Merge tag 'drm-msm-fixes-2021-04-02' of https://gitlab.freedesktop.org/drm/msm into...
[linux-2.6-microblaze.git] / fs / btrfs / file-item.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5
6 #include <linux/bio.h>
7 #include <linux/slab.h>
8 #include <linux/pagemap.h>
9 #include <linux/highmem.h>
10 #include <linux/sched/mm.h>
11 #include <crypto/hash.h>
12 #include "ctree.h"
13 #include "disk-io.h"
14 #include "transaction.h"
15 #include "volumes.h"
16 #include "print-tree.h"
17 #include "compression.h"
18
19 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
20                                    sizeof(struct btrfs_item) * 2) / \
21                                   size) - 1))
22
23 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
24                                        PAGE_SIZE))
25
26 /**
27  * Set inode's size according to filesystem options
28  *
29  * @inode:      inode we want to update the disk_i_size for
30  * @new_i_size: i_size we want to set to, 0 if we use i_size
31  *
32  * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read()
33  * returns as it is perfectly fine with a file that has holes without hole file
34  * extent items.
35  *
36  * However without NO_HOLES we need to only return the area that is contiguous
37  * from the 0 offset of the file.  Otherwise we could end up adjust i_size up
38  * to an extent that has a gap in between.
39  *
40  * Finally new_i_size should only be set in the case of truncate where we're not
41  * ready to use i_size_read() as the limiter yet.
42  */
43 void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size)
44 {
45         struct btrfs_fs_info *fs_info = inode->root->fs_info;
46         u64 start, end, i_size;
47         int ret;
48
49         i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
50         if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
51                 inode->disk_i_size = i_size;
52                 return;
53         }
54
55         spin_lock(&inode->lock);
56         ret = find_contiguous_extent_bit(&inode->file_extent_tree, 0, &start,
57                                          &end, EXTENT_DIRTY);
58         if (!ret && start == 0)
59                 i_size = min(i_size, end + 1);
60         else
61                 i_size = 0;
62         inode->disk_i_size = i_size;
63         spin_unlock(&inode->lock);
64 }
65
66 /**
67  * Mark range within a file as having a new extent inserted
68  *
69  * @inode: inode being modified
70  * @start: start file offset of the file extent we've inserted
71  * @len:   logical length of the file extent item
72  *
73  * Call when we are inserting a new file extent where there was none before.
74  * Does not need to call this in the case where we're replacing an existing file
75  * extent, however if not sure it's fine to call this multiple times.
76  *
77  * The start and len must match the file extent item, so thus must be sectorsize
78  * aligned.
79  */
80 int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
81                                       u64 len)
82 {
83         if (len == 0)
84                 return 0;
85
86         ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
87
88         if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
89                 return 0;
90         return set_extent_bits(&inode->file_extent_tree, start, start + len - 1,
91                                EXTENT_DIRTY);
92 }
93
94 /**
95  * Marks an inode range as not having a backing extent
96  *
97  * @inode: inode being modified
98  * @start: start file offset of the file extent we've inserted
99  * @len:   logical length of the file extent item
100  *
101  * Called when we drop a file extent, for example when we truncate.  Doesn't
102  * need to be called for cases where we're replacing a file extent, like when
103  * we've COWed a file extent.
104  *
105  * The start and len must match the file extent item, so thus must be sectorsize
106  * aligned.
107  */
108 int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
109                                         u64 len)
110 {
111         if (len == 0)
112                 return 0;
113
114         ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
115                len == (u64)-1);
116
117         if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
118                 return 0;
119         return clear_extent_bit(&inode->file_extent_tree, start,
120                                 start + len - 1, EXTENT_DIRTY, 0, 0, NULL);
121 }
122
123 static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info,
124                                         u16 csum_size)
125 {
126         u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size;
127
128         return ncsums * fs_info->sectorsize;
129 }
130
131 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
132                              struct btrfs_root *root,
133                              u64 objectid, u64 pos,
134                              u64 disk_offset, u64 disk_num_bytes,
135                              u64 num_bytes, u64 offset, u64 ram_bytes,
136                              u8 compression, u8 encryption, u16 other_encoding)
137 {
138         int ret = 0;
139         struct btrfs_file_extent_item *item;
140         struct btrfs_key file_key;
141         struct btrfs_path *path;
142         struct extent_buffer *leaf;
143
144         path = btrfs_alloc_path();
145         if (!path)
146                 return -ENOMEM;
147         file_key.objectid = objectid;
148         file_key.offset = pos;
149         file_key.type = BTRFS_EXTENT_DATA_KEY;
150
151         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
152                                       sizeof(*item));
153         if (ret < 0)
154                 goto out;
155         BUG_ON(ret); /* Can't happen */
156         leaf = path->nodes[0];
157         item = btrfs_item_ptr(leaf, path->slots[0],
158                               struct btrfs_file_extent_item);
159         btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
160         btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
161         btrfs_set_file_extent_offset(leaf, item, offset);
162         btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
163         btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
164         btrfs_set_file_extent_generation(leaf, item, trans->transid);
165         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
166         btrfs_set_file_extent_compression(leaf, item, compression);
167         btrfs_set_file_extent_encryption(leaf, item, encryption);
168         btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
169
170         btrfs_mark_buffer_dirty(leaf);
171 out:
172         btrfs_free_path(path);
173         return ret;
174 }
175
176 static struct btrfs_csum_item *
177 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
178                   struct btrfs_root *root,
179                   struct btrfs_path *path,
180                   u64 bytenr, int cow)
181 {
182         struct btrfs_fs_info *fs_info = root->fs_info;
183         int ret;
184         struct btrfs_key file_key;
185         struct btrfs_key found_key;
186         struct btrfs_csum_item *item;
187         struct extent_buffer *leaf;
188         u64 csum_offset = 0;
189         const u32 csum_size = fs_info->csum_size;
190         int csums_in_item;
191
192         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
193         file_key.offset = bytenr;
194         file_key.type = BTRFS_EXTENT_CSUM_KEY;
195         ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
196         if (ret < 0)
197                 goto fail;
198         leaf = path->nodes[0];
199         if (ret > 0) {
200                 ret = 1;
201                 if (path->slots[0] == 0)
202                         goto fail;
203                 path->slots[0]--;
204                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
205                 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
206                         goto fail;
207
208                 csum_offset = (bytenr - found_key.offset) >>
209                                 fs_info->sectorsize_bits;
210                 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
211                 csums_in_item /= csum_size;
212
213                 if (csum_offset == csums_in_item) {
214                         ret = -EFBIG;
215                         goto fail;
216                 } else if (csum_offset > csums_in_item) {
217                         goto fail;
218                 }
219         }
220         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
221         item = (struct btrfs_csum_item *)((unsigned char *)item +
222                                           csum_offset * csum_size);
223         return item;
224 fail:
225         if (ret > 0)
226                 ret = -ENOENT;
227         return ERR_PTR(ret);
228 }
229
230 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
231                              struct btrfs_root *root,
232                              struct btrfs_path *path, u64 objectid,
233                              u64 offset, int mod)
234 {
235         int ret;
236         struct btrfs_key file_key;
237         int ins_len = mod < 0 ? -1 : 0;
238         int cow = mod != 0;
239
240         file_key.objectid = objectid;
241         file_key.offset = offset;
242         file_key.type = BTRFS_EXTENT_DATA_KEY;
243         ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
244         return ret;
245 }
246
247 /*
248  * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
249  * estore the result to @dst.
250  *
251  * Return >0 for the number of sectors we found.
252  * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
253  * for it. Caller may want to try next sector until one range is hit.
254  * Return <0 for fatal error.
255  */
256 static int search_csum_tree(struct btrfs_fs_info *fs_info,
257                             struct btrfs_path *path, u64 disk_bytenr,
258                             u64 len, u8 *dst)
259 {
260         struct btrfs_csum_item *item = NULL;
261         struct btrfs_key key;
262         const u32 sectorsize = fs_info->sectorsize;
263         const u32 csum_size = fs_info->csum_size;
264         u32 itemsize;
265         int ret;
266         u64 csum_start;
267         u64 csum_len;
268
269         ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
270                IS_ALIGNED(len, sectorsize));
271
272         /* Check if the current csum item covers disk_bytenr */
273         if (path->nodes[0]) {
274                 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
275                                       struct btrfs_csum_item);
276                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
277                 itemsize = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
278
279                 csum_start = key.offset;
280                 csum_len = (itemsize / csum_size) * sectorsize;
281
282                 if (in_range(disk_bytenr, csum_start, csum_len))
283                         goto found;
284         }
285
286         /* Current item doesn't contain the desired range, search again */
287         btrfs_release_path(path);
288         item = btrfs_lookup_csum(NULL, fs_info->csum_root, path, disk_bytenr, 0);
289         if (IS_ERR(item)) {
290                 ret = PTR_ERR(item);
291                 goto out;
292         }
293         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
294         itemsize = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
295
296         csum_start = key.offset;
297         csum_len = (itemsize / csum_size) * sectorsize;
298         ASSERT(in_range(disk_bytenr, csum_start, csum_len));
299
300 found:
301         ret = (min(csum_start + csum_len, disk_bytenr + len) -
302                    disk_bytenr) >> fs_info->sectorsize_bits;
303         read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
304                         ret * csum_size);
305 out:
306         if (ret == -ENOENT)
307                 ret = 0;
308         return ret;
309 }
310
311 /*
312  * Locate the file_offset of @cur_disk_bytenr of a @bio.
313  *
314  * Bio of btrfs represents read range of
315  * [bi_sector << 9, bi_sector << 9 + bi_size).
316  * Knowing this, we can iterate through each bvec to locate the page belong to
317  * @cur_disk_bytenr and get the file offset.
318  *
319  * @inode is used to determine if the bvec page really belongs to @inode.
320  *
321  * Return 0 if we can't find the file offset
322  * Return >0 if we find the file offset and restore it to @file_offset_ret
323  */
324 static int search_file_offset_in_bio(struct bio *bio, struct inode *inode,
325                                      u64 disk_bytenr, u64 *file_offset_ret)
326 {
327         struct bvec_iter iter;
328         struct bio_vec bvec;
329         u64 cur = bio->bi_iter.bi_sector << SECTOR_SHIFT;
330         int ret = 0;
331
332         bio_for_each_segment(bvec, bio, iter) {
333                 struct page *page = bvec.bv_page;
334
335                 if (cur > disk_bytenr)
336                         break;
337                 if (cur + bvec.bv_len <= disk_bytenr) {
338                         cur += bvec.bv_len;
339                         continue;
340                 }
341                 ASSERT(in_range(disk_bytenr, cur, bvec.bv_len));
342                 if (page->mapping && page->mapping->host &&
343                     page->mapping->host == inode) {
344                         ret = 1;
345                         *file_offset_ret = page_offset(page) + bvec.bv_offset +
346                                            disk_bytenr - cur;
347                         break;
348                 }
349         }
350         return ret;
351 }
352
353 /**
354  * Lookup the checksum for the read bio in csum tree.
355  *
356  * @inode: inode that the bio is for.
357  * @bio: bio to look up.
358  * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return
359  *       checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If
360  *       NULL, the checksum buffer is allocated and returned in
361  *       btrfs_io_bio(bio)->csum instead.
362  *
363  * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
364  */
365 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst)
366 {
367         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
368         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
369         struct btrfs_path *path;
370         const u32 sectorsize = fs_info->sectorsize;
371         const u32 csum_size = fs_info->csum_size;
372         u32 orig_len = bio->bi_iter.bi_size;
373         u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
374         u64 cur_disk_bytenr;
375         u8 *csum;
376         const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
377         int count = 0;
378
379         if (!fs_info->csum_root || (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
380                 return BLK_STS_OK;
381
382         /*
383          * This function is only called for read bio.
384          *
385          * This means two things:
386          * - All our csums should only be in csum tree
387          *   No ordered extents csums, as ordered extents are only for write
388          *   path.
389          * - No need to bother any other info from bvec
390          *   Since we're looking up csums, the only important info is the
391          *   disk_bytenr and the length, which can be extracted from bi_iter
392          *   directly.
393          */
394         ASSERT(bio_op(bio) == REQ_OP_READ);
395         path = btrfs_alloc_path();
396         if (!path)
397                 return BLK_STS_RESOURCE;
398
399         if (!dst) {
400                 struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
401
402                 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
403                         btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
404                                                         GFP_NOFS);
405                         if (!btrfs_bio->csum) {
406                                 btrfs_free_path(path);
407                                 return BLK_STS_RESOURCE;
408                         }
409                 } else {
410                         btrfs_bio->csum = btrfs_bio->csum_inline;
411                 }
412                 csum = btrfs_bio->csum;
413         } else {
414                 csum = dst;
415         }
416
417         /*
418          * If requested number of sectors is larger than one leaf can contain,
419          * kick the readahead for csum tree.
420          */
421         if (nblocks > fs_info->csums_per_leaf)
422                 path->reada = READA_FORWARD;
423
424         /*
425          * the free space stuff is only read when it hasn't been
426          * updated in the current transaction.  So, we can safely
427          * read from the commit root and sidestep a nasty deadlock
428          * between reading the free space cache and updating the csum tree.
429          */
430         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
431                 path->search_commit_root = 1;
432                 path->skip_locking = 1;
433         }
434
435         for (cur_disk_bytenr = orig_disk_bytenr;
436              cur_disk_bytenr < orig_disk_bytenr + orig_len;
437              cur_disk_bytenr += (count * sectorsize)) {
438                 u64 search_len = orig_disk_bytenr + orig_len - cur_disk_bytenr;
439                 unsigned int sector_offset;
440                 u8 *csum_dst;
441
442                 /*
443                  * Although both cur_disk_bytenr and orig_disk_bytenr is u64,
444                  * we're calculating the offset to the bio start.
445                  *
446                  * Bio size is limited to UINT_MAX, thus unsigned int is large
447                  * enough to contain the raw result, not to mention the right
448                  * shifted result.
449                  */
450                 ASSERT(cur_disk_bytenr - orig_disk_bytenr < UINT_MAX);
451                 sector_offset = (cur_disk_bytenr - orig_disk_bytenr) >>
452                                 fs_info->sectorsize_bits;
453                 csum_dst = csum + sector_offset * csum_size;
454
455                 count = search_csum_tree(fs_info, path, cur_disk_bytenr,
456                                          search_len, csum_dst);
457                 if (count <= 0) {
458                         /*
459                          * Either we hit a critical error or we didn't find
460                          * the csum.
461                          * Either way, we put zero into the csums dst, and skip
462                          * to the next sector.
463                          */
464                         memset(csum_dst, 0, csum_size);
465                         count = 1;
466
467                         /*
468                          * For data reloc inode, we need to mark the range
469                          * NODATASUM so that balance won't report false csum
470                          * error.
471                          */
472                         if (BTRFS_I(inode)->root->root_key.objectid ==
473                             BTRFS_DATA_RELOC_TREE_OBJECTID) {
474                                 u64 file_offset;
475                                 int ret;
476
477                                 ret = search_file_offset_in_bio(bio, inode,
478                                                 cur_disk_bytenr, &file_offset);
479                                 if (ret)
480                                         set_extent_bits(io_tree, file_offset,
481                                                 file_offset + sectorsize - 1,
482                                                 EXTENT_NODATASUM);
483                         } else {
484                                 btrfs_warn_rl(fs_info,
485                         "csum hole found for disk bytenr range [%llu, %llu)",
486                                 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
487                         }
488                 }
489         }
490
491         btrfs_free_path(path);
492         return BLK_STS_OK;
493 }
494
495 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
496                              struct list_head *list, int search_commit)
497 {
498         struct btrfs_fs_info *fs_info = root->fs_info;
499         struct btrfs_key key;
500         struct btrfs_path *path;
501         struct extent_buffer *leaf;
502         struct btrfs_ordered_sum *sums;
503         struct btrfs_csum_item *item;
504         LIST_HEAD(tmplist);
505         unsigned long offset;
506         int ret;
507         size_t size;
508         u64 csum_end;
509         const u32 csum_size = fs_info->csum_size;
510
511         ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
512                IS_ALIGNED(end + 1, fs_info->sectorsize));
513
514         path = btrfs_alloc_path();
515         if (!path)
516                 return -ENOMEM;
517
518         if (search_commit) {
519                 path->skip_locking = 1;
520                 path->reada = READA_FORWARD;
521                 path->search_commit_root = 1;
522         }
523
524         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
525         key.offset = start;
526         key.type = BTRFS_EXTENT_CSUM_KEY;
527
528         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
529         if (ret < 0)
530                 goto fail;
531         if (ret > 0 && path->slots[0] > 0) {
532                 leaf = path->nodes[0];
533                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
534                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
535                     key.type == BTRFS_EXTENT_CSUM_KEY) {
536                         offset = (start - key.offset) >> fs_info->sectorsize_bits;
537                         if (offset * csum_size <
538                             btrfs_item_size_nr(leaf, path->slots[0] - 1))
539                                 path->slots[0]--;
540                 }
541         }
542
543         while (start <= end) {
544                 leaf = path->nodes[0];
545                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
546                         ret = btrfs_next_leaf(root, path);
547                         if (ret < 0)
548                                 goto fail;
549                         if (ret > 0)
550                                 break;
551                         leaf = path->nodes[0];
552                 }
553
554                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
555                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
556                     key.type != BTRFS_EXTENT_CSUM_KEY ||
557                     key.offset > end)
558                         break;
559
560                 if (key.offset > start)
561                         start = key.offset;
562
563                 size = btrfs_item_size_nr(leaf, path->slots[0]);
564                 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
565                 if (csum_end <= start) {
566                         path->slots[0]++;
567                         continue;
568                 }
569
570                 csum_end = min(csum_end, end + 1);
571                 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
572                                       struct btrfs_csum_item);
573                 while (start < csum_end) {
574                         size = min_t(size_t, csum_end - start,
575                                      max_ordered_sum_bytes(fs_info, csum_size));
576                         sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
577                                        GFP_NOFS);
578                         if (!sums) {
579                                 ret = -ENOMEM;
580                                 goto fail;
581                         }
582
583                         sums->bytenr = start;
584                         sums->len = (int)size;
585
586                         offset = (start - key.offset) >> fs_info->sectorsize_bits;
587                         offset *= csum_size;
588                         size >>= fs_info->sectorsize_bits;
589
590                         read_extent_buffer(path->nodes[0],
591                                            sums->sums,
592                                            ((unsigned long)item) + offset,
593                                            csum_size * size);
594
595                         start += fs_info->sectorsize * size;
596                         list_add_tail(&sums->list, &tmplist);
597                 }
598                 path->slots[0]++;
599         }
600         ret = 0;
601 fail:
602         while (ret < 0 && !list_empty(&tmplist)) {
603                 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
604                 list_del(&sums->list);
605                 kfree(sums);
606         }
607         list_splice_tail(&tmplist, list);
608
609         btrfs_free_path(path);
610         return ret;
611 }
612
613 /*
614  * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
615  * @inode:       Owner of the data inside the bio
616  * @bio:         Contains the data to be checksummed
617  * @file_start:  offset in file this bio begins to describe
618  * @contig:      Boolean. If true/1 means all bio vecs in this bio are
619  *               contiguous and they begin at @file_start in the file. False/0
620  *               means this bio can contains potentially discontigous bio vecs
621  *               so the logical offset of each should be calculated separately.
622  */
623 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
624                        u64 file_start, int contig)
625 {
626         struct btrfs_fs_info *fs_info = inode->root->fs_info;
627         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
628         struct btrfs_ordered_sum *sums;
629         struct btrfs_ordered_extent *ordered = NULL;
630         char *data;
631         struct bvec_iter iter;
632         struct bio_vec bvec;
633         int index;
634         int nr_sectors;
635         unsigned long total_bytes = 0;
636         unsigned long this_sum_bytes = 0;
637         int i;
638         u64 offset;
639         unsigned nofs_flag;
640
641         nofs_flag = memalloc_nofs_save();
642         sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
643                        GFP_KERNEL);
644         memalloc_nofs_restore(nofs_flag);
645
646         if (!sums)
647                 return BLK_STS_RESOURCE;
648
649         sums->len = bio->bi_iter.bi_size;
650         INIT_LIST_HEAD(&sums->list);
651
652         if (contig)
653                 offset = file_start;
654         else
655                 offset = 0; /* shut up gcc */
656
657         sums->bytenr = bio->bi_iter.bi_sector << 9;
658         index = 0;
659
660         shash->tfm = fs_info->csum_shash;
661
662         bio_for_each_segment(bvec, bio, iter) {
663                 if (!contig)
664                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
665
666                 if (!ordered) {
667                         ordered = btrfs_lookup_ordered_extent(inode, offset);
668                         BUG_ON(!ordered); /* Logic error */
669                 }
670
671                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
672                                                  bvec.bv_len + fs_info->sectorsize
673                                                  - 1);
674
675                 for (i = 0; i < nr_sectors; i++) {
676                         if (offset >= ordered->file_offset + ordered->num_bytes ||
677                             offset < ordered->file_offset) {
678                                 unsigned long bytes_left;
679
680                                 sums->len = this_sum_bytes;
681                                 this_sum_bytes = 0;
682                                 btrfs_add_ordered_sum(ordered, sums);
683                                 btrfs_put_ordered_extent(ordered);
684
685                                 bytes_left = bio->bi_iter.bi_size - total_bytes;
686
687                                 nofs_flag = memalloc_nofs_save();
688                                 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
689                                                       bytes_left), GFP_KERNEL);
690                                 memalloc_nofs_restore(nofs_flag);
691                                 BUG_ON(!sums); /* -ENOMEM */
692                                 sums->len = bytes_left;
693                                 ordered = btrfs_lookup_ordered_extent(inode,
694                                                                 offset);
695                                 ASSERT(ordered); /* Logic error */
696                                 sums->bytenr = (bio->bi_iter.bi_sector << 9)
697                                         + total_bytes;
698                                 index = 0;
699                         }
700
701                         data = kmap_atomic(bvec.bv_page);
702                         crypto_shash_digest(shash, data + bvec.bv_offset
703                                             + (i * fs_info->sectorsize),
704                                             fs_info->sectorsize,
705                                             sums->sums + index);
706                         kunmap_atomic(data);
707                         index += fs_info->csum_size;
708                         offset += fs_info->sectorsize;
709                         this_sum_bytes += fs_info->sectorsize;
710                         total_bytes += fs_info->sectorsize;
711                 }
712
713         }
714         this_sum_bytes = 0;
715         btrfs_add_ordered_sum(ordered, sums);
716         btrfs_put_ordered_extent(ordered);
717         return 0;
718 }
719
720 /*
721  * helper function for csum removal, this expects the
722  * key to describe the csum pointed to by the path, and it expects
723  * the csum to overlap the range [bytenr, len]
724  *
725  * The csum should not be entirely contained in the range and the
726  * range should not be entirely contained in the csum.
727  *
728  * This calls btrfs_truncate_item with the correct args based on the
729  * overlap, and fixes up the key as required.
730  */
731 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
732                                        struct btrfs_path *path,
733                                        struct btrfs_key *key,
734                                        u64 bytenr, u64 len)
735 {
736         struct extent_buffer *leaf;
737         const u32 csum_size = fs_info->csum_size;
738         u64 csum_end;
739         u64 end_byte = bytenr + len;
740         u32 blocksize_bits = fs_info->sectorsize_bits;
741
742         leaf = path->nodes[0];
743         csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
744         csum_end <<= blocksize_bits;
745         csum_end += key->offset;
746
747         if (key->offset < bytenr && csum_end <= end_byte) {
748                 /*
749                  *         [ bytenr - len ]
750                  *         [   ]
751                  *   [csum     ]
752                  *   A simple truncate off the end of the item
753                  */
754                 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
755                 new_size *= csum_size;
756                 btrfs_truncate_item(path, new_size, 1);
757         } else if (key->offset >= bytenr && csum_end > end_byte &&
758                    end_byte > key->offset) {
759                 /*
760                  *         [ bytenr - len ]
761                  *                 [ ]
762                  *                 [csum     ]
763                  * we need to truncate from the beginning of the csum
764                  */
765                 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
766                 new_size *= csum_size;
767
768                 btrfs_truncate_item(path, new_size, 0);
769
770                 key->offset = end_byte;
771                 btrfs_set_item_key_safe(fs_info, path, key);
772         } else {
773                 BUG();
774         }
775 }
776
777 /*
778  * deletes the csum items from the csum tree for a given
779  * range of bytes.
780  */
781 int btrfs_del_csums(struct btrfs_trans_handle *trans,
782                     struct btrfs_root *root, u64 bytenr, u64 len)
783 {
784         struct btrfs_fs_info *fs_info = trans->fs_info;
785         struct btrfs_path *path;
786         struct btrfs_key key;
787         u64 end_byte = bytenr + len;
788         u64 csum_end;
789         struct extent_buffer *leaf;
790         int ret;
791         const u32 csum_size = fs_info->csum_size;
792         u32 blocksize_bits = fs_info->sectorsize_bits;
793
794         ASSERT(root == fs_info->csum_root ||
795                root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
796
797         path = btrfs_alloc_path();
798         if (!path)
799                 return -ENOMEM;
800
801         while (1) {
802                 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
803                 key.offset = end_byte - 1;
804                 key.type = BTRFS_EXTENT_CSUM_KEY;
805
806                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
807                 if (ret > 0) {
808                         if (path->slots[0] == 0)
809                                 break;
810                         path->slots[0]--;
811                 } else if (ret < 0) {
812                         break;
813                 }
814
815                 leaf = path->nodes[0];
816                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
817
818                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
819                     key.type != BTRFS_EXTENT_CSUM_KEY) {
820                         break;
821                 }
822
823                 if (key.offset >= end_byte)
824                         break;
825
826                 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
827                 csum_end <<= blocksize_bits;
828                 csum_end += key.offset;
829
830                 /* this csum ends before we start, we're done */
831                 if (csum_end <= bytenr)
832                         break;
833
834                 /* delete the entire item, it is inside our range */
835                 if (key.offset >= bytenr && csum_end <= end_byte) {
836                         int del_nr = 1;
837
838                         /*
839                          * Check how many csum items preceding this one in this
840                          * leaf correspond to our range and then delete them all
841                          * at once.
842                          */
843                         if (key.offset > bytenr && path->slots[0] > 0) {
844                                 int slot = path->slots[0] - 1;
845
846                                 while (slot >= 0) {
847                                         struct btrfs_key pk;
848
849                                         btrfs_item_key_to_cpu(leaf, &pk, slot);
850                                         if (pk.offset < bytenr ||
851                                             pk.type != BTRFS_EXTENT_CSUM_KEY ||
852                                             pk.objectid !=
853                                             BTRFS_EXTENT_CSUM_OBJECTID)
854                                                 break;
855                                         path->slots[0] = slot;
856                                         del_nr++;
857                                         key.offset = pk.offset;
858                                         slot--;
859                                 }
860                         }
861                         ret = btrfs_del_items(trans, root, path,
862                                               path->slots[0], del_nr);
863                         if (ret)
864                                 goto out;
865                         if (key.offset == bytenr)
866                                 break;
867                 } else if (key.offset < bytenr && csum_end > end_byte) {
868                         unsigned long offset;
869                         unsigned long shift_len;
870                         unsigned long item_offset;
871                         /*
872                          *        [ bytenr - len ]
873                          *     [csum                ]
874                          *
875                          * Our bytes are in the middle of the csum,
876                          * we need to split this item and insert a new one.
877                          *
878                          * But we can't drop the path because the
879                          * csum could change, get removed, extended etc.
880                          *
881                          * The trick here is the max size of a csum item leaves
882                          * enough room in the tree block for a single
883                          * item header.  So, we split the item in place,
884                          * adding a new header pointing to the existing
885                          * bytes.  Then we loop around again and we have
886                          * a nicely formed csum item that we can neatly
887                          * truncate.
888                          */
889                         offset = (bytenr - key.offset) >> blocksize_bits;
890                         offset *= csum_size;
891
892                         shift_len = (len >> blocksize_bits) * csum_size;
893
894                         item_offset = btrfs_item_ptr_offset(leaf,
895                                                             path->slots[0]);
896
897                         memzero_extent_buffer(leaf, item_offset + offset,
898                                              shift_len);
899                         key.offset = bytenr;
900
901                         /*
902                          * btrfs_split_item returns -EAGAIN when the
903                          * item changed size or key
904                          */
905                         ret = btrfs_split_item(trans, root, path, &key, offset);
906                         if (ret && ret != -EAGAIN) {
907                                 btrfs_abort_transaction(trans, ret);
908                                 goto out;
909                         }
910
911                         key.offset = end_byte - 1;
912                 } else {
913                         truncate_one_csum(fs_info, path, &key, bytenr, len);
914                         if (key.offset < bytenr)
915                                 break;
916                 }
917                 btrfs_release_path(path);
918         }
919         ret = 0;
920 out:
921         btrfs_free_path(path);
922         return ret;
923 }
924
925 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
926                            struct btrfs_root *root,
927                            struct btrfs_ordered_sum *sums)
928 {
929         struct btrfs_fs_info *fs_info = root->fs_info;
930         struct btrfs_key file_key;
931         struct btrfs_key found_key;
932         struct btrfs_path *path;
933         struct btrfs_csum_item *item;
934         struct btrfs_csum_item *item_end;
935         struct extent_buffer *leaf = NULL;
936         u64 next_offset;
937         u64 total_bytes = 0;
938         u64 csum_offset;
939         u64 bytenr;
940         u32 nritems;
941         u32 ins_size;
942         int index = 0;
943         int found_next;
944         int ret;
945         const u32 csum_size = fs_info->csum_size;
946
947         path = btrfs_alloc_path();
948         if (!path)
949                 return -ENOMEM;
950 again:
951         next_offset = (u64)-1;
952         found_next = 0;
953         bytenr = sums->bytenr + total_bytes;
954         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
955         file_key.offset = bytenr;
956         file_key.type = BTRFS_EXTENT_CSUM_KEY;
957
958         item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
959         if (!IS_ERR(item)) {
960                 ret = 0;
961                 leaf = path->nodes[0];
962                 item_end = btrfs_item_ptr(leaf, path->slots[0],
963                                           struct btrfs_csum_item);
964                 item_end = (struct btrfs_csum_item *)((char *)item_end +
965                            btrfs_item_size_nr(leaf, path->slots[0]));
966                 goto found;
967         }
968         ret = PTR_ERR(item);
969         if (ret != -EFBIG && ret != -ENOENT)
970                 goto out;
971
972         if (ret == -EFBIG) {
973                 u32 item_size;
974                 /* we found one, but it isn't big enough yet */
975                 leaf = path->nodes[0];
976                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
977                 if ((item_size / csum_size) >=
978                     MAX_CSUM_ITEMS(fs_info, csum_size)) {
979                         /* already at max size, make a new one */
980                         goto insert;
981                 }
982         } else {
983                 int slot = path->slots[0] + 1;
984                 /* we didn't find a csum item, insert one */
985                 nritems = btrfs_header_nritems(path->nodes[0]);
986                 if (!nritems || (path->slots[0] >= nritems - 1)) {
987                         ret = btrfs_next_leaf(root, path);
988                         if (ret < 0) {
989                                 goto out;
990                         } else if (ret > 0) {
991                                 found_next = 1;
992                                 goto insert;
993                         }
994                         slot = path->slots[0];
995                 }
996                 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
997                 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
998                     found_key.type != BTRFS_EXTENT_CSUM_KEY) {
999                         found_next = 1;
1000                         goto insert;
1001                 }
1002                 next_offset = found_key.offset;
1003                 found_next = 1;
1004                 goto insert;
1005         }
1006
1007         /*
1008          * At this point, we know the tree has a checksum item that ends at an
1009          * offset matching the start of the checksum range we want to insert.
1010          * We try to extend that item as much as possible and then add as many
1011          * checksums to it as they fit.
1012          *
1013          * First check if the leaf has enough free space for at least one
1014          * checksum. If it has go directly to the item extension code, otherwise
1015          * release the path and do a search for insertion before the extension.
1016          */
1017         if (btrfs_leaf_free_space(leaf) >= csum_size) {
1018                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1019                 csum_offset = (bytenr - found_key.offset) >>
1020                         fs_info->sectorsize_bits;
1021                 goto extend_csum;
1022         }
1023
1024         btrfs_release_path(path);
1025         path->search_for_extension = 1;
1026         ret = btrfs_search_slot(trans, root, &file_key, path,
1027                                 csum_size, 1);
1028         path->search_for_extension = 0;
1029         if (ret < 0)
1030                 goto out;
1031
1032         if (ret > 0) {
1033                 if (path->slots[0] == 0)
1034                         goto insert;
1035                 path->slots[0]--;
1036         }
1037
1038         leaf = path->nodes[0];
1039         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1040         csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1041
1042         if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1043             found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1044             csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1045                 goto insert;
1046         }
1047
1048 extend_csum:
1049         if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
1050             csum_size) {
1051                 int extend_nr;
1052                 u64 tmp;
1053                 u32 diff;
1054
1055                 tmp = sums->len - total_bytes;
1056                 tmp >>= fs_info->sectorsize_bits;
1057                 WARN_ON(tmp < 1);
1058
1059                 extend_nr = max_t(int, 1, (int)tmp);
1060                 diff = (csum_offset + extend_nr) * csum_size;
1061                 diff = min(diff,
1062                            MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1063
1064                 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
1065                 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1066                 diff /= csum_size;
1067                 diff *= csum_size;
1068
1069                 btrfs_extend_item(path, diff);
1070                 ret = 0;
1071                 goto csum;
1072         }
1073
1074 insert:
1075         btrfs_release_path(path);
1076         csum_offset = 0;
1077         if (found_next) {
1078                 u64 tmp;
1079
1080                 tmp = sums->len - total_bytes;
1081                 tmp >>= fs_info->sectorsize_bits;
1082                 tmp = min(tmp, (next_offset - file_key.offset) >>
1083                                          fs_info->sectorsize_bits);
1084
1085                 tmp = max_t(u64, 1, tmp);
1086                 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1087                 ins_size = csum_size * tmp;
1088         } else {
1089                 ins_size = csum_size;
1090         }
1091         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1092                                       ins_size);
1093         if (ret < 0)
1094                 goto out;
1095         if (WARN_ON(ret != 0))
1096                 goto out;
1097         leaf = path->nodes[0];
1098 csum:
1099         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1100         item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1101                                       btrfs_item_size_nr(leaf, path->slots[0]));
1102         item = (struct btrfs_csum_item *)((unsigned char *)item +
1103                                           csum_offset * csum_size);
1104 found:
1105         ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1106         ins_size *= csum_size;
1107         ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1108                               ins_size);
1109         write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1110                             ins_size);
1111
1112         index += ins_size;
1113         ins_size /= csum_size;
1114         total_bytes += ins_size * fs_info->sectorsize;
1115
1116         btrfs_mark_buffer_dirty(path->nodes[0]);
1117         if (total_bytes < sums->len) {
1118                 btrfs_release_path(path);
1119                 cond_resched();
1120                 goto again;
1121         }
1122 out:
1123         btrfs_free_path(path);
1124         return ret;
1125 }
1126
1127 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1128                                      const struct btrfs_path *path,
1129                                      struct btrfs_file_extent_item *fi,
1130                                      const bool new_inline,
1131                                      struct extent_map *em)
1132 {
1133         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1134         struct btrfs_root *root = inode->root;
1135         struct extent_buffer *leaf = path->nodes[0];
1136         const int slot = path->slots[0];
1137         struct btrfs_key key;
1138         u64 extent_start, extent_end;
1139         u64 bytenr;
1140         u8 type = btrfs_file_extent_type(leaf, fi);
1141         int compress_type = btrfs_file_extent_compression(leaf, fi);
1142
1143         btrfs_item_key_to_cpu(leaf, &key, slot);
1144         extent_start = key.offset;
1145         extent_end = btrfs_file_extent_end(path);
1146         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1147         if (type == BTRFS_FILE_EXTENT_REG ||
1148             type == BTRFS_FILE_EXTENT_PREALLOC) {
1149                 em->start = extent_start;
1150                 em->len = extent_end - extent_start;
1151                 em->orig_start = extent_start -
1152                         btrfs_file_extent_offset(leaf, fi);
1153                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1154                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1155                 if (bytenr == 0) {
1156                         em->block_start = EXTENT_MAP_HOLE;
1157                         return;
1158                 }
1159                 if (compress_type != BTRFS_COMPRESS_NONE) {
1160                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1161                         em->compress_type = compress_type;
1162                         em->block_start = bytenr;
1163                         em->block_len = em->orig_block_len;
1164                 } else {
1165                         bytenr += btrfs_file_extent_offset(leaf, fi);
1166                         em->block_start = bytenr;
1167                         em->block_len = em->len;
1168                         if (type == BTRFS_FILE_EXTENT_PREALLOC)
1169                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1170                 }
1171         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1172                 em->block_start = EXTENT_MAP_INLINE;
1173                 em->start = extent_start;
1174                 em->len = extent_end - extent_start;
1175                 /*
1176                  * Initialize orig_start and block_len with the same values
1177                  * as in inode.c:btrfs_get_extent().
1178                  */
1179                 em->orig_start = EXTENT_MAP_HOLE;
1180                 em->block_len = (u64)-1;
1181                 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1182                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1183                         em->compress_type = compress_type;
1184                 }
1185         } else {
1186                 btrfs_err(fs_info,
1187                           "unknown file extent item type %d, inode %llu, offset %llu, "
1188                           "root %llu", type, btrfs_ino(inode), extent_start,
1189                           root->root_key.objectid);
1190         }
1191 }
1192
1193 /*
1194  * Returns the end offset (non inclusive) of the file extent item the given path
1195  * points to. If it points to an inline extent, the returned offset is rounded
1196  * up to the sector size.
1197  */
1198 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1199 {
1200         const struct extent_buffer *leaf = path->nodes[0];
1201         const int slot = path->slots[0];
1202         struct btrfs_file_extent_item *fi;
1203         struct btrfs_key key;
1204         u64 end;
1205
1206         btrfs_item_key_to_cpu(leaf, &key, slot);
1207         ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1208         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1209
1210         if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1211                 end = btrfs_file_extent_ram_bytes(leaf, fi);
1212                 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1213         } else {
1214                 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1215         }
1216
1217         return end;
1218 }