]> git.ipfire.org Git - thirdparty/e2fsprogs.git/blame - e2fsck/rehash.c
Makefile.in (uuidgen): Avoid linking lsattr and chattr against
[thirdparty/e2fsprogs.git] / e2fsck / rehash.c
CommitLineData
b7a00563
TT
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 <errno.h>
49#include "e2fsck.h"
50#include "problem.h"
51
52struct fill_dir_struct {
53 char *buf;
54 struct ext2_inode *inode;
55 int err;
56 e2fsck_t ctx;
57 struct hash_entry *harray;
58 int max_array, num_array;
59 int dir_size;
850d05e9 60 int compress;
b7a00563
TT
61 ino_t parent;
62};
63
64struct hash_entry {
65 ext2_dirhash_t hash;
66 ext2_dirhash_t minor_hash;
67 struct ext2_dir_entry *dir;
68};
69
70struct out_dir {
71 int num;
72 int max;
73 char *buf;
74 ext2_dirhash_t *hashes;
75};
76
77static int fill_dir_block(ext2_filsys fs,
78 blk_t *block_nr,
79 e2_blkcnt_t blockcnt,
80 blk_t ref_block,
81 int ref_offset,
82 void *priv_data)
83{
84 struct fill_dir_struct *fd = (struct fill_dir_struct *) priv_data;
85 struct hash_entry *new_array, *ent;
86 struct ext2_dir_entry *dirent;
87 char *dir;
88 int offset, dir_offset;
89
90 if (blockcnt < 0)
91 return 0;
92
93 offset = blockcnt * fs->blocksize;
94 if (offset + fs->blocksize > fd->inode->i_size) {
95 fd->err = EXT2_ET_DIR_CORRUPTED;
96 return BLOCK_ABORT;
97 }
98 dir = (fd->buf+offset);
99 if (HOLE_BLKADDR(*block_nr)) {
100 memset(dir, 0, fs->blocksize);
101 dirent = (struct ext2_dir_entry *) dir;
102 dirent->rec_len = fs->blocksize;
103 } else {
104 fd->err = ext2fs_read_dir_block(fs, *block_nr, dir);
105 if (fd->err)
106 return BLOCK_ABORT;
107 }
108 /* While the directory block is "hot", index it. */
109 dir_offset = 0;
110 while (dir_offset < fs->blocksize) {
111 dirent = (struct ext2_dir_entry *) (dir + dir_offset);
112 if (((dir_offset + dirent->rec_len) > fs->blocksize) ||
113 (dirent->rec_len < 8) ||
114 ((dirent->rec_len % 4) != 0) ||
115 (((dirent->name_len & 0xFF)+8) > dirent->rec_len)) {
116 fd->err = EXT2_ET_DIR_CORRUPTED;
117 return BLOCK_ABORT;
118 }
850d05e9 119 dir_offset += dirent->rec_len;
b7a00563 120 if (dirent->inode == 0)
850d05e9
TT
121 continue;
122 if (!fd->compress && ((dirent->name_len&0xFF) == 1) &&
123 (dirent->name[0] == '.'))
124 continue;
125 if (!fd->compress && ((dirent->name_len&0xFF) == 2) &&
b7a00563
TT
126 (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
127 fd->parent = dirent->inode;
850d05e9 128 continue;
b7a00563
TT
129 }
130 if (fd->num_array >= fd->max_array) {
131 new_array = realloc(fd->harray,
132 sizeof(struct hash_entry) * (fd->max_array+500));
133 if (!new_array) {
134 fd->err = ENOMEM;
135 return BLOCK_ABORT;
136 }
137 fd->harray = new_array;
138 fd->max_array += 500;
139 }
850d05e9 140 ent = fd->harray + fd->num_array++;
b7a00563 141 ent->dir = dirent;
b7a00563 142 fd->dir_size += EXT2_DIR_REC_LEN(dirent->name_len & 0xFF);
850d05e9
TT
143 if (fd->compress)
144 ent->hash = ent->minor_hash = 0;
145 else {
146 fd->err = ext2fs_dirhash(fs->super->s_def_hash_version,
147 dirent->name,
148 dirent->name_len & 0xFF,
149 fs->super->s_hash_seed,
150 &ent->hash, &ent->minor_hash);
151 if (fd->err)
152 return BLOCK_ABORT;
153 }
b7a00563
TT
154 }
155
156 return 0;
157}
158
159/* Used for sorting the hash entry */
b0700a1b 160static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
b7a00563
TT
161{
162 const struct hash_entry *he_a = (const struct hash_entry *) a;
163 const struct hash_entry *he_b = (const struct hash_entry *) b;
164 int ret;
b0700a1b
TT
165 int min_len;
166
167 min_len = he_a->dir->name_len;
168 if (min_len > he_b->dir->name_len)
169 min_len = he_b->dir->name_len;
170
171 ret = strncmp(he_a->dir->name, he_b->dir->name, min_len);
172 if (ret == 0) {
173 if (he_a->dir->name_len > he_b->dir->name_len)
b7a00563 174 ret = 1;
b0700a1b 175 else if (he_a->dir->name_len < he_b->dir->name_len)
b7a00563
TT
176 ret = -1;
177 else
12dd69f5 178 ret = he_b->dir->inode - he_a->dir->inode;
b7a00563
TT
179 }
180 return ret;
181}
182
850d05e9 183/* Used for sorting the hash entry */
b0700a1b 184static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
850d05e9
TT
185{
186 const struct hash_entry *he_a = (const struct hash_entry *) a;
187 const struct hash_entry *he_b = (const struct hash_entry *) b;
188 int ret;
b0700a1b
TT
189
190 if (he_a->hash > he_b->hash)
191 ret = 1;
192 else if (he_a->hash < he_b->hash)
193 ret = -1;
194 else {
195 if (he_a->minor_hash > he_b->minor_hash)
850d05e9 196 ret = 1;
b0700a1b 197 else if (he_a->minor_hash < he_b->minor_hash)
850d05e9
TT
198 ret = -1;
199 else
b0700a1b 200 ret = name_cmp(a, b);
850d05e9
TT
201 }
202 return ret;
203}
204
b7a00563
TT
205static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
206 int blocks)
207{
208 void *new_mem;
209
210 if (outdir->max) {
211 new_mem = realloc(outdir->buf, blocks * fs->blocksize);
212 if (!new_mem)
213 return ENOMEM;
214 outdir->buf = new_mem;
215 new_mem = realloc(outdir->hashes,
216 blocks * sizeof(ext2_dirhash_t));
217 if (!new_mem)
218 return ENOMEM;
219 outdir->hashes = new_mem;
220 } else {
221 outdir->buf = malloc(blocks * fs->blocksize);
222 outdir->hashes = malloc(blocks * sizeof(ext2_dirhash_t));
223 outdir->num = 0;
224 }
225 outdir->max = blocks;
226 return 0;
227}
228
229static void free_out_dir(struct out_dir *outdir)
230{
1d2eef42
TT
231 if (outdir->buf)
232 free(outdir->buf);
233 if (outdir->hashes)
234 free(outdir->hashes);
b7a00563
TT
235 outdir->max = 0;
236 outdir->num =0;
237}
238
850d05e9 239static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
b7a00563
TT
240 char ** ret)
241{
242 errcode_t retval;
243
244 if (outdir->num >= outdir->max) {
245 retval = alloc_size_dir(fs, outdir, outdir->max + 50);
246 if (retval)
247 return retval;
248 }
249 *ret = outdir->buf + (outdir->num++ * fs->blocksize);
850d05e9 250 memset(*ret, 0, fs->blocksize);
b7a00563
TT
251 return 0;
252}
253
b0700a1b
TT
254/*
255 * This function is used to make a unique filename. We do this by
256 * appending ~0, and then incrementing the number. However, we cannot
257 * expand the length of the filename beyond the padding available in
258 * the directory entry.
259 */
260static void mutate_name(char *str, __u16 *len)
261{
262 int i;
263 __u16 l = *len & 0xFF, h = *len & 0xff00;
264
265 /*
266 * First check to see if it looks the name has been mutated
267 * already
268 */
269 for (i = l-1; i > 0; i--) {
270 if (!isdigit(str[i]))
271 break;
272 }
273 if ((i == l-1) || (str[i] != '~')) {
274 if (((l-1) & 3) < 2)
275 l += 2;
276 else
277 l = (l+3) & ~3;
278 str[l-2] = '~';
279 str[l-1] = '0';
280 *len = l | h;
281 return;
282 }
283 for (i = l-1; i >= 0; i--) {
284 if (isdigit(str[i])) {
285 if (str[i] == '9')
286 str[i] = '0';
287 else {
288 str[i]++;
289 return;
290 }
291 continue;
292 }
293 if (i == 1) {
294 if (str[0] == 'z')
295 str[0] = 'A';
296 else if (str[0] == 'Z') {
297 str[0] = '~';
298 str[1] = '0';
299 } else
300 str[0]++;
301 } else if (i > 0) {
302 str[i] = '1';
303 str[i-1] = '~';
304 } else {
305 if (str[0] == '~')
306 str[0] = 'a';
307 else
308 str[0]++;
309 }
310 break;
311 }
312}
313
314static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
315 ext2_ino_t ino,
316 struct fill_dir_struct *fd)
317{
318 struct problem_context pctx;
319 struct hash_entry *ent, *prev;
320 int i, j;
321 int fixed = 0;
322 char new_name[256];
323 __u16 new_len;
324
325 clear_problem_context(&pctx);
326 pctx.ino = ino;
327
328 for (i=1; i < fd->num_array; i++) {
329 ent = fd->harray + i;
330 prev = ent - 1;
331 if (!ent->dir->inode ||
332 ((ent->dir->name_len & 0xFF) !=
333 (prev->dir->name_len & 0xFF)) ||
334 (strncmp(ent->dir->name, prev->dir->name,
335 ent->dir->name_len & 0xFF)))
336 continue;
337 pctx.dirent = ent->dir;
338 if ((ent->dir->inode == prev->dir->inode) &&
339 fix_problem(ctx, PR_2_DUPLICATE_DIRENT, &pctx)) {
340 e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
341 ent->dir->inode = 0;
342 fixed++;
343 continue;
344 }
345 memcpy(new_name, ent->dir->name, ent->dir->name_len & 0xFF);
346 new_len = ent->dir->name_len;
347 mutate_name(new_name, &new_len);
348 for (j=0; j < fd->num_array; j++) {
349 if ((i==j) ||
350 ((ent->dir->name_len & 0xFF) !=
351 (fd->harray[j].dir->name_len & 0xFF)) ||
352 (strncmp(new_name, fd->harray[j].dir->name,
353 new_len & 0xFF)))
354 continue;
355 mutate_name(new_name, &new_len);
356
357 j = -1;
358 }
359 new_name[new_len & 0xFF] = 0;
360 pctx.str = new_name;
361 if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
362 memcpy(ent->dir->name, new_name, new_len & 0xFF);
363 ent->dir->name_len = new_len;
364 ext2fs_dirhash(fs->super->s_def_hash_version,
365 ent->dir->name,
366 ent->dir->name_len & 0xFF,
367 fs->super->s_hash_seed,
368 &ent->hash, &ent->minor_hash);
369 fixed++;
370 }
371 }
372 return fixed;
373}
374
b7a00563 375
850d05e9
TT
376static errcode_t copy_dir_entries(ext2_filsys fs,
377 struct fill_dir_struct *fd,
378 struct out_dir *outdir)
379{
380 errcode_t retval;
381 char *block_start;
382 struct hash_entry *ent;
383 struct ext2_dir_entry *dirent;
384 int i, rec_len, left;
385 ext2_dirhash_t prev_hash;
386 int offset;
387
388 outdir->max = 0;
389 retval = alloc_size_dir(fs, outdir,
390 (fd->dir_size / fs->blocksize) + 2);
391 if (retval)
392 return retval;
393 outdir->num = fd->compress ? 0 : 1;
394 offset = 0;
395 outdir->hashes[0] = 0;
396 prev_hash = 1;
397 if ((retval = get_next_block(fs, outdir, &block_start)))
398 return retval;
399 dirent = (struct ext2_dir_entry *) block_start;
400 left = fs->blocksize;
401 for (i=0; i < fd->num_array; i++) {
402 ent = fd->harray + i;
b0700a1b
TT
403 if (ent->dir->inode == 0)
404 continue;
850d05e9 405 rec_len = EXT2_DIR_REC_LEN(ent->dir->name_len & 0xFF);
850d05e9 406 if (rec_len > left) {
fe5b72d1
TT
407 if (left)
408 dirent->rec_len += left;
850d05e9
TT
409 if ((retval = get_next_block(fs, outdir,
410 &block_start)))
411 return retval;
fe5b72d1 412 offset = 0;
850d05e9 413 }
fe5b72d1
TT
414 left = fs->blocksize - offset;
415 dirent = (struct ext2_dir_entry *) (block_start + offset);
850d05e9
TT
416 if (offset == 0) {
417 if (ent->hash == prev_hash)
418 outdir->hashes[outdir->num-1] = ent->hash | 1;
419 else
420 outdir->hashes[outdir->num-1] = ent->hash;
421 }
422 dirent->inode = ent->dir->inode;
423 dirent->name_len = ent->dir->name_len;
424 dirent->rec_len = rec_len;
425 memcpy(dirent->name, ent->dir->name, dirent->name_len & 0xFF);
426 offset += rec_len;
427 left -= rec_len;
428 if (left < 12) {
429 dirent->rec_len += left;
430 offset += left;
cf3909ed 431 left = 0;
850d05e9
TT
432 }
433 prev_hash = ent->hash;
434 }
435 if (left)
436 dirent->rec_len += left;
437
438 return 0;
439}
440
441
442static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
b7a00563
TT
443 ext2_ino_t ino, ext2_ino_t parent)
444{
445 struct ext2_dir_entry *dir;
446 struct ext2_dx_root_info *root;
447 struct ext2_dx_countlimit *limits;
850d05e9 448 int filetype = 0;
b7a00563
TT
449
450 if (fs->super->s_feature_incompat & EXT2_FEATURE_INCOMPAT_FILETYPE)
451 filetype = EXT2_FT_DIR << 8;
452
453 memset(buf, 0, fs->blocksize);
454 dir = (struct ext2_dir_entry *) buf;
455 dir->inode = ino;
456 dir->name[0] = '.';
457 dir->name_len = 1 | filetype;
458 dir->rec_len = 12;
459 dir = (struct ext2_dir_entry *) (buf + 12);
460 dir->inode = parent;
461 dir->name[0] = '.';
462 dir->name[1] = '.';
463 dir->name_len = 2 | filetype;
464 dir->rec_len = fs->blocksize - 12;
465
466 root = (struct ext2_dx_root_info *) (buf+24);
467 root->reserved_zero = 0;
468 root->hash_version = fs->super->s_def_hash_version;
469 root->info_length = 8;
470 root->indirect_levels = 0;
471 root->unused_flags = 0;
472
473 limits = (struct ext2_dx_countlimit *) (buf+32);
474 limits->limit = (fs->blocksize - 32) / sizeof(struct ext2_dx_entry);
475 limits->count = 0;
476
477 return root;
478}
479
480
850d05e9 481static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
b7a00563
TT
482{
483 struct ext2_dir_entry *dir;
484 struct ext2_dx_countlimit *limits;
485
486 memset(buf, 0, fs->blocksize);
487 dir = (struct ext2_dir_entry *) buf;
488 dir->inode = 0;
489 dir->rec_len = fs->blocksize;
490
491 limits = (struct ext2_dx_countlimit *) (buf+8);
492 limits->limit = (fs->blocksize - 8) / sizeof(struct ext2_dx_entry);
493 limits->count = 0;
494
495 return (struct ext2_dx_entry *) limits;
496}
497
850d05e9
TT
498/*
499 * This function takes the leaf nodes which have been written in
500 * outdir, and populates the root node and any necessary interior nodes.
501 */
502static errcode_t calculate_tree(ext2_filsys fs,
503 struct out_dir *outdir,
504 ext2_ino_t ino,
505 ext2_ino_t parent)
506{
507 struct ext2_dx_root_info *root_info;
508 struct ext2_dx_entry *root, *dx_ent = 0;
509 struct ext2_dx_countlimit *root_limit, *limit;
510 errcode_t retval;
511 char * block_start;
512 int i, c1, c2, nblks;
513 int limit_offset, root_offset;
514
515 root_info = set_root_node(fs, outdir->buf, ino, parent);
516 root_offset = limit_offset = ((char *) root_info - outdir->buf) +
517 root_info->info_length;
518 root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
519 c1 = root_limit->limit;
520 nblks = outdir->num;
521
522 /* Write out the pointer blocks */
523 if (nblks-1 <= c1) {
524 /* Just write out the root block, and we're done */
525 root = (struct ext2_dx_entry *) (outdir->buf + root_offset);
526 for (i=1; i < nblks; i++) {
527 root->block = ext2fs_cpu_to_le32(i);
528 if (i != 1)
529 root->hash =
530 ext2fs_cpu_to_le32(outdir->hashes[i]);
531 root++;
532 c1--;
533 }
534 } else {
535 c2 = 0;
536 limit = 0;
537 root_info->indirect_levels = 1;
538 for (i=1; i < nblks; i++) {
539 if (c1 == 0)
540 return ENOSPC;
541 if (c2 == 0) {
542 if (limit)
543 limit->limit = limit->count =
544 ext2fs_cpu_to_le16(limit->limit);
545 root = (struct ext2_dx_entry *)
546 (outdir->buf + root_offset);
547 root->block = ext2fs_cpu_to_le32(outdir->num);
548 if (i != 1)
549 root->hash =
550 ext2fs_cpu_to_le32(outdir->hashes[i]);
551 if ((retval = get_next_block(fs, outdir,
552 &block_start)))
553 return retval;
554 dx_ent = set_int_node(fs, block_start);
555 limit = (struct ext2_dx_countlimit *) dx_ent;
556 c2 = limit->limit;
557 root_offset += sizeof(struct ext2_dx_entry);
558 c1--;
559 }
560 dx_ent->block = ext2fs_cpu_to_le32(i);
561 if (c2 != limit->limit)
562 dx_ent->hash =
563 ext2fs_cpu_to_le32(outdir->hashes[i]);
564 dx_ent++;
565 c2--;
566 }
567 limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
568 limit->limit = ext2fs_cpu_to_le16(limit->limit);
569 }
570 root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
571 root_limit->count = ext2fs_cpu_to_le16(root_limit->limit - c1);
572 root_limit->limit = ext2fs_cpu_to_le16(root_limit->limit);
573
574 return 0;
575}
b7a00563
TT
576
577struct write_dir_struct {
578 struct out_dir *outdir;
579 errcode_t err;
580 e2fsck_t ctx;
581 int cleared;
582};
583
584/*
850d05e9 585 * Helper function which writes out a directory block.
b7a00563
TT
586 */
587static int write_dir_block(ext2_filsys fs,
588 blk_t *block_nr,
589 e2_blkcnt_t blockcnt,
590 blk_t ref_block,
591 int ref_offset,
592 void *priv_data)
593{
594 struct write_dir_struct *wd = (struct write_dir_struct *) priv_data;
850d05e9 595 blk_t blk;
b7a00563 596 char *dir;
b7a00563
TT
597
598 if (*block_nr == 0)
599 return 0;
600 if (blockcnt >= wd->outdir->num) {
601 e2fsck_read_bitmaps(wd->ctx);
602 blk = *block_nr;
603 ext2fs_unmark_block_bitmap(wd->ctx->block_found_map, blk);
604 ext2fs_block_alloc_stats(fs, blk, -1);
605 *block_nr = 0;
606 wd->cleared++;
607 return BLOCK_CHANGED;
608 }
609 if (blockcnt < 0)
610 return 0;
611
612 dir = wd->outdir->buf + (blockcnt * fs->blocksize);
613 wd->err = ext2fs_write_dir_block(fs, *block_nr, dir);
614 if (wd->err)
615 return BLOCK_ABORT;
616 return 0;
617}
618
b7a00563 619static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
850d05e9
TT
620 struct out_dir *outdir,
621 ext2_ino_t ino, int compress)
b7a00563
TT
622{
623 struct write_dir_struct wd;
624 errcode_t retval;
625 struct ext2_inode inode;
626
627 retval = e2fsck_expand_directory(ctx, ino, -1, outdir->num);
628 if (retval)
629 return retval;
630
631 wd.outdir = outdir;
632 wd.err = 0;
633 wd.ctx = ctx;
634 wd.cleared = 0;
635
636 retval = ext2fs_block_iterate2(fs, ino, 0, 0,
637 write_dir_block, &wd);
638 if (retval)
639 return retval;
640 if (wd.err)
641 return wd.err;
642
643 e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
e70ae99e
TT
644 if (compress)
645 inode.i_flags &= ~EXT2_INDEX_FL;
646 else
850d05e9 647 inode.i_flags |= EXT2_INDEX_FL;
b7a00563
TT
648 inode.i_size = outdir->num * fs->blocksize;
649 inode.i_blocks -= (fs->blocksize / 512) * wd.cleared;
650 e2fsck_write_inode(ctx, ino, &inode, "rehash_dir");
651
652 return 0;
653}
654
655errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino)
656{
657 ext2_filsys fs = ctx->fs;
658 errcode_t retval;
659 struct ext2_inode inode;
850d05e9 660 char *dir_buf = 0;
b7a00563 661 struct fill_dir_struct fd;
b7a00563 662 struct out_dir outdir;
b7a00563 663
1d2eef42
TT
664 outdir.max = outdir.num = 0;
665 outdir.buf = 0;
666 outdir.hashes = 0;
b7a00563 667 e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
b7a00563
TT
668
669 retval = ENOMEM;
670 fd.harray = 0;
671 dir_buf = malloc(inode.i_size);
672 if (!dir_buf)
673 goto errout;
674
675 fd.max_array = inode.i_size / 32;
676 fd.num_array = 0;
677 fd.harray = malloc(fd.max_array * sizeof(struct hash_entry));
678 if (!fd.harray)
679 goto errout;
680
681 fd.ctx = ctx;
682 fd.buf = dir_buf;
683 fd.inode = &inode;
684 fd.err = 0;
685 fd.dir_size = 0;
850d05e9
TT
686 fd.compress = 0;
687 if (!(fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_DIR_INDEX) ||
e70ae99e 688 (inode.i_size / fs->blocksize) < 2)
850d05e9 689 fd.compress = 1;
b7a00563
TT
690 fd.parent = 0;
691
692 /* Read in the entire directory into memory */
693 retval = ext2fs_block_iterate2(fs, ino, 0, 0,
694 fill_dir_block, &fd);
695 if (fd.err) {
696 retval = fd.err;
697 goto errout;
698 }
699
700#if 0
701 printf("%d entries (%d bytes) found in inode %d\n",
702 fd.num_array, fd.dir_size, ino);
703#endif
704
850d05e9 705 /* Sort the list */
b0700a1b 706resort:
850d05e9
TT
707 if (fd.compress)
708 qsort(fd.harray+2, fd.num_array-2,
709 sizeof(struct hash_entry), name_cmp);
710 else
711 qsort(fd.harray, fd.num_array,
712 sizeof(struct hash_entry), hash_cmp);
713
b0700a1b
TT
714 /*
715 * Look for duplicates
716 */
717 if (duplicate_search_and_fix(ctx, fs, ino, &fd))
718 goto resort;
719
1d2eef42
TT
720 if (ctx->options & E2F_OPT_NO) {
721 retval = 0;
722 goto errout;
723 }
724
850d05e9
TT
725 /*
726 * Copy the directory entries. In a htree directory these
727 * will become the leaf nodes.
728 */
729 retval = copy_dir_entries(fs, &fd, &outdir);
b7a00563
TT
730 if (retval)
731 goto errout;
850d05e9 732
b7a00563
TT
733 free(dir_buf); dir_buf = 0;
734
850d05e9
TT
735 if (!fd.compress) {
736 /* Calculate the interior nodes */
737 retval = calculate_tree(fs, &outdir, ino, fd.parent);
738 if (retval)
739 goto errout;
b7a00563 740 }
850d05e9
TT
741
742 retval = write_directory(ctx, fs, &outdir, ino, fd.compress);
b7a00563
TT
743 if (retval)
744 goto errout;
745
746errout:
747 if (dir_buf)
748 free(dir_buf);
850d05e9
TT
749 if (fd.harray)
750 free(fd.harray);
751
b7a00563
TT
752 free_out_dir(&outdir);
753 return retval;
754}
755
756void e2fsck_rehash_directories(e2fsck_t ctx)
757{
b7a00563 758 struct problem_context pctx;
850d05e9
TT
759#ifdef RESOURCE_TRACK
760 struct resource_track rtrack;
761#endif
762 struct dir_info *dir;
763 ext2_u32_iterate iter;
764 ext2_ino_t ino;
765 errcode_t retval;
b0700a1b 766 int i, cur, max, all_dirs, dir_index, first = 1;
850d05e9
TT
767
768#ifdef RESOURCE_TRACK
769 init_resource_track(&rtrack);
770#endif
b7a00563 771
850d05e9
TT
772 all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
773
774 if (!ctx->dirs_to_hash && !all_dirs)
b7a00563
TT
775 return;
776
850d05e9 777 e2fsck_get_lost_and_found(ctx, 0);
b7a00563
TT
778
779 clear_problem_context(&pctx);
850d05e9
TT
780
781 dir_index = ctx->fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_DIR_INDEX;
b0700a1b
TT
782 cur = 0;
783 if (all_dirs) {
850d05e9 784 i = 0;
b0700a1b
TT
785 max = e2fsck_get_num_dirinfo(ctx);
786 } else {
850d05e9
TT
787 retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
788 &iter);
789 if (retval) {
790 pctx.errcode = retval;
791 fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
792 return;
793 }
b0700a1b 794 max = ext2fs_u32_list_count(ctx->dirs_to_hash);
b7a00563 795 }
850d05e9
TT
796 while (1) {
797 if (all_dirs) {
798 if ((dir = e2fsck_dir_info_iter(ctx, &i)) == 0)
799 break;
800 ino = dir->ino;
801 } else {
802 if (!ext2fs_u32_list_iterate(iter, &ino))
803 break;
804 }
805 if (ino == ctx->lost_and_found)
b7a00563
TT
806 continue;
807 pctx.dir = ino;
808 if (first) {
809 fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
810 first = 0;
811 }
b0700a1b 812#if 0
850d05e9 813 fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
b0700a1b 814#endif
b7a00563
TT
815 pctx.errcode = e2fsck_rehash_dir(ctx, ino);
816 if (pctx.errcode) {
850d05e9
TT
817 end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
818 fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
b7a00563 819 }
52734dc5
TT
820 if (ctx->progress && !ctx->progress_fd)
821 e2fsck_simple_progress(ctx, "Rebuilding directory",
1d2eef42 822 100.0 * (float) (++cur) / (float) max, ino);
b7a00563 823 }
850d05e9
TT
824 end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
825 if (!all_dirs)
826 ext2fs_u32_list_iterate_end(iter);
b7a00563 827
850d05e9
TT
828 if (ctx->dirs_to_hash)
829 ext2fs_u32_list_free(ctx->dirs_to_hash);
b7a00563 830 ctx->dirs_to_hash = 0;
850d05e9
TT
831
832#ifdef RESOURCE_TRACK
833 if (ctx->options & E2F_OPT_TIME2) {
834 e2fsck_clear_progbar(ctx);
835 print_resource_track("Pass 3A", &rtrack);
836 }
837#endif
b7a00563 838}