]> git.ipfire.org Git - thirdparty/e2fsprogs.git/blame - e2fsck/rehash.c
Fix build when configured with --disable-threads
[thirdparty/e2fsprogs.git] / e2fsck / rehash.c
CommitLineData
b7a00563
TT
1/*
2 * rehash.c --- rebuild hash tree directories
efc6f628 3 *
b7a00563
TT
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%
efc6f628 10 *
b7a00563
TT
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
efc6f628 39 *
b7a00563
TT
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
d1154eb4 48#include "config.h"
520ead37
TT
49#include <string.h>
50#include <ctype.h>
b7a00563
TT
51#include <errno.h>
52#include "e2fsck.h"
53#include "problem.h"
54
07307114
DW
55/* Schedule a dir to be rebuilt during pass 3A. */
56void 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. */
65int 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
19961cd0
AD
74#undef REHASH_DEBUG
75
b7a00563
TT
76struct fill_dir_struct {
77 char *buf;
78 struct ext2_inode *inode;
62ad2480 79 ext2_ino_t ino;
974d57d3 80 errcode_t err;
b7a00563
TT
81 e2fsck_t ctx;
82 struct hash_entry *harray;
83 int max_array, num_array;
68477355 84 unsigned int dir_size;
850d05e9 85 int compress;
b7a00563 86 ino_t parent;
81683c6a 87 ext2_ino_t dir;
b7a00563
TT
88};
89
90struct hash_entry {
91 ext2_dirhash_t hash;
92 ext2_dirhash_t minor_hash;
d66c3832 93 ino_t ino;
b7a00563
TT
94 struct ext2_dir_entry *dir;
95};
96
97struct out_dir {
98 int num;
99 int max;
100 char *buf;
101 ext2_dirhash_t *hashes;
102};
103
104static int fill_dir_block(ext2_filsys fs,
6dc64392 105 blk64_t *block_nr,
b7a00563 106 e2_blkcnt_t blockcnt,
6dc64392 107 blk64_t ref_block EXT2FS_ATTR((unused)),
54434927 108 int ref_offset EXT2FS_ATTR((unused)),
b7a00563
TT
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;
70f4632b 115 unsigned int offset, dir_offset, rec_len, name_len;
8a480350 116 int hash_alg;
efc6f628 117
b7a00563
TT
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 }
62ad2480 126
b7a00563 127 dir = (fd->buf+offset);
4a05268c 128 if (*block_nr == 0) {
b7a00563
TT
129 memset(dir, 0, fs->blocksize);
130 dirent = (struct ext2_dir_entry *) dir;
8a480350 131 (void) ext2fs_set_rec_len(fs, fs->blocksize, dirent);
b7a00563 132 } else {
2e9d8391 133 int flags = fs->flags;
81683c6a
DW
134 fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
135 fd->err = ext2fs_read_dir_block4(fs, *block_nr, dir, 0,
136 fd->dir);
2e9d8391
DW
137 fs->flags = (flags & EXT2_FLAG_IGNORE_CSUM_ERRORS) |
138 (fs->flags & ~EXT2_FLAG_IGNORE_CSUM_ERRORS);
b7a00563
TT
139 if (fd->err)
140 return BLOCK_ABORT;
141 }
f77704e4
TT
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;
b7a00563
TT
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);
8a480350 150 (void) ext2fs_get_rec_len(fs, dirent, &rec_len);
70f4632b 151 name_len = ext2fs_dirent_name_len(dirent);
5dd77dbe
TT
152 if (((dir_offset + rec_len) > fs->blocksize) ||
153 (rec_len < 8) ||
154 ((rec_len % 4) != 0) ||
70f4632b 155 (name_len + 8 > rec_len)) {
b7a00563
TT
156 fd->err = EXT2_ET_DIR_CORRUPTED;
157 return BLOCK_ABORT;
158 }
5dd77dbe 159 dir_offset += rec_len;
b7a00563 160 if (dirent->inode == 0)
850d05e9 161 continue;
70f4632b 162 if (!fd->compress && (name_len == 1) &&
850d05e9
TT
163 (dirent->name[0] == '.'))
164 continue;
70f4632b 165 if (!fd->compress && (name_len == 2) &&
b7a00563
TT
166 (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
167 fd->parent = dirent->inode;
850d05e9 168 continue;
b7a00563
TT
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 }
850d05e9 180 ent = fd->harray + fd->num_array++;
b7a00563 181 ent->dir = dirent;
70f4632b 182 fd->dir_size += EXT2_DIR_REC_LEN(name_len);
d66c3832 183 ent->ino = dirent->inode;
850d05e9
TT
184 if (fd->compress)
185 ent->hash = ent->minor_hash = 0;
186 else {
437651ad
TT
187 fd->err = ext2fs_dirhash(hash_alg, dirent->name,
188 name_len,
189 fs->super->s_hash_seed,
190 &ent->hash, &ent->minor_hash);
850d05e9
TT
191 if (fd->err)
192 return BLOCK_ABORT;
193 }
b7a00563 194 }
efc6f628 195
b7a00563
TT
196 return 0;
197}
198
d66c3832
TT
199/* Used for sorting the hash entry */
200static 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
b7a00563 208/* Used for sorting the hash entry */
b0700a1b 209static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
b7a00563
TT
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;
62f9bd0e 213 unsigned int he_a_len, he_b_len, min_len;
b7a00563 214 int ret;
b0700a1b 215
70f4632b
JK
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;
b0700a1b
TT
221
222 ret = strncmp(he_a->dir->name, he_b->dir->name, min_len);
223 if (ret == 0) {
70f4632b 224 if (he_a_len > he_b_len)
b7a00563 225 ret = 1;
70f4632b 226 else if (he_a_len < he_b_len)
b7a00563
TT
227 ret = -1;
228 else
12dd69f5 229 ret = he_b->dir->inode - he_a->dir->inode;
b7a00563
TT
230 }
231 return ret;
232}
233
850d05e9 234/* Used for sorting the hash entry */
b0700a1b 235static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
850d05e9
TT
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;
efc6f628 240
b0700a1b
TT
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)
850d05e9 247 ret = 1;
b0700a1b 248 else if (he_a->minor_hash < he_b->minor_hash)
850d05e9
TT
249 ret = -1;
250 else
b0700a1b 251 ret = name_cmp(a, b);
850d05e9
TT
252 }
253 return ret;
254}
255
efc6f628 256static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
b7a00563
TT
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
280static void free_out_dir(struct out_dir *outdir)
281{
45e338f5
JM
282 free(outdir->buf);
283 free(outdir->hashes);
b7a00563
TT
284 outdir->max = 0;
285 outdir->num =0;
286}
287
850d05e9 288static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
b7a00563
TT
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);
850d05e9 299 memset(*ret, 0, fs->blocksize);
b7a00563
TT
300 return 0;
301}
302
b0700a1b
TT
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 */
70f4632b 309static void mutate_name(char *str, unsigned int *len)
b0700a1b 310{
62f9bd0e 311 int i;
70f4632b 312 unsigned int l = *len;
efc6f628 313
b0700a1b
TT
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 }
62f9bd0e 322 if ((i == (int)l - 1) || (str[i] != '~')) {
b0700a1b
TT
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';
70f4632b 329 *len = l;
b0700a1b
TT
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';
efc6f628 356 else
b0700a1b
TT
357 str[0]++;
358 }
359 break;
360 }
361}
362
363static 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];
70f4632b 372 unsigned int new_len;
f77704e4 373 int hash_alg;
efc6f628 374
b0700a1b
TT
375 clear_problem_context(&pctx);
376 pctx.ino = ino;
377
f77704e4
TT
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
b0700a1b
TT
383 for (i=1; i < fd->num_array; i++) {
384 ent = fd->harray + i;
385 prev = ent - 1;
386 if (!ent->dir->inode ||
70f4632b
JK
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)))
b0700a1b
TT
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 }
70f4632b
JK
400 new_len = ext2fs_dirent_name_len(ent->dir);
401 memcpy(new_name, ent->dir->name, new_len);
b0700a1b
TT
402 mutate_name(new_name, &new_len);
403 for (j=0; j < fd->num_array; j++) {
404 if ((i==j) ||
70f4632b 405 (new_len !=
62f9bd0e 406 (unsigned) ext2fs_dirent_name_len(fd->harray[j].dir)) ||
70f4632b 407 strncmp(new_name, fd->harray[j].dir->name, new_len))
b0700a1b
TT
408 continue;
409 mutate_name(new_name, &new_len);
efc6f628 410
b0700a1b
TT
411 j = -1;
412 }
70f4632b 413 new_name[new_len] = 0;
b0700a1b
TT
414 pctx.str = new_name;
415 if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
70f4632b
JK
416 memcpy(ent->dir->name, new_name, new_len);
417 ext2fs_dirent_set_name_len(ent->dir, new_len);
437651ad
TT
418 ext2fs_dirhash(hash_alg, new_name, new_len,
419 fs->super->s_hash_seed,
420 &ent->hash, &ent->minor_hash);
b0700a1b
TT
421 fixed++;
422 }
423 }
424 return fixed;
425}
426
b7a00563 427
7dca4c88 428static errcode_t copy_dir_entries(e2fsck_t ctx,
850d05e9 429 struct fill_dir_struct *fd,
fd9ca825 430 struct out_dir *outdir)
850d05e9 431{
7dca4c88 432 ext2_filsys fs = ctx->fs;
850d05e9
TT
433 errcode_t retval;
434 char *block_start;
435 struct hash_entry *ent;
436 struct ext2_dir_entry *dirent;
68477355
TT
437 unsigned int rec_len, prev_rec_len, left, slack, offset;
438 int i;
850d05e9 439 ext2_dirhash_t prev_hash;
81683c6a
DW
440 int csum_size = 0;
441 struct ext2_dir_entry_tail *t;
7dca4c88
TT
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 }
efc6f628 451
86f3b6cf 452 if (ext2fs_has_feature_metadata_csum(fs->super))
81683c6a
DW
453 csum_size = sizeof(struct ext2_dir_entry_tail);
454
850d05e9
TT
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;
8a480350 467 prev_rec_len = 0;
cf5301d7 468 rec_len = 0;
81683c6a 469 left = fs->blocksize - csum_size;
7dca4c88 470 slack = fd->compress ? 12 :
81683c6a 471 ((fs->blocksize - csum_size) * ctx->htree_slack_percentage)/100;
7dca4c88
TT
472 if (slack < 12)
473 slack = 12;
cf5301d7 474 for (i = 0; i < fd->num_array; i++) {
850d05e9 475 ent = fd->harray + i;
b0700a1b
TT
476 if (ent->dir->inode == 0)
477 continue;
70f4632b 478 rec_len = EXT2_DIR_REC_LEN(ext2fs_dirent_name_len(ent->dir));
850d05e9 479 if (rec_len > left) {
8a480350
TT
480 if (left) {
481 left += prev_rec_len;
482 retval = ext2fs_set_rec_len(fs, left, dirent);
483 if (retval)
484 return retval;
485 }
81683c6a
DW
486 if (csum_size) {
487 t = EXT2_DIRENT_TAIL(block_start,
488 fs->blocksize);
489 ext2fs_initialize_dirent_tail(fs, t);
490 }
850d05e9
TT
491 if ((retval = get_next_block(fs, outdir,
492 &block_start)))
493 return retval;
fe5b72d1 494 offset = 0;
850d05e9 495 }
81683c6a 496 left = (fs->blocksize - csum_size) - offset;
fe5b72d1 497 dirent = (struct ext2_dir_entry *) (block_start + offset);
850d05e9
TT
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;
70f4632b
JK
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));
8a480350
TT
509 retval = ext2fs_set_rec_len(fs, rec_len, dirent);
510 if (retval)
511 return retval;
512 prev_rec_len = rec_len;
70f4632b
JK
513 memcpy(dirent->name, ent->dir->name,
514 ext2fs_dirent_name_len(dirent));
850d05e9
TT
515 offset += rec_len;
516 left -= rec_len;
7dca4c88 517 if (left < slack) {
8a480350
TT
518 prev_rec_len += left;
519 retval = ext2fs_set_rec_len(fs, prev_rec_len, dirent);
520 if (retval)
521 return retval;
850d05e9 522 offset += left;
cf3909ed 523 left = 0;
850d05e9
TT
524 }
525 prev_hash = ent->hash;
526 }
527 if (left)
8a480350 528 retval = ext2fs_set_rec_len(fs, rec_len + left, dirent);
81683c6a
DW
529 if (csum_size) {
530 t = EXT2_DIRENT_TAIL(block_start, fs->blocksize);
531 ext2fs_initialize_dirent_tail(fs, t);
532 }
850d05e9 533
8a480350 534 return retval;
850d05e9
TT
535}
536
537
538static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
b7a00563
TT
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;
850d05e9 544 int filetype = 0;
07307114 545 int csum_size = 0;
b7a00563 546
86f3b6cf 547 if (ext2fs_has_feature_filetype(fs->super))
70f4632b 548 filetype = EXT2_FT_DIR;
efc6f628 549
b7a00563
TT
550 memset(buf, 0, fs->blocksize);
551 dir = (struct ext2_dir_entry *) buf;
552 dir->inode = ino;
553 dir->name[0] = '.';
70f4632b
JK
554 ext2fs_dirent_set_name_len(dir, 1);
555 ext2fs_dirent_set_file_type(dir, filetype);
b7a00563
TT
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] = '.';
70f4632b
JK
561 ext2fs_dirent_set_name_len(dir, 2);
562 ext2fs_dirent_set_file_type(dir, filetype);
b7a00563 563 dir->rec_len = fs->blocksize - 12;
efc6f628 564
b7a00563
TT
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
86f3b6cf 572 if (ext2fs_has_feature_metadata_csum(fs->super))
07307114
DW
573 csum_size = sizeof(struct ext2_dx_tail);
574
b7a00563 575 limits = (struct ext2_dx_countlimit *) (buf+32);
07307114
DW
576 limits->limit = (fs->blocksize - (32 + csum_size)) /
577 sizeof(struct ext2_dx_entry);
b7a00563
TT
578 limits->count = 0;
579
580 return root;
581}
582
583
850d05e9 584static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
b7a00563
TT
585{
586 struct ext2_dir_entry *dir;
587 struct ext2_dx_countlimit *limits;
07307114 588 int csum_size = 0;
b7a00563
TT
589
590 memset(buf, 0, fs->blocksize);
591 dir = (struct ext2_dir_entry *) buf;
592 dir->inode = 0;
8a480350 593 (void) ext2fs_set_rec_len(fs, fs->blocksize, dir);
efc6f628 594
86f3b6cf 595 if (ext2fs_has_feature_metadata_csum(fs->super))
07307114
DW
596 csum_size = sizeof(struct ext2_dx_tail);
597
b7a00563 598 limits = (struct ext2_dx_countlimit *) (buf+8);
07307114
DW
599 limits->limit = (fs->blocksize - (8 + csum_size)) /
600 sizeof(struct ext2_dx_entry);
b7a00563
TT
601 limits->count = 0;
602
603 return (struct ext2_dx_entry *) limits;
604}
605
850d05e9
TT
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 */
610static 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;
efc6f628 622
850d05e9
TT
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)
efc6f628 651 limit->limit = limit->count =
850d05e9
TT
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}
b7a00563
TT
684
685struct write_dir_struct {
686 struct out_dir *outdir;
687 errcode_t err;
19961cd0 688 ext2_ino_t ino;
b7a00563 689 e2fsck_t ctx;
81683c6a 690 ext2_ino_t dir;
b7a00563
TT
691};
692
693/*
850d05e9 694 * Helper function which writes out a directory block.
b7a00563
TT
695 */
696static int write_dir_block(ext2_filsys fs,
6dc64392 697 blk64_t *block_nr,
b7a00563 698 e2_blkcnt_t blockcnt,
6dc64392 699 blk64_t ref_block EXT2FS_ATTR((unused)),
efc6f628 700 int ref_offset EXT2FS_ATTR((unused)),
b7a00563
TT
701 void *priv_data)
702{
703 struct write_dir_struct *wd = (struct write_dir_struct *) priv_data;
82372e32 704 char *dir, *buf = 0;
b7a00563 705
19961cd0
AD
706#ifdef REHASH_DEBUG
707 printf("%u: write_dir_block %lld:%lld", wd->ino, blockcnt, *block_nr);
708#endif
94676ef2 709 if ((*block_nr == 0) || (blockcnt < 0)) {
19961cd0
AD
710#ifdef REHASH_DEBUG
711 printf(" - skip\n");
712#endif
82372e32 713 return 0;
19961cd0 714 }
82372e32
TT
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 {
94676ef2
TT
725 /* Don't free blocks at the end of the directory, they
726 * will be truncated by the caller. */
19961cd0
AD
727#ifdef REHASH_DEBUG
728 printf(" - not freed\n");
729#endif
730 return 0;
b7a00563 731 }
81683c6a 732 wd->err = ext2fs_write_dir_block4(fs, *block_nr, dir, 0, wd->dir);
82372e32
TT
733 if (buf)
734 ext2fs_free_mem(&buf);
735
19961cd0
AD
736#ifdef REHASH_DEBUG
737 printf(" - write (%d)\n", wd->err);
738#endif
b7a00563
TT
739 if (wd->err)
740 return BLOCK_ABORT;
741 return 0;
742}
743
b7a00563 744static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
850d05e9 745 struct out_dir *outdir,
e228d700
DW
746 ext2_ino_t ino, struct ext2_inode *inode,
747 int compress)
b7a00563
TT
748{
749 struct write_dir_struct wd;
750 errcode_t retval;
b7a00563
TT
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;
19961cd0 758 wd.ino = ino;
b7a00563 759 wd.ctx = ctx;
81683c6a 760 wd.dir = ino;
b7a00563 761
19961cd0 762 retval = ext2fs_block_iterate3(fs, ino, 0, NULL,
b7a00563
TT
763 write_dir_block, &wd);
764 if (retval)
765 return retval;
766 if (wd.err)
767 return wd.err;
768
e228d700 769 e2fsck_read_inode(ctx, ino, inode, "rehash_dir");
e70ae99e 770 if (compress)
e228d700 771 inode->i_flags &= ~EXT2_INDEX_FL;
e70ae99e 772 else
e228d700 773 inode->i_flags |= EXT2_INDEX_FL;
19961cd0
AD
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
94676ef2 778 retval = ext2fs_inode_size_set(fs, inode, (ext2_off64_t)outdir->num *
19961cd0 779 fs->blocksize);
97c607b1
DW
780 if (retval)
781 return retval;
b7a00563 782
19961cd0 783 /* ext2fs_punch() calls ext2fs_write_inode() which writes the size */
94676ef2 784 return ext2fs_punch(fs, ino, inode, NULL, outdir->num, ~0ULL);
b7a00563
TT
785}
786
e228d700
DW
787errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino,
788 struct problem_context *pctx)
b7a00563
TT
789{
790 ext2_filsys fs = ctx->fs;
791 errcode_t retval;
792 struct ext2_inode inode;
850d05e9 793 char *dir_buf = 0;
478360f5
TT
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 };
efc6f628 797
b7a00563 798 e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
b7a00563 799
86f3b6cf 800 if (ext2fs_has_feature_inline_data(fs->super) &&
81ac00d0
DW
801 (inode.i_flags & EXT4_INLINE_DATA_FL))
802 return 0;
803
b7a00563 804 retval = ENOMEM;
b7a00563
TT
805 dir_buf = malloc(inode.i_size);
806 if (!dir_buf)
807 goto errout;
808
809 fd.max_array = inode.i_size / 32;
b7a00563
TT
810 fd.harray = malloc(fd.max_array * sizeof(struct hash_entry));
811 if (!fd.harray)
812 goto errout;
813
19961cd0 814 fd.ino = ino;
b7a00563
TT
815 fd.ctx = ctx;
816 fd.buf = dir_buf;
817 fd.inode = &inode;
81683c6a 818 fd.dir = ino;
86f3b6cf 819 if (!ext2fs_has_feature_dir_index(fs->super) ||
e70ae99e 820 (inode.i_size / fs->blocksize) < 2)
850d05e9 821 fd.compress = 1;
b7a00563
TT
822 fd.parent = 0;
823
f4e14505 824retry_nohash:
b7a00563 825 /* Read in the entire directory into memory */
6dc64392 826 retval = ext2fs_block_iterate3(fs, ino, 0, 0,
b7a00563
TT
827 fill_dir_block, &fd);
828 if (fd.err) {
829 retval = fd.err;
830 goto errout;
831 }
832
f4e14505
TT
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
b7a00563
TT
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
850d05e9 849 /* Sort the list */
b0700a1b 850resort:
c23b2cc4 851 if (fd.compress && fd.num_array > 1)
53fbfb2b
TT
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);
850d05e9 857
b0700a1b
TT
858 /*
859 * Look for duplicates
860 */
861 if (duplicate_search_and_fix(ctx, fs, ino, &fd))
862 goto resort;
863
1d2eef42
TT
864 if (ctx->options & E2F_OPT_NO) {
865 retval = 0;
866 goto errout;
867 }
868
b71e0183 869 /* Sort non-hashed directories by inode number */
c23b2cc4 870 if (fd.compress && fd.num_array > 1)
b71e0183
TT
871 qsort(fd.harray+2, fd.num_array-2,
872 sizeof(struct hash_entry), ino_cmp);
873
850d05e9
TT
874 /*
875 * Copy the directory entries. In a htree directory these
876 * will become the leaf nodes.
877 */
fd9ca825 878 retval = copy_dir_entries(ctx, &fd, &outdir);
b7a00563
TT
879 if (retval)
880 goto errout;
efc6f628 881
b7a00563
TT
882 free(dir_buf); dir_buf = 0;
883
850d05e9
TT
884 if (!fd.compress) {
885 /* Calculate the interior nodes */
886 retval = calculate_tree(fs, &outdir, ino, fd.parent);
887 if (retval)
888 goto errout;
b7a00563 889 }
efc6f628 890
e228d700 891 retval = write_directory(ctx, fs, &outdir, ino, &inode, fd.compress);
b7a00563
TT
892 if (retval)
893 goto errout;
894
e228d700
DW
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);
b7a00563 899errout:
45e338f5
JM
900 free(dir_buf);
901 free(fd.harray);
850d05e9 902
b7a00563
TT
903 free_out_dir(&outdir);
904 return retval;
905}
906
907void e2fsck_rehash_directories(e2fsck_t ctx)
908{
b7a00563 909 struct problem_context pctx;
850d05e9
TT
910#ifdef RESOURCE_TRACK
911 struct resource_track rtrack;
912#endif
913 struct dir_info *dir;
914 ext2_u32_iterate iter;
28db82a8 915 struct dir_info_iter * dirinfo_iter = 0;
850d05e9
TT
916 ext2_ino_t ino;
917 errcode_t retval;
e3507739 918 int cur, max, all_dirs, first = 1;
850d05e9 919
6d96b00d 920 init_resource_track(&rtrack, ctx->fs->io);
850d05e9
TT
921 all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
922
923 if (!ctx->dirs_to_hash && !all_dirs)
b7a00563
TT
924 return;
925
82372e32
TT
926 (void) e2fsck_get_lost_and_found(ctx, 0);
927
b7a00563 928 clear_problem_context(&pctx);
850d05e9 929
b0700a1b
TT
930 cur = 0;
931 if (all_dirs) {
28db82a8 932 dirinfo_iter = e2fsck_dir_info_iter_begin(ctx);
b0700a1b
TT
933 max = e2fsck_get_num_dirinfo(ctx);
934 } else {
efc6f628 935 retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
850d05e9
TT
936 &iter);
937 if (retval) {
938 pctx.errcode = retval;
939 fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
940 return;
941 }
b0700a1b 942 max = ext2fs_u32_list_count(ctx->dirs_to_hash);
b7a00563 943 }
850d05e9
TT
944 while (1) {
945 if (all_dirs) {
efc6f628 946 if ((dir = e2fsck_dir_info_iter(ctx,
28db82a8 947 dirinfo_iter)) == 0)
850d05e9
TT
948 break;
949 ino = dir->ino;
950 } else {
951 if (!ext2fs_u32_list_iterate(iter, &ino))
952 break;
953 }
07307114 954
b7a00563
TT
955 pctx.dir = ino;
956 if (first) {
957 fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
958 first = 0;
959 }
b0700a1b 960#if 0
850d05e9 961 fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
b0700a1b 962#endif
e228d700 963 pctx.errcode = e2fsck_rehash_dir(ctx, ino, &pctx);
b7a00563 964 if (pctx.errcode) {
850d05e9
TT
965 end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
966 fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
b7a00563 967 }
52734dc5
TT
968 if (ctx->progress && !ctx->progress_fd)
969 e2fsck_simple_progress(ctx, "Rebuilding directory",
1d2eef42 970 100.0 * (float) (++cur) / (float) max, ino);
b7a00563 971 }
850d05e9 972 end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
28db82a8
TT
973 if (all_dirs)
974 e2fsck_dir_info_iter_end(ctx, dirinfo_iter);
975 else
850d05e9 976 ext2fs_u32_list_iterate_end(iter);
efc6f628 977
850d05e9
TT
978 if (ctx->dirs_to_hash)
979 ext2fs_u32_list_free(ctx->dirs_to_hash);
b7a00563 980 ctx->dirs_to_hash = 0;
850d05e9 981
9facd076 982 print_resource_track(ctx, "Pass 3A", &rtrack, ctx->fs->io);
b7a00563 983}