xref: /openbmc/linux/fs/btrfs/file-item.c (revision 82003e04)
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 #include "compression.h"
29 
30 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
31 				   sizeof(struct btrfs_item) * 2) / \
32 				  size) - 1))
33 
34 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
35 				       PAGE_SIZE))
36 
37 #define MAX_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
38 				   sizeof(struct btrfs_ordered_sum)) / \
39 				   sizeof(u32) * (r)->sectorsize)
40 
41 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
42 			     struct btrfs_root *root,
43 			     u64 objectid, u64 pos,
44 			     u64 disk_offset, u64 disk_num_bytes,
45 			     u64 num_bytes, u64 offset, u64 ram_bytes,
46 			     u8 compression, u8 encryption, u16 other_encoding)
47 {
48 	int ret = 0;
49 	struct btrfs_file_extent_item *item;
50 	struct btrfs_key file_key;
51 	struct btrfs_path *path;
52 	struct extent_buffer *leaf;
53 
54 	path = btrfs_alloc_path();
55 	if (!path)
56 		return -ENOMEM;
57 	file_key.objectid = objectid;
58 	file_key.offset = pos;
59 	file_key.type = BTRFS_EXTENT_DATA_KEY;
60 
61 	path->leave_spinning = 1;
62 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
63 				      sizeof(*item));
64 	if (ret < 0)
65 		goto out;
66 	BUG_ON(ret); /* Can't happen */
67 	leaf = path->nodes[0];
68 	item = btrfs_item_ptr(leaf, path->slots[0],
69 			      struct btrfs_file_extent_item);
70 	btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
71 	btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
72 	btrfs_set_file_extent_offset(leaf, item, offset);
73 	btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
74 	btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
75 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
76 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
77 	btrfs_set_file_extent_compression(leaf, item, compression);
78 	btrfs_set_file_extent_encryption(leaf, item, encryption);
79 	btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
80 
81 	btrfs_mark_buffer_dirty(leaf);
82 out:
83 	btrfs_free_path(path);
84 	return ret;
85 }
86 
87 static struct btrfs_csum_item *
88 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
89 		  struct btrfs_root *root,
90 		  struct btrfs_path *path,
91 		  u64 bytenr, int cow)
92 {
93 	int ret;
94 	struct btrfs_key file_key;
95 	struct btrfs_key found_key;
96 	struct btrfs_csum_item *item;
97 	struct extent_buffer *leaf;
98 	u64 csum_offset = 0;
99 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
100 	int csums_in_item;
101 
102 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
103 	file_key.offset = bytenr;
104 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
105 	ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
106 	if (ret < 0)
107 		goto fail;
108 	leaf = path->nodes[0];
109 	if (ret > 0) {
110 		ret = 1;
111 		if (path->slots[0] == 0)
112 			goto fail;
113 		path->slots[0]--;
114 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
115 		if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
116 			goto fail;
117 
118 		csum_offset = (bytenr - found_key.offset) >>
119 				root->fs_info->sb->s_blocksize_bits;
120 		csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
121 		csums_in_item /= csum_size;
122 
123 		if (csum_offset == csums_in_item) {
124 			ret = -EFBIG;
125 			goto fail;
126 		} else if (csum_offset > csums_in_item) {
127 			goto fail;
128 		}
129 	}
130 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
131 	item = (struct btrfs_csum_item *)((unsigned char *)item +
132 					  csum_offset * csum_size);
133 	return item;
134 fail:
135 	if (ret > 0)
136 		ret = -ENOENT;
137 	return ERR_PTR(ret);
138 }
139 
140 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
141 			     struct btrfs_root *root,
142 			     struct btrfs_path *path, u64 objectid,
143 			     u64 offset, int mod)
144 {
145 	int ret;
146 	struct btrfs_key file_key;
147 	int ins_len = mod < 0 ? -1 : 0;
148 	int cow = mod != 0;
149 
150 	file_key.objectid = objectid;
151 	file_key.offset = offset;
152 	file_key.type = BTRFS_EXTENT_DATA_KEY;
153 	ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
154 	return ret;
155 }
156 
157 static void btrfs_io_bio_endio_readpage(struct btrfs_io_bio *bio, int err)
158 {
159 	kfree(bio->csum_allocated);
160 }
161 
162 static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
163 				   struct inode *inode, struct bio *bio,
164 				   u64 logical_offset, u32 *dst, int dio)
165 {
166 	struct bio_vec *bvec = bio->bi_io_vec;
167 	struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
168 	struct btrfs_csum_item *item = NULL;
169 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
170 	struct btrfs_path *path;
171 	u8 *csum;
172 	u64 offset = 0;
173 	u64 item_start_offset = 0;
174 	u64 item_last_offset = 0;
175 	u64 disk_bytenr;
176 	u64 page_bytes_left;
177 	u32 diff;
178 	int nblocks;
179 	int bio_index = 0;
180 	int count;
181 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
182 
183 	path = btrfs_alloc_path();
184 	if (!path)
185 		return -ENOMEM;
186 
187 	nblocks = bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
188 	if (!dst) {
189 		if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
190 			btrfs_bio->csum_allocated = kmalloc_array(nblocks,
191 					csum_size, GFP_NOFS);
192 			if (!btrfs_bio->csum_allocated) {
193 				btrfs_free_path(path);
194 				return -ENOMEM;
195 			}
196 			btrfs_bio->csum = btrfs_bio->csum_allocated;
197 			btrfs_bio->end_io = btrfs_io_bio_endio_readpage;
198 		} else {
199 			btrfs_bio->csum = btrfs_bio->csum_inline;
200 		}
201 		csum = btrfs_bio->csum;
202 	} else {
203 		csum = (u8 *)dst;
204 	}
205 
206 	if (bio->bi_iter.bi_size > PAGE_SIZE * 8)
207 		path->reada = READA_FORWARD;
208 
209 	WARN_ON(bio->bi_vcnt <= 0);
210 
211 	/*
212 	 * the free space stuff is only read when it hasn't been
213 	 * updated in the current transaction.  So, we can safely
214 	 * read from the commit root and sidestep a nasty deadlock
215 	 * between reading the free space cache and updating the csum tree.
216 	 */
217 	if (btrfs_is_free_space_inode(inode)) {
218 		path->search_commit_root = 1;
219 		path->skip_locking = 1;
220 	}
221 
222 	disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
223 	if (dio)
224 		offset = logical_offset;
225 
226 	page_bytes_left = bvec->bv_len;
227 	while (bio_index < bio->bi_vcnt) {
228 		if (!dio)
229 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
230 		count = btrfs_find_ordered_sum(inode, offset, disk_bytenr,
231 					       (u32 *)csum, nblocks);
232 		if (count)
233 			goto found;
234 
235 		if (!item || disk_bytenr < item_start_offset ||
236 		    disk_bytenr >= item_last_offset) {
237 			struct btrfs_key found_key;
238 			u32 item_size;
239 
240 			if (item)
241 				btrfs_release_path(path);
242 			item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
243 						 path, disk_bytenr, 0);
244 			if (IS_ERR(item)) {
245 				count = 1;
246 				memset(csum, 0, csum_size);
247 				if (BTRFS_I(inode)->root->root_key.objectid ==
248 				    BTRFS_DATA_RELOC_TREE_OBJECTID) {
249 					set_extent_bits(io_tree, offset,
250 						offset + root->sectorsize - 1,
251 						EXTENT_NODATASUM);
252 				} else {
253 					btrfs_info_rl(BTRFS_I(inode)->root->fs_info,
254 						   "no csum found for inode %llu start %llu",
255 					       btrfs_ino(inode), offset);
256 				}
257 				item = NULL;
258 				btrfs_release_path(path);
259 				goto found;
260 			}
261 			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
262 					      path->slots[0]);
263 
264 			item_start_offset = found_key.offset;
265 			item_size = btrfs_item_size_nr(path->nodes[0],
266 						       path->slots[0]);
267 			item_last_offset = item_start_offset +
268 				(item_size / csum_size) *
269 				root->sectorsize;
270 			item = btrfs_item_ptr(path->nodes[0], path->slots[0],
271 					      struct btrfs_csum_item);
272 		}
273 		/*
274 		 * this byte range must be able to fit inside
275 		 * a single leaf so it will also fit inside a u32
276 		 */
277 		diff = disk_bytenr - item_start_offset;
278 		diff = diff / root->sectorsize;
279 		diff = diff * csum_size;
280 		count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
281 					    inode->i_sb->s_blocksize_bits);
282 		read_extent_buffer(path->nodes[0], csum,
283 				   ((unsigned long)item) + diff,
284 				   csum_size * count);
285 found:
286 		csum += count * csum_size;
287 		nblocks -= count;
288 
289 		while (count--) {
290 			disk_bytenr += root->sectorsize;
291 			offset += root->sectorsize;
292 			page_bytes_left -= root->sectorsize;
293 			if (!page_bytes_left) {
294 				bio_index++;
295 				/*
296 				 * make sure we're still inside the
297 				 * bio before we update page_bytes_left
298 				 */
299 				if (bio_index >= bio->bi_vcnt) {
300 					WARN_ON_ONCE(count);
301 					goto done;
302 				}
303 				bvec++;
304 				page_bytes_left = bvec->bv_len;
305 			}
306 
307 		}
308 	}
309 
310 done:
311 	btrfs_free_path(path);
312 	return 0;
313 }
314 
315 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
316 			  struct bio *bio, u32 *dst)
317 {
318 	return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
319 }
320 
321 int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
322 			      struct bio *bio, u64 offset)
323 {
324 	return __btrfs_lookup_bio_sums(root, inode, bio, offset, NULL, 1);
325 }
326 
327 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
328 			     struct list_head *list, int search_commit)
329 {
330 	struct btrfs_key key;
331 	struct btrfs_path *path;
332 	struct extent_buffer *leaf;
333 	struct btrfs_ordered_sum *sums;
334 	struct btrfs_csum_item *item;
335 	LIST_HEAD(tmplist);
336 	unsigned long offset;
337 	int ret;
338 	size_t size;
339 	u64 csum_end;
340 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
341 
342 	ASSERT(IS_ALIGNED(start, root->sectorsize) &&
343 	       IS_ALIGNED(end + 1, root->sectorsize));
344 
345 	path = btrfs_alloc_path();
346 	if (!path)
347 		return -ENOMEM;
348 
349 	if (search_commit) {
350 		path->skip_locking = 1;
351 		path->reada = READA_FORWARD;
352 		path->search_commit_root = 1;
353 	}
354 
355 	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
356 	key.offset = start;
357 	key.type = BTRFS_EXTENT_CSUM_KEY;
358 
359 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
360 	if (ret < 0)
361 		goto fail;
362 	if (ret > 0 && path->slots[0] > 0) {
363 		leaf = path->nodes[0];
364 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
365 		if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
366 		    key.type == BTRFS_EXTENT_CSUM_KEY) {
367 			offset = (start - key.offset) >>
368 				 root->fs_info->sb->s_blocksize_bits;
369 			if (offset * csum_size <
370 			    btrfs_item_size_nr(leaf, path->slots[0] - 1))
371 				path->slots[0]--;
372 		}
373 	}
374 
375 	while (start <= end) {
376 		leaf = path->nodes[0];
377 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
378 			ret = btrfs_next_leaf(root, path);
379 			if (ret < 0)
380 				goto fail;
381 			if (ret > 0)
382 				break;
383 			leaf = path->nodes[0];
384 		}
385 
386 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
387 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
388 		    key.type != BTRFS_EXTENT_CSUM_KEY ||
389 		    key.offset > end)
390 			break;
391 
392 		if (key.offset > start)
393 			start = key.offset;
394 
395 		size = btrfs_item_size_nr(leaf, path->slots[0]);
396 		csum_end = key.offset + (size / csum_size) * root->sectorsize;
397 		if (csum_end <= start) {
398 			path->slots[0]++;
399 			continue;
400 		}
401 
402 		csum_end = min(csum_end, end + 1);
403 		item = btrfs_item_ptr(path->nodes[0], path->slots[0],
404 				      struct btrfs_csum_item);
405 		while (start < csum_end) {
406 			size = min_t(size_t, csum_end - start,
407 				     MAX_ORDERED_SUM_BYTES(root));
408 			sums = kzalloc(btrfs_ordered_sum_size(root, size),
409 				       GFP_NOFS);
410 			if (!sums) {
411 				ret = -ENOMEM;
412 				goto fail;
413 			}
414 
415 			sums->bytenr = start;
416 			sums->len = (int)size;
417 
418 			offset = (start - key.offset) >>
419 				root->fs_info->sb->s_blocksize_bits;
420 			offset *= csum_size;
421 			size >>= root->fs_info->sb->s_blocksize_bits;
422 
423 			read_extent_buffer(path->nodes[0],
424 					   sums->sums,
425 					   ((unsigned long)item) + offset,
426 					   csum_size * size);
427 
428 			start += root->sectorsize * size;
429 			list_add_tail(&sums->list, &tmplist);
430 		}
431 		path->slots[0]++;
432 	}
433 	ret = 0;
434 fail:
435 	while (ret < 0 && !list_empty(&tmplist)) {
436 		sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
437 		list_del(&sums->list);
438 		kfree(sums);
439 	}
440 	list_splice_tail(&tmplist, list);
441 
442 	btrfs_free_path(path);
443 	return ret;
444 }
445 
446 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
447 		       struct bio *bio, u64 file_start, int contig)
448 {
449 	struct btrfs_ordered_sum *sums;
450 	struct btrfs_ordered_extent *ordered;
451 	char *data;
452 	struct bio_vec *bvec = bio->bi_io_vec;
453 	int bio_index = 0;
454 	int index;
455 	int nr_sectors;
456 	int i;
457 	unsigned long total_bytes = 0;
458 	unsigned long this_sum_bytes = 0;
459 	u64 offset;
460 
461 	WARN_ON(bio->bi_vcnt <= 0);
462 	sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_iter.bi_size),
463 		       GFP_NOFS);
464 	if (!sums)
465 		return -ENOMEM;
466 
467 	sums->len = bio->bi_iter.bi_size;
468 	INIT_LIST_HEAD(&sums->list);
469 
470 	if (contig)
471 		offset = file_start;
472 	else
473 		offset = page_offset(bvec->bv_page) + bvec->bv_offset;
474 
475 	ordered = btrfs_lookup_ordered_extent(inode, offset);
476 	BUG_ON(!ordered); /* Logic error */
477 	sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
478 	index = 0;
479 
480 	while (bio_index < bio->bi_vcnt) {
481 		if (!contig)
482 			offset = page_offset(bvec->bv_page) + bvec->bv_offset;
483 
484 		data = kmap_atomic(bvec->bv_page);
485 
486 		nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info,
487 						bvec->bv_len + root->sectorsize
488 						- 1);
489 
490 		for (i = 0; i < nr_sectors; i++) {
491 			if (offset >= ordered->file_offset + ordered->len ||
492 				offset < ordered->file_offset) {
493 				unsigned long bytes_left;
494 
495 				kunmap_atomic(data);
496 				sums->len = this_sum_bytes;
497 				this_sum_bytes = 0;
498 				btrfs_add_ordered_sum(inode, ordered, sums);
499 				btrfs_put_ordered_extent(ordered);
500 
501 				bytes_left = bio->bi_iter.bi_size - total_bytes;
502 
503 				sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
504 					GFP_NOFS);
505 				BUG_ON(!sums); /* -ENOMEM */
506 				sums->len = bytes_left;
507 				ordered = btrfs_lookup_ordered_extent(inode,
508 								offset);
509 				ASSERT(ordered); /* Logic error */
510 				sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
511 					+ total_bytes;
512 				index = 0;
513 
514 				data = kmap_atomic(bvec->bv_page);
515 			}
516 
517 			sums->sums[index] = ~(u32)0;
518 			sums->sums[index]
519 				= btrfs_csum_data(data + bvec->bv_offset
520 						+ (i * root->sectorsize),
521 						sums->sums[index],
522 						root->sectorsize);
523 			btrfs_csum_final(sums->sums[index],
524 					(char *)(sums->sums + index));
525 			index++;
526 			offset += root->sectorsize;
527 			this_sum_bytes += root->sectorsize;
528 			total_bytes += root->sectorsize;
529 		}
530 
531 		kunmap_atomic(data);
532 
533 		bio_index++;
534 		bvec++;
535 	}
536 	this_sum_bytes = 0;
537 	btrfs_add_ordered_sum(inode, ordered, sums);
538 	btrfs_put_ordered_extent(ordered);
539 	return 0;
540 }
541 
542 /*
543  * helper function for csum removal, this expects the
544  * key to describe the csum pointed to by the path, and it expects
545  * the csum to overlap the range [bytenr, len]
546  *
547  * The csum should not be entirely contained in the range and the
548  * range should not be entirely contained in the csum.
549  *
550  * This calls btrfs_truncate_item with the correct args based on the
551  * overlap, and fixes up the key as required.
552  */
553 static noinline void truncate_one_csum(struct btrfs_root *root,
554 				       struct btrfs_path *path,
555 				       struct btrfs_key *key,
556 				       u64 bytenr, u64 len)
557 {
558 	struct extent_buffer *leaf;
559 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
560 	u64 csum_end;
561 	u64 end_byte = bytenr + len;
562 	u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
563 
564 	leaf = path->nodes[0];
565 	csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
566 	csum_end <<= root->fs_info->sb->s_blocksize_bits;
567 	csum_end += key->offset;
568 
569 	if (key->offset < bytenr && csum_end <= end_byte) {
570 		/*
571 		 *         [ bytenr - len ]
572 		 *         [   ]
573 		 *   [csum     ]
574 		 *   A simple truncate off the end of the item
575 		 */
576 		u32 new_size = (bytenr - key->offset) >> blocksize_bits;
577 		new_size *= csum_size;
578 		btrfs_truncate_item(root, path, new_size, 1);
579 	} else if (key->offset >= bytenr && csum_end > end_byte &&
580 		   end_byte > key->offset) {
581 		/*
582 		 *         [ bytenr - len ]
583 		 *                 [ ]
584 		 *                 [csum     ]
585 		 * we need to truncate from the beginning of the csum
586 		 */
587 		u32 new_size = (csum_end - end_byte) >> blocksize_bits;
588 		new_size *= csum_size;
589 
590 		btrfs_truncate_item(root, path, new_size, 0);
591 
592 		key->offset = end_byte;
593 		btrfs_set_item_key_safe(root->fs_info, path, key);
594 	} else {
595 		BUG();
596 	}
597 }
598 
599 /*
600  * deletes the csum items from the csum tree for a given
601  * range of bytes.
602  */
603 int btrfs_del_csums(struct btrfs_trans_handle *trans,
604 		    struct btrfs_root *root, u64 bytenr, u64 len)
605 {
606 	struct btrfs_path *path;
607 	struct btrfs_key key;
608 	u64 end_byte = bytenr + len;
609 	u64 csum_end;
610 	struct extent_buffer *leaf;
611 	int ret;
612 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
613 	int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
614 
615 	root = root->fs_info->csum_root;
616 
617 	path = btrfs_alloc_path();
618 	if (!path)
619 		return -ENOMEM;
620 
621 	while (1) {
622 		key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
623 		key.offset = end_byte - 1;
624 		key.type = BTRFS_EXTENT_CSUM_KEY;
625 
626 		path->leave_spinning = 1;
627 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
628 		if (ret > 0) {
629 			if (path->slots[0] == 0)
630 				break;
631 			path->slots[0]--;
632 		} else if (ret < 0) {
633 			break;
634 		}
635 
636 		leaf = path->nodes[0];
637 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
638 
639 		if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
640 		    key.type != BTRFS_EXTENT_CSUM_KEY) {
641 			break;
642 		}
643 
644 		if (key.offset >= end_byte)
645 			break;
646 
647 		csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
648 		csum_end <<= blocksize_bits;
649 		csum_end += key.offset;
650 
651 		/* this csum ends before we start, we're done */
652 		if (csum_end <= bytenr)
653 			break;
654 
655 		/* delete the entire item, it is inside our range */
656 		if (key.offset >= bytenr && csum_end <= end_byte) {
657 			ret = btrfs_del_item(trans, root, path);
658 			if (ret)
659 				goto out;
660 			if (key.offset == bytenr)
661 				break;
662 		} else if (key.offset < bytenr && csum_end > end_byte) {
663 			unsigned long offset;
664 			unsigned long shift_len;
665 			unsigned long item_offset;
666 			/*
667 			 *        [ bytenr - len ]
668 			 *     [csum                ]
669 			 *
670 			 * Our bytes are in the middle of the csum,
671 			 * we need to split this item and insert a new one.
672 			 *
673 			 * But we can't drop the path because the
674 			 * csum could change, get removed, extended etc.
675 			 *
676 			 * The trick here is the max size of a csum item leaves
677 			 * enough room in the tree block for a single
678 			 * item header.  So, we split the item in place,
679 			 * adding a new header pointing to the existing
680 			 * bytes.  Then we loop around again and we have
681 			 * a nicely formed csum item that we can neatly
682 			 * truncate.
683 			 */
684 			offset = (bytenr - key.offset) >> blocksize_bits;
685 			offset *= csum_size;
686 
687 			shift_len = (len >> blocksize_bits) * csum_size;
688 
689 			item_offset = btrfs_item_ptr_offset(leaf,
690 							    path->slots[0]);
691 
692 			memset_extent_buffer(leaf, 0, item_offset + offset,
693 					     shift_len);
694 			key.offset = bytenr;
695 
696 			/*
697 			 * btrfs_split_item returns -EAGAIN when the
698 			 * item changed size or key
699 			 */
700 			ret = btrfs_split_item(trans, root, path, &key, offset);
701 			if (ret && ret != -EAGAIN) {
702 				btrfs_abort_transaction(trans, ret);
703 				goto out;
704 			}
705 
706 			key.offset = end_byte - 1;
707 		} else {
708 			truncate_one_csum(root, path, &key, bytenr, len);
709 			if (key.offset < bytenr)
710 				break;
711 		}
712 		btrfs_release_path(path);
713 	}
714 	ret = 0;
715 out:
716 	btrfs_free_path(path);
717 	return ret;
718 }
719 
720 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
721 			   struct btrfs_root *root,
722 			   struct btrfs_ordered_sum *sums)
723 {
724 	struct btrfs_key file_key;
725 	struct btrfs_key found_key;
726 	struct btrfs_path *path;
727 	struct btrfs_csum_item *item;
728 	struct btrfs_csum_item *item_end;
729 	struct extent_buffer *leaf = NULL;
730 	u64 next_offset;
731 	u64 total_bytes = 0;
732 	u64 csum_offset;
733 	u64 bytenr;
734 	u32 nritems;
735 	u32 ins_size;
736 	int index = 0;
737 	int found_next;
738 	int ret;
739 	u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
740 
741 	path = btrfs_alloc_path();
742 	if (!path)
743 		return -ENOMEM;
744 again:
745 	next_offset = (u64)-1;
746 	found_next = 0;
747 	bytenr = sums->bytenr + total_bytes;
748 	file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
749 	file_key.offset = bytenr;
750 	file_key.type = BTRFS_EXTENT_CSUM_KEY;
751 
752 	item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
753 	if (!IS_ERR(item)) {
754 		ret = 0;
755 		leaf = path->nodes[0];
756 		item_end = btrfs_item_ptr(leaf, path->slots[0],
757 					  struct btrfs_csum_item);
758 		item_end = (struct btrfs_csum_item *)((char *)item_end +
759 			   btrfs_item_size_nr(leaf, path->slots[0]));
760 		goto found;
761 	}
762 	ret = PTR_ERR(item);
763 	if (ret != -EFBIG && ret != -ENOENT)
764 		goto fail_unlock;
765 
766 	if (ret == -EFBIG) {
767 		u32 item_size;
768 		/* we found one, but it isn't big enough yet */
769 		leaf = path->nodes[0];
770 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
771 		if ((item_size / csum_size) >=
772 		    MAX_CSUM_ITEMS(root, csum_size)) {
773 			/* already at max size, make a new one */
774 			goto insert;
775 		}
776 	} else {
777 		int slot = path->slots[0] + 1;
778 		/* we didn't find a csum item, insert one */
779 		nritems = btrfs_header_nritems(path->nodes[0]);
780 		if (!nritems || (path->slots[0] >= nritems - 1)) {
781 			ret = btrfs_next_leaf(root, path);
782 			if (ret == 1)
783 				found_next = 1;
784 			if (ret != 0)
785 				goto insert;
786 			slot = path->slots[0];
787 		}
788 		btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
789 		if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
790 		    found_key.type != BTRFS_EXTENT_CSUM_KEY) {
791 			found_next = 1;
792 			goto insert;
793 		}
794 		next_offset = found_key.offset;
795 		found_next = 1;
796 		goto insert;
797 	}
798 
799 	/*
800 	 * at this point, we know the tree has an item, but it isn't big
801 	 * enough yet to put our csum in.  Grow it
802 	 */
803 	btrfs_release_path(path);
804 	ret = btrfs_search_slot(trans, root, &file_key, path,
805 				csum_size, 1);
806 	if (ret < 0)
807 		goto fail_unlock;
808 
809 	if (ret > 0) {
810 		if (path->slots[0] == 0)
811 			goto insert;
812 		path->slots[0]--;
813 	}
814 
815 	leaf = path->nodes[0];
816 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
817 	csum_offset = (bytenr - found_key.offset) >>
818 			root->fs_info->sb->s_blocksize_bits;
819 
820 	if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
821 	    found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
822 	    csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
823 		goto insert;
824 	}
825 
826 	if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
827 	    csum_size) {
828 		int extend_nr;
829 		u64 tmp;
830 		u32 diff;
831 		u32 free_space;
832 
833 		if (btrfs_leaf_free_space(root, leaf) <
834 				 sizeof(struct btrfs_item) + csum_size * 2)
835 			goto insert;
836 
837 		free_space = btrfs_leaf_free_space(root, leaf) -
838 					 sizeof(struct btrfs_item) - csum_size;
839 		tmp = sums->len - total_bytes;
840 		tmp >>= root->fs_info->sb->s_blocksize_bits;
841 		WARN_ON(tmp < 1);
842 
843 		extend_nr = max_t(int, 1, (int)tmp);
844 		diff = (csum_offset + extend_nr) * csum_size;
845 		diff = min(diff, MAX_CSUM_ITEMS(root, csum_size) * csum_size);
846 
847 		diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
848 		diff = min(free_space, diff);
849 		diff /= csum_size;
850 		diff *= csum_size;
851 
852 		btrfs_extend_item(root, path, diff);
853 		ret = 0;
854 		goto csum;
855 	}
856 
857 insert:
858 	btrfs_release_path(path);
859 	csum_offset = 0;
860 	if (found_next) {
861 		u64 tmp;
862 
863 		tmp = sums->len - total_bytes;
864 		tmp >>= root->fs_info->sb->s_blocksize_bits;
865 		tmp = min(tmp, (next_offset - file_key.offset) >>
866 					 root->fs_info->sb->s_blocksize_bits);
867 
868 		tmp = max((u64)1, tmp);
869 		tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
870 		ins_size = csum_size * tmp;
871 	} else {
872 		ins_size = csum_size;
873 	}
874 	path->leave_spinning = 1;
875 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
876 				      ins_size);
877 	path->leave_spinning = 0;
878 	if (ret < 0)
879 		goto fail_unlock;
880 	if (WARN_ON(ret != 0))
881 		goto fail_unlock;
882 	leaf = path->nodes[0];
883 csum:
884 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
885 	item_end = (struct btrfs_csum_item *)((unsigned char *)item +
886 				      btrfs_item_size_nr(leaf, path->slots[0]));
887 	item = (struct btrfs_csum_item *)((unsigned char *)item +
888 					  csum_offset * csum_size);
889 found:
890 	ins_size = (u32)(sums->len - total_bytes) >>
891 		   root->fs_info->sb->s_blocksize_bits;
892 	ins_size *= csum_size;
893 	ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
894 			      ins_size);
895 	write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
896 			    ins_size);
897 
898 	ins_size /= csum_size;
899 	total_bytes += ins_size * root->sectorsize;
900 	index += ins_size;
901 
902 	btrfs_mark_buffer_dirty(path->nodes[0]);
903 	if (total_bytes < sums->len) {
904 		btrfs_release_path(path);
905 		cond_resched();
906 		goto again;
907 	}
908 out:
909 	btrfs_free_path(path);
910 	return ret;
911 
912 fail_unlock:
913 	goto out;
914 }
915 
916 void btrfs_extent_item_to_extent_map(struct inode *inode,
917 				     const struct btrfs_path *path,
918 				     struct btrfs_file_extent_item *fi,
919 				     const bool new_inline,
920 				     struct extent_map *em)
921 {
922 	struct btrfs_root *root = BTRFS_I(inode)->root;
923 	struct extent_buffer *leaf = path->nodes[0];
924 	const int slot = path->slots[0];
925 	struct btrfs_key key;
926 	u64 extent_start, extent_end;
927 	u64 bytenr;
928 	u8 type = btrfs_file_extent_type(leaf, fi);
929 	int compress_type = btrfs_file_extent_compression(leaf, fi);
930 
931 	em->bdev = root->fs_info->fs_devices->latest_bdev;
932 	btrfs_item_key_to_cpu(leaf, &key, slot);
933 	extent_start = key.offset;
934 
935 	if (type == BTRFS_FILE_EXTENT_REG ||
936 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
937 		extent_end = extent_start +
938 			btrfs_file_extent_num_bytes(leaf, fi);
939 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
940 		size_t size;
941 		size = btrfs_file_extent_inline_len(leaf, slot, fi);
942 		extent_end = ALIGN(extent_start + size, root->sectorsize);
943 	}
944 
945 	em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
946 	if (type == BTRFS_FILE_EXTENT_REG ||
947 	    type == BTRFS_FILE_EXTENT_PREALLOC) {
948 		em->start = extent_start;
949 		em->len = extent_end - extent_start;
950 		em->orig_start = extent_start -
951 			btrfs_file_extent_offset(leaf, fi);
952 		em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
953 		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
954 		if (bytenr == 0) {
955 			em->block_start = EXTENT_MAP_HOLE;
956 			return;
957 		}
958 		if (compress_type != BTRFS_COMPRESS_NONE) {
959 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
960 			em->compress_type = compress_type;
961 			em->block_start = bytenr;
962 			em->block_len = em->orig_block_len;
963 		} else {
964 			bytenr += btrfs_file_extent_offset(leaf, fi);
965 			em->block_start = bytenr;
966 			em->block_len = em->len;
967 			if (type == BTRFS_FILE_EXTENT_PREALLOC)
968 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
969 		}
970 	} else if (type == BTRFS_FILE_EXTENT_INLINE) {
971 		em->block_start = EXTENT_MAP_INLINE;
972 		em->start = extent_start;
973 		em->len = extent_end - extent_start;
974 		/*
975 		 * Initialize orig_start and block_len with the same values
976 		 * as in inode.c:btrfs_get_extent().
977 		 */
978 		em->orig_start = EXTENT_MAP_HOLE;
979 		em->block_len = (u64)-1;
980 		if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
981 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
982 			em->compress_type = compress_type;
983 		}
984 	} else {
985 		btrfs_err(root->fs_info,
986 			  "unknown file extent item type %d, inode %llu, offset %llu, root %llu",
987 			  type, btrfs_ino(inode), extent_start,
988 			  root->root_key.objectid);
989 	}
990 }
991