2 * rehash.c --- rebuild hash tree directories
4 * Copyright (C) 2002 Theodore Ts'o
7 * This file may be redistributed under the terms of the GNU Public
11 * This algorithm is designed for simplicity of implementation and to
12 * pack the directory as much as possible. It however requires twice
13 * as much memory as the size of the directory. The maximum size
14 * directory supported using a 4k blocksize is roughly a gigabyte, and
15 * so there may very well be problems with machines that don't have
16 * virtual memory, and obscenely large directories.
18 * An alternate algorithm which is much more disk intensive could be
19 * written, and probably will need to be written in the future. The
20 * design goals of such an algorithm are: (a) use (roughly) constant
21 * amounts of memory, no matter how large the directory, (b) the
22 * directory must be safe at all times, even if e2fsck is interrupted
23 * in the middle, (c) we must use minimal amounts of extra disk
24 * blocks. This pretty much requires an incremental approach, where
25 * we are reading from one part of the directory, and inserting into
26 * the front half. So the algorithm will have to keep track of a
27 * moving block boundary between the new tree and the old tree, and
28 * files will need to be moved from the old directory and inserted
29 * into the new tree. If the new directory requires space which isn't
30 * yet available, blocks from the beginning part of the old directory
31 * may need to be moved to the end of the directory to make room for
34 * --------------------------------------------------------
35 * | new tree | | old tree |
36 * --------------------------------------------------------
40 * This is going to be a pain in the tuckus to implement, and will
41 * require a lot more disk accesses. So I'm going to skip it for now;
42 * it's only really going to be an issue for really, really big
43 * filesystems (when we reach the level of tens of millions of files
44 * in a single directory). It will probably be easier to simply
45 * require that e2fsck use VM first.
54 #include "support/sort_r.h"
56 /* Schedule a dir to be rebuilt during pass 3A. */
57 void e2fsck_rehash_dir_later(e2fsck_t ctx, ext2_ino_t ino)
59 if (!ctx->dirs_to_hash)
60 ext2fs_u32_list_create(&ctx->dirs_to_hash, 50);
61 if (ctx->dirs_to_hash)
62 ext2fs_u32_list_add(ctx->dirs_to_hash, ino);
65 /* Ask if a dir will be rebuilt during pass 3A. */
66 int e2fsck_dir_will_be_rehashed(e2fsck_t ctx, ext2_ino_t ino)
68 if (ctx->options & E2F_OPT_COMPRESS_DIRS)
70 if (!ctx->dirs_to_hash)
72 return ext2fs_u32_list_test(ctx->dirs_to_hash, ino);
77 struct fill_dir_struct {
79 struct ext2_inode *inode;
83 struct hash_entry *harray;
84 blk_t max_array, num_array;
85 ext2_off64_t dir_size;
93 ext2_dirhash_t minor_hash;
95 struct ext2_dir_entry *dir;
102 ext2_dirhash_t *hashes;
105 #define DOTDOT_OFFSET 12
107 static int is_fake_entry(ext2_filsys fs, int lblk, unsigned int offset)
109 /* Entries in the first block before this value refer to . or .. */
110 if (lblk == 0 && offset <= DOTDOT_OFFSET)
112 /* Check if this is likely the csum entry */
113 if (ext2fs_has_feature_metadata_csum(fs->super) &&
114 (offset & (fs->blocksize - 1)) ==
115 fs->blocksize - sizeof(struct ext2_dir_entry_tail))
120 static int fill_dir_block(ext2_filsys fs,
122 e2_blkcnt_t blockcnt,
123 blk64_t ref_block EXT2FS_ATTR((unused)),
124 int ref_offset EXT2FS_ATTR((unused)),
127 struct fill_dir_struct *fd = (struct fill_dir_struct *) priv_data;
128 struct hash_entry *ent;
129 struct ext2_dir_entry *dirent;
131 unsigned int offset, dir_offset, rec_len, name_len;
132 int hash_alg, hash_flags, hash_in_entry;
137 offset = blockcnt * fs->blocksize;
138 if (offset + fs->blocksize > fd->inode->i_size) {
139 fd->err = EXT2_ET_DIR_CORRUPTED;
143 dir = (fd->buf+offset);
144 if (*block_nr == 0) {
145 memset(dir, 0, fs->blocksize);
146 dirent = (struct ext2_dir_entry *) dir;
147 (void) ext2fs_set_rec_len(fs, fs->blocksize, dirent);
149 int flags = fs->flags;
150 fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
151 fd->err = ext2fs_read_dir_block4(fs, *block_nr, dir, 0,
153 fs->flags = (flags & EXT2_FLAG_IGNORE_CSUM_ERRORS) |
154 (fs->flags & ~EXT2_FLAG_IGNORE_CSUM_ERRORS);
158 hash_flags = fd->inode->i_flags & EXT4_CASEFOLD_FL;
159 hash_in_entry = ext4_hash_in_dirent(fd->inode);
160 hash_alg = fs->super->s_def_hash_version;
161 if ((hash_alg <= EXT2_HASH_TEA) &&
162 (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
164 /* While the directory block is "hot", index it. */
166 while (dir_offset < fs->blocksize) {
167 unsigned int min_rec = EXT2_DIR_ENTRY_HEADER_LEN;
168 int extended = hash_in_entry && !is_fake_entry(fs, blockcnt, dir_offset);
171 min_rec += EXT2_DIR_ENTRY_HASH_LEN;
172 dirent = (struct ext2_dir_entry *) (dir + dir_offset);
173 (void) ext2fs_get_rec_len(fs, dirent, &rec_len);
174 name_len = ext2fs_dirent_name_len(dirent);
175 if (((dir_offset + rec_len) > fs->blocksize) ||
176 (rec_len < min_rec) ||
177 ((rec_len % 4) != 0) ||
178 (name_len + min_rec > rec_len)) {
179 fd->err = EXT2_ET_DIR_CORRUPTED;
182 dir_offset += rec_len;
183 if (dirent->inode == 0)
185 if ((name_len) == 0) {
186 fd->err = EXT2_ET_DIR_CORRUPTED;
189 if (!fd->compress && (name_len == 1) &&
190 (dirent->name[0] == '.'))
192 if (!fd->compress && (name_len == 2) &&
193 (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
194 fd->parent = dirent->inode;
197 if (fd->num_array >= fd->max_array) {
200 retval = ext2fs_resize_array(sizeof(struct hash_entry),
208 fd->max_array += 500;
210 ent = fd->harray + fd->num_array++;
212 fd->dir_size += ext2fs_dir_rec_len(name_len, extended);
213 ent->ino = dirent->inode;
215 ent->hash = EXT2_DIRENT_HASH(dirent);
216 ent->minor_hash = EXT2_DIRENT_MINOR_HASH(dirent);
217 } else if (fd->compress) {
218 ent->hash = ent->minor_hash = 0;
220 fd->err = ext2fs_dirhash2(hash_alg,
221 dirent->name, name_len,
222 fs->encoding, hash_flags,
223 fs->super->s_hash_seed,
224 &ent->hash, &ent->minor_hash);
233 /* Used for sorting the hash entry */
234 static EXT2_QSORT_TYPE ino_cmp(const void *a, const void *b)
236 const struct hash_entry *he_a = (const struct hash_entry *) a;
237 const struct hash_entry *he_b = (const struct hash_entry *) b;
239 return (he_a->ino - he_b->ino);
245 const struct ext2fs_nls_table *tbl;
248 static int same_name(const struct name_cmp_ctx *cmp_ctx, char *s1,
249 int len1, char *s2, int len2)
251 if (!cmp_ctx->casefold)
252 return (len1 == len2 && !memcmp(s1, s2, len1));
254 return !ext2fs_casefold_cmp(cmp_ctx->tbl,
255 (unsigned char *) s1, len1,
256 (unsigned char *) s2, len2);
259 /* Used for sorting the hash entry */
260 static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
262 const struct hash_entry *he_a = (const struct hash_entry *) a;
263 const struct hash_entry *he_b = (const struct hash_entry *) b;
264 unsigned int he_a_len, he_b_len, min_len;
267 he_a_len = ext2fs_dirent_name_len(he_a->dir);
268 he_b_len = ext2fs_dirent_name_len(he_b->dir);
270 if (min_len > he_b_len)
273 ret = memcmp(he_a->dir->name, he_b->dir->name, min_len);
275 if (he_a_len > he_b_len)
277 else if (he_a_len < he_b_len)
280 ret = he_b->dir->inode - he_a->dir->inode;
285 static EXT2_QSORT_TYPE name_cf_cmp(const struct name_cmp_ctx *ctx,
286 const void *a, const void *b)
288 const struct hash_entry *he_a = (const struct hash_entry *) a;
289 const struct hash_entry *he_b = (const struct hash_entry *) b;
290 unsigned int he_a_len, he_b_len;
293 he_a_len = ext2fs_dirent_name_len(he_a->dir);
294 he_b_len = ext2fs_dirent_name_len(he_b->dir);
296 ret = ext2fs_casefold_cmp(ctx->tbl,
297 (unsigned char *) he_a->dir->name, he_a_len,
298 (unsigned char *) he_b->dir->name, he_b_len);
300 if (he_a_len > he_b_len)
302 else if (he_a_len < he_b_len)
305 ret = he_b->dir->inode - he_a->dir->inode;
310 /* Used for sorting the hash entry */
311 static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b, void *arg)
313 const struct name_cmp_ctx *ctx = (struct name_cmp_ctx *) arg;
314 const struct hash_entry *he_a = (const struct hash_entry *) a;
315 const struct hash_entry *he_b = (const struct hash_entry *) b;
318 if (he_a->hash > he_b->hash)
320 else if (he_a->hash < he_b->hash)
323 if (he_a->minor_hash > he_b->minor_hash)
325 else if (he_a->minor_hash < he_b->minor_hash)
329 ret = name_cf_cmp(ctx, a, b);
331 ret = name_cmp(a, b);
337 static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
343 retval = ext2fs_resize_array(fs->blocksize, outdir->max, blocks,
347 retval = ext2fs_resize_array(sizeof(ext2_dirhash_t),
353 retval = ext2fs_get_array(fs->blocksize, blocks, &outdir->buf);
356 retval = ext2fs_get_array(sizeof(ext2_dirhash_t), blocks,
362 outdir->max = blocks;
366 static void free_out_dir(struct out_dir *outdir)
369 free(outdir->hashes);
374 static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
379 if (outdir->num >= outdir->max) {
380 int increment = outdir->max / 10;
384 retval = alloc_size_dir(fs, outdir, outdir->max + increment);
388 *ret = outdir->buf + (size_t)outdir->num++ * fs->blocksize;
389 memset(*ret, 0, fs->blocksize);
394 * This function is used to make a unique filename. We do this by
395 * appending ~0, and then incrementing the number. However, we cannot
396 * expand the length of the filename beyond the padding available in
397 * the directory entry.
399 static void mutate_name(char *str, unsigned int *len)
402 unsigned int l = *len;
405 * First check to see if it looks the name has been mutated
408 for (i = l-1; i > 0; i--) {
409 if (!isdigit(str[i]))
412 if ((i == (int)l - 1) || (str[i] != '~')) {
422 for (i = l-1; i >= 0; i--) {
423 if (isdigit(str[i])) {
435 else if (str[0] == 'Z') {
453 static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
455 struct fill_dir_struct *fd,
456 const struct name_cmp_ctx *cmp_ctx)
458 struct problem_context pctx;
459 struct hash_entry *ent, *prev;
463 unsigned int new_len;
465 int hash_flags = fd->inode->i_flags & EXT4_CASEFOLD_FL;
467 clear_problem_context(&pctx);
470 hash_alg = fs->super->s_def_hash_version;
471 if ((hash_alg <= EXT2_HASH_TEA) &&
472 (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
475 for (i=1; i < fd->num_array; i++) {
476 ent = fd->harray + i;
478 if (!ent->dir->inode ||
479 !same_name(cmp_ctx, ent->dir->name,
480 ext2fs_dirent_name_len(ent->dir),
482 ext2fs_dirent_name_len(prev->dir)))
484 pctx.dirent = ent->dir;
485 if ((ent->dir->inode == prev->dir->inode) &&
486 fix_problem(ctx, PR_2_DUPLICATE_DIRENT, &pctx)) {
487 e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
492 /* Can't alter encrypted name without key, so just drop it */
493 if (fd->inode->i_flags & EXT4_ENCRYPT_FL) {
494 if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE_NO_RENAME, &pctx)) {
495 e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
501 new_len = ext2fs_dirent_name_len(ent->dir);
503 /* should never happen */
504 ext2fs_unmark_valid(fs);
507 memcpy(new_name, ent->dir->name, new_len);
508 mutate_name(new_name, &new_len);
509 for (j=0; j < fd->num_array; j++) {
511 !same_name(cmp_ctx, new_name, new_len,
512 fd->harray[j].dir->name,
513 ext2fs_dirent_name_len(fd->harray[j].dir))) {
516 mutate_name(new_name, &new_len);
520 new_name[new_len] = 0;
522 if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
523 memcpy(ent->dir->name, new_name, new_len);
524 ext2fs_dirent_set_name_len(ent->dir, new_len);
525 ext2fs_dirhash2(hash_alg, new_name, new_len,
526 fs->encoding, hash_flags,
527 fs->super->s_hash_seed,
528 &ent->hash, &ent->minor_hash);
536 static errcode_t copy_dir_entries(e2fsck_t ctx,
537 struct fill_dir_struct *fd,
538 struct out_dir *outdir)
540 ext2_filsys fs = ctx->fs;
543 struct hash_entry *ent;
544 struct ext2_dir_entry *dirent;
545 unsigned int rec_len, prev_rec_len, left, slack, offset;
547 ext2_dirhash_t prev_hash;
549 struct ext2_dir_entry_tail *t;
550 int hash_in_entry = ext4_hash_in_dirent(fd->inode);
551 unsigned int min_rec_len = ext2fs_dir_rec_len(1, hash_in_entry);
553 if (ctx->htree_slack_percentage == 255) {
554 profile_get_uint(ctx->profile, "options",
555 "indexed_dir_slack_percentage",
557 &ctx->htree_slack_percentage);
558 if (ctx->htree_slack_percentage > 100)
559 ctx->htree_slack_percentage = 20;
562 if (ext2fs_has_feature_metadata_csum(fs->super))
563 csum_size = sizeof(struct ext2_dir_entry_tail);
566 retval = alloc_size_dir(fs, outdir,
567 (fd->dir_size / fs->blocksize) + 2);
570 outdir->num = fd->compress ? 0 : 1;
572 outdir->hashes[0] = 0;
574 if ((retval = get_next_block(fs, outdir, &block_start)))
576 dirent = (struct ext2_dir_entry *) block_start;
579 left = fs->blocksize - csum_size;
580 slack = fd->compress ? min_rec_len :
581 ((fs->blocksize - csum_size) * ctx->htree_slack_percentage)/100;
582 if (slack < min_rec_len)
584 for (i = 0; i < fd->num_array; i++) {
585 ent = fd->harray + i;
586 if (ent->dir->inode == 0)
588 rec_len = ext2fs_dir_rec_len(ext2fs_dirent_name_len(ent->dir),
590 if (rec_len > left) {
592 left += prev_rec_len;
593 retval = ext2fs_set_rec_len(fs, left, dirent);
598 t = EXT2_DIRENT_TAIL(block_start,
600 ext2fs_initialize_dirent_tail(fs, t);
602 if ((retval = get_next_block(fs, outdir,
607 left = (fs->blocksize - csum_size) - offset;
608 dirent = (struct ext2_dir_entry *) (block_start + offset);
610 if (ent->hash == prev_hash)
611 outdir->hashes[outdir->num-1] = ent->hash | 1;
613 outdir->hashes[outdir->num-1] = ent->hash;
615 dirent->inode = ent->dir->inode;
616 ext2fs_dirent_set_name_len(dirent,
617 ext2fs_dirent_name_len(ent->dir));
618 ext2fs_dirent_set_file_type(dirent,
619 ext2fs_dirent_file_type(ent->dir));
620 retval = ext2fs_set_rec_len(fs, rec_len, dirent);
623 prev_rec_len = rec_len;
624 memcpy(dirent->name, ent->dir->name,
625 ext2fs_dirent_name_len(dirent));
627 EXT2_DIRENT_HASHES(dirent)->hash = ext2fs_cpu_to_le32(ent->hash);
628 EXT2_DIRENT_HASHES(dirent)->minor_hash =
629 ext2fs_cpu_to_le32(ent->minor_hash);
634 prev_rec_len += left;
635 retval = ext2fs_set_rec_len(fs, prev_rec_len, dirent);
641 prev_hash = ent->hash;
644 retval = ext2fs_set_rec_len(fs, rec_len + left, dirent);
646 t = EXT2_DIRENT_TAIL(block_start, fs->blocksize);
647 ext2fs_initialize_dirent_tail(fs, t);
654 static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
655 ext2_ino_t ino, ext2_ino_t parent,
656 struct ext2_inode *inode)
658 struct ext2_dir_entry *dir;
659 struct ext2_dx_root_info *root;
660 struct ext2_dx_countlimit *limits;
664 if (ext2fs_has_feature_filetype(fs->super))
665 filetype = EXT2_FT_DIR;
667 memset(buf, 0, fs->blocksize);
668 dir = (struct ext2_dir_entry *) buf;
671 ext2fs_dirent_set_name_len(dir, 1);
672 ext2fs_dirent_set_file_type(dir, filetype);
674 dir = (struct ext2_dir_entry *) (buf + 12);
678 ext2fs_dirent_set_name_len(dir, 2);
679 ext2fs_dirent_set_file_type(dir, filetype);
680 dir->rec_len = fs->blocksize - 12;
682 root = (struct ext2_dx_root_info *) (buf+24);
683 root->reserved_zero = 0;
684 if (ext4_hash_in_dirent(inode))
685 root->hash_version = EXT2_HASH_SIPHASH;
687 root->hash_version = fs->super->s_def_hash_version;
688 root->info_length = 8;
689 root->indirect_levels = 0;
690 root->unused_flags = 0;
692 if (ext2fs_has_feature_metadata_csum(fs->super))
693 csum_size = sizeof(struct ext2_dx_tail);
695 limits = (struct ext2_dx_countlimit *) (buf+32);
696 limits->limit = (fs->blocksize - (32 + csum_size)) /
697 sizeof(struct ext2_dx_entry);
704 static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
706 struct ext2_dir_entry *dir;
707 struct ext2_dx_countlimit *limits;
710 memset(buf, 0, fs->blocksize);
711 dir = (struct ext2_dir_entry *) buf;
713 (void) ext2fs_set_rec_len(fs, fs->blocksize, dir);
715 if (ext2fs_has_feature_metadata_csum(fs->super))
716 csum_size = sizeof(struct ext2_dx_tail);
718 limits = (struct ext2_dx_countlimit *) (buf+8);
719 limits->limit = (fs->blocksize - (8 + csum_size)) /
720 sizeof(struct ext2_dx_entry);
723 return (struct ext2_dx_entry *) limits;
726 static int alloc_blocks(ext2_filsys fs,
727 struct ext2_dx_countlimit **limit,
728 struct ext2_dx_entry **prev_ent,
729 struct ext2_dx_entry **next_ent,
730 int *prev_offset, int *next_offset,
731 struct out_dir *outdir, int i,
732 int *prev_count, int *next_count)
738 (*limit)->limit = (*limit)->count =
739 ext2fs_cpu_to_le16((*limit)->limit);
740 *prev_ent = (struct ext2_dx_entry *) (outdir->buf + *prev_offset);
741 (*prev_ent)->block = ext2fs_cpu_to_le32(outdir->num);
745 ext2fs_cpu_to_le32(outdir->hashes[i]);
747 retval = get_next_block(fs, outdir, &block_start);
751 /* outdir->buf might be reallocated */
752 *prev_ent = (struct ext2_dx_entry *) (outdir->buf + *prev_offset);
754 *next_ent = set_int_node(fs, block_start);
755 *limit = (struct ext2_dx_countlimit *)(*next_ent);
757 *next_offset = ((char *) *next_ent - outdir->buf);
759 *next_count = (*limit)->limit;
760 (*prev_offset) += sizeof(struct ext2_dx_entry);
767 * This function takes the leaf nodes which have been written in
768 * outdir, and populates the root node and any necessary interior nodes.
770 static errcode_t calculate_tree(ext2_filsys fs,
771 struct out_dir *outdir,
774 struct ext2_inode *inode)
776 struct ext2_dx_root_info *root_info;
777 struct ext2_dx_entry *root, *int_ent, *dx_ent = 0;
778 struct ext2_dx_countlimit *root_limit, *int_limit, *limit;
780 int i, c1, c2, c3, nblks;
781 int limit_offset, int_offset, root_offset;
783 root_info = set_root_node(fs, outdir->buf, ino, parent, inode);
784 root_offset = limit_offset = ((char *) root_info - outdir->buf) +
785 root_info->info_length;
786 root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
787 c1 = root_limit->limit;
790 /* Write out the pointer blocks */
791 if (nblks - 1 <= c1) {
792 /* Just write out the root block, and we're done */
793 root = (struct ext2_dx_entry *) (outdir->buf + root_offset);
794 for (i=1; i < nblks; i++) {
795 root->block = ext2fs_cpu_to_le32(i);
798 ext2fs_cpu_to_le32(outdir->hashes[i]);
802 } else if (nblks - 1 <= ext2fs_htree_intnode_maxrecs(fs, c1)) {
805 root_info->indirect_levels = 1;
806 for (i=1; i < nblks; i++) {
807 if (c2 == 0 && c1 == 0)
810 retval = alloc_blocks(fs, &limit, &root,
811 &dx_ent, &root_offset,
812 NULL, outdir, i, &c1,
817 dx_ent->block = ext2fs_cpu_to_le32(i);
818 if (c2 != limit->limit)
820 ext2fs_cpu_to_le32(outdir->hashes[i]);
824 limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
825 limit->limit = ext2fs_cpu_to_le16(limit->limit);
831 root_info->indirect_levels = 2;
832 for (i = 1; i < nblks; i++) {
833 if (c3 == 0 && c2 == 0 && c1 == 0)
835 if (c3 == 0 && c2 == 0) {
836 retval = alloc_blocks(fs, &int_limit, &root,
837 &int_ent, &root_offset,
838 &int_offset, outdir, i,
844 int delta1 = (char *)int_limit - outdir->buf;
845 int delta2 = (char *)root - outdir->buf;
847 retval = alloc_blocks(fs, &limit, &int_ent,
848 &dx_ent, &int_offset,
849 NULL, outdir, i, &c2,
854 /* outdir->buf might be reallocated */
855 int_limit = (struct ext2_dx_countlimit *)
856 (outdir->buf + delta1);
857 root = (struct ext2_dx_entry *)
858 (outdir->buf + delta2);
860 dx_ent->block = ext2fs_cpu_to_le32(i);
861 if (c3 != limit->limit)
863 ext2fs_cpu_to_le32(outdir->hashes[i]);
867 int_limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
868 int_limit->limit = ext2fs_cpu_to_le16(limit->limit);
870 limit->count = ext2fs_cpu_to_le16(limit->limit - c3);
871 limit->limit = ext2fs_cpu_to_le16(limit->limit);
874 root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
875 root_limit->count = ext2fs_cpu_to_le16(root_limit->limit - c1);
876 root_limit->limit = ext2fs_cpu_to_le16(root_limit->limit);
881 struct write_dir_struct {
882 struct out_dir *outdir;
890 * Helper function which writes out a directory block.
892 static int write_dir_block(ext2_filsys fs,
894 e2_blkcnt_t blockcnt,
895 blk64_t ref_block EXT2FS_ATTR((unused)),
896 int ref_offset EXT2FS_ATTR((unused)),
899 struct write_dir_struct *wd = (struct write_dir_struct *) priv_data;
903 printf("%u: write_dir_block %lld:%lld", wd->ino, blockcnt, *block_nr);
905 if ((*block_nr == 0) || (blockcnt < 0)) {
911 if (blockcnt < wd->outdir->num)
912 dir = wd->outdir->buf + (blockcnt * fs->blocksize);
913 else if (wd->ctx->lost_and_found == wd->dir) {
914 /* Don't release any extra directory blocks for lost+found */
915 wd->err = ext2fs_new_dir_block(fs, 0, 0, &buf);
921 /* Don't free blocks at the end of the directory, they
922 * will be truncated by the caller. */
924 printf(" - not freed\n");
928 wd->err = ext2fs_write_dir_block4(fs, *block_nr, dir, 0, wd->dir);
930 ext2fs_free_mem(&buf);
933 printf(" - write (%d)\n", wd->err);
940 static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
941 struct out_dir *outdir,
942 ext2_ino_t ino, struct ext2_inode *inode,
945 struct write_dir_struct wd;
948 retval = e2fsck_expand_directory(ctx, ino, -1, outdir->num);
958 retval = ext2fs_block_iterate3(fs, ino, 0, NULL,
959 write_dir_block, &wd);
965 e2fsck_read_inode(ctx, ino, inode, "rehash_dir");
967 inode->i_flags &= ~EXT2_INDEX_FL;
969 inode->i_flags |= EXT2_INDEX_FL;
971 printf("%u: set inode size to %u blocks = %u bytes\n",
972 ino, outdir->num, outdir->num * fs->blocksize);
974 retval = ext2fs_inode_size_set(fs, inode, (ext2_off64_t)outdir->num *
979 /* ext2fs_punch() calls ext2fs_write_inode() which writes the size */
980 return ext2fs_punch(fs, ino, inode, NULL, outdir->num, ~0ULL);
983 errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino,
984 struct problem_context *pctx)
986 ext2_filsys fs = ctx->fs;
988 struct ext2_inode inode;
990 struct fill_dir_struct fd = { NULL, NULL, 0, 0, 0, NULL,
992 struct out_dir outdir = { 0, 0, 0, 0 };
993 struct name_cmp_ctx name_cmp_ctx = {0, NULL};
995 e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
997 if (ext2fs_has_feature_inline_data(fs->super) &&
998 (inode.i_flags & EXT4_INLINE_DATA_FL))
1001 retval = ext2fs_get_mem(inode.i_size, &dir_buf);
1005 fd.max_array = inode.i_size / 32;
1006 retval = ext2fs_get_array(sizeof(struct hash_entry),
1007 fd.max_array, &fd.harray);
1016 if (!ext2fs_has_feature_dir_index(fs->super) ||
1017 (inode.i_size / fs->blocksize) < 2)
1021 if (fs->encoding && (inode.i_flags & EXT4_CASEFOLD_FL)) {
1022 name_cmp_ctx.casefold = 1;
1023 name_cmp_ctx.tbl = fs->encoding;
1027 /* Read in the entire directory into memory */
1028 retval = ext2fs_block_iterate3(fs, ino, 0, 0,
1029 fill_dir_block, &fd);
1036 * If the entries read are less than a block, then don't index
1039 if (!fd.compress && (fd.dir_size < (fs->blocksize - 24))) {
1047 printf("%d entries (%d bytes) found in inode %d\n",
1048 fd.num_array, fd.dir_size, ino);
1053 if (fd.compress && fd.num_array > 1)
1054 sort_r_simple(fd.harray+2, fd.num_array-2,
1055 sizeof(struct hash_entry),
1056 hash_cmp, &name_cmp_ctx);
1058 sort_r_simple(fd.harray, fd.num_array,
1059 sizeof(struct hash_entry),
1060 hash_cmp, &name_cmp_ctx);
1063 * Look for duplicates
1065 if (duplicate_search_and_fix(ctx, fs, ino, &fd, &name_cmp_ctx))
1068 if (ctx->options & E2F_OPT_NO) {
1073 /* Sort non-hashed directories by inode number */
1074 if (fd.compress && fd.num_array > 1)
1075 qsort(fd.harray+2, fd.num_array-2,
1076 sizeof(struct hash_entry), ino_cmp);
1079 * Copy the directory entries. In a htree directory these
1080 * will become the leaf nodes.
1082 retval = copy_dir_entries(ctx, &fd, &outdir);
1086 free(dir_buf); dir_buf = 0;
1089 /* Calculate the interior nodes */
1090 retval = calculate_tree(fs, &outdir, ino, fd.parent, fd.inode);
1095 retval = write_directory(ctx, fs, &outdir, ino, &inode, fd.compress);
1099 if (ctx->options & E2F_OPT_CONVERT_BMAP)
1100 retval = e2fsck_rebuild_extents_later(ctx, ino);
1102 retval = e2fsck_check_rebuild_extents(ctx, ino, &inode, pctx);
1104 ext2fs_free_mem(&dir_buf);
1105 ext2fs_free_mem(&fd.harray);
1107 free_out_dir(&outdir);
1111 void e2fsck_rehash_directories(e2fsck_t ctx)
1113 struct problem_context pctx;
1114 #ifdef RESOURCE_TRACK
1115 struct resource_track rtrack;
1117 struct dir_info *dir;
1118 ext2_u32_iterate iter;
1119 struct dir_info_iter * dirinfo_iter = 0;
1122 int cur, max, all_dirs, first = 1;
1124 init_resource_track(&rtrack, ctx->fs->io);
1125 all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
1127 if (!ctx->dirs_to_hash && !all_dirs)
1130 (void) e2fsck_get_lost_and_found(ctx, 0);
1132 clear_problem_context(&pctx);
1136 dirinfo_iter = e2fsck_dir_info_iter_begin(ctx);
1137 max = e2fsck_get_num_dirinfo(ctx);
1139 retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
1142 pctx.errcode = retval;
1143 fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
1146 max = ext2fs_u32_list_count(ctx->dirs_to_hash);
1150 if ((dir = e2fsck_dir_info_iter(ctx,
1151 dirinfo_iter)) == 0)
1155 if (!ext2fs_u32_list_iterate(iter, &ino))
1158 if (!ext2fs_test_inode_bitmap2(ctx->inode_dir_map, ino))
1163 fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
1167 fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
1169 pctx.errcode = e2fsck_rehash_dir(ctx, ino, &pctx);
1171 end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
1172 fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
1174 if (ctx->progress && !ctx->progress_fd)
1175 e2fsck_simple_progress(ctx, "Rebuilding directory",
1176 100.0 * (float) (++cur) / (float) max, ino);
1178 end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
1180 e2fsck_dir_info_iter_end(ctx, dirinfo_iter);
1182 ext2fs_u32_list_iterate_end(iter);
1184 if (ctx->dirs_to_hash)
1185 ext2fs_u32_list_free(ctx->dirs_to_hash);
1186 ctx->dirs_to_hash = 0;
1188 print_resource_track(ctx, "Pass 3A", &rtrack, ctx->fs->io);