]> git.ipfire.org Git - thirdparty/e2fsprogs.git/blob - e2fsck/rehash.c
projects / thirdparty / e2fsprogs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'maint' into next
[thirdparty/e2fsprogs.git] / e2fsck / rehash.c
1 /*
2 * rehash.c --- rebuild hash tree directories
3 *
4 * Copyright (C) 2002 Theodore Ts'o
5 *
6 * %Begin-Header%
7 * This file may be redistributed under the terms of the GNU Public
8 * License.
9 * %End-Header%
10 *
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.
17 *
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
32 * the new tree:
33 *
34 * --------------------------------------------------------
35 * | new tree | | old tree |
36 * --------------------------------------------------------
37 * ^ ptr ^ptr
38 * tail new head old
39 *
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.
46 */
47
48 #include "config.h"
49 #include <string.h>
50 #include <ctype.h>
51 #include <errno.h>
52 #include "e2fsck.h"
53 #include "problem.h"
54
55 /* Schedule a dir to be rebuilt during pass 3A. */
56 void e2fsck_rehash_dir_later(e2fsck_t ctx, ext2_ino_t ino)
57 {
58 if (!ctx->dirs_to_hash)
59 ext2fs_u32_list_create(&ctx->dirs_to_hash, 50);
60 if (ctx->dirs_to_hash)
61 ext2fs_u32_list_add(ctx->dirs_to_hash, ino);
62 }
63
64 /* Ask if a dir will be rebuilt during pass 3A. */
65 int e2fsck_dir_will_be_rehashed(e2fsck_t ctx, ext2_ino_t ino)
66 {
67 if (ctx->options & E2F_OPT_COMPRESS_DIRS)
68 return 1;
69 if (!ctx->dirs_to_hash)
70 return 0;
71 return ext2fs_u32_list_test(ctx->dirs_to_hash, ino);
72 }
73
74 #undef REHASH_DEBUG
75
76 struct fill_dir_struct {
77 char *buf;
78 struct ext2_inode *inode;
79 ext2_ino_t ino;
80 errcode_t err;
81 e2fsck_t ctx;
82 struct hash_entry *harray;
83 int max_array, num_array;
84 unsigned int dir_size;
85 int compress;
86 ino_t parent;
87 ext2_ino_t dir;
88 };
89
90 struct hash_entry {
91 ext2_dirhash_t hash;
92 ext2_dirhash_t minor_hash;
93 ino_t ino;
94 struct ext2_dir_entry *dir;
95 };
96
97 struct out_dir {
98 int num;
99 int max;
100 char *buf;
101 ext2_dirhash_t *hashes;
102 };
103
104 static int fill_dir_block(ext2_filsys fs,
105 blk64_t *block_nr,
106 e2_blkcnt_t blockcnt,
107 blk64_t ref_block EXT2FS_ATTR((unused)),
108 int ref_offset EXT2FS_ATTR((unused)),
109 void *priv_data)
110 {
111 struct fill_dir_struct *fd = (struct fill_dir_struct *) priv_data;
112 struct hash_entry *new_array, *ent;
113 struct ext2_dir_entry *dirent;
114 char *dir;
115 unsigned int offset, dir_offset, rec_len, name_len;
116 int hash_alg;
117
118 if (blockcnt < 0)
119 return 0;
120
121 offset = blockcnt * fs->blocksize;
122 if (offset + fs->blocksize > fd->inode->i_size) {
123 fd->err = EXT2_ET_DIR_CORRUPTED;
124 return BLOCK_ABORT;
125 }
126
127 dir = (fd->buf+offset);
128 if (*block_nr == 0) {
129 memset(dir, 0, fs->blocksize);
130 dirent = (struct ext2_dir_entry *) dir;
131 (void) ext2fs_set_rec_len(fs, fs->blocksize, dirent);
132 } else {
133 int flags = fs->flags;
134 fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
135 fd->err = ext2fs_read_dir_block4(fs, *block_nr, dir, 0,
136 fd->dir);
137 fs->flags = (flags & EXT2_FLAG_IGNORE_CSUM_ERRORS) |
138 (fs->flags & ~EXT2_FLAG_IGNORE_CSUM_ERRORS);
139 if (fd->err)
140 return BLOCK_ABORT;
141 }
142 hash_alg = fs->super->s_def_hash_version;
143 if ((hash_alg <= EXT2_HASH_TEA) &&
144 (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
145 hash_alg += 3;
146 /* While the directory block is "hot", index it. */
147 dir_offset = 0;
148 while (dir_offset < fs->blocksize) {
149 dirent = (struct ext2_dir_entry *) (dir + dir_offset);
150 (void) ext2fs_get_rec_len(fs, dirent, &rec_len);
151 name_len = ext2fs_dirent_name_len(dirent);
152 if (((dir_offset + rec_len) > fs->blocksize) ||
153 (rec_len < 8) ||
154 ((rec_len % 4) != 0) ||
155 (name_len + 8 > rec_len)) {
156 fd->err = EXT2_ET_DIR_CORRUPTED;
157 return BLOCK_ABORT;
158 }
159 dir_offset += rec_len;
160 if (dirent->inode == 0)
161 continue;
162 if (!fd->compress && (name_len == 1) &&
163 (dirent->name[0] == '.'))
164 continue;
165 if (!fd->compress && (name_len == 2) &&
166 (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
167 fd->parent = dirent->inode;
168 continue;
169 }
170 if (fd->num_array >= fd->max_array) {
171 new_array = realloc(fd->harray,
172 sizeof(struct hash_entry) * (fd->max_array+500));
173 if (!new_array) {
174 fd->err = ENOMEM;
175 return BLOCK_ABORT;
176 }
177 fd->harray = new_array;
178 fd->max_array += 500;
179 }
180 ent = fd->harray + fd->num_array++;
181 ent->dir = dirent;
182 fd->dir_size += EXT2_DIR_REC_LEN(name_len);
183 ent->ino = dirent->inode;
184 if (fd->compress)
185 ent->hash = ent->minor_hash = 0;
186 else {
187 fd->err = ext2fs_dirhash(hash_alg, dirent->name,
188 name_len,
189 fs->super->s_hash_seed,
190 &ent->hash, &ent->minor_hash);
191 if (fd->err)
192 return BLOCK_ABORT;
193 }
194 }
195
196 return 0;
197 }
198
199 /* Used for sorting the hash entry */
200 static EXT2_QSORT_TYPE ino_cmp(const void *a, const void *b)
201 {
202 const struct hash_entry *he_a = (const struct hash_entry *) a;
203 const struct hash_entry *he_b = (const struct hash_entry *) b;
204
205 return (he_a->ino - he_b->ino);
206 }
207
208 /* Used for sorting the hash entry */
209 static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
210 {
211 const struct hash_entry *he_a = (const struct hash_entry *) a;
212 const struct hash_entry *he_b = (const struct hash_entry *) b;
213 unsigned int he_a_len, he_b_len;
214 int ret;
215 int min_len;
216
217 he_a_len = ext2fs_dirent_name_len(he_a->dir);
218 he_b_len = ext2fs_dirent_name_len(he_b->dir);
219 min_len = he_a_len;
220 if (min_len > he_b_len)
221 min_len = he_b_len;
222
223 ret = strncmp(he_a->dir->name, he_b->dir->name, min_len);
224 if (ret == 0) {
225 if (he_a_len > he_b_len)
226 ret = 1;
227 else if (he_a_len < he_b_len)
228 ret = -1;
229 else
230 ret = he_b->dir->inode - he_a->dir->inode;
231 }
232 return ret;
233 }
234
235 /* Used for sorting the hash entry */
236 static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
237 {
238 const struct hash_entry *he_a = (const struct hash_entry *) a;
239 const struct hash_entry *he_b = (const struct hash_entry *) b;
240 int ret;
241
242 if (he_a->hash > he_b->hash)
243 ret = 1;
244 else if (he_a->hash < he_b->hash)
245 ret = -1;
246 else {
247 if (he_a->minor_hash > he_b->minor_hash)
248 ret = 1;
249 else if (he_a->minor_hash < he_b->minor_hash)
250 ret = -1;
251 else
252 ret = name_cmp(a, b);
253 }
254 return ret;
255 }
256
257 static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
258 int blocks)
259 {
260 void *new_mem;
261
262 if (outdir->max) {
263 new_mem = realloc(outdir->buf, blocks * fs->blocksize);
264 if (!new_mem)
265 return ENOMEM;
266 outdir->buf = new_mem;
267 new_mem = realloc(outdir->hashes,
268 blocks * sizeof(ext2_dirhash_t));
269 if (!new_mem)
270 return ENOMEM;
271 outdir->hashes = new_mem;
272 } else {
273 outdir->buf = malloc(blocks * fs->blocksize);
274 outdir->hashes = malloc(blocks * sizeof(ext2_dirhash_t));
275 outdir->num = 0;
276 }
277 outdir->max = blocks;
278 return 0;
279 }
280
281 static void free_out_dir(struct out_dir *outdir)
282 {
283 free(outdir->buf);
284 free(outdir->hashes);
285 outdir->max = 0;
286 outdir->num =0;
287 }
288
289 static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
290 char ** ret)
291 {
292 errcode_t retval;
293
294 if (outdir->num >= outdir->max) {
295 retval = alloc_size_dir(fs, outdir, outdir->max + 50);
296 if (retval)
297 return retval;
298 }
299 *ret = outdir->buf + (outdir->num++ * fs->blocksize);
300 memset(*ret, 0, fs->blocksize);
301 return 0;
302 }
303
304 /*
305 * This function is used to make a unique filename. We do this by
306 * appending ~0, and then incrementing the number. However, we cannot
307 * expand the length of the filename beyond the padding available in
308 * the directory entry.
309 */
310 static void mutate_name(char *str, unsigned int *len)
311 {
312 int i;
313 unsigned int l = *len;
314
315 /*
316 * First check to see if it looks the name has been mutated
317 * already
318 */
319 for (i = l-1; i > 0; i--) {
320 if (!isdigit(str[i]))
321 break;
322 }
323 if ((i == l-1) || (str[i] != '~')) {
324 if (((l-1) & 3) < 2)
325 l += 2;
326 else
327 l = (l+3) & ~3;
328 str[l-2] = '~';
329 str[l-1] = '0';
330 *len = l;
331 return;
332 }
333 for (i = l-1; i >= 0; i--) {
334 if (isdigit(str[i])) {
335 if (str[i] == '9')
336 str[i] = '0';
337 else {
338 str[i]++;
339 return;
340 }
341 continue;
342 }
343 if (i == 1) {
344 if (str[0] == 'z')
345 str[0] = 'A';
346 else if (str[0] == 'Z') {
347 str[0] = '~';
348 str[1] = '0';
349 } else
350 str[0]++;
351 } else if (i > 0) {
352 str[i] = '1';
353 str[i-1] = '~';
354 } else {
355 if (str[0] == '~')
356 str[0] = 'a';
357 else
358 str[0]++;
359 }
360 break;
361 }
362 }
363
364 static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
365 ext2_ino_t ino,
366 struct fill_dir_struct *fd)
367 {
368 struct problem_context pctx;
369 struct hash_entry *ent, *prev;
370 int i, j;
371 int fixed = 0;
372 char new_name[256];
373 unsigned int new_len;
374 int hash_alg;
375
376 clear_problem_context(&pctx);
377 pctx.ino = ino;
378
379 hash_alg = fs->super->s_def_hash_version;
380 if ((hash_alg <= EXT2_HASH_TEA) &&
381 (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
382 hash_alg += 3;
383
384 for (i=1; i < fd->num_array; i++) {
385 ent = fd->harray + i;
386 prev = ent - 1;
387 if (!ent->dir->inode ||
388 (ext2fs_dirent_name_len(ent->dir) !=
389 ext2fs_dirent_name_len(prev->dir)) ||
390 strncmp(ent->dir->name, prev->dir->name,
391 ext2fs_dirent_name_len(ent->dir)))
392 continue;
393 pctx.dirent = ent->dir;
394 if ((ent->dir->inode == prev->dir->inode) &&
395 fix_problem(ctx, PR_2_DUPLICATE_DIRENT, &pctx)) {
396 e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
397 ent->dir->inode = 0;
398 fixed++;
399 continue;
400 }
401 new_len = ext2fs_dirent_name_len(ent->dir);
402 memcpy(new_name, ent->dir->name, new_len);
403 mutate_name(new_name, &new_len);
404 for (j=0; j < fd->num_array; j++) {
405 if ((i==j) ||
406 (new_len !=
407 ext2fs_dirent_name_len(fd->harray[j].dir)) ||
408 strncmp(new_name, fd->harray[j].dir->name, new_len))
409 continue;
410 mutate_name(new_name, &new_len);
411
412 j = -1;
413 }
414 new_name[new_len] = 0;
415 pctx.str = new_name;
416 if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
417 memcpy(ent->dir->name, new_name, new_len);
418 ext2fs_dirent_set_name_len(ent->dir, new_len);
419 ext2fs_dirhash(hash_alg, new_name, new_len,
420 fs->super->s_hash_seed,
421 &ent->hash, &ent->minor_hash);
422 fixed++;
423 }
424 }
425 return fixed;
426 }
427
428
429 static errcode_t copy_dir_entries(e2fsck_t ctx,
430 struct fill_dir_struct *fd,
431 struct out_dir *outdir)
432 {
433 ext2_filsys fs = ctx->fs;
434 errcode_t retval;
435 char *block_start;
436 struct hash_entry *ent;
437 struct ext2_dir_entry *dirent;
438 unsigned int rec_len, prev_rec_len, left, slack, offset;
439 int i;
440 ext2_dirhash_t prev_hash;
441 int csum_size = 0;
442 struct ext2_dir_entry_tail *t;
443
444 if (ctx->htree_slack_percentage == 255) {
445 profile_get_uint(ctx->profile, "options",
446 "indexed_dir_slack_percentage",
447 0, 20,
448 &ctx->htree_slack_percentage);
449 if (ctx->htree_slack_percentage > 100)
450 ctx->htree_slack_percentage = 20;
451 }
452
453 if (ext2fs_has_feature_metadata_csum(fs->super))
454 csum_size = sizeof(struct ext2_dir_entry_tail);
455
456 outdir->max = 0;
457 retval = alloc_size_dir(fs, outdir,
458 (fd->dir_size / fs->blocksize) + 2);
459 if (retval)
460 return retval;
461 outdir->num = fd->compress ? 0 : 1;
462 offset = 0;
463 outdir->hashes[0] = 0;
464 prev_hash = 1;
465 if ((retval = get_next_block(fs, outdir, &block_start)))
466 return retval;
467 dirent = (struct ext2_dir_entry *) block_start;
468 prev_rec_len = 0;
469 rec_len = 0;
470 left = fs->blocksize - csum_size;
471 slack = fd->compress ? 12 :
472 ((fs->blocksize - csum_size) * ctx->htree_slack_percentage)/100;
473 if (slack < 12)
474 slack = 12;
475 for (i = 0; i < fd->num_array; i++) {
476 ent = fd->harray + i;
477 if (ent->dir->inode == 0)
478 continue;
479 rec_len = EXT2_DIR_REC_LEN(ext2fs_dirent_name_len(ent->dir));
480 if (rec_len > left) {
481 if (left) {
482 left += prev_rec_len;
483 retval = ext2fs_set_rec_len(fs, left, dirent);
484 if (retval)
485 return retval;
486 }
487 if (csum_size) {
488 t = EXT2_DIRENT_TAIL(block_start,
489 fs->blocksize);
490 ext2fs_initialize_dirent_tail(fs, t);
491 }
492 if ((retval = get_next_block(fs, outdir,
493 &block_start)))
494 return retval;
495 offset = 0;
496 }
497 left = (fs->blocksize - csum_size) - offset;
498 dirent = (struct ext2_dir_entry *) (block_start + offset);
499 if (offset == 0) {
500 if (ent->hash == prev_hash)
501 outdir->hashes[outdir->num-1] = ent->hash | 1;
502 else
503 outdir->hashes[outdir->num-1] = ent->hash;
504 }
505 dirent->inode = ent->dir->inode;
506 ext2fs_dirent_set_name_len(dirent,
507 ext2fs_dirent_name_len(ent->dir));
508 ext2fs_dirent_set_file_type(dirent,
509 ext2fs_dirent_file_type(ent->dir));
510 retval = ext2fs_set_rec_len(fs, rec_len, dirent);
511 if (retval)
512 return retval;
513 prev_rec_len = rec_len;
514 memcpy(dirent->name, ent->dir->name,
515 ext2fs_dirent_name_len(dirent));
516 offset += rec_len;
517 left -= rec_len;
518 if (left < slack) {
519 prev_rec_len += left;
520 retval = ext2fs_set_rec_len(fs, prev_rec_len, dirent);
521 if (retval)
522 return retval;
523 offset += left;
524 left = 0;
525 }
526 prev_hash = ent->hash;
527 }
528 if (left)
529 retval = ext2fs_set_rec_len(fs, rec_len + left, dirent);
530 if (csum_size) {
531 t = EXT2_DIRENT_TAIL(block_start, fs->blocksize);
532 ext2fs_initialize_dirent_tail(fs, t);
533 }
534
535 return retval;
536 }
537
538
539 static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
540 ext2_ino_t ino, ext2_ino_t parent)
541 {
542 struct ext2_dir_entry *dir;
543 struct ext2_dx_root_info *root;
544 struct ext2_dx_countlimit *limits;
545 int filetype = 0;
546 int csum_size = 0;
547
548 if (ext2fs_has_feature_filetype(fs->super))
549 filetype = EXT2_FT_DIR;
550
551 memset(buf, 0, fs->blocksize);
552 dir = (struct ext2_dir_entry *) buf;
553 dir->inode = ino;
554 dir->name[0] = '.';
555 ext2fs_dirent_set_name_len(dir, 1);
556 ext2fs_dirent_set_file_type(dir, filetype);
557 dir->rec_len = 12;
558 dir = (struct ext2_dir_entry *) (buf + 12);
559 dir->inode = parent;
560 dir->name[0] = '.';
561 dir->name[1] = '.';
562 ext2fs_dirent_set_name_len(dir, 2);
563 ext2fs_dirent_set_file_type(dir, filetype);
564 dir->rec_len = fs->blocksize - 12;
565
566 root = (struct ext2_dx_root_info *) (buf+24);
567 root->reserved_zero = 0;
568 root->hash_version = fs->super->s_def_hash_version;
569 root->info_length = 8;
570 root->indirect_levels = 0;
571 root->unused_flags = 0;
572
573 if (ext2fs_has_feature_metadata_csum(fs->super))
574 csum_size = sizeof(struct ext2_dx_tail);
575
576 limits = (struct ext2_dx_countlimit *) (buf+32);
577 limits->limit = (fs->blocksize - (32 + csum_size)) /
578 sizeof(struct ext2_dx_entry);
579 limits->count = 0;
580
581 return root;
582 }
583
584
585 static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
586 {
587 struct ext2_dir_entry *dir;
588 struct ext2_dx_countlimit *limits;
589 int csum_size = 0;
590
591 memset(buf, 0, fs->blocksize);
592 dir = (struct ext2_dir_entry *) buf;
593 dir->inode = 0;
594 (void) ext2fs_set_rec_len(fs, fs->blocksize, dir);
595
596 if (ext2fs_has_feature_metadata_csum(fs->super))
597 csum_size = sizeof(struct ext2_dx_tail);
598
599 limits = (struct ext2_dx_countlimit *) (buf+8);
600 limits->limit = (fs->blocksize - (8 + csum_size)) /
601 sizeof(struct ext2_dx_entry);
602 limits->count = 0;
603
604 return (struct ext2_dx_entry *) limits;
605 }
606
607 /*
608 * This function takes the leaf nodes which have been written in
609 * outdir, and populates the root node and any necessary interior nodes.
610 */
611 static errcode_t calculate_tree(ext2_filsys fs,
612 struct out_dir *outdir,
613 ext2_ino_t ino,
614 ext2_ino_t parent)
615 {
616 struct ext2_dx_root_info *root_info;
617 struct ext2_dx_entry *root, *dx_ent = 0;
618 struct ext2_dx_countlimit *root_limit, *limit;
619 errcode_t retval;
620 char * block_start;
621 int i, c1, c2, nblks;
622 int limit_offset, root_offset;
623
624 root_info = set_root_node(fs, outdir->buf, ino, parent);
625 root_offset = limit_offset = ((char *) root_info - outdir->buf) +
626 root_info->info_length;
627 root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
628 c1 = root_limit->limit;
629 nblks = outdir->num;
630
631 /* Write out the pointer blocks */
632 if (nblks-1 <= c1) {
633 /* Just write out the root block, and we're done */
634 root = (struct ext2_dx_entry *) (outdir->buf + root_offset);
635 for (i=1; i < nblks; i++) {
636 root->block = ext2fs_cpu_to_le32(i);
637 if (i != 1)
638 root->hash =
639 ext2fs_cpu_to_le32(outdir->hashes[i]);
640 root++;
641 c1--;
642 }
643 } else {
644 c2 = 0;
645 limit = 0;
646 root_info->indirect_levels = 1;
647 for (i=1; i < nblks; i++) {
648 if (c1 == 0)
649 return ENOSPC;
650 if (c2 == 0) {
651 if (limit)
652 limit->limit = limit->count =
653 ext2fs_cpu_to_le16(limit->limit);
654 root = (struct ext2_dx_entry *)
655 (outdir->buf + root_offset);
656 root->block = ext2fs_cpu_to_le32(outdir->num);
657 if (i != 1)
658 root->hash =
659 ext2fs_cpu_to_le32(outdir->hashes[i]);
660 if ((retval = get_next_block(fs, outdir,
661 &block_start)))
662 return retval;
663 dx_ent = set_int_node(fs, block_start);
664 limit = (struct ext2_dx_countlimit *) dx_ent;
665 c2 = limit->limit;
666 root_offset += sizeof(struct ext2_dx_entry);
667 c1--;
668 }
669 dx_ent->block = ext2fs_cpu_to_le32(i);
670 if (c2 != limit->limit)
671 dx_ent->hash =
672 ext2fs_cpu_to_le32(outdir->hashes[i]);
673 dx_ent++;
674 c2--;
675 }
676 limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
677 limit->limit = ext2fs_cpu_to_le16(limit->limit);
678 }
679 root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
680 root_limit->count = ext2fs_cpu_to_le16(root_limit->limit - c1);
681 root_limit->limit = ext2fs_cpu_to_le16(root_limit->limit);
682
683 return 0;
684 }
685
686 struct write_dir_struct {
687 struct out_dir *outdir;
688 errcode_t err;
689 ext2_ino_t ino;
690 e2fsck_t ctx;
691 ext2_ino_t dir;
692 };
693
694 /*
695 * Helper function which writes out a directory block.
696 */
697 static int write_dir_block(ext2_filsys fs,
698 blk64_t *block_nr,
699 e2_blkcnt_t blockcnt,
700 blk64_t ref_block EXT2FS_ATTR((unused)),
701 int ref_offset EXT2FS_ATTR((unused)),
702 void *priv_data)
703 {
704 struct write_dir_struct *wd = (struct write_dir_struct *) priv_data;
705 blk64_t blk;
706 char *dir, *buf = 0;
707
708 #ifdef REHASH_DEBUG
709 printf("%u: write_dir_block %lld:%lld", wd->ino, blockcnt, *block_nr);
710 #endif
711 if ((*block_nr == 0) || (blockcnt < 0)) {
712 #ifdef REHASH_DEBUG
713 printf(" - skip\n");
714 #endif
715 return 0;
716 }
717 if (blockcnt < wd->outdir->num)
718 dir = wd->outdir->buf + (blockcnt * fs->blocksize);
719 else if (wd->ctx->lost_and_found == wd->dir) {
720 /* Don't release any extra directory blocks for lost+found */
721 wd->err = ext2fs_new_dir_block(fs, 0, 0, &buf);
722 if (wd->err)
723 return BLOCK_ABORT;
724 dir = buf;
725 wd->outdir->num++;
726 } else {
727 /* Don't free blocks at the end of the directory, they
728 * will be truncated by the caller. */
729 #ifdef REHASH_DEBUG
730 printf(" - not freed\n");
731 #endif
732 return 0;
733 }
734 wd->err = ext2fs_write_dir_block4(fs, *block_nr, dir, 0, wd->dir);
735 if (buf)
736 ext2fs_free_mem(&buf);
737
738 #ifdef REHASH_DEBUG
739 printf(" - write (%d)\n", wd->err);
740 #endif
741 if (wd->err)
742 return BLOCK_ABORT;
743 return 0;
744 }
745
746 static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
747 struct out_dir *outdir,
748 ext2_ino_t ino, struct ext2_inode *inode,
749 int compress)
750 {
751 struct write_dir_struct wd;
752 errcode_t retval;
753
754 retval = e2fsck_expand_directory(ctx, ino, -1, outdir->num);
755 if (retval)
756 return retval;
757
758 wd.outdir = outdir;
759 wd.err = 0;
760 wd.ino = ino;
761 wd.ctx = ctx;
762 wd.dir = ino;
763
764 retval = ext2fs_block_iterate3(fs, ino, 0, NULL,
765 write_dir_block, &wd);
766 if (retval)
767 return retval;
768 if (wd.err)
769 return wd.err;
770
771 e2fsck_read_inode(ctx, ino, inode, "rehash_dir");
772 if (compress)
773 inode->i_flags &= ~EXT2_INDEX_FL;
774 else
775 inode->i_flags |= EXT2_INDEX_FL;
776 #ifdef REHASH_DEBUG
777 printf("%u: set inode size to %u blocks = %u bytes\n",
778 ino, outdir->num, outdir->num * fs->blocksize);
779 #endif
780 retval = ext2fs_inode_size_set(fs, inode, (ext2_off64_t)outdir->num *
781 fs->blocksize);
782 if (retval)
783 return retval;
784
785 /* ext2fs_punch() calls ext2fs_write_inode() which writes the size */
786 return ext2fs_punch(fs, ino, inode, NULL, outdir->num, ~0ULL);
787 }
788
789 errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino,
790 struct problem_context *pctx)
791 {
792 ext2_filsys fs = ctx->fs;
793 errcode_t retval;
794 struct ext2_inode inode;
795 char *dir_buf = 0;
796 struct fill_dir_struct fd = { NULL };
797 struct out_dir outdir = { 0 };
798
799 e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
800
801 if (ext2fs_has_feature_inline_data(fs->super) &&
802 (inode.i_flags & EXT4_INLINE_DATA_FL))
803 return 0;
804
805 retval = ENOMEM;
806 dir_buf = malloc(inode.i_size);
807 if (!dir_buf)
808 goto errout;
809
810 fd.max_array = inode.i_size / 32;
811 fd.harray = malloc(fd.max_array * sizeof(struct hash_entry));
812 if (!fd.harray)
813 goto errout;
814
815 fd.ino = ino;
816 fd.ctx = ctx;
817 fd.buf = dir_buf;
818 fd.inode = &inode;
819 fd.dir = ino;
820 if (!ext2fs_has_feature_dir_index(fs->super) ||
821 (inode.i_size / fs->blocksize) < 2)
822 fd.compress = 1;
823 fd.parent = 0;
824
825 retry_nohash:
826 /* Read in the entire directory into memory */
827 retval = ext2fs_block_iterate3(fs, ino, 0, 0,
828 fill_dir_block, &fd);
829 if (fd.err) {
830 retval = fd.err;
831 goto errout;
832 }
833
834 /*
835 * If the entries read are less than a block, then don't index
836 * the directory
837 */
838 if (!fd.compress && (fd.dir_size < (fs->blocksize - 24))) {
839 fd.compress = 1;
840 fd.dir_size = 0;
841 fd.num_array = 0;
842 goto retry_nohash;
843 }
844
845 #if 0
846 printf("%d entries (%d bytes) found in inode %d\n",
847 fd.num_array, fd.dir_size, ino);
848 #endif
849
850 /* Sort the list */
851 resort:
852 if (fd.compress && fd.num_array > 1)
853 qsort(fd.harray+2, fd.num_array-2, sizeof(struct hash_entry),
854 hash_cmp);
855 else
856 qsort(fd.harray, fd.num_array, sizeof(struct hash_entry),
857 hash_cmp);
858
859 /*
860 * Look for duplicates
861 */
862 if (duplicate_search_and_fix(ctx, fs, ino, &fd))
863 goto resort;
864
865 if (ctx->options & E2F_OPT_NO) {
866 retval = 0;
867 goto errout;
868 }
869
870 /* Sort non-hashed directories by inode number */
871 if (fd.compress && fd.num_array > 1)
872 qsort(fd.harray+2, fd.num_array-2,
873 sizeof(struct hash_entry), ino_cmp);
874
875 /*
876 * Copy the directory entries. In a htree directory these
877 * will become the leaf nodes.
878 */
879 retval = copy_dir_entries(ctx, &fd, &outdir);
880 if (retval)
881 goto errout;
882
883 free(dir_buf); dir_buf = 0;
884
885 if (!fd.compress) {
886 /* Calculate the interior nodes */
887 retval = calculate_tree(fs, &outdir, ino, fd.parent);
888 if (retval)
889 goto errout;
890 }
891
892 retval = write_directory(ctx, fs, &outdir, ino, &inode, fd.compress);
893 if (retval)
894 goto errout;
895
896 if (ctx->options & E2F_OPT_CONVERT_BMAP)
897 retval = e2fsck_rebuild_extents_later(ctx, ino);
898 else
899 retval = e2fsck_check_rebuild_extents(ctx, ino, &inode, pctx);
900 errout:
901 free(dir_buf);
902 free(fd.harray);
903
904 free_out_dir(&outdir);
905 return retval;
906 }
907
908 void e2fsck_rehash_directories(e2fsck_t ctx)
909 {
910 struct problem_context pctx;
911 #ifdef RESOURCE_TRACK
912 struct resource_track rtrack;
913 #endif
914 struct dir_info *dir;
915 ext2_u32_iterate iter;
916 struct dir_info_iter * dirinfo_iter = 0;
917 ext2_ino_t ino;
918 errcode_t retval;
919 int cur, max, all_dirs, first = 1;
920
921 init_resource_track(&rtrack, ctx->fs->io);
922 all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
923
924 if (!ctx->dirs_to_hash && !all_dirs)
925 return;
926
927 (void) e2fsck_get_lost_and_found(ctx, 0);
928
929 clear_problem_context(&pctx);
930
931 cur = 0;
932 if (all_dirs) {
933 dirinfo_iter = e2fsck_dir_info_iter_begin(ctx);
934 max = e2fsck_get_num_dirinfo(ctx);
935 } else {
936 retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
937 &iter);
938 if (retval) {
939 pctx.errcode = retval;
940 fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
941 return;
942 }
943 max = ext2fs_u32_list_count(ctx->dirs_to_hash);
944 }
945 while (1) {
946 if (all_dirs) {
947 if ((dir = e2fsck_dir_info_iter(ctx,
948 dirinfo_iter)) == 0)
949 break;
950 ino = dir->ino;
951 } else {
952 if (!ext2fs_u32_list_iterate(iter, &ino))
953 break;
954 }
955
956 pctx.dir = ino;
957 if (first) {
958 fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
959 first = 0;
960 }
961 #if 0
962 fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
963 #endif
964 pctx.errcode = e2fsck_rehash_dir(ctx, ino, &pctx);
965 if (pctx.errcode) {
966 end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
967 fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
968 }
969 if (ctx->progress && !ctx->progress_fd)
970 e2fsck_simple_progress(ctx, "Rebuilding directory",
971 100.0 * (float) (++cur) / (float) max, ino);
972 }
973 end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
974 if (all_dirs)
975 e2fsck_dir_info_iter_end(ctx, dirinfo_iter);
976 else
977 ext2fs_u32_list_iterate_end(iter);
978
979 if (ctx->dirs_to_hash)
980 ext2fs_u32_list_free(ctx->dirs_to_hash);
981 ctx->dirs_to_hash = 0;
982
983 print_resource_track(ctx, "Pass 3A", &rtrack, ctx->fs->io);
984 }
Unnamed repository; edit this file 'description' to name the repository.