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 
54 struct fill_dir_struct {
55 	char *buf;
56 	struct ext2_inode *inode;
57 	errcode_t err;
58 	e2fsck_t ctx;
59 	struct hash_entry *harray;
60 	int max_array, num_array;
61 	unsigned int dir_size;
62 	int compress;
63 	ino_t parent;
64 };
65 
66 struct 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 
73 struct out_dir {
74 	int		num;
75 	int		max;
76 	char		*buf;
77 	ext2_dirhash_t	*hashes;
78 };
79 
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)80 static int fill_dir_block(ext2_filsys fs,
81 			  blk64_t *block_nr,
82 			  e2_blkcnt_t blockcnt,
83 			  blk64_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, rec_len;
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 		(void) ext2fs_set_rec_len(fs, fs->blocksize, dirent);
107 	} else {
108 		fd->err = ext2fs_read_dir_block3(fs, *block_nr, dir, 0);
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 		(void) ext2fs_get_rec_len(fs, dirent, &rec_len);
121 		if (((dir_offset + rec_len) > fs->blocksize) ||
122 		    (rec_len < 8) ||
123 		    ((rec_len % 4) != 0) ||
124 		    (((dirent->name_len & 0xFF)+8U) > rec_len)) {
125 			fd->err = EXT2_ET_DIR_CORRUPTED;
126 			return BLOCK_ABORT;
127 		}
128 		dir_offset += rec_len;
129 		if (dirent->inode == 0)
130 			continue;
131 		if (!fd->compress && ((dirent->name_len&0xFF) == 1) &&
132 		    (dirent->name[0] == '.'))
133 			continue;
134 		if (!fd->compress && ((dirent->name_len&0xFF) == 2) &&
135 		    (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
136 			fd->parent = dirent->inode;
137 			continue;
138 		}
139 		if (fd->num_array >= fd->max_array) {
140 			new_array = realloc(fd->harray,
141 			    sizeof(struct hash_entry) * (fd->max_array+500));
142 			if (!new_array) {
143 				fd->err = ENOMEM;
144 				return BLOCK_ABORT;
145 			}
146 			fd->harray = new_array;
147 			fd->max_array += 500;
148 		}
149 		ent = fd->harray + fd->num_array++;
150 		ent->dir = dirent;
151 		fd->dir_size += EXT2_DIR_REC_LEN(dirent->name_len & 0xFF);
152 		ent->ino = dirent->inode;
153 		if (fd->compress)
154 			ent->hash = ent->minor_hash = 0;
155 		else {
156 			fd->err = ext2fs_dirhash(hash_alg, dirent->name,
157 						 dirent->name_len & 0xFF,
158 						 fs->super->s_hash_seed,
159 						 &ent->hash, &ent->minor_hash);
160 			if (fd->err)
161 				return BLOCK_ABORT;
162 		}
163 	}
164 
165 	return 0;
166 }
167 
168 /* Used for sorting the hash entry */
ino_cmp(const void * a,const void * b)169 static EXT2_QSORT_TYPE ino_cmp(const void *a, const void *b)
170 {
171 	const struct hash_entry *he_a = (const struct hash_entry *) a;
172 	const struct hash_entry *he_b = (const struct hash_entry *) b;
173 
174 	return (he_a->ino - he_b->ino);
175 }
176 
177 /* Used for sorting the hash entry */
name_cmp(const void * a,const void * b)178 static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
179 {
180 	const struct hash_entry *he_a = (const struct hash_entry *) a;
181 	const struct hash_entry *he_b = (const struct hash_entry *) b;
182 	int	ret;
183 	int	min_len;
184 
185 	min_len = he_a->dir->name_len;
186 	if (min_len > he_b->dir->name_len)
187 		min_len = he_b->dir->name_len;
188 
189 	ret = strncmp(he_a->dir->name, he_b->dir->name, min_len);
190 	if (ret == 0) {
191 		if (he_a->dir->name_len > he_b->dir->name_len)
192 			ret = 1;
193 		else if (he_a->dir->name_len < he_b->dir->name_len)
194 			ret = -1;
195 		else
196 			ret = he_b->dir->inode - he_a->dir->inode;
197 	}
198 	return ret;
199 }
200 
201 /* Used for sorting the hash entry */
hash_cmp(const void * a,const void * b)202 static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
203 {
204 	const struct hash_entry *he_a = (const struct hash_entry *) a;
205 	const struct hash_entry *he_b = (const struct hash_entry *) b;
206 	int	ret;
207 
208 	if (he_a->hash > he_b->hash)
209 		ret = 1;
210 	else if (he_a->hash < he_b->hash)
211 		ret = -1;
212 	else {
213 		if (he_a->minor_hash > he_b->minor_hash)
214 			ret = 1;
215 		else if (he_a->minor_hash < he_b->minor_hash)
216 			ret = -1;
217 		else
218 			ret = name_cmp(a, b);
219 	}
220 	return ret;
221 }
222 
alloc_size_dir(ext2_filsys fs,struct out_dir * outdir,int blocks)223 static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
224 				int blocks)
225 {
226 	void			*new_mem;
227 
228 	if (outdir->max) {
229 		new_mem = realloc(outdir->buf, blocks * fs->blocksize);
230 		if (!new_mem)
231 			return ENOMEM;
232 		outdir->buf = new_mem;
233 		new_mem = realloc(outdir->hashes,
234 				  blocks * sizeof(ext2_dirhash_t));
235 		if (!new_mem)
236 			return ENOMEM;
237 		outdir->hashes = new_mem;
238 	} else {
239 		outdir->buf = malloc(blocks * fs->blocksize);
240 		outdir->hashes = malloc(blocks * sizeof(ext2_dirhash_t));
241 		outdir->num = 0;
242 	}
243 	outdir->max = blocks;
244 	return 0;
245 }
246 
free_out_dir(struct out_dir * outdir)247 static void free_out_dir(struct out_dir *outdir)
248 {
249 	free(outdir->buf);
250 	free(outdir->hashes);
251 	outdir->max = 0;
252 	outdir->num =0;
253 }
254 
get_next_block(ext2_filsys fs,struct out_dir * outdir,char ** ret)255 static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
256 			 char ** ret)
257 {
258 	errcode_t	retval;
259 
260 	if (outdir->num >= outdir->max) {
261 		retval = alloc_size_dir(fs, outdir, outdir->max + 50);
262 		if (retval)
263 			return retval;
264 	}
265 	*ret = outdir->buf + (outdir->num++ * fs->blocksize);
266 	memset(*ret, 0, fs->blocksize);
267 	return 0;
268 }
269 
270 /*
271  * This function is used to make a unique filename.  We do this by
272  * appending ~0, and then incrementing the number.  However, we cannot
273  * expand the length of the filename beyond the padding available in
274  * the directory entry.
275  */
mutate_name(char * str,__u16 * len)276 static void mutate_name(char *str, __u16 *len)
277 {
278 	int	i;
279 	__u16	l = *len & 0xFF, h = *len & 0xff00;
280 
281 	/*
282 	 * First check to see if it looks the name has been mutated
283 	 * already
284 	 */
285 	for (i = l-1; i > 0; i--) {
286 		if (!isdigit(str[i]))
287 			break;
288 	}
289 	if ((i == l-1) || (str[i] != '~')) {
290 		if (((l-1) & 3) < 2)
291 			l += 2;
292 		else
293 			l = (l+3) & ~3;
294 		str[l-2] = '~';
295 		str[l-1] = '0';
296 		*len = l | h;
297 		return;
298 	}
299 	for (i = l-1; i >= 0; i--) {
300 		if (isdigit(str[i])) {
301 			if (str[i] == '9')
302 				str[i] = '0';
303 			else {
304 				str[i]++;
305 				return;
306 			}
307 			continue;
308 		}
309 		if (i == 1) {
310 			if (str[0] == 'z')
311 				str[0] = 'A';
312 			else if (str[0] == 'Z') {
313 				str[0] = '~';
314 				str[1] = '0';
315 			} else
316 				str[0]++;
317 		} else if (i > 0) {
318 			str[i] = '1';
319 			str[i-1] = '~';
320 		} else {
321 			if (str[0] == '~')
322 				str[0] = 'a';
323 			else
324 				str[0]++;
325 		}
326 		break;
327 	}
328 }
329 
duplicate_search_and_fix(e2fsck_t ctx,ext2_filsys fs,ext2_ino_t ino,struct fill_dir_struct * fd)330 static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
331 				    ext2_ino_t ino,
332 				    struct fill_dir_struct *fd)
333 {
334 	struct problem_context	pctx;
335 	struct hash_entry 	*ent, *prev;
336 	int			i, j;
337 	int			fixed = 0;
338 	char			new_name[256];
339 	__u16			new_len;
340 	int			hash_alg;
341 
342 	clear_problem_context(&pctx);
343 	pctx.ino = ino;
344 
345 	hash_alg = fs->super->s_def_hash_version;
346 	if ((hash_alg <= EXT2_HASH_TEA) &&
347 	    (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
348 		hash_alg += 3;
349 
350 	for (i=1; i < fd->num_array; i++) {
351 		ent = fd->harray + i;
352 		prev = ent - 1;
353 		if (!ent->dir->inode ||
354 		    ((ent->dir->name_len & 0xFF) !=
355 		     (prev->dir->name_len & 0xFF)) ||
356 		    (strncmp(ent->dir->name, prev->dir->name,
357 			     ent->dir->name_len & 0xFF)))
358 			continue;
359 		pctx.dirent = ent->dir;
360 		if ((ent->dir->inode == prev->dir->inode) &&
361 		    fix_problem(ctx, PR_2_DUPLICATE_DIRENT, &pctx)) {
362 			e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
363 			ent->dir->inode = 0;
364 			fixed++;
365 			continue;
366 		}
367 		memcpy(new_name, ent->dir->name, ent->dir->name_len & 0xFF);
368 		new_len = ent->dir->name_len;
369 		mutate_name(new_name, &new_len);
370 		for (j=0; j < fd->num_array; j++) {
371 			if ((i==j) ||
372 			    ((new_len & 0xFF) !=
373 			     (fd->harray[j].dir->name_len & 0xFF)) ||
374 			    (strncmp(new_name, fd->harray[j].dir->name,
375 				     new_len & 0xFF)))
376 				continue;
377 			mutate_name(new_name, &new_len);
378 
379 			j = -1;
380 		}
381 		new_name[new_len & 0xFF] = 0;
382 		pctx.str = new_name;
383 		if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
384 			memcpy(ent->dir->name, new_name, new_len & 0xFF);
385 			ent->dir->name_len = new_len;
386 			ext2fs_dirhash(hash_alg, ent->dir->name,
387 				       ent->dir->name_len & 0xFF,
388 				       fs->super->s_hash_seed,
389 				       &ent->hash, &ent->minor_hash);
390 			fixed++;
391 		}
392 	}
393 	return fixed;
394 }
395 
396 
copy_dir_entries(e2fsck_t ctx,struct fill_dir_struct * fd,struct out_dir * outdir)397 static errcode_t copy_dir_entries(e2fsck_t ctx,
398 				  struct fill_dir_struct *fd,
399 				  struct out_dir *outdir)
400 {
401 	ext2_filsys 		fs = ctx->fs;
402 	errcode_t		retval;
403 	char			*block_start;
404 	struct hash_entry 	*ent;
405 	struct ext2_dir_entry	*dirent;
406 	unsigned int		rec_len, prev_rec_len, left, slack, offset;
407 	int			i;
408 	ext2_dirhash_t		prev_hash;
409 
410 	if (ctx->htree_slack_percentage == 255) {
411 		profile_get_uint(ctx->profile, "options",
412 				 "indexed_dir_slack_percentage",
413 				 0, 20,
414 				 &ctx->htree_slack_percentage);
415 		if (ctx->htree_slack_percentage > 100)
416 			ctx->htree_slack_percentage = 20;
417 	}
418 
419 	outdir->max = 0;
420 	retval = alloc_size_dir(fs, outdir,
421 				(fd->dir_size / fs->blocksize) + 2);
422 	if (retval)
423 		return retval;
424 	outdir->num = fd->compress ? 0 : 1;
425 	offset = 0;
426 	outdir->hashes[0] = 0;
427 	prev_hash = 1;
428 	if ((retval = get_next_block(fs, outdir, &block_start)))
429 		return retval;
430 	dirent = (struct ext2_dir_entry *) block_start;
431 	prev_rec_len = 0;
432 	rec_len = 0;
433 	left = fs->blocksize;
434 	slack = fd->compress ? 12 :
435 		(fs->blocksize * ctx->htree_slack_percentage)/100;
436 	if (slack < 12)
437 		slack = 12;
438 	for (i = 0; i < fd->num_array; i++) {
439 		ent = fd->harray + i;
440 		if (ent->dir->inode == 0)
441 			continue;
442 		rec_len = EXT2_DIR_REC_LEN(ent->dir->name_len & 0xFF);
443 		if (rec_len > left) {
444 			if (left) {
445 				left += prev_rec_len;
446 				retval = ext2fs_set_rec_len(fs, left, dirent);
447 				if (retval)
448 					return retval;
449 			}
450 			if ((retval = get_next_block(fs, outdir,
451 						      &block_start)))
452 				return retval;
453 			offset = 0;
454 		}
455 		left = fs->blocksize - offset;
456 		dirent = (struct ext2_dir_entry *) (block_start + offset);
457 		if (offset == 0) {
458 			if (ent->hash == prev_hash)
459 				outdir->hashes[outdir->num-1] = ent->hash | 1;
460 			else
461 				outdir->hashes[outdir->num-1] = ent->hash;
462 		}
463 		dirent->inode = ent->dir->inode;
464 		dirent->name_len = ent->dir->name_len;
465 		retval = ext2fs_set_rec_len(fs, rec_len, dirent);
466 		if (retval)
467 			return retval;
468 		prev_rec_len = rec_len;
469 		memcpy(dirent->name, ent->dir->name, dirent->name_len & 0xFF);
470 		offset += rec_len;
471 		left -= rec_len;
472 		if (left < slack) {
473 			prev_rec_len += left;
474 			retval = ext2fs_set_rec_len(fs, prev_rec_len, dirent);
475 			if (retval)
476 				return retval;
477 			offset += left;
478 			left = 0;
479 		}
480 		prev_hash = ent->hash;
481 	}
482 	if (left)
483 		retval = ext2fs_set_rec_len(fs, rec_len + left, dirent);
484 
485 	return retval;
486 }
487 
488 
set_root_node(ext2_filsys fs,char * buf,ext2_ino_t ino,ext2_ino_t parent)489 static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
490 				    ext2_ino_t ino, ext2_ino_t parent)
491 {
492 	struct ext2_dir_entry 		*dir;
493 	struct ext2_dx_root_info  	*root;
494 	struct ext2_dx_countlimit	*limits;
495 	int				filetype = 0;
496 
497 	if (fs->super->s_feature_incompat & EXT2_FEATURE_INCOMPAT_FILETYPE)
498 		filetype = EXT2_FT_DIR << 8;
499 
500 	memset(buf, 0, fs->blocksize);
501 	dir = (struct ext2_dir_entry *) buf;
502 	dir->inode = ino;
503 	dir->name[0] = '.';
504 	dir->name_len = 1 | filetype;
505 	dir->rec_len = 12;
506 	dir = (struct ext2_dir_entry *) (buf + 12);
507 	dir->inode = parent;
508 	dir->name[0] = '.';
509 	dir->name[1] = '.';
510 	dir->name_len = 2 | filetype;
511 	dir->rec_len = fs->blocksize - 12;
512 
513 	root = (struct ext2_dx_root_info *) (buf+24);
514 	root->reserved_zero = 0;
515 	root->hash_version = fs->super->s_def_hash_version;
516 	root->info_length = 8;
517 	root->indirect_levels = 0;
518 	root->unused_flags = 0;
519 
520 	limits = (struct ext2_dx_countlimit *) (buf+32);
521 	limits->limit = (fs->blocksize - 32) / sizeof(struct ext2_dx_entry);
522 	limits->count = 0;
523 
524 	return root;
525 }
526 
527 
set_int_node(ext2_filsys fs,char * buf)528 static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
529 {
530 	struct ext2_dir_entry 		*dir;
531 	struct ext2_dx_countlimit	*limits;
532 
533 	memset(buf, 0, fs->blocksize);
534 	dir = (struct ext2_dir_entry *) buf;
535 	dir->inode = 0;
536 	(void) ext2fs_set_rec_len(fs, fs->blocksize, dir);
537 
538 	limits = (struct ext2_dx_countlimit *) (buf+8);
539 	limits->limit = (fs->blocksize - 8) / sizeof(struct ext2_dx_entry);
540 	limits->count = 0;
541 
542 	return (struct ext2_dx_entry *) limits;
543 }
544 
545 /*
546  * This function takes the leaf nodes which have been written in
547  * outdir, and populates the root node and any necessary interior nodes.
548  */
calculate_tree(ext2_filsys fs,struct out_dir * outdir,ext2_ino_t ino,ext2_ino_t parent)549 static errcode_t calculate_tree(ext2_filsys fs,
550 				struct out_dir *outdir,
551 				ext2_ino_t ino,
552 				ext2_ino_t parent)
553 {
554 	struct ext2_dx_root_info  	*root_info;
555 	struct ext2_dx_entry 		*root, *dx_ent = 0;
556 	struct ext2_dx_countlimit	*root_limit, *limit;
557 	errcode_t			retval;
558 	char				* block_start;
559 	int				i, c1, c2, nblks;
560 	int				limit_offset, root_offset;
561 
562 	root_info = set_root_node(fs, outdir->buf, ino, parent);
563 	root_offset = limit_offset = ((char *) root_info - outdir->buf) +
564 		root_info->info_length;
565 	root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
566 	c1 = root_limit->limit;
567 	nblks = outdir->num;
568 
569 	/* Write out the pointer blocks */
570 	if (nblks-1 <= c1) {
571 		/* Just write out the root block, and we're done */
572 		root = (struct ext2_dx_entry *) (outdir->buf + root_offset);
573 		for (i=1; i < nblks; i++) {
574 			root->block = ext2fs_cpu_to_le32(i);
575 			if (i != 1)
576 				root->hash =
577 					ext2fs_cpu_to_le32(outdir->hashes[i]);
578 			root++;
579 			c1--;
580 		}
581 	} else {
582 		c2 = 0;
583 		limit = 0;
584 		root_info->indirect_levels = 1;
585 		for (i=1; i < nblks; i++) {
586 			if (c1 == 0)
587 				return ENOSPC;
588 			if (c2 == 0) {
589 				if (limit)
590 					limit->limit = limit->count =
591 		ext2fs_cpu_to_le16(limit->limit);
592 				root = (struct ext2_dx_entry *)
593 					(outdir->buf + root_offset);
594 				root->block = ext2fs_cpu_to_le32(outdir->num);
595 				if (i != 1)
596 					root->hash =
597 			ext2fs_cpu_to_le32(outdir->hashes[i]);
598 				if ((retval =  get_next_block(fs, outdir,
599 							      &block_start)))
600 					return retval;
601 				dx_ent = set_int_node(fs, block_start);
602 				limit = (struct ext2_dx_countlimit *) dx_ent;
603 				c2 = limit->limit;
604 				root_offset += sizeof(struct ext2_dx_entry);
605 				c1--;
606 			}
607 			dx_ent->block = ext2fs_cpu_to_le32(i);
608 			if (c2 != limit->limit)
609 				dx_ent->hash =
610 					ext2fs_cpu_to_le32(outdir->hashes[i]);
611 			dx_ent++;
612 			c2--;
613 		}
614 		limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
615 		limit->limit = ext2fs_cpu_to_le16(limit->limit);
616 	}
617 	root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
618 	root_limit->count = ext2fs_cpu_to_le16(root_limit->limit - c1);
619 	root_limit->limit = ext2fs_cpu_to_le16(root_limit->limit);
620 
621 	return 0;
622 }
623 
624 struct write_dir_struct {
625 	struct out_dir *outdir;
626 	errcode_t	err;
627 	e2fsck_t	ctx;
628 	blk64_t		cleared;
629 };
630 
631 /*
632  * Helper function which writes out a directory block.
633  */
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)634 static int write_dir_block(ext2_filsys fs,
635 			   blk64_t *block_nr,
636 			   e2_blkcnt_t blockcnt,
637 			   blk64_t ref_block EXT2FS_ATTR((unused)),
638 			   int ref_offset EXT2FS_ATTR((unused)),
639 			   void *priv_data)
640 {
641 	struct write_dir_struct	*wd = (struct write_dir_struct *) priv_data;
642 	blk64_t	blk;
643 	char	*dir;
644 
645 	if (*block_nr == 0)
646 		return 0;
647 	if (blockcnt >= wd->outdir->num) {
648 		e2fsck_read_bitmaps(wd->ctx);
649 		blk = *block_nr;
650 		/*
651 		 * In theory, we only release blocks from the end of the
652 		 * directory file, so it's fine to clobber a whole cluster at
653 		 * once.
654 		 */
655 		if (blk % EXT2FS_CLUSTER_RATIO(fs) == 0) {
656 			ext2fs_unmark_block_bitmap2(wd->ctx->block_found_map,
657 						    blk);
658 			ext2fs_block_alloc_stats2(fs, blk, -1);
659 			wd->cleared++;
660 		}
661 		*block_nr = 0;
662 		return BLOCK_CHANGED;
663 	}
664 	if (blockcnt < 0)
665 		return 0;
666 
667 	dir = wd->outdir->buf + (blockcnt * fs->blocksize);
668 	wd->err = ext2fs_write_dir_block3(fs, *block_nr, dir, 0);
669 	if (wd->err)
670 		return BLOCK_ABORT;
671 	return 0;
672 }
673 
write_directory(e2fsck_t ctx,ext2_filsys fs,struct out_dir * outdir,ext2_ino_t ino,int compress)674 static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
675 				 struct out_dir *outdir,
676 				 ext2_ino_t ino, int compress)
677 {
678 	struct write_dir_struct wd;
679 	errcode_t	retval;
680 	struct ext2_inode 	inode;
681 
682 	retval = e2fsck_expand_directory(ctx, ino, -1, outdir->num);
683 	if (retval)
684 		return retval;
685 
686 	wd.outdir = outdir;
687 	wd.err = 0;
688 	wd.ctx = ctx;
689 	wd.cleared = 0;
690 
691 	retval = ext2fs_block_iterate3(fs, ino, 0, 0,
692 				       write_dir_block, &wd);
693 	if (retval)
694 		return retval;
695 	if (wd.err)
696 		return wd.err;
697 
698 	e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
699 	if (compress)
700 		inode.i_flags &= ~EXT2_INDEX_FL;
701 	else
702 		inode.i_flags |= EXT2_INDEX_FL;
703 	inode.i_size = outdir->num * fs->blocksize;
704 	ext2fs_iblk_sub_blocks(fs, &inode, wd.cleared);
705 	e2fsck_write_inode(ctx, ino, &inode, "rehash_dir");
706 
707 	return 0;
708 }
709 
e2fsck_rehash_dir(e2fsck_t ctx,ext2_ino_t ino)710 errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino)
711 {
712 	ext2_filsys 		fs = ctx->fs;
713 	errcode_t		retval;
714 	struct ext2_inode 	inode;
715 	char			*dir_buf = 0;
716 	struct fill_dir_struct	fd;
717 	struct out_dir		outdir;
718 
719 	outdir.max = outdir.num = 0;
720 	outdir.buf = 0;
721 	outdir.hashes = 0;
722 	e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
723 
724 	retval = ENOMEM;
725 	fd.harray = 0;
726 	dir_buf = malloc(inode.i_size);
727 	if (!dir_buf)
728 		goto errout;
729 
730 	fd.max_array = inode.i_size / 32;
731 	fd.num_array = 0;
732 	fd.harray = malloc(fd.max_array * sizeof(struct hash_entry));
733 	if (!fd.harray)
734 		goto errout;
735 
736 	fd.ctx = ctx;
737 	fd.buf = dir_buf;
738 	fd.inode = &inode;
739 	fd.err = 0;
740 	fd.dir_size = 0;
741 	fd.compress = 0;
742 	if (!(fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_DIR_INDEX) ||
743 	    (inode.i_size / fs->blocksize) < 2)
744 		fd.compress = 1;
745 	fd.parent = 0;
746 
747 retry_nohash:
748 	/* Read in the entire directory into memory */
749 	retval = ext2fs_block_iterate3(fs, ino, 0, 0,
750 				       fill_dir_block, &fd);
751 	if (fd.err) {
752 		retval = fd.err;
753 		goto errout;
754 	}
755 
756 	/*
757 	 * If the entries read are less than a block, then don't index
758 	 * the directory
759 	 */
760 	if (!fd.compress && (fd.dir_size < (fs->blocksize - 24))) {
761 		fd.compress = 1;
762 		fd.dir_size = 0;
763 		fd.num_array = 0;
764 		goto retry_nohash;
765 	}
766 
767 #if 0
768 	printf("%d entries (%d bytes) found in inode %d\n",
769 	       fd.num_array, fd.dir_size, ino);
770 #endif
771 
772 	/* Sort the list */
773 resort:
774 	if (fd.compress)
775 		qsort(fd.harray+2, fd.num_array-2, sizeof(struct hash_entry),
776 		      hash_cmp);
777 	else
778 		qsort(fd.harray, fd.num_array, sizeof(struct hash_entry),
779 		      hash_cmp);
780 
781 	/*
782 	 * Look for duplicates
783 	 */
784 	if (duplicate_search_and_fix(ctx, fs, ino, &fd))
785 		goto resort;
786 
787 	if (ctx->options & E2F_OPT_NO) {
788 		retval = 0;
789 		goto errout;
790 	}
791 
792 	/* Sort non-hashed directories by inode number */
793 	if (fd.compress)
794 		qsort(fd.harray+2, fd.num_array-2,
795 		      sizeof(struct hash_entry), ino_cmp);
796 
797 	/*
798 	 * Copy the directory entries.  In a htree directory these
799 	 * will become the leaf nodes.
800 	 */
801 	retval = copy_dir_entries(ctx, &fd, &outdir);
802 	if (retval)
803 		goto errout;
804 
805 	free(dir_buf); dir_buf = 0;
806 
807 	if (!fd.compress) {
808 		/* Calculate the interior nodes */
809 		retval = calculate_tree(fs, &outdir, ino, fd.parent);
810 		if (retval)
811 			goto errout;
812 	}
813 
814 	retval = write_directory(ctx, fs, &outdir, ino, fd.compress);
815 	if (retval)
816 		goto errout;
817 
818 errout:
819 	free(dir_buf);
820 	free(fd.harray);
821 
822 	free_out_dir(&outdir);
823 	return retval;
824 }
825 
e2fsck_rehash_directories(e2fsck_t ctx)826 void e2fsck_rehash_directories(e2fsck_t ctx)
827 {
828 	struct problem_context	pctx;
829 #ifdef RESOURCE_TRACK
830 	struct resource_track	rtrack;
831 #endif
832 	struct dir_info		*dir;
833 	ext2_u32_iterate 	iter;
834 	struct dir_info_iter *	dirinfo_iter = 0;
835 	ext2_ino_t		ino;
836 	errcode_t		retval;
837 	int			cur, max, all_dirs, first = 1;
838 
839 	init_resource_track(&rtrack, ctx->fs->io);
840 	all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
841 
842 	if (!ctx->dirs_to_hash && !all_dirs)
843 		return;
844 
845 	e2fsck_get_lost_and_found(ctx, 0);
846 
847 	clear_problem_context(&pctx);
848 
849 	cur = 0;
850 	if (all_dirs) {
851 		dirinfo_iter = e2fsck_dir_info_iter_begin(ctx);
852 		max = e2fsck_get_num_dirinfo(ctx);
853 	} else {
854 		retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
855 						       &iter);
856 		if (retval) {
857 			pctx.errcode = retval;
858 			fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
859 			return;
860 		}
861 		max = ext2fs_u32_list_count(ctx->dirs_to_hash);
862 	}
863 	while (1) {
864 		if (all_dirs) {
865 			if ((dir = e2fsck_dir_info_iter(ctx,
866 							dirinfo_iter)) == 0)
867 				break;
868 			ino = dir->ino;
869 		} else {
870 			if (!ext2fs_u32_list_iterate(iter, &ino))
871 				break;
872 		}
873 		if (ino == ctx->lost_and_found)
874 			continue;
875 		pctx.dir = ino;
876 		if (first) {
877 			fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
878 			first = 0;
879 		}
880 #if 0
881 		fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
882 #endif
883 		pctx.errcode = e2fsck_rehash_dir(ctx, ino);
884 		if (pctx.errcode) {
885 			end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
886 			fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
887 		}
888 		if (ctx->progress && !ctx->progress_fd)
889 			e2fsck_simple_progress(ctx, "Rebuilding directory",
890 			       100.0 * (float) (++cur) / (float) max, ino);
891 	}
892 	end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
893 	if (all_dirs)
894 		e2fsck_dir_info_iter_end(ctx, dirinfo_iter);
895 	else
896 		ext2fs_u32_list_iterate_end(iter);
897 
898 	if (ctx->dirs_to_hash)
899 		ext2fs_u32_list_free(ctx->dirs_to_hash);
900 	ctx->dirs_to_hash = 0;
901 
902 	print_resource_track(ctx, "Pass 3A", &rtrack, ctx->fs->io);
903 }
904