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