64 len)
578 {
579 	struct btrfs_path *path;
580 	struct btrfs_key key;
581 	u64 end_byte = bytenr + len;
582 	u64 csum_end;
583 	struct extent_buffer *leaf;
584 	int ret;
585 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
586 	int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
587 
588 	root = root->fs_info->csum_root;
589 
590 	path = btrfs_alloc_path();
591 	if (!path)
592 		return -ENOMEM;
593 
594 	while (1) {
595 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
596 		key.offset = end_byte - 1;
597 		key.type = BTRFS_EXTENT_CSUM_KEY;
598 
599 		path->leave_spinning = 1;
600 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
601 		if (ret > 0) {
602 			if (path->slots[0] == 0)
603 				break;
604 			path->slots[0]--;
605 		} else if (ret < 0) {
606 			break;
607 		}
608 
609 		leaf = path->nodes[0];
610 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
611 
612 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
613 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
614 			break;
615 		}
616 
617 		if (key.offset >= end_byte)
618 			break;
619 
620 		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
621 		csum_end <<= blocksize_bits;
622 		csum_end += key.offset;
623 
624 		/* this csum ends before we start, we're done */
625 		if (csum_end <= bytenr)
626 			break;
627 
628 		/* delete the entire item, it is inside our range */
629 		if (key.offset >= bytenr && csum_end <= end_byte) {
630 			ret = btrfs_del_item(trans, root, path);
631 			if (ret)
632 				goto out;
633 			if (key.offset == bytenr)
634 				break;
635 		} else if (key.offset < bytenr && csum_end > end_byte) {
636 			unsigned long offset;
637 			unsigned long shift_len;
638 			unsigned long item_offset;
639 			/*
640 			 *        [ bytenr - len ]
641 			 *     [csum                ]
642 			 *
643 			 * Our bytes are in the middle of the csum,
644 			 * we need to split this item and insert a new one.
645 			 *
646 			 * But we can't drop the path because the
647 			 * csum could change, get removed, extended etc.
648 			 *
649 			 * The trick here is the max size of a csum item leaves
650 			 * enough room in the tree block for a single
651 			 * item header.  So, we split the item in place,
652 			 * adding a new header pointing to the existing
653 			 * bytes.  Then we loop around again and we have
654 			 * a nicely formed csum item that we can neatly
655 			 * truncate.
656 			 */
657 			offset = (bytenr - key.offset) >> blocksize_bits;
658 			offset *= csum_size;
659 
660 			shift_len = (len >> blocksize_bits) * csum_size;
661 
662 			item_offset = btrfs_item_ptr_offset(leaf,
663 							    path->slots[0]);
664 
665 			memset_extent_buffer(leaf, 0, item_offset + offset,
666 					     shift_len);
667 			key.offset = bytenr;
668 
669 			/*
670 			 * btrfs_split_item returns -EAGAIN when the
671 			 * item changed size or key
672 			 */
673 			ret = btrfs_split_item(trans, root, path, &key, offset);
674 			if (ret && ret != -EAGAIN) {
675 				btrfs_abort_transaction(trans, root, ret);
676 				goto out;
677 			}
678 
679 			key.offset = end_byte - 1;
680 		} else {
681 			truncate_one_csum(root, path, &key, bytenr, len);
682 			if (key.offset < bytenr)
683 				break;
684 		}
685 		btrfs_release_path(path);
686 	}
687 	ret = 0;
688 out:
689 	btrfs_free_path(path);
690 	return ret;
691 }
692 
693 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
694 			   struct btrfs_root *root,
695 			   struct btrfs_ordered_sum *sums)
696 {
697 	struct btrfs_key file_key;
698 	struct btrfs_key found_key;
699 	struct btrfs_path *path;
700 	struct btrfs_csum_item *item;
701 	struct btrfs_csum_item *item_end;
702 	struct extent_buffer *leaf = NULL;
703 	u64 next_offset;
704 	u64 total_bytes = 0;
705 	u64 csum_offset;
706 	u64 bytenr;
707 	u32 nritems;
708 	u32 ins_size;
709 	int index = 0;
710 	int found_next;
711 	int ret;
712 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
713 
714 	path = btrfs_alloc_path();
715 	if (!path)
716 		return -ENOMEM;
717 again:
718 	next_offset = (u64)-1;
719 	found_next = 0;
720 	bytenr = sums->bytenr + total_bytes;
721 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
722 	file_key.offset = bytenr;
723 	btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
724 
725 	item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
726 	if (!IS_ERR(item)) {
727 		ret = 0;
728 		leaf = path->nodes[0];
729 		item_end = btrfs_item_ptr(leaf, path->slots[0],
730 					  struct btrfs_csum_item);
731 		item_end = (struct btrfs_csum_item *)((char *)item_end +
732 			   btrfs_item_size_nr(leaf, path->slots[0]));
733 		goto found;
734 	}
735 	ret = PTR_ERR(item);
736 	if (ret != -EFBIG && ret != -ENOENT)
737 		goto fail_unlock;
738 
739 	if (ret == -EFBIG) {
740 		u32 item_size;
741 		/* we found one, but it isn't big enough yet */
742 		leaf = path->nodes[0];
743 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
744 		if ((item_size / csum_size) >=
745 		    MAX_CSUM_ITEMS(root, csum_size)) {
746 			/* already at max size, make a new one */
747 			goto insert;
748 		}
749 	} else {
750 		int slot = path->slots[0] + 1;
751 		/* we didn't find a csum item, insert one */
752 		nritems = btrfs_header_nritems(path->nodes[0]);
753 		if (!nritems || (path->slots[0] >= nritems - 1)) {
754 			ret = btrfs_next_leaf(root, path);
755 			if (ret == 1)
756 				found_next = 1;
757 			if (ret != 0)
758 				goto insert;
759 			slot = 0;
760 		}
761 		btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
762 		if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
763 		    found_key.type != BTRFS_EXTENT_CSUM_KEY) {
764 			found_next = 1;
765 			goto insert;
766 		}
767 		next_offset = found_key.offset;
768 		found_next = 1;
769 		goto insert;
770 	}
771 
772 	/*
773 	 * at this point, we know the tree has an item, but it isn't big
774 	 * enough yet to put our csum in.  Grow it
775 	 */
776 	btrfs_release_path(path);
777 	ret = btrfs_search_slot(trans, root, &file_key, path,
778 				csum_size, 1);
779 	if (ret < 0)
780 		goto fail_unlock;
781 
782 	if (ret > 0) {
783 		if (path->slots[0] == 0)
784 			goto insert;
785 		path->slots[0]--;
786 	}
787 
788 	leaf = path->nodes[0];
789 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
790 	csum_offset = (bytenr - found_key.offset) >>
791 			root->fs_info->sb->s_blocksize_bits;
792 
793 	if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
794 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
795 	    csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
796 		goto insert;
797 	}
798 
799 	if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
800 	    csum_size) {
801 		int extend_nr;
802 		u64 tmp;
803 		u32 diff;
804 		u32 free_space;
805 
806 		if (btrfs_leaf_free_space(root, leaf) <
807 				 sizeof(struct btrfs_item) + csum_size * 2)
808 			goto insert;
809 
810 		free_space = btrfs_leaf_free_space(root, leaf) -
811 					 sizeof(struct btrfs_item) - csum_size;
812 		tmp = sums->len - total_bytes;
813 		tmp >>= root->fs_info->sb->s_blocksize_bits;
814 		WARN_ON(tmp < 1);
815 
816 		extend_nr = max_t(int, 1, (int)tmp);
817 		diff = (csum_offset + extend_nr) * csum_size;
818 		diff = min(diff, MAX_CSUM_ITEMS(root, csum_size) * csum_size);
819 
820 		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
821 		diff = min(free_space, diff);
822 		diff /= csum_size;
823 		diff *= csum_size;
824 
825 		btrfs_extend_item(root, path, diff);
826 		ret = 0;
827 		goto csum;
828 	}
829 
830 insert:
831 	btrfs_release_path(path);
832 	csum_offset = 0;
833 	if (found_next) {
834 		u64 tmp;
835 
836 		tmp = sums->len - total_bytes;
837 		tmp >>= root->fs_info->sb->s_blocksize_bits;
838 		tmp = min(tmp, (next_offset - file_key.offset) >>
839 					 root->fs_info->sb->s_blocksize_bits);
840 
841 		tmp = max((u64)1, tmp);
842 		tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
843 		ins_size = csum_size * tmp;
844 	} else {
845 		ins_size = csum_size;
846 	}
847 	path->leave_spinning = 1;
848 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
849 				      ins_size);
850 	path->leave_spinning = 0;
851 	if (ret < 0)
852 		goto fail_unlock;
853 	if (WARN_ON(ret != 0))
854 		goto fail_unlock;
855 	leaf = path->nodes[0];
856 csum:
857 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
858 	item_end = (struct btrfs_csum_item *)((unsigned char *)item +
859 				      btrfs_item_size_nr(leaf, path->slots[0]));
860 	item = (struct btrfs_csum_item *)((unsigned char *)item +
861 					  csum_offset * csum_size);
862 found:
863 	ins_size = (u32)(sums->len - total_bytes) >>
864 		   root->fs_info->sb->s_blocksize_bits;
865 	ins_size *= csum_size;
866 	ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
867 			      ins_size);
868 	write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
869 			    ins_size);
870 
871 	ins_size /= csum_size;
872 	total_bytes += ins_size * root->sectorsize;
873 	index += ins_size;
874 
875 	btrfs_mark_buffer_dirty(path->nodes[0]);
876 	if (total_bytes < sums->len) {
877 		btrfs_release_path(path);
878 		cond_resched();
879 		goto again;
880 	}
881 out:
882 	btrfs_free_path(path);
883 	return ret;
884 
885 fail_unlock:
886 	goto out;
887 }
888 
889 void btrfs_extent_item_to_extent_map(struct inode *inode,
890 				     const struct btrfs_path *path,
891 				     struct btrfs_file_extent_item *fi,
892 				     const bool new_inline,
893 				     struct extent_map *em)
894 {
895 	struct btrfs_root *root = BTRFS_I(inode)->root;
896 	struct extent_buffer *leaf = path->nodes[0];
897 	const int slot = path->slots[0];
898 	struct btrfs_key key;
899 	u64 extent_start, extent_end;
900 	u64 bytenr;
901 	u8 type = btrfs_file_extent_type(leaf, fi);
902 	int compress_type = btrfs_file_extent_compression(leaf, fi);
903 
904 	em->bdev = root->fs_info->fs_devices->latest_bdev;
905 	btrfs_item_key_to_cpu(leaf, &key, slot);
906 	extent_start = key.offset;
907 
908 	if (type == BTRFS_FILE_EXTENT_REG ||
909 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
910 		extent_end = extent_start +
911 			btrfs_file_extent_num_bytes(leaf, fi);
912 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
913 		size_t size;
914 		size = btrfs_file_extent_inline_len(leaf, slot, fi);
915 		extent_end = ALIGN(extent_start + size, root->sectorsize);
916 	}
917 
918 	em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
919 	if (type == BTRFS_FILE_EXTENT_REG ||
920 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
921 		em->start = extent_start;
922 		em->len = extent_end - extent_start;
923 		em->orig_start = extent_start -
924 			btrfs_file_extent_offset(leaf, fi);
925 		em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
926 		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
927 		if (bytenr == 0) {
928 			em->block_start = EXTENT_MAP_HOLE;
929 			return;
930 		}
931 		if (compress_type != BTRFS_COMPRESS_NONE) {
932 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
933 			em->compress_type = compress_type;
934 			em->block_start = bytenr;
935 			em->block_len = em->orig_block_len;
936 		} else {
937 			bytenr += btrfs_file_extent_offset(leaf, fi);
938 			em->block_start = bytenr;
939 			em->block_len = em->len;
940 			if (type == BTRFS_FILE_EXTENT_PREALLOC)
941 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
942 		}
943 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
944 		em->block_start = EXTENT_MAP_INLINE;
945 		em->start = extent_start;
946 		em->len = extent_end - extent_start;
947 		/*
948 		 * Initialize orig_start and block_len with the same values
949 		 * as in inode.c:btrfs_get_extent().
950 		 */
951 		em->orig_start = EXTENT_MAP_HOLE;
952 		em->block_len = (u64)-1;
953 		if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
954 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
955 			em->compress_type = compress_type;
956 		}
957 	} else {
958 		btrfs_err(root->fs_info,
959 			  "unknown file extent item type %d, inode %llu, offset %llu, root %llu",
960 			  type, btrfs_ino(inode), extent_start,
961 			  root->root_key.objectid);
962 	}
963 }
964