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 blk_t max_array, num_array;
84 ext2_off64_t dir_size;
85 int compress;
86 ext2_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 blk_t num;
99 blk_t 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, hash_flags;
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_flags = fd->inode->i_flags & EXT4_CASEFOLD_FL;
143 hash_alg = fs->super->s_def_hash_version;
144 if ((hash_alg <= EXT2_HASH_TEA) &&
145 (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
146 hash_alg += 3;
147 /* While the directory block is "hot", index it. */
148 dir_offset = 0;
149 while (dir_offset < fs->blocksize) {
150 dirent = (struct ext2_dir_entry *) (dir + dir_offset);
151 (void) ext2fs_get_rec_len(fs, dirent, &rec_len);
152 name_len = ext2fs_dirent_name_len(dirent);
153 if (((dir_offset + rec_len) > fs->blocksize) ||
154 (rec_len < 8) ||
155 ((rec_len % 4) != 0) ||
156 (name_len + 8 > rec_len)) {
157 fd->err = EXT2_ET_DIR_CORRUPTED;
158 return BLOCK_ABORT;
159 }
160 dir_offset += rec_len;
161 if (dirent->inode == 0)
162 continue;
163 if ((name_len) == 0) {
164 fd->err = EXT2_ET_DIR_CORRUPTED;
165 return BLOCK_ABORT;
166 }
167 if (!fd->compress && (name_len == 1) &&
168 (dirent->name[0] == '.'))
169 continue;
170 if (!fd->compress && (name_len == 2) &&
171 (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
172 fd->parent = dirent->inode;
173 continue;
174 }
175 if (fd->num_array >= fd->max_array) {
176 errcode_t retval;
177
178 retval = ext2fs_resize_array(sizeof(struct hash_entry),
179 fd->max_array,
180 fd->max_array + 500,
181 &fd->harray);
182 if (retval) {
183 fd->err = retval;
184 return BLOCK_ABORT;
185 }
186 fd->max_array += 500;
187 }
188 ent = fd->harray + fd->num_array++;
189 ent->dir = dirent;
190 fd->dir_size += EXT2_DIR_REC_LEN(name_len);
191 ent->ino = dirent->inode;
192 if (fd->compress)
193 ent->hash = ent->minor_hash = 0;
194 else {
195 fd->err = ext2fs_dirhash2(hash_alg,
196 dirent->name, name_len,
197 fs->encoding, hash_flags,
198 fs->super->s_hash_seed,
199 &ent->hash, &ent->minor_hash);
200 if (fd->err)
201 return BLOCK_ABORT;
202 }
203 }
204
205 return 0;
206 }
207
208 /* Used for sorting the hash entry */
ino_cmp(const void * a,const void * b)209 static EXT2_QSORT_TYPE ino_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
214 return (he_a->ino - he_b->ino);
215 }
216
217 /* Used for sorting the hash entry */
name_cmp(const void * a,const void * b)218 static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
219 {
220 const struct hash_entry *he_a = (const struct hash_entry *) a;
221 const struct hash_entry *he_b = (const struct hash_entry *) b;
222 unsigned int he_a_len, he_b_len, min_len;
223 int ret;
224
225 he_a_len = ext2fs_dirent_name_len(he_a->dir);
226 he_b_len = ext2fs_dirent_name_len(he_b->dir);
227 min_len = he_a_len;
228 if (min_len > he_b_len)
229 min_len = he_b_len;
230
231 ret = memcmp(he_a->dir->name, he_b->dir->name, min_len);
232 if (ret == 0) {
233 if (he_a_len > he_b_len)
234 ret = 1;
235 else if (he_a_len < he_b_len)
236 ret = -1;
237 else
238 ret = he_b->dir->inode - he_a->dir->inode;
239 }
240 return ret;
241 }
242
243 /* Used for sorting the hash entry */
hash_cmp(const void * a,const void * b)244 static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
245 {
246 const struct hash_entry *he_a = (const struct hash_entry *) a;
247 const struct hash_entry *he_b = (const struct hash_entry *) b;
248 int ret;
249
250 if (he_a->hash > he_b->hash)
251 ret = 1;
252 else if (he_a->hash < he_b->hash)
253 ret = -1;
254 else {
255 if (he_a->minor_hash > he_b->minor_hash)
256 ret = 1;
257 else if (he_a->minor_hash < he_b->minor_hash)
258 ret = -1;
259 else
260 ret = name_cmp(a, b);
261 }
262 return ret;
263 }
264
alloc_size_dir(ext2_filsys fs,struct out_dir * outdir,blk_t blocks)265 static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
266 blk_t blocks)
267 {
268 errcode_t retval;
269
270 if (outdir->max) {
271 retval = ext2fs_resize_array(fs->blocksize, outdir->max, blocks,
272 &outdir->buf);
273 if (retval)
274 return retval;
275 retval = ext2fs_resize_array(sizeof(ext2_dirhash_t),
276 outdir->max, blocks,
277 &outdir->hashes);
278 if (retval)
279 return retval;
280 } else {
281 retval = ext2fs_get_array(fs->blocksize, blocks, &outdir->buf);
282 if (retval)
283 return retval;
284 retval = ext2fs_get_array(sizeof(ext2_dirhash_t), blocks,
285 &outdir->hashes);
286 if (retval)
287 return retval;
288 outdir->num = 0;
289 }
290 outdir->max = blocks;
291 return 0;
292 }
293
free_out_dir(struct out_dir * outdir)294 static void free_out_dir(struct out_dir *outdir)
295 {
296 free(outdir->buf);
297 free(outdir->hashes);
298 outdir->max = 0;
299 outdir->num =0;
300 }
301
get_next_block(ext2_filsys fs,struct out_dir * outdir,char ** ret)302 static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
303 char ** ret)
304 {
305 errcode_t retval;
306
307 if (outdir->num >= outdir->max) {
308 int increment = outdir->max / 10;
309
310 if (increment < 50)
311 increment = 50;
312 retval = alloc_size_dir(fs, outdir, outdir->max + increment);
313 if (retval)
314 return retval;
315 }
316 *ret = outdir->buf + (size_t)outdir->num++ * fs->blocksize;
317 memset(*ret, 0, fs->blocksize);
318 return 0;
319 }
320
321 /*
322 * This function is used to make a unique filename. We do this by
323 * appending ~0, and then incrementing the number. However, we cannot
324 * expand the length of the filename beyond the padding available in
325 * the directory entry.
326 */
mutate_name(char * str,unsigned int * len)327 static void mutate_name(char *str, unsigned int *len)
328 {
329 int i;
330 unsigned int l = *len;
331
332 /*
333 * First check to see if it looks the name has been mutated
334 * already
335 */
336 for (i = l-1; i > 0; i--) {
337 if (!isdigit(str[i]))
338 break;
339 }
340 if ((i == (int)l - 1) || (str[i] != '~')) {
341 if (((l-1) & 3) < 2)
342 l += 2;
343 else
344 l = (l+3) & ~3;
345 str[l-2] = '~';
346 str[l-1] = '0';
347 *len = l;
348 return;
349 }
350 for (i = l-1; i >= 0; i--) {
351 if (isdigit(str[i])) {
352 if (str[i] == '9')
353 str[i] = '0';
354 else {
355 str[i]++;
356 return;
357 }
358 continue;
359 }
360 if (i == 1) {
361 if (str[0] == 'z')
362 str[0] = 'A';
363 else if (str[0] == 'Z') {
364 str[0] = '~';
365 str[1] = '0';
366 } else
367 str[0]++;
368 } else if (i > 0) {
369 str[i] = '1';
370 str[i-1] = '~';
371 } else {
372 if (str[0] == '~')
373 str[0] = 'a';
374 else
375 str[0]++;
376 }
377 break;
378 }
379 }
380
duplicate_search_and_fix(e2fsck_t ctx,ext2_filsys fs,ext2_ino_t ino,struct fill_dir_struct * fd)381 static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
382 ext2_ino_t ino,
383 struct fill_dir_struct *fd)
384 {
385 struct problem_context pctx;
386 struct hash_entry *ent, *prev;
387 blk_t i, j;
388 int fixed = 0;
389 char new_name[256];
390 unsigned int new_len;
391 int hash_alg;
392 int hash_flags = fd->inode->i_flags & EXT4_CASEFOLD_FL;
393
394 clear_problem_context(&pctx);
395 pctx.ino = ino;
396
397 hash_alg = fs->super->s_def_hash_version;
398 if ((hash_alg <= EXT2_HASH_TEA) &&
399 (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
400 hash_alg += 3;
401
402 for (i=1; i < fd->num_array; i++) {
403 ent = fd->harray + i;
404 prev = ent - 1;
405 if (!ent->dir->inode ||
406 (ext2fs_dirent_name_len(ent->dir) !=
407 ext2fs_dirent_name_len(prev->dir)) ||
408 memcmp(ent->dir->name, prev->dir->name,
409 ext2fs_dirent_name_len(ent->dir)))
410 continue;
411 pctx.dirent = ent->dir;
412 if ((ent->dir->inode == prev->dir->inode) &&
413 fix_problem(ctx, PR_2_DUPLICATE_DIRENT, &pctx)) {
414 e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
415 ent->dir->inode = 0;
416 fixed++;
417 continue;
418 }
419 new_len = ext2fs_dirent_name_len(ent->dir);
420 if (new_len == 0) {
421 /* should never happen */
422 ext2fs_unmark_valid(fs);
423 continue;
424 }
425 memcpy(new_name, ent->dir->name, new_len);
426 mutate_name(new_name, &new_len);
427 for (j=0; j < fd->num_array; j++) {
428 if ((i==j) ||
429 (new_len !=
430 (unsigned) ext2fs_dirent_name_len(fd->harray[j].dir)) ||
431 memcmp(new_name, fd->harray[j].dir->name, new_len))
432 continue;
433 mutate_name(new_name, &new_len);
434
435 j = -1;
436 }
437 new_name[new_len] = 0;
438 pctx.str = new_name;
439 if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
440 memcpy(ent->dir->name, new_name, new_len);
441 ext2fs_dirent_set_name_len(ent->dir, new_len);
442 ext2fs_dirhash2(hash_alg, new_name, new_len,
443 fs->encoding, hash_flags,
444 fs->super->s_hash_seed,
445 &ent->hash, &ent->minor_hash);
446 fixed++;
447 }
448 }
449 return fixed;
450 }
451
452
copy_dir_entries(e2fsck_t ctx,struct fill_dir_struct * fd,struct out_dir * outdir)453 static errcode_t copy_dir_entries(e2fsck_t ctx,
454 struct fill_dir_struct *fd,
455 struct out_dir *outdir)
456 {
457 ext2_filsys fs = ctx->fs;
458 errcode_t retval;
459 char *block_start;
460 struct hash_entry *ent;
461 struct ext2_dir_entry *dirent;
462 unsigned int rec_len, prev_rec_len, left, slack, offset;
463 int i;
464 ext2_dirhash_t prev_hash;
465 int csum_size = 0;
466 struct ext2_dir_entry_tail *t;
467
468 if (ctx->htree_slack_percentage == 255) {
469 profile_get_uint(ctx->profile, "options",
470 "indexed_dir_slack_percentage",
471 0, 20,
472 &ctx->htree_slack_percentage);
473 if (ctx->htree_slack_percentage > 100)
474 ctx->htree_slack_percentage = 20;
475 }
476
477 if (ext2fs_has_feature_metadata_csum(fs->super))
478 csum_size = sizeof(struct ext2_dir_entry_tail);
479
480 outdir->max = 0;
481 retval = alloc_size_dir(fs, outdir,
482 (fd->dir_size / fs->blocksize) + 2);
483 if (retval)
484 return retval;
485 outdir->num = fd->compress ? 0 : 1;
486 offset = 0;
487 outdir->hashes[0] = 0;
488 prev_hash = 1;
489 if ((retval = get_next_block(fs, outdir, &block_start)))
490 return retval;
491 dirent = (struct ext2_dir_entry *) block_start;
492 prev_rec_len = 0;
493 rec_len = 0;
494 left = fs->blocksize - csum_size;
495 slack = fd->compress ? 12 :
496 ((fs->blocksize - csum_size) * ctx->htree_slack_percentage)/100;
497 if (slack < 12)
498 slack = 12;
499 for (i = 0; i < fd->num_array; i++) {
500 ent = fd->harray + i;
501 if (ent->dir->inode == 0)
502 continue;
503 rec_len = EXT2_DIR_REC_LEN(ext2fs_dirent_name_len(ent->dir));
504 if (rec_len > left) {
505 if (left) {
506 left += prev_rec_len;
507 retval = ext2fs_set_rec_len(fs, left, dirent);
508 if (retval)
509 return retval;
510 }
511 if (csum_size) {
512 t = EXT2_DIRENT_TAIL(block_start,
513 fs->blocksize);
514 ext2fs_initialize_dirent_tail(fs, t);
515 }
516 if ((retval = get_next_block(fs, outdir,
517 &block_start)))
518 return retval;
519 offset = 0;
520 }
521 left = (fs->blocksize - csum_size) - offset;
522 dirent = (struct ext2_dir_entry *) (block_start + offset);
523 if (offset == 0) {
524 if (ent->hash == prev_hash)
525 outdir->hashes[outdir->num-1] = ent->hash | 1;
526 else
527 outdir->hashes[outdir->num-1] = ent->hash;
528 }
529 dirent->inode = ent->dir->inode;
530 ext2fs_dirent_set_name_len(dirent,
531 ext2fs_dirent_name_len(ent->dir));
532 ext2fs_dirent_set_file_type(dirent,
533 ext2fs_dirent_file_type(ent->dir));
534 retval = ext2fs_set_rec_len(fs, rec_len, dirent);
535 if (retval)
536 return retval;
537 prev_rec_len = rec_len;
538 memcpy(dirent->name, ent->dir->name,
539 ext2fs_dirent_name_len(dirent));
540 offset += rec_len;
541 left -= rec_len;
542 if (left < slack) {
543 prev_rec_len += left;
544 retval = ext2fs_set_rec_len(fs, prev_rec_len, dirent);
545 if (retval)
546 return retval;
547 offset += left;
548 left = 0;
549 }
550 prev_hash = ent->hash;
551 }
552 if (left)
553 retval = ext2fs_set_rec_len(fs, rec_len + left, dirent);
554 if (csum_size) {
555 t = EXT2_DIRENT_TAIL(block_start, fs->blocksize);
556 ext2fs_initialize_dirent_tail(fs, t);
557 }
558
559 return retval;
560 }
561
562
set_root_node(ext2_filsys fs,char * buf,ext2_ino_t ino,ext2_ino_t parent)563 static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
564 ext2_ino_t ino, ext2_ino_t parent)
565 {
566 struct ext2_dir_entry *dir;
567 struct ext2_dx_root_info *root;
568 struct ext2_dx_countlimit *limits;
569 int filetype = 0;
570 int csum_size = 0;
571
572 if (ext2fs_has_feature_filetype(fs->super))
573 filetype = EXT2_FT_DIR;
574
575 memset(buf, 0, fs->blocksize);
576 dir = (struct ext2_dir_entry *) buf;
577 dir->inode = ino;
578 dir->name[0] = '.';
579 ext2fs_dirent_set_name_len(dir, 1);
580 ext2fs_dirent_set_file_type(dir, filetype);
581 dir->rec_len = 12;
582 dir = (struct ext2_dir_entry *) (buf + 12);
583 dir->inode = parent;
584 dir->name[0] = '.';
585 dir->name[1] = '.';
586 ext2fs_dirent_set_name_len(dir, 2);
587 ext2fs_dirent_set_file_type(dir, filetype);
588 dir->rec_len = fs->blocksize - 12;
589
590 root = (struct ext2_dx_root_info *) (buf+24);
591 root->reserved_zero = 0;
592 root->hash_version = fs->super->s_def_hash_version;
593 root->info_length = 8;
594 root->indirect_levels = 0;
595 root->unused_flags = 0;
596
597 if (ext2fs_has_feature_metadata_csum(fs->super))
598 csum_size = sizeof(struct ext2_dx_tail);
599
600 limits = (struct ext2_dx_countlimit *) (buf+32);
601 limits->limit = (fs->blocksize - (32 + csum_size)) /
602 sizeof(struct ext2_dx_entry);
603 limits->count = 0;
604
605 return root;
606 }
607
608
set_int_node(ext2_filsys fs,char * buf)609 static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
610 {
611 struct ext2_dir_entry *dir;
612 struct ext2_dx_countlimit *limits;
613 int csum_size = 0;
614
615 memset(buf, 0, fs->blocksize);
616 dir = (struct ext2_dir_entry *) buf;
617 dir->inode = 0;
618 (void) ext2fs_set_rec_len(fs, fs->blocksize, dir);
619
620 if (ext2fs_has_feature_metadata_csum(fs->super))
621 csum_size = sizeof(struct ext2_dx_tail);
622
623 limits = (struct ext2_dx_countlimit *) (buf+8);
624 limits->limit = (fs->blocksize - (8 + csum_size)) /
625 sizeof(struct ext2_dx_entry);
626 limits->count = 0;
627
628 return (struct ext2_dx_entry *) limits;
629 }
630
alloc_blocks(ext2_filsys fs,struct ext2_dx_countlimit ** limit,struct ext2_dx_entry ** prev_ent,struct ext2_dx_entry ** next_ent,int * prev_offset,int * next_offset,struct out_dir * outdir,int i,int * prev_count,int * next_count)631 static int alloc_blocks(ext2_filsys fs,
632 struct ext2_dx_countlimit **limit,
633 struct ext2_dx_entry **prev_ent,
634 struct ext2_dx_entry **next_ent,
635 int *prev_offset, int *next_offset,
636 struct out_dir *outdir, int i,
637 int *prev_count, int *next_count)
638 {
639 errcode_t retval;
640 char *block_start;
641
642 if (*limit)
643 (*limit)->limit = (*limit)->count =
644 ext2fs_cpu_to_le16((*limit)->limit);
645 *prev_ent = (struct ext2_dx_entry *) (outdir->buf + *prev_offset);
646 (*prev_ent)->block = ext2fs_cpu_to_le32(outdir->num);
647
648 if (i != 1)
649 (*prev_ent)->hash =
650 ext2fs_cpu_to_le32(outdir->hashes[i]);
651
652 retval = get_next_block(fs, outdir, &block_start);
653 if (retval)
654 return retval;
655
656 /* outdir->buf might be reallocated */
657 *prev_ent = (struct ext2_dx_entry *) (outdir->buf + *prev_offset);
658
659 *next_ent = set_int_node(fs, block_start);
660 *limit = (struct ext2_dx_countlimit *)(*next_ent);
661 if (next_offset)
662 *next_offset = ((char *) *next_ent - outdir->buf);
663
664 *next_count = (*limit)->limit;
665 (*prev_offset) += sizeof(struct ext2_dx_entry);
666 (*prev_count)--;
667
668 return 0;
669 }
670
671 /*
672 * This function takes the leaf nodes which have been written in
673 * outdir, and populates the root node and any necessary interior nodes.
674 */
calculate_tree(ext2_filsys fs,struct out_dir * outdir,ext2_ino_t ino,ext2_ino_t parent)675 static errcode_t calculate_tree(ext2_filsys fs,
676 struct out_dir *outdir,
677 ext2_ino_t ino,
678 ext2_ino_t parent)
679 {
680 struct ext2_dx_root_info *root_info;
681 struct ext2_dx_entry *root, *int_ent, *dx_ent = 0;
682 struct ext2_dx_countlimit *root_limit, *int_limit, *limit;
683 errcode_t retval;
684 int i, c1, c2, c3, nblks;
685 int limit_offset, int_offset, root_offset;
686
687 root_info = set_root_node(fs, outdir->buf, ino, parent);
688 root_offset = limit_offset = ((char *) root_info - outdir->buf) +
689 root_info->info_length;
690 root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
691 c1 = root_limit->limit;
692 nblks = outdir->num;
693
694 /* Write out the pointer blocks */
695 if (nblks - 1 <= c1) {
696 /* Just write out the root block, and we're done */
697 root = (struct ext2_dx_entry *) (outdir->buf + root_offset);
698 for (i=1; i < nblks; i++) {
699 root->block = ext2fs_cpu_to_le32(i);
700 if (i != 1)
701 root->hash =
702 ext2fs_cpu_to_le32(outdir->hashes[i]);
703 root++;
704 c1--;
705 }
706 } else if (nblks - 1 <= ext2fs_htree_intnode_maxrecs(fs, c1)) {
707 c2 = 0;
708 limit = NULL;
709 root_info->indirect_levels = 1;
710 for (i=1; i < nblks; i++) {
711 if (c2 == 0 && c1 == 0)
712 return ENOSPC;
713 if (c2 == 0) {
714 retval = alloc_blocks(fs, &limit, &root,
715 &dx_ent, &root_offset,
716 NULL, outdir, i, &c1,
717 &c2);
718 if (retval)
719 return retval;
720 }
721 dx_ent->block = ext2fs_cpu_to_le32(i);
722 if (c2 != limit->limit)
723 dx_ent->hash =
724 ext2fs_cpu_to_le32(outdir->hashes[i]);
725 dx_ent++;
726 c2--;
727 }
728 limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
729 limit->limit = ext2fs_cpu_to_le16(limit->limit);
730 } else {
731 c2 = 0;
732 c3 = 0;
733 limit = NULL;
734 int_limit = 0;
735 root_info->indirect_levels = 2;
736 for (i = 1; i < nblks; i++) {
737 if (c3 == 0 && c2 == 0 && c1 == 0)
738 return ENOSPC;
739 if (c3 == 0 && c2 == 0) {
740 retval = alloc_blocks(fs, &int_limit, &root,
741 &int_ent, &root_offset,
742 &int_offset, outdir, i,
743 &c1, &c2);
744 if (retval)
745 return retval;
746 }
747 if (c3 == 0) {
748 int delta1 = (char *)int_limit - outdir->buf;
749 int delta2 = (char *)root - outdir->buf;
750
751 retval = alloc_blocks(fs, &limit, &int_ent,
752 &dx_ent, &int_offset,
753 NULL, outdir, i, &c2,
754 &c3);
755 if (retval)
756 return retval;
757
758 /* outdir->buf might be reallocated */
759 int_limit = (struct ext2_dx_countlimit *)
760 (outdir->buf + delta1);
761 root = (struct ext2_dx_entry *)
762 (outdir->buf + delta2);
763 }
764 dx_ent->block = ext2fs_cpu_to_le32(i);
765 if (c3 != limit->limit)
766 dx_ent->hash =
767 ext2fs_cpu_to_le32(outdir->hashes[i]);
768 dx_ent++;
769 c3--;
770 }
771 int_limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
772 int_limit->limit = ext2fs_cpu_to_le16(limit->limit);
773
774 limit->count = ext2fs_cpu_to_le16(limit->limit - c3);
775 limit->limit = ext2fs_cpu_to_le16(limit->limit);
776
777 }
778 root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
779 root_limit->count = ext2fs_cpu_to_le16(root_limit->limit - c1);
780 root_limit->limit = ext2fs_cpu_to_le16(root_limit->limit);
781
782 return 0;
783 }
784
785 struct write_dir_struct {
786 struct out_dir *outdir;
787 errcode_t err;
788 ext2_ino_t ino;
789 e2fsck_t ctx;
790 ext2_ino_t dir;
791 };
792
793 /*
794 * Helper function which writes out a directory block.
795 */
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)796 static int write_dir_block(ext2_filsys fs,
797 blk64_t *block_nr,
798 e2_blkcnt_t blockcnt,
799 blk64_t ref_block EXT2FS_ATTR((unused)),
800 int ref_offset EXT2FS_ATTR((unused)),
801 void *priv_data)
802 {
803 struct write_dir_struct *wd = (struct write_dir_struct *) priv_data;
804 char *dir, *buf = 0;
805
806 #ifdef REHASH_DEBUG
807 printf("%u: write_dir_block %lld:%lld", wd->ino, blockcnt, *block_nr);
808 #endif
809 if ((*block_nr == 0) || (blockcnt < 0)) {
810 #ifdef REHASH_DEBUG
811 printf(" - skip\n");
812 #endif
813 return 0;
814 }
815 if (blockcnt < wd->outdir->num)
816 dir = wd->outdir->buf + (blockcnt * fs->blocksize);
817 else if (wd->ctx->lost_and_found == wd->dir) {
818 /* Don't release any extra directory blocks for lost+found */
819 wd->err = ext2fs_new_dir_block(fs, 0, 0, &buf);
820 if (wd->err)
821 return BLOCK_ABORT;
822 dir = buf;
823 wd->outdir->num++;
824 } else {
825 /* Don't free blocks at the end of the directory, they
826 * will be truncated by the caller. */
827 #ifdef REHASH_DEBUG
828 printf(" - not freed\n");
829 #endif
830 return 0;
831 }
832 wd->err = ext2fs_write_dir_block4(fs, *block_nr, dir, 0, wd->dir);
833 if (buf)
834 ext2fs_free_mem(&buf);
835
836 #ifdef REHASH_DEBUG
837 printf(" - write (%d)\n", wd->err);
838 #endif
839 if (wd->err)
840 return BLOCK_ABORT;
841 return 0;
842 }
843
write_directory(e2fsck_t ctx,ext2_filsys fs,struct out_dir * outdir,ext2_ino_t ino,struct ext2_inode * inode,int compress)844 static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
845 struct out_dir *outdir,
846 ext2_ino_t ino, struct ext2_inode *inode,
847 int compress)
848 {
849 struct write_dir_struct wd;
850 errcode_t retval;
851
852 retval = e2fsck_expand_directory(ctx, ino, -1, outdir->num);
853 if (retval)
854 return retval;
855
856 wd.outdir = outdir;
857 wd.err = 0;
858 wd.ino = ino;
859 wd.ctx = ctx;
860 wd.dir = ino;
861
862 retval = ext2fs_block_iterate3(fs, ino, 0, NULL,
863 write_dir_block, &wd);
864 if (retval)
865 return retval;
866 if (wd.err)
867 return wd.err;
868
869 e2fsck_read_inode(ctx, ino, inode, "rehash_dir");
870 if (compress)
871 inode->i_flags &= ~EXT2_INDEX_FL;
872 else
873 inode->i_flags |= EXT2_INDEX_FL;
874 #ifdef REHASH_DEBUG
875 printf("%u: set inode size to %u blocks = %u bytes\n",
876 ino, outdir->num, outdir->num * fs->blocksize);
877 #endif
878 retval = ext2fs_inode_size_set(fs, inode, (ext2_off64_t)outdir->num *
879 fs->blocksize);
880 if (retval)
881 return retval;
882
883 /* ext2fs_punch() calls ext2fs_write_inode() which writes the size */
884 return ext2fs_punch(fs, ino, inode, NULL, outdir->num, ~0ULL);
885 }
886
e2fsck_rehash_dir(e2fsck_t ctx,ext2_ino_t ino,struct problem_context * pctx)887 errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino,
888 struct problem_context *pctx)
889 {
890 ext2_filsys fs = ctx->fs;
891 errcode_t retval;
892 struct ext2_inode inode;
893 char *dir_buf = 0;
894 struct fill_dir_struct fd = { NULL, NULL, 0, 0, 0, NULL,
895 0, 0, 0, 0, 0, 0 };
896 struct out_dir outdir = { 0, 0, 0, 0 };
897
898 e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
899
900 if (ext2fs_has_feature_inline_data(fs->super) &&
901 (inode.i_flags & EXT4_INLINE_DATA_FL))
902 return 0;
903
904 retval = ext2fs_get_mem(inode.i_size, &dir_buf);
905 if (retval)
906 goto errout;
907
908 fd.max_array = inode.i_size / 32;
909 retval = ext2fs_get_array(sizeof(struct hash_entry),
910 fd.max_array, &fd.harray);
911 if (retval)
912 goto errout;
913
914 fd.ino = ino;
915 fd.ctx = ctx;
916 fd.buf = dir_buf;
917 fd.inode = &inode;
918 fd.dir = ino;
919 if (!ext2fs_has_feature_dir_index(fs->super) ||
920 (inode.i_size / fs->blocksize) < 2)
921 fd.compress = 1;
922 fd.parent = 0;
923
924 retry_nohash:
925 /* Read in the entire directory into memory */
926 retval = ext2fs_block_iterate3(fs, ino, 0, 0,
927 fill_dir_block, &fd);
928 if (fd.err) {
929 retval = fd.err;
930 goto errout;
931 }
932
933 /*
934 * If the entries read are less than a block, then don't index
935 * the directory
936 */
937 if (!fd.compress && (fd.dir_size < (fs->blocksize - 24))) {
938 fd.compress = 1;
939 fd.dir_size = 0;
940 fd.num_array = 0;
941 goto retry_nohash;
942 }
943
944 #if 0
945 printf("%d entries (%d bytes) found in inode %d\n",
946 fd.num_array, fd.dir_size, ino);
947 #endif
948
949 /* Sort the list */
950 resort:
951 if (fd.compress && fd.num_array > 1)
952 qsort(fd.harray+2, fd.num_array-2, sizeof(struct hash_entry),
953 hash_cmp);
954 else
955 qsort(fd.harray, fd.num_array, sizeof(struct hash_entry),
956 hash_cmp);
957
958 /*
959 * Look for duplicates
960 */
961 if (duplicate_search_and_fix(ctx, fs, ino, &fd))
962 goto resort;
963
964 if (ctx->options & E2F_OPT_NO) {
965 retval = 0;
966 goto errout;
967 }
968
969 /* Sort non-hashed directories by inode number */
970 if (fd.compress && fd.num_array > 1)
971 qsort(fd.harray+2, fd.num_array-2,
972 sizeof(struct hash_entry), ino_cmp);
973
974 /*
975 * Copy the directory entries. In a htree directory these
976 * will become the leaf nodes.
977 */
978 retval = copy_dir_entries(ctx, &fd, &outdir);
979 if (retval)
980 goto errout;
981
982 free(dir_buf); dir_buf = 0;
983
984 if (!fd.compress) {
985 /* Calculate the interior nodes */
986 retval = calculate_tree(fs, &outdir, ino, fd.parent);
987 if (retval)
988 goto errout;
989 }
990
991 retval = write_directory(ctx, fs, &outdir, ino, &inode, fd.compress);
992 if (retval)
993 goto errout;
994
995 if (ctx->options & E2F_OPT_CONVERT_BMAP)
996 retval = e2fsck_rebuild_extents_later(ctx, ino);
997 else
998 retval = e2fsck_check_rebuild_extents(ctx, ino, &inode, pctx);
999 errout:
1000 ext2fs_free_mem(&dir_buf);
1001 ext2fs_free_mem(&fd.harray);
1002
1003 free_out_dir(&outdir);
1004 return retval;
1005 }
1006
e2fsck_rehash_directories(e2fsck_t ctx)1007 void e2fsck_rehash_directories(e2fsck_t ctx)
1008 {
1009 struct problem_context pctx;
1010 #ifdef RESOURCE_TRACK
1011 struct resource_track rtrack;
1012 #endif
1013 struct dir_info *dir;
1014 ext2_u32_iterate iter;
1015 struct dir_info_iter * dirinfo_iter = 0;
1016 ext2_ino_t ino;
1017 errcode_t retval;
1018 int cur, max, all_dirs, first = 1;
1019
1020 init_resource_track(&rtrack, ctx->fs->io);
1021 all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
1022
1023 if (!ctx->dirs_to_hash && !all_dirs)
1024 return;
1025
1026 (void) e2fsck_get_lost_and_found(ctx, 0);
1027
1028 clear_problem_context(&pctx);
1029
1030 cur = 0;
1031 if (all_dirs) {
1032 dirinfo_iter = e2fsck_dir_info_iter_begin(ctx);
1033 max = e2fsck_get_num_dirinfo(ctx);
1034 } else {
1035 retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
1036 &iter);
1037 if (retval) {
1038 pctx.errcode = retval;
1039 fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
1040 return;
1041 }
1042 max = ext2fs_u32_list_count(ctx->dirs_to_hash);
1043 }
1044 while (1) {
1045 if (all_dirs) {
1046 if ((dir = e2fsck_dir_info_iter(ctx,
1047 dirinfo_iter)) == 0)
1048 break;
1049 ino = dir->ino;
1050 } else {
1051 if (!ext2fs_u32_list_iterate(iter, &ino))
1052 break;
1053 }
1054 if (!ext2fs_test_inode_bitmap2(ctx->inode_dir_map, ino))
1055 continue;
1056
1057 pctx.dir = ino;
1058 if (first) {
1059 fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
1060 first = 0;
1061 }
1062 #if 0
1063 fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
1064 #endif
1065 pctx.errcode = e2fsck_rehash_dir(ctx, ino, &pctx);
1066 if (pctx.errcode) {
1067 end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
1068 fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
1069 }
1070 if (ctx->progress && !ctx->progress_fd)
1071 e2fsck_simple_progress(ctx, "Rebuilding directory",
1072 100.0 * (float) (++cur) / (float) max, ino);
1073 }
1074 end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
1075 if (all_dirs)
1076 e2fsck_dir_info_iter_end(ctx, dirinfo_iter);
1077 else
1078 ext2fs_u32_list_iterate_end(iter);
1079
1080 if (ctx->dirs_to_hash)
1081 ext2fs_u32_list_free(ctx->dirs_to_hash);
1082 ctx->dirs_to_hash = 0;
1083
1084 print_resource_track(ctx, "Pass 3A", &rtrack, ctx->fs->io);
1085 }
1086