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