Return -E_OSL_NOENT if stat/open fails with errno == ENOENT.
[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 * Paraslash's version of 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 * Paraslash's version of calloc().
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 * Paraslash's 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 ret = osl_open(old_idx_filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
581 if (ret < 0)
582 goto out_free;
583 fd = ret;
584 ret = _write_all(fd, t->index_map.data, t->index_map.size);
585 close(fd);
586 if (ret < 0)
587 goto out_free;
588 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
589 if (!lost_bytes[i])
590 continue;
591 col_filenames[i] = column_filename(t, i);
592 new_col_filenames[i] = make_message("%s.fsck", col_filenames[i]);
593 ret = osl_open(new_col_filenames[i], O_WRONLY | O_CREAT | O_EXCL, 0644);
594 if (ret < 0)
595 goto out_unlink_data;
596 fd = ret;
597 ret = prune_mapped_column(t, i, fd);
598 close(fd);
599 if (ret < 0)
600 goto out_unlink_data;
601 }
602 ret = unmap_table(t, OSL_MARK_CLEAN);
603 if (ret < 0)
604 goto out_unlink_data;
605 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
606 if (!lost_bytes[i])
607 continue;
608 ret = osl_rename(new_col_filenames[i], col_filenames[i]);
609 if (ret < 0) { /* we're kinda screwed here */
610 CRIT_LOG("rename of col %i failed: %s\n", i,
611 osl_strerror(errno));
612 goto out_free;
613 }
614 }
615 unlink(old_idx_filename);
616 ret = map_table(t, 0);
617 goto out_free;
618 out_unlink_data:
619 FOR_EACH_MAPPED_COLUMN(i, t, cd)
620 unlink(new_col_filenames[i]);
621 out_free:
622 free(old_idx_filename);
623 free(idx_filename);
624 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
625 free(col_filenames[i]);
626 free(new_col_filenames[i]);
627 }
628 free(col_filenames);
629 free(new_col_filenames);
630 return ret;
631 }
632
633 static struct osl_column_description hash_tree_table_cols[] = {
634 {
635 .storage_type = OSL_NO_STORAGE,
636 .storage_flags = OSL_RBTREE | OSL_FIXED_SIZE | OSL_UNIQUE,
637 .name = "hash",
638 .compare_function = uint32_compare,
639 .data_size = HASH_SIZE
640 },
641 };
642
643 static const struct osl_table_description hash_tree_table_desc = {
644 .dir = "/", /* irrelevant */
645 .name = "hash_tree",
646 .num_columns = 1,
647 .flags = 0,
648 .column_descriptions = hash_tree_table_cols
649 };
650
651 /**
652 * The hash_tree table contains all hashes of the disk storage name column.
653 * of each row. It is used for checking if a disk storage file has a reference
654 * in the table.
655 */
656 static struct osl_table *hash_tree_table;
657 static HASH_TYPE *hashes;
658
659 static int check_disk_storage_column(struct osl_table *t, int row_num,
660 int col_num, char *ds_name, unsigned *num_missing_objects)
661 {
662 int ret;
663 struct stat statbuf;
664 char *path = disk_storage_path(t, col_num, ds_name);
665 unsigned dsnc = t->disk_storage_name_column;
666 struct osl_object obj;
667
668 DEBUG_LOG("checking if %s is a regular file\n", path);
669 ret = stat(path, &statbuf);
670 if (ret < 0 && errno == ENOENT) {
671 struct osl_row *row;
672 (*num_missing_objects)++;
673 ERROR_LOG("row %d: object %s is missing\n", row_num, path);
674 NOTICE_LOG("trying to delete row %d\n", row_num);
675 ret = osl_get_row(t, dsnc, &obj, &row);
676 if (ret < 0) {
677 CRIT_LOG("unable to get row %d\n", row_num);
678 mark_row_invalid(t, row_num);
679 CRIT_LOG("Please re-run fsck\n");
680 goto out;
681 }
682 ret = osl_del_row(t, row);
683 if (ret < 0)
684 goto out;
685 }
686 out:
687 free(path);
688 if (ret < 0)
689 return ret;
690 ret = -E_FSCK_NOT_A_REGULAR_FILE;
691 if (!(S_IFREG & statbuf.st_mode))
692 return ret;
693 return 1;
694 }
695
696 static int check_disk_storage_presence(struct osl_table *t)
697 {
698 int ret, i, j;
699 struct osl_object obj, hash_obj = {.size = HASH_SIZE};
700 char *ds_name;
701 const struct osl_column_description *cd;
702 unsigned dsnc = t->disk_storage_name_column, missing_objects = 0;
703
704 if (!t->num_rows)
705 return 1;
706 hashes = fsck_malloc(t->num_rows * HASH_SIZE);
707 INFO_LOG("looking for missing disk storage objects\n");
708 for (i = 0; i < t->num_rows; i++) {
709 if (row_is_invalid(t, i))
710 continue;
711 ret = get_mapped_object(t, dsnc, i, &obj);
712 if (ret < 0)
713 return ret;
714 hash_object(&obj, hashes + i * HASH_SIZE);
715 hash_obj.data = hashes + i * HASH_SIZE;
716 osl_add_row(hash_tree_table, &hash_obj);
717 ds_name = disk_storage_name_of_hash(t, hashes + i * HASH_SIZE);
718 FOR_EACH_DISK_STORAGE_COLUMN(j, t, cd) {
719 ret = check_disk_storage_column(t, i, j, ds_name,
720 &missing_objects);
721 if (ret < 0)
722 goto err;
723 }
724 free(ds_name);
725 }
726 if (!missing_objects)
727 INFO_LOG("all referenced disk storage objects exist, good\n");
728 else
729 NOTICE_LOG("%d missing object(s)\n", missing_objects);
730 return missing_objects;
731 err:
732 free(ds_name);
733 return ret;
734 }
735
736 static int dummy_compare(const struct osl_object *obj1, const struct osl_object *obj2)
737 {
738 if (obj1 < obj2)
739 return -1;
740 if (obj1 > obj2)
741 return 1;
742 return 0;
743 }
744
745 static unsigned files_pruned;
746
747 int prune_disk_storage_file(const char *path, void *private_data)
748 {
749 HASH_TYPE hash[HASH_SIZE];
750 uint8_t flags = *(uint8_t *)private_data;
751 struct osl_object obj = {.data = hash, .size = HASH_SIZE};
752 struct osl_row *row;
753 int ret = -1;
754 size_t len = strlen(path);
755
756
757 DEBUG_LOG("path: %s\n", path);
758 if (flags & OSL_LARGE_TABLE) {
759 if (len < HASH_SIZE * 2 + 2)
760 goto invalid;
761 // NOTICE_LOG("p: %s\n", path + len - 2 * HASH_SIZE - 1);
762 ret = asc_to_hash(path + len - 2 * HASH_SIZE - 1, 1, hash);
763 if (ret < 0)
764 goto invalid;
765 ret = asc_to_hash(path + len - 2 * HASH_SIZE + 2, HASH_SIZE - 1,
766 hash + 1);
767 if (ret < 0)
768 goto invalid;
769 // INFO_LOG("high: %x, low: %x, hash: %x\n", high, low, hash);
770 } else {
771 if (len < 2 * HASH_SIZE + 1)
772 goto invalid;
773 ret = asc_to_hash(path + len - 2 * HASH_SIZE, 2 * HASH_SIZE, hash);
774 if (ret < 0)
775 goto invalid;
776 // INFO_LOG("hash: %x\n", hash);
777 }
778 #if 0
779 {
780 char asc[2 * HASH_SIZE + 1];
781 hash_to_asc(hash, asc);
782 NOTICE_LOG("before: %s\nafter: %s\n", path, asc);
783 }
784 #endif
785 ret = osl_get_row(hash_tree_table, 0, &obj, &row);
786 if (ret >= 0)
787 return 1;
788 NOTICE_LOG("unreferenced file in hash dir: %s\n", path);
789 goto remove;
790 invalid:
791 ERROR_LOG("could not read hash value of %s\n", path);
792 remove:
793 NOTICE_LOG("removing %s\n", path);
794 unlink(path);
795 files_pruned++;
796 return 1;
797 }
798
799 static int prune_disk_storage_files(struct osl_table *t)
800 {
801 int i, ret = 1;
802 const struct osl_column_description *cd;
803
804 INFO_LOG("looking for unreferenced disk storage files\n");
805 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
806 char *dirname = column_filename(t, i);
807 uint8_t flags = t->desc->flags;
808 ret = for_each_file_in_dir(dirname, prune_disk_storage_file,
809 &flags);
810 free(dirname);
811 }
812 if (files_pruned)
813 NOTICE_LOG("%u disk storage files deleted\n",
814 files_pruned);
815 else
816 INFO_LOG("all files are are referenced, good\n");
817 return ret;
818 }
819
820 static int check_disk_storage_columns(struct osl_table *t)
821 {
822 int ret, i;
823 const struct osl_column_description *cd;
824
825 if (!t->num_disk_storage_columns) {
826 INFO_LOG("no disk storage columns in table '%s', "
827 "skipping checks\n", t->desc->name);
828 return 1;
829 }
830 FOR_EACH_COLUMN(i, t->desc, cd)
831 t->desc->column_descriptions[i].compare_function = dummy_compare;
832 ret = init_rbtrees(t);
833 if (ret < 0)
834 return ret;
835 INFO_LOG("creating rbtree for disk storage hash values\n");
836 ret = osl_open_table(&hash_tree_table_desc, &hash_tree_table);
837 if (ret < 0)
838 goto out;
839 ret = check_disk_storage_presence(t);
840 if (ret < 0)
841 goto out_close_hash_tree;
842 ret = prune_disk_storage_files(t);
843 out_close_hash_tree:
844 osl_close_table(hash_tree_table, 0);
845 free(hashes);
846 hashes = NULL;
847 out:
848 clear_rbtrees(t); /* TODO why are we doing that here? Seems odd */
849 return ret;
850 }
851
852 static void set_dummy_contents(struct osl_table_description *desc)
853 {
854 int i;
855 struct osl_column_description *cd;
856
857 for (i = 0; i < desc->num_columns; i++) {
858 cd = get_column_description(desc, i);
859 cd->compare_function = dummy_compare;
860 }
861 }
862
863 static int fsck_init(struct osl_table_description *desc, struct osl_table **t)
864 {
865 struct osl_object map;
866 int ret = map_index(desc, &map);
867
868 if (ret < 0)
869 goto out;
870 ret = read_table_desc(&map, desc);
871 if (ret < 0) {
872 osl_munmap(map.data, map.size);
873 goto out;
874 }
875 set_dummy_contents(desc);
876 ret = init_table_structure(desc, t);
877 if (ret < 0) {
878 osl_munmap(map.data, map.size);
879 goto out;
880 }
881 DEBUG_LOG("unmapping index\n");
882 osl_munmap(map.data, map.size);
883 if (conf.force_given)
884 ret = map_table(*t, (MAP_TBL_FL_IGNORE_DIRTY));
885 else
886 ret = map_table(*t, 0);
887 if (ret >= 0)
888 (*t)->num_rows = table_num_rows(*t);
889 out:
890 return ret;
891 }
892
893 static void fsck_cleanup(struct osl_table *t)
894 {
895 int i;
896
897 if (!t)
898 return;
899 if (t->desc->column_descriptions) {
900 struct osl_column_description *cd;
901 for (i = 0; i < t->desc->num_columns; i++) {
902 cd = get_column_description(t->desc, i);
903 free((char*)cd->name);
904 }
905 free(t->desc->column_descriptions);
906 }
907 free(t->columns);
908 free(t);
909
910 }
911
912 #define ST_CASE(st) case st: return #st
913
914 const char *get_asc_storage_type(enum osl_storage_type st)
915 {
916 switch (st) {
917 ST_CASE(OSL_MAPPED_STORAGE);
918 ST_CASE(OSL_DISK_STORAGE);
919 ST_CASE(OSL_NO_STORAGE);
920 }
921 return NULL;
922 }
923
924 #define APPEND_ASC_SF(sf, flag, str) do { if (sf & flag) { \
925 if (str) str = make_message("%s%s", str, " | " # flag); \
926 else str = fsck_strdup(#flag); }} while (0)
927
928
929 char *get_asc_storage_flags(enum osl_storage_type sf)
930 {
931 char *asc_sf = NULL;
932
933 APPEND_ASC_SF(sf, OSL_RBTREE, asc_sf);
934 APPEND_ASC_SF(sf, OSL_FIXED_SIZE, asc_sf);
935 APPEND_ASC_SF(sf, OSL_UNIQUE, asc_sf);
936 return asc_sf;
937 }
938
939 static int dump_table_desc(struct osl_table *t, int fd)
940 {
941 const struct osl_table_description *desc = t->desc;
942 int ret, i;
943 struct osl_column_description *cd;
944 char *msg = make_message("static struct osl_column_description cols[] = {\n");
945 ret = _write_all(fd, msg, strlen(msg));
946 if (ret < 0)
947 return ret;
948 free(msg);
949 FOR_EACH_COLUMN(i, desc, cd) {
950 const char *asc_st;
951 msg = make_message("\t[%d] = {\n", i);
952 ret = _write_all(fd, msg, strlen(msg));
953 if (ret < 0)
954 return ret;
955 free(msg);
956 asc_st = get_asc_storage_type(cd->storage_type);
957 msg = make_message("\t\t.storage_type = %s,\n", asc_st);
958 ret = _write_all(fd, msg, strlen(msg));
959 if (ret < 0)
960 return ret;
961 free(msg);
962 if (cd->storage_flags) {
963 char *asc_sf = get_asc_storage_flags(cd->storage_flags);
964 msg = make_message("\t\t,storage_flags = %s,\n", asc_sf);
965 free(asc_sf);
966 ret = _write_all(fd, msg, strlen(msg));
967 if (ret < 0)
968 return ret;
969 free(msg);
970 }
971 if (cd->storage_flags & OSL_FIXED_SIZE) {
972 msg = make_message("\t\t.data_size = %u,\n", cd->data_size);
973 ret = _write_all(fd, msg, strlen(msg));
974 if (ret < 0)
975 return ret;
976 free(msg);
977 }
978 msg = make_message("\t\t.name = \"%s\",\n", cd->name);
979 ret = _write_all(fd, msg, strlen(msg));
980 if (ret < 0)
981 return ret;
982 free(msg);
983 if (cd->storage_flags & OSL_RBTREE) {
984 msg = make_message("\t\t.compare_function = compare_func,\n");
985 ret = _write_all(fd, msg, strlen(msg));
986 if (ret < 0)
987 return ret;
988 free(msg);
989 }
990 msg = make_message("\t},\n");
991 ret = _write_all(fd, msg, strlen(msg));
992 if (ret < 0)
993 return ret;
994 free(msg);
995 }
996 msg = make_message("};\n");
997 ret = _write_all(fd, msg, strlen(msg));
998 if (ret < 0)
999 return ret;
1000 free(msg);
1001 return 1;
1002 }
1003
1004 static int dump_row(struct osl_table *t, unsigned row_num, const char *row_dir)
1005 {
1006 int ret, i;
1007 const struct osl_column_description *cd;
1008 unsigned dsnc;
1009 struct osl_object obj;
1010 char *ds_name;
1011 HASH_TYPE hash[HASH_SIZE];
1012 char *filename;
1013
1014 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
1015 ret = get_mapped_object(t, i, row_num, &obj);
1016 if (ret < 0)
1017 return ret;
1018 filename = make_message("%s/col_%03u", row_dir, i);
1019 ret = write_file(filename, obj.data, obj.size);
1020 free(filename);
1021 if (ret < 0)
1022 return ret;
1023 }
1024 if (!t->num_disk_storage_columns)
1025 return 1;
1026 dsnc = t->disk_storage_name_column;
1027 ret = get_mapped_object(t, dsnc, row_num, &obj);
1028 if (ret < 0)
1029 return ret;
1030 hash_object(&obj, hash);
1031 ds_name = disk_storage_name_of_hash(t, hash);
1032 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1033 filename = disk_storage_path(t, i, ds_name);
1034 ret = mmap_full_file(filename, O_RDONLY, &obj.data, &obj.size, NULL);
1035 free(filename);
1036 if (ret < 0)
1037 goto out;
1038 filename = make_message("%s/col_%03u", row_dir, i);
1039 ret = write_file(filename, obj.data, obj.size);
1040 free(filename);
1041 if (ret < 0)
1042 goto out;
1043 }
1044 ret = 1;
1045 out:
1046 free(ds_name);
1047 return ret;
1048 }
1049
1050 static int dump_rows(char *dump_dir, struct osl_table *t)
1051 {
1052 unsigned i;
1053 char *current_dir = NULL;
1054 int ret = 0;
1055
1056 for (i = 0; i < t->num_rows; i++) {
1057 char *row_dir;
1058 if (row_is_invalid(t, i))
1059 continue;
1060 if (!(i % 1000)) {
1061 free(current_dir);
1062 current_dir = make_message("%s/rows_%u-%u", dump_dir, i, i + 999);
1063 NOTICE_LOG("dumping rows %u - %u\n", i, i + 999);
1064 ret = osl_mkdir(current_dir, 0777);
1065 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1066 goto out;
1067 }
1068 row_dir = make_message("%s/row_%03u", current_dir, i);
1069 ret = osl_mkdir(row_dir, 0777);
1070 if (ret < 0 && ret != -E_OSL_DIR_EXISTS) {
1071 free(row_dir);
1072 goto out;
1073 }
1074 ret = dump_row(t, i, row_dir);
1075 free(row_dir);
1076 if (ret < 0)
1077 goto out;
1078 }
1079 out:
1080 free(current_dir);
1081 return ret;
1082 }
1083
1084 static int dump_table(char *dump_dir, struct osl_table_description *desc)
1085 {
1086 struct osl_table *t = NULL;
1087 int fd, ret = fsck_init(desc, &t);
1088 char *desc_file;
1089 char *table_dump_dir = NULL;
1090
1091 if (ret < 0)
1092 goto out;
1093 ret = osl_mkdir(dump_dir, 0777);
1094 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1095 goto out;
1096 table_dump_dir = make_message("%s/%s", dump_dir, desc->name);
1097 ret = osl_mkdir(table_dump_dir, 0777);
1098 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1099 goto out;
1100 desc_file = make_message("%s/table_description.c", table_dump_dir);
1101 ret = osl_open(desc_file, O_WRONLY | O_CREAT | O_EXCL, 0644);
1102 free(desc_file);
1103 if (ret < 0)
1104 goto out;
1105 fd = ret;
1106 ret = dump_table_desc(t, fd);
1107 close(fd);
1108 if (ret < 0)
1109 goto out;
1110 ret = dump_rows(table_dump_dir, t);
1111 out:
1112 free(table_dump_dir);
1113 fsck_cleanup(t);
1114 return ret;
1115 }
1116
1117 static int fsck(struct osl_table_description *desc)
1118 {
1119 int ret;
1120 struct osl_table *t = NULL;
1121 uint32_t *lost_bytes = NULL;
1122
1123 ret = fsck_init(desc, &t);
1124 if (ret < 0)
1125 goto out;
1126 ret = check_index_ranges(t);
1127 if (ret < 0)
1128 goto out_unmap;
1129 ret = check_disk_storage_columns(t);
1130 if (ret < 0)
1131 goto out_unmap;
1132 ret = prune_invalid_rows_from_index(t);
1133 if (ret < 0)
1134 goto out_unmap;
1135 ret = check_for_invalid_objects(t, &lost_bytes);
1136 if (ret < 0)
1137 goto out_unmap;
1138 if (ret > 0) { /* at least one mapped data file needs pruning */
1139 ret = prune_objects(t, lost_bytes);
1140 if (ret < 0)
1141 goto out_unmap;
1142 }
1143 free(lost_bytes);
1144 out_unmap:
1145 unmap_table(t, OSL_MARK_CLEAN);
1146 out:
1147 fsck_cleanup(t);
1148 return ret;
1149 }
1150
1151 static int check_table(char *db_dir, char *table_name)
1152 {
1153 struct osl_table_description desc = {
1154 .column_descriptions = NULL,
1155 .dir = db_dir,
1156 .name = table_name
1157 };
1158 int ret;
1159
1160 INFO_LOG("checking table %s\n", table_name);
1161 if (!conf.no_fsck_given) {
1162 ret = fsck(&desc);
1163 if (ret < 0)
1164 goto out;
1165 }
1166 ret = 1;
1167 if (!conf.dump_dir_given || !*conf.dump_dir_arg)
1168 goto out;
1169 ret = dump_table(conf.dump_dir_arg, &desc);
1170 out:
1171 if (ret < 0)
1172 ERROR_LOG("failed to check table %s\n", table_name);
1173 else
1174 NOTICE_LOG("successfully checked table %s\n", table_name);
1175 return ret;
1176 }
1177
1178 static int check_all_tables(char *db_dir)
1179 {
1180 DIR *dir;
1181 struct dirent *entry;
1182 int cwd_fd, ret2, ret = fsck_opendir(db_dir, &dir, &cwd_fd);
1183
1184 if (ret < 0)
1185 return ret;
1186 while ((entry = readdir(dir))) {
1187 mode_t m;
1188 struct stat s;
1189 if (!strcmp(entry->d_name, "."))
1190 continue;
1191 if (!strcmp(entry->d_name, ".."))
1192 continue;
1193 if (lstat(entry->d_name, &s) == -1)
1194 continue;
1195 m = s.st_mode;
1196 if (!S_ISDIR(m))
1197 continue;
1198 ret = check_table(db_dir, entry->d_name);
1199 if (ret < 0)
1200 break;
1201 }
1202 closedir(dir);
1203 ret2 = __fchdir(cwd_fd);
1204 if (ret2 < 0 && ret >= 0)
1205 ret = ret2;
1206 close(cwd_fd);
1207 return ret;
1208 }
1209
1210 int main(int argc, char **argv)
1211 {
1212 int i, ret;
1213 struct fsck_cmdline_parser_params params = {
1214 .override = 0,
1215 .initialize = 1,
1216 .check_required = 1,
1217 .check_ambiguity = 1,
1218 .print_errors = 1
1219 };
1220 ret = fsck_cmdline_parser_ext(argc, argv, &conf, &params);
1221 if (ret < 0) {
1222 loglevel = EMERG;
1223 ret = -E_FSCK_SYNTAX;
1224 goto out;
1225 }
1226 loglevel = conf.loglevel_arg;
1227 HANDLE_VERSION_FLAG("fsck", conf);
1228 INFO_LOG("database dir: %s\n", conf.database_dir_arg);
1229 if (!conf.inputs_num) {
1230 ret = check_all_tables(conf.database_dir_arg);
1231 goto out;
1232 }
1233 for (i = 0; i < conf.inputs_num; i++) {
1234 ret = check_table(conf.database_dir_arg, conf.inputs[i]);
1235 if (ret < 0)
1236 break;
1237 }
1238 out:
1239 if (ret < 0) {
1240 ERROR_LOG("%s\n", fsck_strerror(-ret));
1241 if (conf.loglevel_arg > 1)
1242 EMERG_LOG("re-run with \"--loglevel %d\" to increase verbosity\n",
1243 conf.loglevel_arg - 1);
1244 } else
1245 NOTICE_LOG("success\n");
1246 return ret < 0? EXIT_FAILURE : EXIT_SUCCESS;
1247 }