fsck: Overwrite the backup index file if it exists.
[osl.git] / fsck.c
1 /*
2 * Copyright (C) 2007-2008 Andre Noll <maan@systemlinux.org>
3 *
4 * Licensed under the GPL v2. For licencing details see COPYING.
5 */
6
7 /** \file fsck.c The program used to check an osl table. */
8
9 #include <sys/stat.h>
10 #include <sys/types.h>
11 #include <dirent.h>
12 #include <assert.h>
13 #include <pwd.h>
14
15 #include "log.h"
16 #include "osl.h"
17 #include "util.h"
18 #include "osl_core.h"
19 #include "fsck.cmdline.h"
20
21 /** version text used by various commands if -V switch was given */
22 #define VERSION_TEXT(prefix) "osl_" prefix " " VERSION " " "\n" \
23 "Copyright (C) 2008 Andre Noll\n" \
24 "This is free software with ABSOLUTELY NO WARRANTY." \
25 " See COPYING for details.\n" \
26 "Written by Andre Noll.\n" \
27 "Report bugs to <maan@systemlinux.org>.\n"
28
29 /** print out \p VERSION_TEXT and exit if version flag was given */
30 #define HANDLE_VERSION_FLAG(_prefix, _args_info_struct) \
31 if (_args_info_struct.version_given) { \
32 printf("%s", VERSION_TEXT(_prefix)); \
33 exit(EXIT_SUCCESS); \
34 }
35
36 static struct fsck_args_info conf;
37
38 #define FSCK_ERRORS \
39 FSCK_ERROR(RANGE_VIOLATION, "range violation detected, very bad"), \
40 FSCK_ERROR(NOT_A_REGULAR_FILE, "not a regular file"), \
41 FSCK_ERROR(SYNTAX, "fsck syntax error"), \
42 FSCK_ERROR(ACCESS, "permission denied"), \
43 FSCK_ERROR(CHDIR, "could not change directory"), \
44 FSCK_ERROR(OPENDIR, "could not open directory"),
45
46 #define FSCK_ERROR_BIT 29
47 #define FSCK_ERROR(num, txt) E_FSCK_ ## num
48 enum {
49 FSCK_DUMMY = (1 << FSCK_ERROR_BIT) - 1,
50 FSCK_ERRORS
51 };
52 #undef FSCK_ERROR
53 #define FSCK_ERROR(num, txt) txt
54 const char const *fsck_errors[] = {
55 FSCK_ERRORS
56 };
57
58 static const char *fsck_strerror(int num)
59 {
60 if (num & (1 << FSCK_ERROR_BIT))
61 return fsck_errors[num & ((1 << FSCK_ERROR_BIT) - 1)];
62 return osl_strerror(num);
63 }
64
65 extern int loglevel;
66
67 /* taken from git */
68 signed char hexval_table[256] = {
69 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
71 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
73 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
74 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
75 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
76 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
77 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
78 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
79 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
80 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
81 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
82 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
83 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
84 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
85 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
86 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
87 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
88 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
89 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
90 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
91 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
92 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
93 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
94 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
95 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
96 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
97 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
98 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
99 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
100 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
101 };
102
103 int asc_to_hash(const char *asc_hash, int len, HASH_TYPE *hash)
104 {
105 int i = 0;
106 const unsigned char *asc = (const unsigned char *) asc_hash;
107
108 while (*asc && i++ < len) {
109 unsigned int val = (hexval_table[asc[0]] << 4) | hexval_table[asc[1]];
110 if (val & ~0xff)
111 return -1;
112 *hash++ = val;
113 asc += 2;
114
115 }
116 return 1;
117 }
118
119 static int _write_all(int fd, const char *buf, size_t len)
120 {
121 return write_all(fd, buf, &len);
122 }
123
124 /*
125 * Wrapper for malloc().
126 *
127 * \param size The desired new size.
128 *
129 * A wrapper for malloc(3) which exits on errors.
130 *
131 * \return A pointer to the allocated memory, which is suitably aligned for any
132 * kind of variable.
133 *
134 * \sa malloc(3).
135 */
136 __must_check __malloc static void *fsck_malloc(size_t size)
137 {
138 assert(size);
139 void *p = malloc(size);
140
141 if (!p) {
142 EMERG_LOG("malloc failed (size = %zu), aborting\n",
143 size);
144 exit(EXIT_FAILURE);
145 }
146 return p;
147 }
148
149 /**
150 * Allocate memory and fill with zeros.
151 *
152 * \param size The desired new size.
153 *
154 * A wrapper for calloc(3) which exits on errors.
155 *
156 * \return A pointer to the allocated and zeroed-out memory, which is suitably
157 * aligned for any kind of variable.
158 *
159 * \sa calloc(3)
160 */
161 __must_check __malloc static void *fsck_calloc(size_t size)
162 {
163 void *ret = fsck_malloc(size);
164
165 memset(ret, 0, size);
166 return ret;
167 }
168
169 /**
170 * Save version of strdup().
171 *
172 * \param s The string to be duplicated.
173 *
174 * A wrapper for strdup(3). It calls \p exit(EXIT_FAILURE) on errors, i.e.
175 * there is no need to check the return value in the caller.
176 *
177 * \return A pointer to the duplicated string. If \p s was the NULL pointer,
178 * an pointer to an empty string is returned.
179 *
180 * \sa strdup(3)
181 */
182 __must_check __malloc static char *fsck_strdup(const char *s)
183 {
184 char *ret;
185
186 if ((ret = strdup(s? s: "")))
187 return ret;
188 EMERG_LOG("strdup failed, aborting\n");
189 exit(EXIT_FAILURE);
190 }
191
192 /**
193 * Compare two osl objects pointing to unsigned integers of 32 bit size.
194 *
195 * \param obj1 Pointer to the first integer.
196 * \param obj2 Pointer to the second integer.
197 *
198 * \return The values required for an osl compare function.
199 *
200 * \sa osl_compare_func, osl_hash_compare().
201 */
202 static int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
203 {
204 uint32_t d1 = read_u32((const char *)obj1->data);
205 uint32_t d2 = read_u32((const char *)obj2->data);
206
207 if (d1 < d2)
208 return 1;
209 if (d1 > d2)
210 return -1;
211 return 0;
212 }
213
214 /**
215 * A wrapper for fchdir().
216 *
217 * \param fd An open file descriptor.
218 *
219 * \return Standard.
220 */
221 static inline int __fchdir(int fd)
222 {
223 if (fchdir(fd) >= 0)
224 return 1;
225 return errno == EACCES? -E_FSCK_ACCESS : -E_FSCK_CHDIR;
226 }
227
228 /**
229 * Wrapper for chdir(2).
230 *
231 * \param path The specified directory.
232 *
233 * \return Standard.
234 */
235 _static_inline_ int __chdir(const char *path)
236 {
237 if (chdir(path) >= 0)
238 return 1;
239 return errno == EACCES? -E_FSCK_ACCESS : -E_FSCK_CHDIR;
240 }
241
242 /**
243 * Save the cwd and open a given directory.
244 *
245 * \param dirname Path to the directory to open.
246 * \param dir Result pointer.
247 * \param cwd File descriptor of the current working directory.
248 *
249 * \return Standard.
250 *
251 * Opening the current directory (".") and calling fchdir() to return is
252 * usually faster and more reliable than saving cwd in some buffer and calling
253 * chdir() afterwards.
254 *
255 * If \a cwd is not \p NULL "." is opened and the resulting file descriptor is
256 * stored in \a cwd. If the function returns success, and \a cwd is not \p
257 * NULL, the caller must close this file descriptor (probably after calling
258 * fchdir(*cwd)).
259 *
260 * On errors, the function undos everything, so the caller needs neither close
261 * any files, nor change back to the original working directory.
262 *
263 * \sa getcwd(3).
264 *
265 */
266 static int fsck_opendir(const char *dirname, DIR **dir, int *cwd)
267 {
268 int ret;
269
270 if (cwd) {
271 ret = osl_open(".", O_RDONLY, 0);
272 if (ret < 0)
273 return ret;
274 *cwd = ret;
275 }
276 ret = __chdir(dirname);
277 if (ret < 0)
278 goto close_cwd;
279 *dir = opendir(".");
280 if (*dir)
281 return 1;
282 ret = errno == EACCES? -E_FSCK_ACCESS : -E_FSCK_OPENDIR;
283 /* Ignore return value of fchdir() and close(). We're busted anyway. */
284 if (cwd)
285 fchdir(*cwd);
286 close_cwd:
287 if (cwd)
288 close(*cwd);
289 return ret;
290 }
291
292
293
294 /**
295 * Traverse the given directory recursively.
296 *
297 * \param dirname The directory to traverse.
298 * \param func The function to call for each entry.
299 * \param private_data Pointer to an arbitrary data structure.
300 *
301 * For each regular file under \a dirname, the supplied function \a func is
302 * called. The full path of the regular file and the \a private_data pointer
303 * are passed to \a func. Directories for which the calling process has no
304 * permissions to change to are silently ignored.
305 *
306 * \return Standard.
307 */
308 static int for_each_file_in_dir(const char *dirname,
309 int (*func)(const char *, void *), void *private_data)
310 {
311 DIR *dir;
312 struct dirent *entry;
313 int cwd_fd, ret2, ret = fsck_opendir(dirname, &dir, &cwd_fd);
314
315 if (ret < 0)
316 return ret == -E_FSCK_ACCESS? 1 : ret;
317 /* scan cwd recursively */
318 while ((entry = readdir(dir))) {
319 mode_t m;
320 char *tmp;
321 struct stat s;
322
323 if (!strcmp(entry->d_name, "."))
324 continue;
325 if (!strcmp(entry->d_name, ".."))
326 continue;
327 if (lstat(entry->d_name, &s) == -1)
328 continue;
329 m = s.st_mode;
330 if (!S_ISREG(m) && !S_ISDIR(m))
331 continue;
332 tmp = make_message("%s/%s", dirname, entry->d_name);
333 if (!S_ISDIR(m)) {
334 ret = func(tmp, private_data);
335 free(tmp);
336 if (ret < 0)
337 goto out;
338 continue;
339 }
340 /* directory */
341 ret = for_each_file_in_dir(tmp, func, private_data);
342 free(tmp);
343 if (ret < 0)
344 goto out;
345 }
346 ret = 1;
347 out:
348 closedir(dir);
349 ret2 = __fchdir(cwd_fd);
350 if (ret2 < 0 && ret >= 0)
351 ret = ret2;
352 close(cwd_fd);
353 return ret;
354 }
355
356 /*
357 * check for object boundary violations
358 *
359 * test whether the range pointed to by the index entry for a given cell is
360 * contained in mapped data file. This should always be the case. Otherwise
361 * we are in real trouble.
362 */
363 static int check_range(struct osl_table *t, uint32_t row_num, uint32_t col_num)
364 {
365 char *index_entry;
366 struct osl_object obj;
367 struct osl_column *col;
368 int ret;
369 char *map_start, *obj_start;
370
371 ret = get_cell_index(t, row_num, col_num, &index_entry);
372 if (ret < 0)
373 return ret;
374 ret = get_mapped_object(t, col_num, row_num, &obj);
375 if (ret < 0)
376 return ret;
377 col = t->columns + col_num;
378 obj_start = obj.data;
379 map_start = col->data_map.data;
380 // INFO_LOG("obj: %p..%p\n", obj_start, obj_start + obj.size);
381 // INFO_LOG("map: %p..%p\n", map_start, map_start + col->data_map.size);
382 if (obj_start < map_start || obj_start + obj.size > map_start + col->data_map.size) {
383 CRIT_LOG("range violation in row %u, col %u\n", row_num,
384 col_num);
385 return -E_FSCK_RANGE_VIOLATION;
386 }
387 DEBUG_LOG("col %u: ok\n", col_num);
388 return 1;
389 }
390
391 /*
392 * check all cells of the given table for boundary violations
393 */
394 static int check_index_ranges(struct osl_table *t)
395 {
396 int i, j, ret;
397
398 INFO_LOG("checking for range violations in index\n");
399 //DEBUG_LOG("%d rows. %d columns\n", t->num_rows, t->desc->num_columns);
400 t->num_invalid_rows = 0;
401 for (i = 0; i < t->num_rows; i++) {
402 if (row_is_invalid(t, i)) {
403 t->num_invalid_rows++;
404 continue;
405 }
406 for (j = 0; j < t->desc->num_columns; j++) { /* FXIME */
407 const struct osl_column_description *cd =
408 get_column_description(t->desc, j);
409 if (cd->storage_type != OSL_MAPPED_STORAGE)
410 continue;
411 ret = check_range(t, i, j);
412 if (ret < 0) {
413 if (ret != -E_FSCK_RANGE_VIOLATION)
414 goto err;
415 ret = mark_row_invalid(t, i);
416 if (ret < 0)
417 goto err;
418 t->num_invalid_rows++;
419 break;
420 }
421 }
422
423 }
424 if (t->num_invalid_rows)
425 NOTICE_LOG("ranges OK. %d invalid row(s) detected\n",
426 t->num_invalid_rows);
427 else
428 INFO_LOG("no invalid rows, no range violations, good\n");
429 return 1;
430 err:
431 return ret;
432 }
433
434 static int move_index_entry(struct osl_table *t, uint32_t dest, uint32_t src)
435 {
436 char *dest_ie, *src_ie;
437 int ret = get_row_index(t, dest, &dest_ie);
438
439 if (ret < 0)
440 return ret;
441 ret = get_row_index(t, src, &src_ie);
442 if (ret < 0)
443 return ret;
444 INFO_LOG("moving entry #%u to position %u\n", src, dest);
445 memcpy(dest_ie, src_ie, t->row_index_size);
446 return 1;
447 }
448
449 static int map_index(const struct osl_table_description *desc, struct osl_object *map)
450 {
451 char *filename = index_filename(desc);
452 int ret;
453
454 ret = mmap_full_file(filename, O_RDWR, &map->data, &map->size, NULL);
455 DEBUG_LOG("mapping index %s: ret: %d, size: %zu\n", filename, ret, map->size);
456 free(filename);
457 return ret;
458 }
459
460 static int prune_invalid_rows_from_index(struct osl_table *t)
461 {
462 uint32_t top = 0, bottom;
463 char *filename;
464 int ret;
465
466 if (!t->num_invalid_rows) {
467 INFO_LOG("all rows are valid, good\n");
468 return 1;
469 }
470 NOTICE_LOG("deleting %u invalid row(s) (%d bytes) from index\n",
471 t->num_invalid_rows, t->row_index_size * t->num_invalid_rows);
472 bottom = t->num_rows - 1;
473 while (top < bottom) {
474 if (!row_is_invalid(t, top)) {
475 top++;
476 continue;
477 }
478 while (bottom > top) {
479 if (row_is_invalid(t, bottom)) {
480 bottom--;
481 continue;
482 }
483 /* move bottom index entry to top */
484 move_index_entry(t, top, bottom);
485 bottom--;
486 top++;
487 break;
488 }
489 }
490 DEBUG_LOG("unmapping index\n");
491 osl_munmap(t->index_map.data, t->index_map.size);
492 filename = index_filename(t->desc);
493 ret = truncate_file(filename, t->row_index_size
494 * t->num_invalid_rows);
495 free(filename);
496 if (ret < 0)
497 return ret;
498 ret = map_index(t->desc, &t->index_map);
499 if (ret < 0)
500 return ret;
501 t->num_rows = table_num_rows(t);
502 return 1;
503 }
504
505 static int check_for_invalid_objects(struct osl_table *t, uint32_t **lost_bytes)
506 {
507 int i, j, ret;
508 const struct osl_column_description *cd;
509 uint32_t *loss = fsck_malloc(sizeof(uint32_t) * t->desc->num_columns);
510
511 INFO_LOG("looking for mapped objects not contained in index\n");
512 /* first count used bytes */
513 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
514 loss[i] = t->columns[i].data_map.size;
515 for (j = 0; j < t->num_rows; j++) {
516 struct osl_object obj;
517 ret = get_mapped_object(t, i, j, &obj);
518 if (ret < 0)
519 goto err;
520 loss[i] -= obj.size;
521 }
522 }
523 ret = 0;
524 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
525 if (loss[i]) {
526 NOTICE_LOG("column %u contains %u lost bytes\n",
527 i, loss[i]);
528 ret = 1;
529 }
530 }
531 if (!ret)
532 INFO_LOG("all mapped objects are valid, good\n");
533 *lost_bytes = loss;
534 return ret;
535 err:
536 free(loss);
537 return ret;
538 }
539
540 /* prune_invalid_rows() must be run on the table before calling this */
541 static int prune_mapped_column(struct osl_table *t, uint32_t col_num, int fd)
542 {
543 int i, ret;
544 uint32_t written = 0;
545 struct osl_column *col = t->columns + col_num;
546
547 INFO_LOG("pruning col %u\n", col_num);
548 for (i = 0; i < t->num_rows; i++) {
549 struct osl_object obj;
550 char *index_entry;
551
552 DEBUG_LOG("checking row %u/%u\n", i, t->num_rows);
553 ret = get_mapped_object(t, col_num, i, &obj);
554 if (ret < 0)
555 return ret;
556 ret = _write_all(fd, (char *)(obj.data) - 1, obj.size + 1);
557 if (ret < 0)
558 return ret;
559 written += obj.size + 1;
560 ret = get_row_index(t, i, &index_entry);
561 if (ret < 0)
562 return ret;
563 update_cell_index(index_entry, col, written, obj.size);
564 }
565 return 1;
566 }
567
568 static int prune_objects(struct osl_table *t, uint32_t *lost_bytes)
569 {
570 int i, ret;
571 const struct osl_column_description *cd;
572 char **col_filenames = fsck_calloc(t->desc->num_columns * sizeof(char *));
573 char **new_col_filenames = fsck_calloc(t->desc->num_columns * sizeof(char *));
574 char *idx_filename = index_filename(t->desc);
575 char *old_idx_filename = make_message("%s.bak", idx_filename);
576 int fd;
577
578 NOTICE_LOG("removing unreferenced objects from data files\n");
579 /* first make a copy of the index */
580 DEBUG_LOG("opening %s\n", old_idx_filename);
581 ret = osl_open(old_idx_filename, O_WRONLY | O_CREAT | O_TRUNC, 0644);
582 if (ret < 0)
583 goto out_free;
584 fd = ret;
585 ret = _write_all(fd, t->index_map.data, t->index_map.size);
586 close(fd);
587 if (ret < 0)
588 goto out_free;
589 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
590 if (!lost_bytes[i])
591 continue;
592 col_filenames[i] = column_filename(t, i);
593 new_col_filenames[i] = make_message("%s.fsck", col_filenames[i]);
594 ret = osl_open(new_col_filenames[i], O_WRONLY | O_CREAT | O_EXCL, 0644);
595 if (ret < 0)
596 goto out_unlink_data;
597 fd = ret;
598 ret = prune_mapped_column(t, i, fd);
599 close(fd);
600 if (ret < 0)
601 goto out_unlink_data;
602 }
603 ret = unmap_table(t, OSL_MARK_CLEAN);
604 if (ret < 0)
605 goto out_unlink_data;
606 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
607 if (!lost_bytes[i])
608 continue;
609 ret = osl_rename(new_col_filenames[i], col_filenames[i]);
610 if (ret < 0) { /* we're kinda screwed here */
611 CRIT_LOG("rename of col %i failed: %s\n", i,
612 osl_strerror(errno));
613 goto out_free;
614 }
615 }
616 unlink(old_idx_filename);
617 ret = map_table(t, 0);
618 goto out_free;
619 out_unlink_data:
620 FOR_EACH_MAPPED_COLUMN(i, t, cd)
621 unlink(new_col_filenames[i]);
622 out_free:
623 free(old_idx_filename);
624 free(idx_filename);
625 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
626 free(col_filenames[i]);
627 free(new_col_filenames[i]);
628 }
629 free(col_filenames);
630 free(new_col_filenames);
631 return ret;
632 }
633
634 static struct osl_column_description hash_tree_table_cols[] = {
635 {
636 .storage_type = OSL_NO_STORAGE,
637 .storage_flags = OSL_RBTREE | OSL_FIXED_SIZE | OSL_UNIQUE,
638 .name = "hash",
639 .compare_function = uint32_compare,
640 .data_size = HASH_SIZE
641 },
642 };
643
644 static const struct osl_table_description hash_tree_table_desc = {
645 .dir = "/", /* irrelevant */
646 .name = "hash_tree",
647 .num_columns = 1,
648 .flags = 0,
649 .column_descriptions = hash_tree_table_cols
650 };
651
652 /**
653 * The hash_tree table contains all hashes of the disk storage name column.
654 * of each row. It is used for checking if a disk storage file has a reference
655 * in the table.
656 */
657 static struct osl_table *hash_tree_table;
658 static HASH_TYPE *hashes;
659
660 static int check_disk_storage_column(struct osl_table *t, int row_num,
661 int col_num, char *ds_name, unsigned *num_missing_objects)
662 {
663 int ret;
664 struct stat statbuf;
665 char *path = disk_storage_path(t, col_num, ds_name);
666 unsigned dsnc = t->disk_storage_name_column;
667 struct osl_object obj;
668
669 DEBUG_LOG("checking if %s is a regular file\n", path);
670 ret = stat(path, &statbuf);
671 if (ret < 0 && errno == ENOENT) {
672 struct osl_row *row;
673 (*num_missing_objects)++;
674 ERROR_LOG("row %d: object %s is missing\n", row_num, path);
675 NOTICE_LOG("trying to delete row %d\n", row_num);
676 ret = osl_get_row(t, dsnc, &obj, &row);
677 if (ret < 0) {
678 CRIT_LOG("unable to get row %d\n", row_num);
679 mark_row_invalid(t, row_num);
680 CRIT_LOG("Please re-run fsck\n");
681 goto out;
682 }
683 ret = osl_del_row(t, row);
684 if (ret < 0)
685 goto out;
686 }
687 out:
688 free(path);
689 if (ret < 0)
690 return ret;
691 ret = -E_FSCK_NOT_A_REGULAR_FILE;
692 if (!(S_IFREG & statbuf.st_mode))
693 return ret;
694 return 1;
695 }
696
697 static int check_disk_storage_presence(struct osl_table *t)
698 {
699 int ret, i, j;
700 struct osl_object obj, hash_obj = {.size = HASH_SIZE};
701 char *ds_name;
702 const struct osl_column_description *cd;
703 unsigned dsnc = t->disk_storage_name_column, missing_objects = 0;
704
705 if (!t->num_rows)
706 return 1;
707 hashes = fsck_malloc(t->num_rows * HASH_SIZE);
708 INFO_LOG("looking for missing disk storage objects\n");
709 for (i = 0; i < t->num_rows; i++) {
710 if (row_is_invalid(t, i))
711 continue;
712 ret = get_mapped_object(t, dsnc, i, &obj);
713 if (ret < 0)
714 return ret;
715 hash_object(&obj, hashes + i * HASH_SIZE);
716 hash_obj.data = hashes + i * HASH_SIZE;
717 osl_add_row(hash_tree_table, &hash_obj);
718 ds_name = disk_storage_name_of_hash(t, hashes + i * HASH_SIZE);
719 FOR_EACH_DISK_STORAGE_COLUMN(j, t, cd) {
720 ret = check_disk_storage_column(t, i, j, ds_name,
721 &missing_objects);
722 if (ret < 0)
723 goto err;
724 }
725 free(ds_name);
726 }
727 if (!missing_objects)
728 INFO_LOG("all referenced disk storage objects exist, good\n");
729 else
730 NOTICE_LOG("%d missing object(s)\n", missing_objects);
731 return missing_objects;
732 err:
733 free(ds_name);
734 return ret;
735 }
736
737 static int dummy_compare(const struct osl_object *obj1, const struct osl_object *obj2)
738 {
739 if (obj1 < obj2)
740 return -1;
741 if (obj1 > obj2)
742 return 1;
743 return 0;
744 }
745
746 static unsigned files_pruned;
747
748 int prune_disk_storage_file(const char *path, void *private_data)
749 {
750 HASH_TYPE hash[HASH_SIZE];
751 uint8_t flags = *(uint8_t *)private_data;
752 struct osl_object obj = {.data = hash, .size = HASH_SIZE};
753 struct osl_row *row;
754 int ret = -1;
755 size_t len = strlen(path);
756
757
758 DEBUG_LOG("path: %s\n", path);
759 if (flags & OSL_LARGE_TABLE) {
760 if (len < HASH_SIZE * 2 + 2)
761 goto invalid;
762 // NOTICE_LOG("p: %s\n", path + len - 2 * HASH_SIZE - 1);
763 ret = asc_to_hash(path + len - 2 * HASH_SIZE - 1, 1, hash);
764 if (ret < 0)
765 goto invalid;
766 ret = asc_to_hash(path + len - 2 * HASH_SIZE + 2, HASH_SIZE - 1,
767 hash + 1);
768 if (ret < 0)
769 goto invalid;
770 // INFO_LOG("high: %x, low: %x, hash: %x\n", high, low, hash);
771 } else {
772 if (len < 2 * HASH_SIZE + 1)
773 goto invalid;
774 ret = asc_to_hash(path + len - 2 * HASH_SIZE, 2 * HASH_SIZE, hash);
775 if (ret < 0)
776 goto invalid;
777 // INFO_LOG("hash: %x\n", hash);
778 }
779 #if 0
780 {
781 char asc[2 * HASH_SIZE + 1];
782 hash_to_asc(hash, asc);
783 NOTICE_LOG("before: %s\nafter: %s\n", path, asc);
784 }
785 #endif
786 ret = osl_get_row(hash_tree_table, 0, &obj, &row);
787 if (ret >= 0)
788 return 1;
789 NOTICE_LOG("unreferenced file in hash dir: %s\n", path);
790 goto remove;
791 invalid:
792 ERROR_LOG("could not read hash value of %s\n", path);
793 remove:
794 NOTICE_LOG("removing %s\n", path);
795 unlink(path);
796 files_pruned++;
797 return 1;
798 }
799
800 static int prune_disk_storage_files(struct osl_table *t)
801 {
802 int i, ret = 1;
803 const struct osl_column_description *cd;
804
805 INFO_LOG("looking for unreferenced disk storage files\n");
806 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
807 char *dirname = column_filename(t, i);
808 uint8_t flags = t->desc->flags;
809 ret = for_each_file_in_dir(dirname, prune_disk_storage_file,
810 &flags);
811 free(dirname);
812 }
813 if (files_pruned)
814 NOTICE_LOG("%u disk storage files deleted\n",
815 files_pruned);
816 else
817 INFO_LOG("all files are are referenced, good\n");
818 return ret;
819 }
820
821 static int check_disk_storage_columns(struct osl_table *t)
822 {
823 int ret, i;
824 const struct osl_column_description *cd;
825
826 if (!t->num_disk_storage_columns) {
827 INFO_LOG("no disk storage columns in table '%s', "
828 "skipping checks\n", t->desc->name);
829 return 1;
830 }
831 FOR_EACH_COLUMN(i, t->desc, cd)
832 t->desc->column_descriptions[i].compare_function = dummy_compare;
833 ret = init_rbtrees(t);
834 if (ret < 0)
835 return ret;
836 INFO_LOG("creating rbtree for disk storage hash values\n");
837 ret = osl_open_table(&hash_tree_table_desc, &hash_tree_table);
838 if (ret < 0)
839 goto out;
840 ret = check_disk_storage_presence(t);
841 if (ret < 0)
842 goto out_close_hash_tree;
843 ret = prune_disk_storage_files(t);
844 out_close_hash_tree:
845 osl_close_table(hash_tree_table, 0);
846 free(hashes);
847 hashes = NULL;
848 out:
849 clear_rbtrees(t); /* TODO why are we doing that here? Seems odd */
850 return ret;
851 }
852
853 static void set_dummy_contents(struct osl_table_description *desc)
854 {
855 int i;
856 struct osl_column_description *cd;
857
858 for (i = 0; i < desc->num_columns; i++) {
859 cd = get_column_description(desc, i);
860 cd->compare_function = dummy_compare;
861 }
862 }
863
864 static int fsck_init(struct osl_table_description *desc, struct osl_table **t)
865 {
866 struct osl_object map;
867 int ret = map_index(desc, &map);
868
869 if (ret < 0)
870 goto out;
871 ret = read_table_desc(&map, desc);
872 if (ret < 0) {
873 osl_munmap(map.data, map.size);
874 goto out;
875 }
876 set_dummy_contents(desc);
877 ret = init_table_structure(desc, t);
878 if (ret < 0) {
879 osl_munmap(map.data, map.size);
880 goto out;
881 }
882 DEBUG_LOG("unmapping index\n");
883 osl_munmap(map.data, map.size);
884 if (conf.force_given)
885 ret = map_table(*t, (MAP_TBL_FL_IGNORE_DIRTY));
886 else
887 ret = map_table(*t, 0);
888 if (ret >= 0)
889 (*t)->num_rows = table_num_rows(*t);
890 out:
891 return ret;
892 }
893
894 static void fsck_cleanup(struct osl_table *t)
895 {
896 int i;
897
898 if (!t)
899 return;
900 if (t->desc->column_descriptions) {
901 struct osl_column_description *cd;
902 for (i = 0; i < t->desc->num_columns; i++) {
903 cd = get_column_description(t->desc, i);
904 free((char*)cd->name);
905 }
906 free(t->desc->column_descriptions);
907 }
908 free(t->columns);
909 free(t);
910
911 }
912
913 #define ST_CASE(st) case st: return #st
914
915 const char *get_asc_storage_type(enum osl_storage_type st)
916 {
917 switch (st) {
918 ST_CASE(OSL_MAPPED_STORAGE);
919 ST_CASE(OSL_DISK_STORAGE);
920 ST_CASE(OSL_NO_STORAGE);
921 }
922 return NULL;
923 }
924
925 #define APPEND_ASC_SF(sf, flag, str) do { if (sf & flag) { \
926 if (str) str = make_message("%s%s", str, " | " # flag); \
927 else str = fsck_strdup(#flag); }} while (0)
928
929
930 char *get_asc_storage_flags(enum osl_storage_type sf)
931 {
932 char *asc_sf = NULL;
933
934 APPEND_ASC_SF(sf, OSL_RBTREE, asc_sf);
935 APPEND_ASC_SF(sf, OSL_FIXED_SIZE, asc_sf);
936 APPEND_ASC_SF(sf, OSL_UNIQUE, asc_sf);
937 return asc_sf;
938 }
939
940 static int dump_table_desc(struct osl_table *t, int fd)
941 {
942 const struct osl_table_description *desc = t->desc;
943 int ret, i;
944 struct osl_column_description *cd;
945 char *msg = make_message("static struct osl_column_description cols[] = {\n");
946 ret = _write_all(fd, msg, strlen(msg));
947 if (ret < 0)
948 return ret;
949 free(msg);
950 FOR_EACH_COLUMN(i, desc, cd) {
951 const char *asc_st;
952 msg = make_message("\t[%d] = {\n", i);
953 ret = _write_all(fd, msg, strlen(msg));
954 if (ret < 0)
955 return ret;
956 free(msg);
957 asc_st = get_asc_storage_type(cd->storage_type);
958 msg = make_message("\t\t.storage_type = %s,\n", asc_st);
959 ret = _write_all(fd, msg, strlen(msg));
960 if (ret < 0)
961 return ret;
962 free(msg);
963 if (cd->storage_flags) {
964 char *asc_sf = get_asc_storage_flags(cd->storage_flags);
965 msg = make_message("\t\t,storage_flags = %s,\n", asc_sf);
966 free(asc_sf);
967 ret = _write_all(fd, msg, strlen(msg));
968 if (ret < 0)
969 return ret;
970 free(msg);
971 }
972 if (cd->storage_flags & OSL_FIXED_SIZE) {
973 msg = make_message("\t\t.data_size = %u,\n", cd->data_size);
974 ret = _write_all(fd, msg, strlen(msg));
975 if (ret < 0)
976 return ret;
977 free(msg);
978 }
979 msg = make_message("\t\t.name = \"%s\",\n", cd->name);
980 ret = _write_all(fd, msg, strlen(msg));
981 if (ret < 0)
982 return ret;
983 free(msg);
984 if (cd->storage_flags & OSL_RBTREE) {
985 msg = make_message("\t\t.compare_function = compare_func,\n");
986 ret = _write_all(fd, msg, strlen(msg));
987 if (ret < 0)
988 return ret;
989 free(msg);
990 }
991 msg = make_message("\t},\n");
992 ret = _write_all(fd, msg, strlen(msg));
993 if (ret < 0)
994 return ret;
995 free(msg);
996 }
997 msg = make_message("};\n");
998 ret = _write_all(fd, msg, strlen(msg));
999 if (ret < 0)
1000 return ret;
1001 free(msg);
1002 return 1;
1003 }
1004
1005 static int dump_row(struct osl_table *t, unsigned row_num, const char *row_dir)
1006 {
1007 int ret, i;
1008 const struct osl_column_description *cd;
1009 unsigned dsnc;
1010 struct osl_object obj;
1011 char *ds_name;
1012 HASH_TYPE hash[HASH_SIZE];
1013 char *filename;
1014
1015 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
1016 ret = get_mapped_object(t, i, row_num, &obj);
1017 if (ret < 0)
1018 return ret;
1019 filename = make_message("%s/col_%03u", row_dir, i);
1020 ret = write_file(filename, obj.data, obj.size);
1021 free(filename);
1022 if (ret < 0)
1023 return ret;
1024 }
1025 if (!t->num_disk_storage_columns)
1026 return 1;
1027 dsnc = t->disk_storage_name_column;
1028 ret = get_mapped_object(t, dsnc, row_num, &obj);
1029 if (ret < 0)
1030 return ret;
1031 hash_object(&obj, hash);
1032 ds_name = disk_storage_name_of_hash(t, hash);
1033 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1034 filename = disk_storage_path(t, i, ds_name);
1035 ret = mmap_full_file(filename, O_RDONLY, &obj.data, &obj.size, NULL);
1036 free(filename);
1037 if (ret < 0)
1038 goto out;
1039 filename = make_message("%s/col_%03u", row_dir, i);
1040 ret = write_file(filename, obj.data, obj.size);
1041 free(filename);
1042 if (ret < 0)
1043 goto out;
1044 }
1045 ret = 1;
1046 out:
1047 free(ds_name);
1048 return ret;
1049 }
1050
1051 static int dump_rows(char *dump_dir, struct osl_table *t)
1052 {
1053 unsigned i;
1054 char *current_dir = NULL;
1055 int ret = 0;
1056
1057 for (i = 0; i < t->num_rows; i++) {
1058 char *row_dir;
1059 if (row_is_invalid(t, i))
1060 continue;
1061 if (!(i % 1000)) {
1062 free(current_dir);
1063 current_dir = make_message("%s/rows_%u-%u", dump_dir, i, i + 999);
1064 NOTICE_LOG("dumping rows %u - %u\n", i, i + 999);
1065 ret = osl_mkdir(current_dir, 0777);
1066 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1067 goto out;
1068 }
1069 row_dir = make_message("%s/row_%03u", current_dir, i);
1070 ret = osl_mkdir(row_dir, 0777);
1071 if (ret < 0 && ret != -E_OSL_DIR_EXISTS) {
1072 free(row_dir);
1073 goto out;
1074 }
1075 ret = dump_row(t, i, row_dir);
1076 free(row_dir);
1077 if (ret < 0)
1078 goto out;
1079 }
1080 out:
1081 free(current_dir);
1082 return ret;
1083 }
1084
1085 static int dump_table(char *dump_dir, struct osl_table_description *desc)
1086 {
1087 struct osl_table *t = NULL;
1088 int fd, ret = fsck_init(desc, &t);
1089 char *desc_file;
1090 char *table_dump_dir = NULL;
1091
1092 if (ret < 0)
1093 goto out;
1094 ret = osl_mkdir(dump_dir, 0777);
1095 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1096 goto out;
1097 table_dump_dir = make_message("%s/%s", dump_dir, desc->name);
1098 ret = osl_mkdir(table_dump_dir, 0777);
1099 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1100 goto out;
1101 desc_file = make_message("%s/table_description.c", table_dump_dir);
1102 ret = osl_open(desc_file, O_WRONLY | O_CREAT | O_EXCL, 0644);
1103 free(desc_file);
1104 if (ret < 0)
1105 goto out;
1106 fd = ret;
1107 ret = dump_table_desc(t, fd);
1108 close(fd);
1109 if (ret < 0)
1110 goto out;
1111 ret = dump_rows(table_dump_dir, t);
1112 out:
1113 free(table_dump_dir);
1114 fsck_cleanup(t);
1115 return ret;
1116 }
1117
1118 static int fsck(struct osl_table_description *desc)
1119 {
1120 int ret;
1121 struct osl_table *t = NULL;
1122 uint32_t *lost_bytes = NULL;
1123
1124 ret = fsck_init(desc, &t);
1125 if (ret < 0)
1126 goto out;
1127 ret = check_index_ranges(t);
1128 if (ret < 0)
1129 goto out_unmap;
1130 ret = check_disk_storage_columns(t);
1131 if (ret < 0)
1132 goto out_unmap;
1133 ret = prune_invalid_rows_from_index(t);
1134 if (ret < 0)
1135 goto out_unmap;
1136 ret = check_for_invalid_objects(t, &lost_bytes);
1137 if (ret < 0)
1138 goto out_unmap;
1139 if (ret > 0) { /* at least one mapped data file needs pruning */
1140 ret = prune_objects(t, lost_bytes);
1141 if (ret < 0)
1142 goto out_unmap;
1143 }
1144 free(lost_bytes);
1145 out_unmap:
1146 unmap_table(t, OSL_MARK_CLEAN);
1147 out:
1148 fsck_cleanup(t);
1149 return ret;
1150 }
1151
1152 static int check_table(char *db_dir, char *table_name)
1153 {
1154 struct osl_table_description desc = {
1155 .column_descriptions = NULL,
1156 .dir = db_dir,
1157 .name = table_name
1158 };
1159 int ret;
1160
1161 INFO_LOG("checking table %s\n", table_name);
1162 if (!conf.no_fsck_given) {
1163 ret = fsck(&desc);
1164 if (ret < 0)
1165 goto out;
1166 }
1167 ret = 1;
1168 if (!conf.dump_dir_given || !*conf.dump_dir_arg)
1169 goto out;
1170 ret = dump_table(conf.dump_dir_arg, &desc);
1171 out:
1172 if (ret < 0)
1173 ERROR_LOG("failed to check table %s: %s\n", table_name,
1174 fsck_strerror(-ret));
1175 else
1176 NOTICE_LOG("successfully checked table %s\n", table_name);
1177 return ret;
1178 }
1179
1180 static int check_all_tables(char *db_dir)
1181 {
1182 DIR *dir;
1183 struct dirent *entry;
1184 int cwd_fd, ret2, ret = fsck_opendir(db_dir, &dir, &cwd_fd);
1185
1186 if (ret < 0)
1187 return ret;
1188 while ((entry = readdir(dir))) {
1189 mode_t m;
1190 struct stat s;
1191 if (!strcmp(entry->d_name, "."))
1192 continue;
1193 if (!strcmp(entry->d_name, ".."))
1194 continue;
1195 if (lstat(entry->d_name, &s) == -1)
1196 continue;
1197 m = s.st_mode;
1198 if (!S_ISDIR(m))
1199 continue;
1200 ret = check_table(db_dir, entry->d_name);
1201 if (ret < 0)
1202 break;
1203 }
1204 closedir(dir);
1205 ret2 = __fchdir(cwd_fd);
1206 if (ret2 < 0 && ret >= 0)
1207 ret = ret2;
1208 close(cwd_fd);
1209 return ret;
1210 }
1211
1212 int main(int argc, char **argv)
1213 {
1214 int i, ret;
1215 struct fsck_cmdline_parser_params params = {
1216 .override = 0,
1217 .initialize = 1,
1218 .check_required = 1,
1219 .check_ambiguity = 1,
1220 .print_errors = 1
1221 };
1222 ret = fsck_cmdline_parser_ext(argc, argv, &conf, &params);
1223 if (ret < 0) {
1224 loglevel = EMERG;
1225 ret = -E_FSCK_SYNTAX;
1226 goto out;
1227 }
1228 loglevel = conf.loglevel_arg;
1229 HANDLE_VERSION_FLAG("fsck", conf);
1230 INFO_LOG("database dir: %s\n", conf.database_dir_arg);
1231 if (!conf.inputs_num) {
1232 ret = check_all_tables(conf.database_dir_arg);
1233 goto out;
1234 }
1235 for (i = 0; i < conf.inputs_num; i++) {
1236 ret = check_table(conf.database_dir_arg, conf.inputs[i]);
1237 if (ret < 0)
1238 break;
1239 }
1240 out:
1241 if (ret < 0) {
1242 ERROR_LOG("%s\n", fsck_strerror(-ret));
1243 if (conf.loglevel_arg > 1)
1244 EMERG_LOG("re-run with \"--loglevel %d\" to increase verbosity\n",
1245 conf.loglevel_arg - 1);
1246 } else
1247 NOTICE_LOG("success\n");
1248 return ret < 0? EXIT_FAILURE : EXIT_SUCCESS;
1249 }