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 <string.h>
49#include <ctype.h>
50#include <errno.h>
51#include "e2fsck.h"
52#include "problem.h"
53
54struct fill_dir_struct {
55	char *buf;
56	struct ext2_inode *inode;
57	int err;
58	e2fsck_t ctx;
59	struct hash_entry *harray;
60	int max_array, num_array;
61	int dir_size;
62	int compress;
63	ino_t parent;
64};
65
66struct hash_entry {
67	ext2_dirhash_t	hash;
68	ext2_dirhash_t	minor_hash;
69	ino_t		ino;
70	struct ext2_dir_entry	*dir;
71};
72
73struct out_dir {
74	int		num;
75	int		max;
76	char		*buf;
77	ext2_dirhash_t	*hashes;
78};
79
80static int fill_dir_block(ext2_filsys fs,
81			  blk_t	*block_nr,
82			  e2_blkcnt_t blockcnt,
83			  blk_t ref_block EXT2FS_ATTR((unused)),
84			  int ref_offset EXT2FS_ATTR((unused)),
85			  void *priv_data)
86{
87	struct fill_dir_struct	*fd = (struct fill_dir_struct *) priv_data;
88	struct hash_entry 	*new_array, *ent;
89	struct ext2_dir_entry 	*dirent;
90	char			*dir;
91	unsigned int		offset, dir_offset;
92	int			hash_alg;
93
94	if (blockcnt < 0)
95		return 0;
96
97	offset = blockcnt * fs->blocksize;
98	if (offset + fs->blocksize > fd->inode->i_size) {
99		fd->err = EXT2_ET_DIR_CORRUPTED;
100		return BLOCK_ABORT;
101	}
102	dir = (fd->buf+offset);
103	if (HOLE_BLKADDR(*block_nr)) {
104		memset(dir, 0, fs->blocksize);
105		dirent = (struct ext2_dir_entry *) dir;
106		dirent->rec_len = fs->blocksize;
107	} else {
108		fd->err = ext2fs_read_dir_block(fs, *block_nr, dir);
109		if (fd->err)
110			return BLOCK_ABORT;
111	}
112	hash_alg = fs->super->s_def_hash_version;
113	if ((hash_alg <= EXT2_HASH_TEA) &&
114	    (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
115		hash_alg += 3;
116	/* While the directory block is "hot", index it. */
117	dir_offset = 0;
118	while (dir_offset < fs->blocksize) {
119		dirent = (struct ext2_dir_entry *) (dir + dir_offset);
120		if (((dir_offset + dirent->rec_len) > fs->blocksize) ||
121		    (dirent->rec_len < 8) ||
122		    ((dirent->rec_len % 4) != 0) ||
123		    (((dirent->name_len & 0xFF)+8) > dirent->rec_len)) {
124			fd->err = EXT2_ET_DIR_CORRUPTED;
125			return BLOCK_ABORT;
126		}
127		dir_offset += dirent->rec_len;
128		if (dirent->inode == 0)
129			continue;
130		if (!fd->compress && ((dirent->name_len&0xFF) == 1) &&
131		    (dirent->name[0] == '.'))
132			continue;
133		if (!fd->compress && ((dirent->name_len&0xFF) == 2) &&
134		    (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
135			fd->parent = dirent->inode;
136			continue;
137		}
138		if (fd->num_array >= fd->max_array) {
139			new_array = realloc(fd->harray,
140			    sizeof(struct hash_entry) * (fd->max_array+500));
141			if (!new_array) {
142				fd->err = ENOMEM;
143				return BLOCK_ABORT;
144			}
145			fd->harray = new_array;
146			fd->max_array += 500;
147		}
148		ent = fd->harray + fd->num_array++;
149		ent->dir = dirent;
150		fd->dir_size += EXT2_DIR_REC_LEN(dirent->name_len & 0xFF);
151		ent->ino = dirent->inode;
152		if (fd->compress)
153			ent->hash = ent->minor_hash = 0;
154		else {
155			fd->err = ext2fs_dirhash(hash_alg, dirent->name,
156						 dirent->name_len & 0xFF,
157						 fs->super->s_hash_seed,
158						 &ent->hash, &ent->minor_hash);
159			if (fd->err)
160				return BLOCK_ABORT;
161		}
162	}
163
164	return 0;
165}
166
167/* Used for sorting the hash entry */
168static EXT2_QSORT_TYPE ino_cmp(const void *a, const void *b)
169{
170	const struct hash_entry *he_a = (const struct hash_entry *) a;
171	const struct hash_entry *he_b = (const struct hash_entry *) b;
172
173	return (he_a->ino - he_b->ino);
174}
175
176/* Used for sorting the hash entry */
177static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
178{
179	const struct hash_entry *he_a = (const struct hash_entry *) a;
180	const struct hash_entry *he_b = (const struct hash_entry *) b;
181	int	ret;
182	int	min_len;
183
184	min_len = he_a->dir->name_len;
185	if (min_len > he_b->dir->name_len)
186		min_len = he_b->dir->name_len;
187
188	ret = strncmp(he_a->dir->name, he_b->dir->name, min_len);
189	if (ret == 0) {
190		if (he_a->dir->name_len > he_b->dir->name_len)
191			ret = 1;
192		else if (he_a->dir->name_len < he_b->dir->name_len)
193			ret = -1;
194		else
195			ret = he_b->dir->inode - he_a->dir->inode;
196	}
197	return ret;
198}
199
200/* Used for sorting the hash entry */
201static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
202{
203	const struct hash_entry *he_a = (const struct hash_entry *) a;
204	const struct hash_entry *he_b = (const struct hash_entry *) b;
205	int	ret;
206
207	if (he_a->hash > he_b->hash)
208		ret = 1;
209	else if (he_a->hash < he_b->hash)
210		ret = -1;
211	else {
212		if (he_a->minor_hash > he_b->minor_hash)
213			ret = 1;
214		else if (he_a->minor_hash < he_b->minor_hash)
215			ret = -1;
216		else
217			ret = name_cmp(a, b);
218	}
219	return ret;
220}
221
222static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
223				int blocks)
224{
225	void			*new_mem;
226
227	if (outdir->max) {
228		new_mem = realloc(outdir->buf, blocks * fs->blocksize);
229		if (!new_mem)
230			return ENOMEM;
231		outdir->buf = new_mem;
232		new_mem = realloc(outdir->hashes,
233				  blocks * sizeof(ext2_dirhash_t));
234		if (!new_mem)
235			return ENOMEM;
236		outdir->hashes = new_mem;
237	} else {
238		outdir->buf = malloc(blocks * fs->blocksize);
239		outdir->hashes = malloc(blocks * sizeof(ext2_dirhash_t));
240		outdir->num = 0;
241	}
242	outdir->max = blocks;
243	return 0;
244}
245
246static void free_out_dir(struct out_dir *outdir)
247{
248	if (outdir->buf)
249		free(outdir->buf);
250	if (outdir->hashes)
251		free(outdir->hashes);
252	outdir->max = 0;
253	outdir->num =0;
254}
255
256static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
257			 char ** ret)
258{
259	errcode_t	retval;
260
261	if (outdir->num >= outdir->max) {
262		retval = alloc_size_dir(fs, outdir, outdir->max + 50);
263		if (retval)
264			return retval;
265	}
266	*ret = outdir->buf + (outdir->num++ * fs->blocksize);
267	memset(*ret, 0, fs->blocksize);
268	return 0;
269}
270
271/*
272 * This function is used to make a unique filename.  We do this by
273 * appending ~0, and then incrementing the number.  However, we cannot
274 * expand the length of the filename beyond the padding available in
275 * the directory entry.
276 */
277static void mutate_name(char *str, __u16 *len)
278{
279	int	i;
280	__u16	l = *len & 0xFF, h = *len & 0xff00;
281
282	/*
283	 * First check to see if it looks the name has been mutated
284	 * already
285	 */
286	for (i = l-1; i > 0; i--) {
287		if (!isdigit(str[i]))
288			break;
289	}
290	if ((i == l-1) || (str[i] != '~')) {
291		if (((l-1) & 3) < 2)
292			l += 2;
293		else
294			l = (l+3) & ~3;
295		str[l-2] = '~';
296		str[l-1] = '0';
297		*len = l | h;
298		return;
299	}
300	for (i = l-1; i >= 0; i--) {
301		if (isdigit(str[i])) {
302			if (str[i] == '9')
303				str[i] = '0';
304			else {
305				str[i]++;
306				return;
307			}
308			continue;
309		}
310		if (i == 1) {
311			if (str[0] == 'z')
312				str[0] = 'A';
313			else if (str[0] == 'Z') {
314				str[0] = '~';
315				str[1] = '0';
316			} else
317				str[0]++;
318		} else if (i > 0) {
319			str[i] = '1';
320			str[i-1] = '~';
321		} else {
322			if (str[0] == '~')
323				str[0] = 'a';
324			else
325				str[0]++;
326		}
327		break;
328	}
329}
330
331static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
332				    ext2_ino_t ino,
333				    struct fill_dir_struct *fd)
334{
335	struct problem_context	pctx;
336	struct hash_entry 	*ent, *prev;
337	int			i, j;
338	int			fixed = 0;
339	char			new_name[256];
340	__u16			new_len;
341	int			hash_alg;
342
343	clear_problem_context(&pctx);
344	pctx.ino = ino;
345
346	hash_alg = fs->super->s_def_hash_version;
347	if ((hash_alg <= EXT2_HASH_TEA) &&
348	    (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
349		hash_alg += 3;
350
351	for (i=1; i < fd->num_array; i++) {
352		ent = fd->harray + i;
353		prev = ent - 1;
354		if (!ent->dir->inode ||
355		    ((ent->dir->name_len & 0xFF) !=
356		     (prev->dir->name_len & 0xFF)) ||
357		    (strncmp(ent->dir->name, prev->dir->name,
358			     ent->dir->name_len & 0xFF)))
359			continue;
360		pctx.dirent = ent->dir;
361		if ((ent->dir->inode == prev->dir->inode) &&
362		    fix_problem(ctx, PR_2_DUPLICATE_DIRENT, &pctx)) {
363			e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
364			ent->dir->inode = 0;
365			fixed++;
366			continue;
367		}
368		memcpy(new_name, ent->dir->name, ent->dir->name_len & 0xFF);
369		new_len = ent->dir->name_len;
370		mutate_name(new_name, &new_len);
371		for (j=0; j < fd->num_array; j++) {
372			if ((i==j) ||
373			    ((ent->dir->name_len & 0xFF) !=
374			     (fd->harray[j].dir->name_len & 0xFF)) ||
375			    (strncmp(new_name, fd->harray[j].dir->name,
376				     new_len & 0xFF)))
377				continue;
378			mutate_name(new_name, &new_len);
379
380			j = -1;
381		}
382		new_name[new_len & 0xFF] = 0;
383		pctx.str = new_name;
384		if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
385			memcpy(ent->dir->name, new_name, new_len & 0xFF);
386			ent->dir->name_len = new_len;
387			ext2fs_dirhash(hash_alg, ent->dir->name,
388				       ent->dir->name_len & 0xFF,
389				       fs->super->s_hash_seed,
390				       &ent->hash, &ent->minor_hash);
391			fixed++;
392		}
393	}
394	return fixed;
395}
396
397
398static errcode_t copy_dir_entries(ext2_filsys fs,
399				  struct fill_dir_struct *fd,
400				  struct out_dir *outdir)
401{
402	errcode_t		retval;
403	char			*block_start;
404	struct hash_entry 	*ent;
405	struct ext2_dir_entry	*dirent;
406	int			i, rec_len, left;
407	ext2_dirhash_t		prev_hash;
408	int			offset;
409
410	outdir->max = 0;
411	retval = alloc_size_dir(fs, outdir,
412				(fd->dir_size / fs->blocksize) + 2);
413	if (retval)
414		return retval;
415	outdir->num = fd->compress ? 0 : 1;
416	offset = 0;
417	outdir->hashes[0] = 0;
418	prev_hash = 1;
419	if ((retval = get_next_block(fs, outdir, &block_start)))
420		return retval;
421	dirent = (struct ext2_dir_entry *) block_start;
422	left = fs->blocksize;
423	for (i=0; i < fd->num_array; i++) {
424		ent = fd->harray + i;
425		if (ent->dir->inode == 0)
426			continue;
427		rec_len = EXT2_DIR_REC_LEN(ent->dir->name_len & 0xFF);
428		if (rec_len > left) {
429			if (left)
430				dirent->rec_len += left;
431			if ((retval = get_next_block(fs, outdir,
432						      &block_start)))
433				return retval;
434			offset = 0;
435		}
436		left = fs->blocksize - offset;
437		dirent = (struct ext2_dir_entry *) (block_start + offset);
438		if (offset == 0) {
439			if (ent->hash == prev_hash)
440				outdir->hashes[outdir->num-1] = ent->hash | 1;
441			else
442				outdir->hashes[outdir->num-1] = ent->hash;
443		}
444		dirent->inode = ent->dir->inode;
445		dirent->name_len = ent->dir->name_len;
446		dirent->rec_len = rec_len;
447		memcpy(dirent->name, ent->dir->name, dirent->name_len & 0xFF);
448		offset += rec_len;
449		left -= rec_len;
450		if (left < 12) {
451			dirent->rec_len += left;
452			offset += left;
453			left = 0;
454		}
455		prev_hash = ent->hash;
456	}
457	if (left)
458		dirent->rec_len += left;
459
460	return 0;
461}
462
463
464static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
465				    ext2_ino_t ino, ext2_ino_t parent)
466{
467	struct ext2_dir_entry 		*dir;
468	struct ext2_dx_root_info  	*root;
469	struct ext2_dx_countlimit	*limits;
470	int				filetype = 0;
471
472	if (fs->super->s_feature_incompat & EXT2_FEATURE_INCOMPAT_FILETYPE)
473		filetype = EXT2_FT_DIR << 8;
474
475	memset(buf, 0, fs->blocksize);
476	dir = (struct ext2_dir_entry *) buf;
477	dir->inode = ino;
478	dir->name[0] = '.';
479	dir->name_len = 1 | filetype;
480	dir->rec_len = 12;
481	dir = (struct ext2_dir_entry *) (buf + 12);
482	dir->inode = parent;
483	dir->name[0] = '.';
484	dir->name[1] = '.';
485	dir->name_len = 2 | filetype;
486	dir->rec_len = fs->blocksize - 12;
487
488	root = (struct ext2_dx_root_info *) (buf+24);
489	root->reserved_zero = 0;
490	root->hash_version = fs->super->s_def_hash_version;
491	root->info_length = 8;
492	root->indirect_levels = 0;
493	root->unused_flags = 0;
494
495	limits = (struct ext2_dx_countlimit *) (buf+32);
496	limits->limit = (fs->blocksize - 32) / sizeof(struct ext2_dx_entry);
497	limits->count = 0;
498
499	return root;
500}
501
502
503static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
504{
505	struct ext2_dir_entry 		*dir;
506	struct ext2_dx_countlimit	*limits;
507
508	memset(buf, 0, fs->blocksize);
509	dir = (struct ext2_dir_entry *) buf;
510	dir->inode = 0;
511	dir->rec_len = fs->blocksize;
512
513	limits = (struct ext2_dx_countlimit *) (buf+8);
514	limits->limit = (fs->blocksize - 8) / sizeof(struct ext2_dx_entry);
515	limits->count = 0;
516
517	return (struct ext2_dx_entry *) limits;
518}
519
520/*
521 * This function takes the leaf nodes which have been written in
522 * outdir, and populates the root node and any necessary interior nodes.
523 */
524static errcode_t calculate_tree(ext2_filsys fs,
525				struct out_dir *outdir,
526				ext2_ino_t ino,
527				ext2_ino_t parent)
528{
529	struct ext2_dx_root_info  	*root_info;
530	struct ext2_dx_entry 		*root, *dx_ent = 0;
531	struct ext2_dx_countlimit	*root_limit, *limit;
532	errcode_t			retval;
533	char				* block_start;
534	int				i, c1, c2, nblks;
535	int				limit_offset, root_offset;
536
537	root_info = set_root_node(fs, outdir->buf, ino, parent);
538	root_offset = limit_offset = ((char *) root_info - outdir->buf) +
539		root_info->info_length;
540	root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
541	c1 = root_limit->limit;
542	nblks = outdir->num;
543
544	/* Write out the pointer blocks */
545	if (nblks-1 <= c1) {
546		/* Just write out the root block, and we're done */
547		root = (struct ext2_dx_entry *) (outdir->buf + root_offset);
548		for (i=1; i < nblks; i++) {
549			root->block = ext2fs_cpu_to_le32(i);
550			if (i != 1)
551				root->hash =
552					ext2fs_cpu_to_le32(outdir->hashes[i]);
553			root++;
554			c1--;
555		}
556	} else {
557		c2 = 0;
558		limit = 0;
559		root_info->indirect_levels = 1;
560		for (i=1; i < nblks; i++) {
561			if (c1 == 0)
562				return ENOSPC;
563			if (c2 == 0) {
564				if (limit)
565					limit->limit = limit->count =
566		ext2fs_cpu_to_le16(limit->limit);
567				root = (struct ext2_dx_entry *)
568					(outdir->buf + root_offset);
569				root->block = ext2fs_cpu_to_le32(outdir->num);
570				if (i != 1)
571					root->hash =
572			ext2fs_cpu_to_le32(outdir->hashes[i]);
573				if ((retval =  get_next_block(fs, outdir,
574							      &block_start)))
575					return retval;
576				dx_ent = set_int_node(fs, block_start);
577				limit = (struct ext2_dx_countlimit *) dx_ent;
578				c2 = limit->limit;
579				root_offset += sizeof(struct ext2_dx_entry);
580				c1--;
581			}
582			dx_ent->block = ext2fs_cpu_to_le32(i);
583			if (c2 != limit->limit)
584				dx_ent->hash =
585					ext2fs_cpu_to_le32(outdir->hashes[i]);
586			dx_ent++;
587			c2--;
588		}
589		limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
590		limit->limit = ext2fs_cpu_to_le16(limit->limit);
591	}
592	root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
593	root_limit->count = ext2fs_cpu_to_le16(root_limit->limit - c1);
594	root_limit->limit = ext2fs_cpu_to_le16(root_limit->limit);
595
596	return 0;
597}
598
599struct write_dir_struct {
600	struct out_dir *outdir;
601	errcode_t	err;
602	e2fsck_t	ctx;
603	int		cleared;
604};
605
606/*
607 * Helper function which writes out a directory block.
608 */
609static int write_dir_block(ext2_filsys fs,
610			   blk_t	*block_nr,
611			   e2_blkcnt_t blockcnt,
612			   blk_t ref_block EXT2FS_ATTR((unused)),
613			   int ref_offset EXT2FS_ATTR((unused)),
614			   void *priv_data)
615{
616	struct write_dir_struct	*wd = (struct write_dir_struct *) priv_data;
617	blk_t	blk;
618	char	*dir;
619
620	if (*block_nr == 0)
621		return 0;
622	if (blockcnt >= wd->outdir->num) {
623		e2fsck_read_bitmaps(wd->ctx);
624		blk = *block_nr;
625		ext2fs_unmark_block_bitmap(wd->ctx->block_found_map, blk);
626		ext2fs_block_alloc_stats(fs, blk, -1);
627		*block_nr = 0;
628		wd->cleared++;
629		return BLOCK_CHANGED;
630	}
631	if (blockcnt < 0)
632		return 0;
633
634	dir = wd->outdir->buf + (blockcnt * fs->blocksize);
635	wd->err = ext2fs_write_dir_block(fs, *block_nr, dir);
636	if (wd->err)
637		return BLOCK_ABORT;
638	return 0;
639}
640
641static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
642				 struct out_dir *outdir,
643				 ext2_ino_t ino, int compress)
644{
645	struct write_dir_struct wd;
646	errcode_t	retval;
647	struct ext2_inode 	inode;
648
649	retval = e2fsck_expand_directory(ctx, ino, -1, outdir->num);
650	if (retval)
651		return retval;
652
653	wd.outdir = outdir;
654	wd.err = 0;
655	wd.ctx = ctx;
656	wd.cleared = 0;
657
658	retval = ext2fs_block_iterate2(fs, ino, 0, 0,
659				       write_dir_block, &wd);
660	if (retval)
661		return retval;
662	if (wd.err)
663		return wd.err;
664
665	e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
666	if (compress)
667		inode.i_flags &= ~EXT2_INDEX_FL;
668	else
669		inode.i_flags |= EXT2_INDEX_FL;
670	inode.i_size = outdir->num * fs->blocksize;
671	inode.i_blocks -= (fs->blocksize / 512) * wd.cleared;
672	e2fsck_write_inode(ctx, ino, &inode, "rehash_dir");
673
674	return 0;
675}
676
677errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino)
678{
679	ext2_filsys 		fs = ctx->fs;
680	errcode_t		retval;
681	struct ext2_inode 	inode;
682	char			*dir_buf = 0;
683	struct fill_dir_struct	fd;
684	struct out_dir		outdir;
685
686	outdir.max = outdir.num = 0;
687	outdir.buf = 0;
688	outdir.hashes = 0;
689	e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
690
691	retval = ENOMEM;
692	fd.harray = 0;
693	dir_buf = malloc(inode.i_size);
694	if (!dir_buf)
695		goto errout;
696
697	fd.max_array = inode.i_size / 32;
698	fd.num_array = 0;
699	fd.harray = malloc(fd.max_array * sizeof(struct hash_entry));
700	if (!fd.harray)
701		goto errout;
702
703	fd.ctx = ctx;
704	fd.buf = dir_buf;
705	fd.inode = &inode;
706	fd.err = 0;
707	fd.dir_size = 0;
708	fd.compress = 0;
709	if (!(fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_DIR_INDEX) ||
710	    (inode.i_size / fs->blocksize) < 2)
711		fd.compress = 1;
712	fd.parent = 0;
713
714	/* Read in the entire directory into memory */
715	retval = ext2fs_block_iterate2(fs, ino, 0, 0,
716				       fill_dir_block, &fd);
717	if (fd.err) {
718		retval = fd.err;
719		goto errout;
720	}
721
722#if 0
723	printf("%d entries (%d bytes) found in inode %d\n",
724	       fd.num_array, fd.dir_size, ino);
725#endif
726
727	/* Sort the list */
728resort:
729	if (fd.compress)
730		qsort(fd.harray+2, fd.num_array-2,
731		      sizeof(struct hash_entry), ino_cmp);
732	else
733		qsort(fd.harray, fd.num_array,
734		      sizeof(struct hash_entry), hash_cmp);
735
736	/*
737	 * Look for duplicates
738	 */
739	if (duplicate_search_and_fix(ctx, fs, ino, &fd))
740		goto resort;
741
742	if (ctx->options & E2F_OPT_NO) {
743		retval = 0;
744		goto errout;
745	}
746
747	/*
748	 * Copy the directory entries.  In a htree directory these
749	 * will become the leaf nodes.
750	 */
751	retval = copy_dir_entries(fs, &fd, &outdir);
752	if (retval)
753		goto errout;
754
755	free(dir_buf); dir_buf = 0;
756
757	if (!fd.compress) {
758		/* Calculate the interior nodes */
759		retval = calculate_tree(fs, &outdir, ino, fd.parent);
760		if (retval)
761			goto errout;
762	}
763
764	retval = write_directory(ctx, fs, &outdir, ino, fd.compress);
765	if (retval)
766		goto errout;
767
768errout:
769	if (dir_buf)
770		free(dir_buf);
771	if (fd.harray)
772		free(fd.harray);
773
774	free_out_dir(&outdir);
775	return retval;
776}
777
778void e2fsck_rehash_directories(e2fsck_t ctx)
779{
780	struct problem_context	pctx;
781#ifdef RESOURCE_TRACK
782	struct resource_track	rtrack;
783#endif
784	struct dir_info		*dir;
785	ext2_u32_iterate 	iter;
786	struct dir_info_iter *	dirinfo_iter = 0;
787	ext2_ino_t		ino;
788	errcode_t		retval;
789	int			cur, max, all_dirs, dir_index, first = 1;
790
791#ifdef RESOURCE_TRACK
792	init_resource_track(&rtrack);
793#endif
794
795	all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
796
797	if (!ctx->dirs_to_hash && !all_dirs)
798		return;
799
800	e2fsck_get_lost_and_found(ctx, 0);
801
802	clear_problem_context(&pctx);
803
804	dir_index = ctx->fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_DIR_INDEX;
805	cur = 0;
806	if (all_dirs) {
807		dirinfo_iter = e2fsck_dir_info_iter_begin(ctx);
808		max = e2fsck_get_num_dirinfo(ctx);
809	} else {
810		retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
811						       &iter);
812		if (retval) {
813			pctx.errcode = retval;
814			fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
815			return;
816		}
817		max = ext2fs_u32_list_count(ctx->dirs_to_hash);
818	}
819	while (1) {
820		if (all_dirs) {
821			if ((dir = e2fsck_dir_info_iter(ctx,
822							dirinfo_iter)) == 0)
823				break;
824			ino = dir->ino;
825		} else {
826			if (!ext2fs_u32_list_iterate(iter, &ino))
827				break;
828		}
829		if (ino == ctx->lost_and_found)
830			continue;
831		pctx.dir = ino;
832		if (first) {
833			fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
834			first = 0;
835		}
836#if 0
837		fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
838#endif
839		pctx.errcode = e2fsck_rehash_dir(ctx, ino);
840		if (pctx.errcode) {
841			end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
842			fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
843		}
844		if (ctx->progress && !ctx->progress_fd)
845			e2fsck_simple_progress(ctx, "Rebuilding directory",
846			       100.0 * (float) (++cur) / (float) max, ino);
847	}
848	end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
849	if (all_dirs)
850		e2fsck_dir_info_iter_end(ctx, dirinfo_iter);
851	else
852		ext2fs_u32_list_iterate_end(iter);
853
854	if (ctx->dirs_to_hash)
855		ext2fs_u32_list_free(ctx->dirs_to_hash);
856	ctx->dirs_to_hash = 0;
857
858#ifdef RESOURCE_TRACK
859	if (ctx->options & E2F_OPT_TIME2) {
860		e2fsck_clear_progbar(ctx);
861		print_resource_track("Pass 3A", &rtrack);
862	}
863#endif
864}
865