Btrfs: don't cache the csum value into the extent state tree
[linux-2.6-block.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         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_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_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_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                                         printk(KERN_INFO "btrfs no csum found "
250                                                "for inode %llu start %llu\n",
251                                                (unsigned long long)
252                                                btrfs_ino(inode),
253                                                (unsigned long long)offset);
254                                 }
255                                 item = NULL;
256                                 btrfs_release_path(path);
257                                 goto found;
258                         }
259                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
260                                               path->slots[0]);
261
262                         item_start_offset = found_key.offset;
263                         item_size = btrfs_item_size_nr(path->nodes[0],
264                                                        path->slots[0]);
265                         item_last_offset = item_start_offset +
266                                 (item_size / csum_size) *
267                                 root->sectorsize;
268                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
269                                               struct btrfs_csum_item);
270                 }
271                 /*
272                  * this byte range must be able to fit inside
273                  * a single leaf so it will also fit inside a u32
274                  */
275                 diff = disk_bytenr - item_start_offset;
276                 diff = diff / root->sectorsize;
277                 diff = diff * csum_size;
278                 count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
279                                             inode->i_sb->s_blocksize_bits);
280                 read_extent_buffer(path->nodes[0], csum,
281                                    ((unsigned long)item) + diff,
282                                    csum_size * count);
283 found:
284                 csum += count * csum_size;
285                 nblocks -= count;
286                 while (count--) {
287                         disk_bytenr += bvec->bv_len;
288                         offset += bvec->bv_len;
289                         bio_index++;
290                         bvec++;
291                 }
292         }
293         btrfs_free_path(path);
294         return 0;
295 }
296
297 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
298                           struct bio *bio, u32 *dst)
299 {
300         return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
301 }
302
303 int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
304                               struct btrfs_dio_private *dip, struct bio *bio,
305                               u64 offset)
306 {
307         int len = (bio->bi_sector << 9) - dip->disk_bytenr;
308         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
309         int ret;
310
311         len >>= inode->i_sb->s_blocksize_bits;
312         len *= csum_size;
313
314         ret = __btrfs_lookup_bio_sums(root, inode, bio, offset,
315                                       (u32 *)(dip->csum + len), 1);
316         return ret;
317 }
318
319 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
320                              struct list_head *list, int search_commit)
321 {
322         struct btrfs_key key;
323         struct btrfs_path *path;
324         struct extent_buffer *leaf;
325         struct btrfs_ordered_sum *sums;
326         struct btrfs_csum_item *item;
327         LIST_HEAD(tmplist);
328         unsigned long offset;
329         int ret;
330         size_t size;
331         u64 csum_end;
332         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
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 = 2;
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, 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         unsigned long total_bytes = 0;
445         unsigned long this_sum_bytes = 0;
446         u64 offset;
447
448         WARN_ON(bio->bi_vcnt <= 0);
449         sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
450         if (!sums)
451                 return -ENOMEM;
452
453         sums->len = bio->bi_size;
454         INIT_LIST_HEAD(&sums->list);
455
456         if (contig)
457                 offset = file_start;
458         else
459                 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
460
461         ordered = btrfs_lookup_ordered_extent(inode, offset);
462         BUG_ON(!ordered); /* Logic error */
463         sums->bytenr = (u64)bio->bi_sector << 9;
464         index = 0;
465
466         while (bio_index < bio->bi_vcnt) {
467                 if (!contig)
468                         offset = page_offset(bvec->bv_page) + bvec->bv_offset;
469
470                 if (offset >= ordered->file_offset + ordered->len ||
471                     offset < ordered->file_offset) {
472                         unsigned long bytes_left;
473                         sums->len = this_sum_bytes;
474                         this_sum_bytes = 0;
475                         btrfs_add_ordered_sum(inode, ordered, sums);
476                         btrfs_put_ordered_extent(ordered);
477
478                         bytes_left = bio->bi_size - total_bytes;
479
480                         sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
481                                        GFP_NOFS);
482                         BUG_ON(!sums); /* -ENOMEM */
483                         sums->len = bytes_left;
484                         ordered = btrfs_lookup_ordered_extent(inode, offset);
485                         BUG_ON(!ordered); /* Logic error */
486                         sums->bytenr = ((u64)bio->bi_sector << 9) +
487                                        total_bytes;
488                         index = 0;
489                 }
490
491                 data = kmap_atomic(bvec->bv_page);
492                 sums->sums[index] = ~(u32)0;
493                 sums->sums[index] = btrfs_csum_data(data + bvec->bv_offset,
494                                                     sums->sums[index],
495                                                     bvec->bv_len);
496                 kunmap_atomic(data);
497                 btrfs_csum_final(sums->sums[index],
498                                  (char *)(sums->sums + index));
499
500                 bio_index++;
501                 index++;
502                 total_bytes += bvec->bv_len;
503                 this_sum_bytes += bvec->bv_len;
504                 offset += bvec->bv_len;
505                 bvec++;
506         }
507         this_sum_bytes = 0;
508         btrfs_add_ordered_sum(inode, ordered, sums);
509         btrfs_put_ordered_extent(ordered);
510         return 0;
511 }
512
513 /*
514  * helper function for csum removal, this expects the
515  * key to describe the csum pointed to by the path, and it expects
516  * the csum to overlap the range [bytenr, len]
517  *
518  * The csum should not be entirely contained in the range and the
519  * range should not be entirely contained in the csum.
520  *
521  * This calls btrfs_truncate_item with the correct args based on the
522  * overlap, and fixes up the key as required.
523  */
524 static noinline void truncate_one_csum(struct btrfs_root *root,
525                                        struct btrfs_path *path,
526                                        struct btrfs_key *key,
527                                        u64 bytenr, u64 len)
528 {
529         struct extent_buffer *leaf;
530         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
531         u64 csum_end;
532         u64 end_byte = bytenr + len;
533         u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
534
535         leaf = path->nodes[0];
536         csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
537         csum_end <<= root->fs_info->sb->s_blocksize_bits;
538         csum_end += key->offset;
539
540         if (key->offset < bytenr && csum_end <= end_byte) {
541                 /*
542                  *         [ bytenr - len ]
543                  *         [   ]
544                  *   [csum     ]
545                  *   A simple truncate off the end of the item
546                  */
547                 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
548                 new_size *= csum_size;
549                 btrfs_truncate_item(root, path, new_size, 1);
550         } else if (key->offset >= bytenr && csum_end > end_byte &&
551                    end_byte > key->offset) {
552                 /*
553                  *         [ bytenr - len ]
554                  *                 [ ]
555                  *                 [csum     ]
556                  * we need to truncate from the beginning of the csum
557                  */
558                 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
559                 new_size *= csum_size;
560
561                 btrfs_truncate_item(root, path, new_size, 0);
562
563                 key->offset = end_byte;
564                 btrfs_set_item_key_safe(root, path, key);
565         } else {
566                 BUG();
567         }
568 }
569
570 /*
571  * deletes the csum items from the csum tree for a given
572  * range of bytes.
573  */
574 int btrfs_del_csums(struct btrfs_trans_handle *trans,
575                     struct btrfs_root *root, u64 bytenr, u64 len)
576 {
577         struct btrfs_path *path;
578         struct btrfs_key key;
579         u64 end_byte = bytenr + len;
580         u64 csum_end;
581         struct extent_buffer *leaf;
582         int ret;
583         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
584         int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
585
586         root = root->fs_info->csum_root;
587
588         path = btrfs_alloc_path();
589         if (!path)
590                 return -ENOMEM;
591
592         while (1) {
593                 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
594                 key.offset = end_byte - 1;
595                 key.type = BTRFS_EXTENT_CSUM_KEY;
596
597                 path->leave_spinning = 1;
598                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
599                 if (ret > 0) {
600                         if (path->slots[0] == 0)
601                                 break;
602                         path->slots[0]--;
603                 } else if (ret < 0) {
604                         break;
605                 }
606
607                 leaf = path->nodes[0];
608                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
609
610                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
611                     key.type != BTRFS_EXTENT_CSUM_KEY) {
612                         break;
613                 }
614
615                 if (key.offset >= end_byte)
616                         break;
617
618                 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
619                 csum_end <<= blocksize_bits;
620                 csum_end += key.offset;
621
622                 /* this csum ends before we start, we're done */
623                 if (csum_end <= bytenr)
624                         break;
625
626                 /* delete the entire item, it is inside our range */
627                 if (key.offset >= bytenr && csum_end <= end_byte) {
628                         ret = btrfs_del_item(trans, root, path);
629                         if (ret)
630                                 goto out;
631                         if (key.offset == bytenr)
632                                 break;
633                 } else if (key.offset < bytenr && csum_end > end_byte) {
634                         unsigned long offset;
635                         unsigned long shift_len;
636                         unsigned long item_offset;
637                         /*
638                          *        [ bytenr - len ]
639                          *     [csum                ]
640                          *
641                          * Our bytes are in the middle of the csum,
642                          * we need to split this item and insert a new one.
643                          *
644                          * But we can't drop the path because the
645                          * csum could change, get removed, extended etc.
646                          *
647                          * The trick here is the max size of a csum item leaves
648                          * enough room in the tree block for a single
649                          * item header.  So, we split the item in place,
650                          * adding a new header pointing to the existing
651                          * bytes.  Then we loop around again and we have
652                          * a nicely formed csum item that we can neatly
653                          * truncate.
654                          */
655                         offset = (bytenr - key.offset) >> blocksize_bits;
656                         offset *= csum_size;
657
658                         shift_len = (len >> blocksize_bits) * csum_size;
659
660                         item_offset = btrfs_item_ptr_offset(leaf,
661                                                             path->slots[0]);
662
663                         memset_extent_buffer(leaf, 0, item_offset + offset,
664                                              shift_len);
665                         key.offset = bytenr;
666
667                         /*
668                          * btrfs_split_item returns -EAGAIN when the
669                          * item changed size or key
670                          */
671                         ret = btrfs_split_item(trans, root, path, &key, offset);
672                         if (ret && ret != -EAGAIN) {
673                                 btrfs_abort_transaction(trans, root, ret);
674                                 goto out;
675                         }
676
677                         key.offset = end_byte - 1;
678                 } else {
679                         truncate_one_csum(root, path, &key, bytenr, len);
680                         if (key.offset < bytenr)
681                                 break;
682                 }
683                 btrfs_release_path(path);
684         }
685         ret = 0;
686 out:
687         btrfs_free_path(path);
688         return ret;
689 }
690
691 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
692                            struct btrfs_root *root,
693                            struct btrfs_ordered_sum *sums)
694 {
695         struct btrfs_key file_key;
696         struct btrfs_key found_key;
697         struct btrfs_path *path;
698         struct btrfs_csum_item *item;
699         struct btrfs_csum_item *item_end;
700         struct extent_buffer *leaf = NULL;
701         u64 next_offset;
702         u64 total_bytes = 0;
703         u64 csum_offset;
704         u64 bytenr;
705         u32 nritems;
706         u32 ins_size;
707         int index = 0;
708         int found_next;
709         int ret;
710         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
711
712         path = btrfs_alloc_path();
713         if (!path)
714                 return -ENOMEM;
715 again:
716         next_offset = (u64)-1;
717         found_next = 0;
718         bytenr = sums->bytenr + total_bytes;
719         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
720         file_key.offset = bytenr;
721         btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
722
723         item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
724         if (!IS_ERR(item)) {
725                 ret = 0;
726                 leaf = path->nodes[0];
727                 item_end = btrfs_item_ptr(leaf, path->slots[0],
728                                           struct btrfs_csum_item);
729                 item_end = (struct btrfs_csum_item *)((char *)item_end +
730                            btrfs_item_size_nr(leaf, path->slots[0]));
731                 goto found;
732         }
733         ret = PTR_ERR(item);
734         if (ret != -EFBIG && ret != -ENOENT)
735                 goto fail_unlock;
736
737         if (ret == -EFBIG) {
738                 u32 item_size;
739                 /* we found one, but it isn't big enough yet */
740                 leaf = path->nodes[0];
741                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
742                 if ((item_size / csum_size) >=
743                     MAX_CSUM_ITEMS(root, csum_size)) {
744                         /* already at max size, make a new one */
745                         goto insert;
746                 }
747         } else {
748                 int slot = path->slots[0] + 1;
749                 /* we didn't find a csum item, insert one */
750                 nritems = btrfs_header_nritems(path->nodes[0]);
751                 if (path->slots[0] >= nritems - 1) {
752                         ret = btrfs_next_leaf(root, path);
753                         if (ret == 1)
754                                 found_next = 1;
755                         if (ret != 0)
756                                 goto insert;
757                         slot = 0;
758                 }
759                 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
760                 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
761                     found_key.type != BTRFS_EXTENT_CSUM_KEY) {
762                         found_next = 1;
763                         goto insert;
764                 }
765                 next_offset = found_key.offset;
766                 found_next = 1;
767                 goto insert;
768         }
769
770         /*
771          * at this point, we know the tree has an item, but it isn't big
772          * enough yet to put our csum in.  Grow it
773          */
774         btrfs_release_path(path);
775         ret = btrfs_search_slot(trans, root, &file_key, path,
776                                 csum_size, 1);
777         if (ret < 0)
778                 goto fail_unlock;
779
780         if (ret > 0) {
781                 if (path->slots[0] == 0)
782                         goto insert;
783                 path->slots[0]--;
784         }
785
786         leaf = path->nodes[0];
787         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
788         csum_offset = (bytenr - found_key.offset) >>
789                         root->fs_info->sb->s_blocksize_bits;
790
791         if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
792             found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
793             csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
794                 goto insert;
795         }
796
797         if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
798             csum_size) {
799                 int extend_nr;
800                 u64 tmp;
801                 u32 diff;
802                 u32 free_space;
803
804                 if (btrfs_leaf_free_space(root, leaf) <
805                                  sizeof(struct btrfs_item) + csum_size * 2)
806                         goto insert;
807
808                 free_space = btrfs_leaf_free_space(root, leaf) -
809                                          sizeof(struct btrfs_item) - csum_size;
810                 tmp = sums->len - total_bytes;
811                 tmp >>= root->fs_info->sb->s_blocksize_bits;
812                 WARN_ON(tmp < 1);
813
814                 extend_nr = max_t(int, 1, (int)tmp);
815                 diff = (csum_offset + extend_nr) * csum_size;
816                 diff = min(diff, MAX_CSUM_ITEMS(root, csum_size) * csum_size);
817
818                 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
819                 diff = min(free_space, diff);
820                 diff /= csum_size;
821                 diff *= csum_size;
822
823                 btrfs_extend_item(root, path, diff);
824                 ret = 0;
825                 goto csum;
826         }
827
828 insert:
829         btrfs_release_path(path);
830         csum_offset = 0;
831         if (found_next) {
832                 u64 tmp;
833
834                 tmp = sums->len - total_bytes;
835                 tmp >>= root->fs_info->sb->s_blocksize_bits;
836                 tmp = min(tmp, (next_offset - file_key.offset) >>
837                                          root->fs_info->sb->s_blocksize_bits);
838
839                 tmp = max((u64)1, tmp);
840                 tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
841                 ins_size = csum_size * tmp;
842         } else {
843                 ins_size = csum_size;
844         }
845         path->leave_spinning = 1;
846         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
847                                       ins_size);
848         path->leave_spinning = 0;
849         if (ret < 0)
850                 goto fail_unlock;
851         if (ret != 0) {
852                 WARN_ON(1);
853                 goto fail_unlock;
854         }
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 }