Add oslfsck.
[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 /* FIXME: check error codes of make_message or write wrapper */
10
11
12
13 #include <sys/stat.h>
14 #include <sys/types.h>
15 #include <dirent.h>
16 #include <assert.h>
17 #include <pwd.h>
18
19 #include "log.h"
20 #include "osl.h"
21 #include "error.h"
22 #include "fd.h"
23 #include "osl_core.h"
24 #include "fsck.cmdline.h"
25
26 /** version text used by various commands if -V switch was given */
27 #define VERSION_TEXT(prefix) "osl_" prefix " " VERSION " " "\n" \
28 "Copyright (C) 2008 Andre Noll\n" \
29 "This is free software with ABSOLUTELY NO WARRANTY." \
30 " See COPYING for details.\n" \
31 "Written by Andre Noll.\n" \
32 "Report bugs to <maan@systemlinux.org>.\n"
33
34 /** print out \p VERSION_TEXT and exit if version flag was given */
35 #define HANDLE_VERSION_FLAG(_prefix, _args_info_struct) \
36 if (_args_info_struct.version_given) { \
37 printf("%s", VERSION_TEXT(_prefix)); \
38 exit(EXIT_SUCCESS); \
39 }
40
41 static struct fsck_args_info conf;
42
43 enum fsck_errors {
44 E_RANGE_VIOLATION,
45 E_INVALID_OBJECT,
46 E_NOT_A_REGULAR_FILE,
47 E_FSCK_SYNTAX,
48 };
49
50 __printf_2_3 void para_log(int ll, const char* fmt,...)
51 {
52 va_list argp;
53
54 if (ll < conf.loglevel_arg)
55 return;
56 va_start(argp, fmt);
57 vfprintf(stderr, fmt, argp);
58 va_end(argp);
59 }
60
61 /* taken from git */
62 signed char hexval_table[256] = {
63 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
69 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
70 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
71 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
73 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
74 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
75 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
76 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
77 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
78 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
79 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
80 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
81 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
82 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
83 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
84 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
85 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
86 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
87 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
88 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
89 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
90 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
91 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
92 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
93 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
94 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
95 };
96
97 int asc_to_hash(const char *asc_hash, int len, HASH_TYPE *hash)
98 {
99 int i = 0;
100 const unsigned char *asc = (const unsigned char *) asc_hash;
101
102 while (*asc && i++ < len) {
103 unsigned int val = (hexval_table[asc[0]] << 4) | hexval_table[asc[1]];
104 if (val & ~0xff)
105 return -1;
106 *hash++ = val;
107 asc += 2;
108
109 }
110 return 1;
111 }
112
113 static int _write_all(int fd, const char *buf, size_t len)
114 {
115 return write_all(fd, buf, &len);
116 }
117
118 /**
119 * Paraslash's version of malloc().
120 *
121 * \param size The desired new size.
122 *
123 * A wrapper for malloc(3) which exits on errors.
124 *
125 * \return A pointer to the allocated memory, which is suitably aligned for any
126 * kind of variable.
127 *
128 * \sa malloc(3).
129 */
130 __must_check __malloc static void *para_malloc(size_t size)
131 {
132 assert(size);
133 void *p = malloc(size);
134
135 if (!p) {
136 EMERG_LOG("malloc failed (size = %zu), aborting\n",
137 size);
138 exit(EXIT_FAILURE);
139 }
140 return p;
141 }
142
143 /**
144 * Paraslash's version of calloc().
145 *
146 * \param size The desired new size.
147 *
148 * A wrapper for calloc(3) which exits on errors.
149 *
150 * \return A pointer to the allocated and zeroed-out memory, which is suitably
151 * aligned for any kind of variable.
152 *
153 * \sa calloc(3)
154 */
155 __must_check __malloc static void *para_calloc(size_t size)
156 {
157 void *ret = para_malloc(size);
158
159 memset(ret, 0, size);
160 return ret;
161 }
162
163 /**
164 * Paraslash's version of strdup().
165 *
166 * \param s The string to be duplicated.
167 *
168 * A wrapper for strdup(3). It calls \p exit(EXIT_FAILURE) on errors, i.e.
169 * there is no need to check the return value in the caller.
170 *
171 * \return A pointer to the duplicated string. If \p s was the NULL pointer,
172 * an pointer to an empty string is returned.
173 *
174 * \sa strdup(3)
175 */
176 __must_check __malloc static char *para_strdup(const char *s)
177 {
178 char *ret;
179
180 if ((ret = strdup(s? s: "")))
181 return ret;
182 EMERG_LOG("strdup failed, aborting\n");
183 exit(EXIT_FAILURE);
184 }
185
186 /**
187 * Get the home directory of the current user.
188 *
189 * \return A dynammically allocated string that must be freed by the caller. If
190 * the home directory could not be found, this function returns "/tmp".
191 */
192 __must_check __malloc static char *para_homedir(void)
193 {
194 struct passwd *pw = getpwuid(getuid());
195 return para_strdup(pw? pw->pw_dir : "/tmp");
196 }
197
198 /**
199 * Compare two osl objects pointing to unsigned integers of 32 bit size.
200 *
201 * \param obj1 Pointer to the first integer.
202 * \param obj2 Pointer to the second integer.
203 *
204 * \return The values required for an osl compare function.
205 *
206 * \sa osl_compare_func, osl_hash_compare().
207 */
208 static int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
209 {
210 uint32_t d1 = read_u32((const char *)obj1->data);
211 uint32_t d2 = read_u32((const char *)obj2->data);
212
213 if (d1 < d2)
214 return 1;
215 if (d1 > d2)
216 return -1;
217 return 0;
218 }
219
220 /**
221 * Traverse the given directory recursively.
222 *
223 * \param dirname The directory to traverse.
224 * \param func The function to call for each entry.
225 * \param private_data Pointer to an arbitrary data structure.
226 *
227 * For each regular file under \a dirname, the supplied function \a func is
228 * called. The full path of the regular file and the \a private_data pointer
229 * are passed to \a func. Directories for which the calling process has no
230 * permissions to change to are silently ignored.
231 *
232 * \return Standard.
233 */
234 static int for_each_file_in_dir(const char *dirname,
235 int (*func)(const char *, void *), void *private_data)
236 {
237 DIR *dir;
238 struct dirent *entry;
239 int cwd_fd, ret2, ret = para_opendir(dirname, &dir, &cwd_fd);
240
241 if (ret < 0)
242 return ret == -ERRNO_TO_ERROR(EACCES)? 1 : ret;
243 /* scan cwd recursively */
244 while ((entry = readdir(dir))) {
245 mode_t m;
246 char *tmp;
247 struct stat s;
248
249 if (!strcmp(entry->d_name, "."))
250 continue;
251 if (!strcmp(entry->d_name, ".."))
252 continue;
253 if (lstat(entry->d_name, &s) == -1)
254 continue;
255 m = s.st_mode;
256 if (!S_ISREG(m) && !S_ISDIR(m))
257 continue;
258 tmp = make_message("%s/%s", dirname, entry->d_name);
259 if (!S_ISDIR(m)) {
260 ret = func(tmp, private_data);
261 free(tmp);
262 if (ret < 0)
263 goto out;
264 continue;
265 }
266 /* directory */
267 ret = for_each_file_in_dir(tmp, func, private_data);
268 free(tmp);
269 if (ret < 0)
270 goto out;
271 }
272 ret = 1;
273 out:
274 closedir(dir);
275 ret2 = para_fchdir(cwd_fd);
276 if (ret2 < 0 && ret >= 0)
277 ret = ret2;
278 close(cwd_fd);
279 return ret;
280 }
281
282 /*
283 * check for object boundary violations
284 *
285 * test whether the range pointed to by the index entry for a given cell is
286 * contained in mapped data file. This should always be the case. Otherwise
287 * we are in real trouble.
288 */
289 static int check_range(struct osl_table *t, uint32_t row_num, uint32_t col_num)
290 {
291 char *index_entry;
292 struct osl_object obj;
293 struct osl_column *col;
294 int ret;
295 char *map_start, *obj_start;
296
297 ret = get_cell_index(t, row_num, col_num, &index_entry);
298 if (ret < 0)
299 return ret;
300 ret = get_mapped_object(t, col_num, row_num, &obj);
301 if (ret < 0)
302 return ret;
303 col = t->columns + col_num;
304 obj_start = obj.data;
305 map_start = col->data_map.data;
306 // INFO_LOG("obj: %p..%p\n", obj_start, obj_start + obj.size);
307 // INFO_LOG("map: %p..%p\n", map_start, map_start + col->data_map.size);
308 if (obj_start < map_start || obj_start + obj.size > map_start + col->data_map.size) {
309 CRIT_LOG("range violation in row %u, col %u\n", row_num,
310 col_num);
311 return -E_RANGE_VIOLATION;
312 }
313 DEBUG_LOG("col %u: ok\n", col_num);
314 return 1;
315 }
316
317 /*
318 * check all cells of the given table for boundary violations
319 */
320 static int check_index_ranges(struct osl_table *t)
321 {
322 int i, j, ret;
323
324 INFO_LOG("checking for range violations in index\n");
325 //DEBUG_LOG("%d rows. %d columns\n", t->num_rows, t->desc->num_columns);
326 t->num_invalid_rows = 0;
327 for (i = 0; i < t->num_rows; i++) {
328 if (row_is_invalid(t, i)) {
329 t->num_invalid_rows++;
330 continue;
331 }
332 for (j = 0; j < t->desc->num_columns; j++) { /* FXIME */
333 const struct osl_column_description *cd =
334 get_column_description(t->desc, j);
335 if (cd->storage_type != OSL_MAPPED_STORAGE)
336 continue;
337 ret = check_range(t, i, j);
338 if (ret < 0) {
339 if (ret != -E_INVALID_OBJECT &&
340 ret != -E_RANGE_VIOLATION)
341 goto err;
342 if (ret == -E_INVALID_OBJECT) {
343 CRIT_LOG("row %d, col %d maps to an "
344 "invalid object\n", i, j);
345 }
346 ret = mark_row_invalid(t, i);
347 if (ret < 0)
348 goto err;
349 t->num_invalid_rows++;
350 break;
351 }
352 }
353
354 }
355 if (t->num_invalid_rows)
356 NOTICE_LOG("ranges OK. %d invalid row(s) detected\n",
357 t->num_invalid_rows);
358 else
359 INFO_LOG("no invalid rows, no range violations, good\n");
360 return 1;
361 err:
362 return ret;
363 }
364
365 static int move_index_entry(struct osl_table *t, uint32_t dest, uint32_t src)
366 {
367 char *dest_ie, *src_ie;
368 int ret = get_row_index(t, dest, &dest_ie);
369
370 if (ret < 0)
371 return ret;
372 ret = get_row_index(t, src, &src_ie);
373 if (ret < 0)
374 return ret;
375 INFO_LOG("moving entry #%u to position %u\n", src, dest);
376 memcpy(dest_ie, src_ie, t->row_index_size);
377 return 1;
378 }
379
380 static int map_index(const struct osl_table_description *desc, struct osl_object *map)
381 {
382 char *filename = index_filename(desc);
383 int ret;
384
385 ret = mmap_full_file(filename, O_RDWR, &map->data, &map->size, NULL);
386 DEBUG_LOG("mapping index %s: ret: %d, size: %zu\n", filename, ret, map->size);
387 free(filename);
388 return ret;
389 }
390
391 static int prune_invalid_rows_from_index(struct osl_table *t)
392 {
393 uint32_t top = 0, bottom;
394 char *filename;
395 int ret;
396
397 if (!t->num_invalid_rows) {
398 INFO_LOG("all rows are valid, good\n");
399 return 1;
400 }
401 NOTICE_LOG("deleting %u invalid row(s) (%d bytes) from index\n",
402 t->num_invalid_rows, t->row_index_size * t->num_invalid_rows);
403 bottom = t->num_rows - 1;
404 while (top < bottom) {
405 if (!row_is_invalid(t, top)) {
406 top++;
407 continue;
408 }
409 while (bottom > top) {
410 if (row_is_invalid(t, bottom)) {
411 bottom--;
412 continue;
413 }
414 /* move bottom index entry to top */
415 move_index_entry(t, top, bottom);
416 bottom--;
417 top++;
418 break;
419 }
420 }
421 DEBUG_LOG("unmapping index\n");
422 osl_munmap(t->index_map.data, t->index_map.size);
423 filename = index_filename(t->desc);
424 ret = para_truncate(filename, t->row_index_size
425 * t->num_invalid_rows);
426 free(filename);
427 if (ret < 0)
428 return ret;
429 ret = map_index(t->desc, &t->index_map);
430 if (ret < 0)
431 return ret;
432 t->num_rows = table_num_rows(t);
433 return 1;
434 }
435
436 static int check_for_invalid_objects(struct osl_table *t, uint32_t **lost_bytes)
437 {
438 int i, j, ret;
439 const struct osl_column_description *cd;
440 uint32_t *loss = para_malloc(sizeof(uint32_t) * t->desc->num_columns);
441
442 INFO_LOG("looking for mapped objects not contained in index\n");
443 /* first count used bytes */
444 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
445 loss[i] = t->columns[i].data_map.size;
446 for (j = 0; j < t->num_rows; j++) {
447 struct osl_object obj;
448 ret = get_mapped_object(t, i, j, &obj);
449 if (ret >= 0) {
450 loss[i] -= obj.size + 1; /* add one for header byte */
451 continue;
452 }
453 if (ret != -E_INVALID_OBJECT)
454 goto err;
455 CRIT_LOG("row %d, col %d points to an invalid "
456 "mapped object, bad\n", j, i);
457 }
458 }
459 ret = 0;
460 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
461 if (loss[i]) {
462 NOTICE_LOG("column %u contains %u lost bytes\n",
463 i, loss[i]);
464 ret = 1;
465 }
466 }
467 if (!ret)
468 INFO_LOG("all mapped objects are valid, good\n");
469 *lost_bytes = loss;
470 return ret;
471 err:
472 free(loss);
473 return ret;
474 }
475
476 /* prune_invalid_rows() must be run on the table before calling this */
477 static int prune_mapped_column(struct osl_table *t, uint32_t col_num, int fd)
478 {
479 int i, ret;
480 uint32_t written = 0;
481 struct osl_column *col = t->columns + col_num;
482
483 INFO_LOG("pruning col %u\n", col_num);
484 for (i = 0; i < t->num_rows; i++) {
485 struct osl_object obj;
486 char *index_entry;
487
488 DEBUG_LOG("checking row %u/%u\n", i, t->num_rows);
489 ret = get_mapped_object(t, col_num, i, &obj);
490 if (ret < 0)
491 return ret;
492 ret = _write_all(fd, (char *)(obj.data) - 1, obj.size + 1);
493 if (ret < 0)
494 return ret;
495 written += obj.size + 1;
496 ret = get_row_index(t, i, &index_entry);
497 if (ret < 0)
498 return ret;
499 update_cell_index(index_entry, col, written, obj.size);
500 }
501 return 1;
502 }
503
504 static int prune_objects(struct osl_table *t, uint32_t *lost_bytes)
505 {
506 int i, ret;
507 const struct osl_column_description *cd;
508 char **col_filenames = para_calloc(t->desc->num_columns * sizeof(char *));
509 char **new_col_filenames = para_calloc(t->desc->num_columns * sizeof(char *));
510 char *idx_filename = index_filename(t->desc);
511 char *old_idx_filename = make_message("%s.bak", idx_filename);
512 int fd;
513
514 NOTICE_LOG("removing unreferenced objects from data files\n");
515 /* first make a copy of the index */
516 ret = osl_open(old_idx_filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
517 if (ret < 0)
518 goto out_free;
519 fd = ret;
520 ret = _write_all(fd, t->index_map.data, t->index_map.size);
521 close(fd);
522 if (ret < 0)
523 goto out_free;
524 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
525 if (!lost_bytes[i])
526 continue;
527 col_filenames[i] = column_filename(t, i);
528 new_col_filenames[i] = make_message("%s.fsck", col_filenames[i]);
529 ret = osl_open(new_col_filenames[i], O_WRONLY | O_CREAT | O_EXCL, 0644);
530 if (ret < 0)
531 goto out_unlink_data;
532 fd = ret;
533 ret = prune_mapped_column(t, i, fd);
534 close(fd);
535 if (ret < 0)
536 goto out_unlink_data;
537 }
538 ret = unmap_table(t, OSL_MARK_CLEAN);
539 if (ret < 0)
540 goto out_unlink_data;
541 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
542 if (!lost_bytes[i])
543 continue;
544 ret = osl_rename(new_col_filenames[i], col_filenames[i]);
545 if (ret < 0) { /* we're kinda screwed here */
546 CRIT_LOG("rename of col %i failed: %s\n", i,
547 strerror(errno));
548 goto out_free;
549 }
550 }
551 unlink(old_idx_filename);
552 ret = map_table(t, 0);
553 goto out_free;
554 out_unlink_data:
555 FOR_EACH_MAPPED_COLUMN(i, t, cd)
556 unlink(new_col_filenames[i]);
557 out_free:
558 free(old_idx_filename);
559 free(idx_filename);
560 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
561 free(col_filenames[i]);
562 free(new_col_filenames[i]);
563 }
564 free(col_filenames);
565 free(new_col_filenames);
566 return ret;
567 }
568
569 static struct osl_column_description hash_tree_table_cols[] = {
570 {
571 .storage_type = OSL_NO_STORAGE,
572 .storage_flags = OSL_RBTREE | OSL_FIXED_SIZE | OSL_UNIQUE,
573 .name = "hash",
574 .compare_function = uint32_compare,
575 .data_size = HASH_SIZE
576 },
577 };
578
579 static const struct osl_table_description hash_tree_table_desc = {
580 .dir = "/", /* irrelevant */
581 .name = "hash_tree",
582 .num_columns = 1,
583 .flags = 0,
584 .column_descriptions = hash_tree_table_cols
585 };
586
587 /**
588 * The hash_tree table contains all hashes of the disk storage name column.
589 * of each row. It is used for checking if a disk storage file has a reference
590 * in the table.
591 */
592 static struct osl_table *hash_tree_table;
593 static HASH_TYPE *hashes;
594
595 static int check_disk_storage_column(struct osl_table *t, int row_num,
596 int col_num, char *ds_name, unsigned *num_missing_objects)
597 {
598 int ret;
599 struct stat statbuf;
600 char *path = disk_storage_path(t, col_num, ds_name);
601 unsigned dsnc = t->disk_storage_name_column;
602 struct osl_object obj;
603
604 DEBUG_LOG("checking if %s is a regular file\n", path);
605 ret = stat(path, &statbuf);
606 if (ret < 0 && errno == ENOENT) {
607 struct osl_row *row;
608 (*num_missing_objects)++;
609 ERROR_LOG("row %d: object %s is missing\n", row_num, path);
610 NOTICE_LOG("trying to delete row %d\n", row_num);
611 ret = osl_get_row(t, dsnc, &obj, &row);
612 if (ret < 0) {
613 CRIT_LOG("unable to get row %d\n", row_num);
614 mark_row_invalid(t, row_num);
615 CRIT_LOG("Please re-run fsck\n");
616 goto out;
617 }
618 ret = osl_del_row(t, row);
619 if (ret < 0)
620 goto out;
621 }
622 out:
623 free(path);
624 if (ret < 0)
625 return ret;
626 ret = -E_NOT_A_REGULAR_FILE;
627 if (!(S_IFREG & statbuf.st_mode))
628 return ret;
629 return 1;
630 }
631
632 static int check_disk_storage_presence(struct osl_table *t)
633 {
634 int ret, i, j;
635 struct osl_object obj, hash_obj = {.size = HASH_SIZE};
636 char *ds_name;
637 const struct osl_column_description *cd;
638 unsigned dsnc = t->disk_storage_name_column, missing_objects = 0;
639
640 if (!t->num_rows)
641 return 1;
642 hashes = para_malloc(t->num_rows * HASH_SIZE);
643 INFO_LOG("looking for missing disk storage objects\n");
644 for (i = 0; i < t->num_rows; i++) {
645 if (row_is_invalid(t, i))
646 continue;
647 ret = get_mapped_object(t, dsnc, i, &obj);
648 if (ret < 0)
649 return ret;
650 hash_object(&obj, hashes + i * HASH_SIZE);
651 hash_obj.data = hashes + i * HASH_SIZE;
652 osl_add_row(hash_tree_table, &hash_obj);
653 ds_name = disk_storage_name_of_hash(t, hashes + i * HASH_SIZE);
654 FOR_EACH_DISK_STORAGE_COLUMN(j, t, cd) {
655 ret = check_disk_storage_column(t, i, j, ds_name,
656 &missing_objects);
657 if (ret < 0)
658 goto err;
659 }
660 free(ds_name);
661 }
662 if (!missing_objects)
663 INFO_LOG("all referenced disk storage objects exist, good\n");
664 else
665 NOTICE_LOG("%d missing object(s)\n", missing_objects);
666 return missing_objects;
667 err:
668 free(ds_name);
669 return ret;
670 }
671
672 static int dummy_compare(const struct osl_object *obj1, const struct osl_object *obj2)
673 {
674 if (obj1 < obj2)
675 return -1;
676 if (obj1 > obj2)
677 return 1;
678 return 0;
679 }
680
681 static unsigned files_pruned;
682
683 int prune_disk_storage_file(const char *path, void *private_data)
684 {
685 HASH_TYPE hash[HASH_SIZE];
686 unsigned flags = *(unsigned *)private_data;
687 struct osl_object obj = {.data = hash, .size = HASH_SIZE};
688 struct osl_row *row;
689 int ret = -1;
690 size_t len = strlen(path);
691
692
693 DEBUG_LOG("path: %s\n", path);
694 if (flags & OSL_LARGE_TABLE) {
695 if (len < HASH_SIZE * 2 + 2)
696 goto invalid;
697 // NOTICE_LOG("p: %s\n", path + len - 2 * HASH_SIZE - 1);
698 ret = asc_to_hash(path + len - 2 * HASH_SIZE - 1, 1, hash);
699 if (ret < 0)
700 goto invalid;
701 ret = asc_to_hash(path + len - 2 * HASH_SIZE + 2, HASH_SIZE - 1,
702 hash + 1);
703 if (ret < 0)
704 goto invalid;
705 // INFO_LOG("high: %x, low: %x, hash: %x\n", high, low, hash);
706 } else {
707 if (len < 2 * HASH_SIZE + 1)
708 goto invalid;
709 ret = asc_to_hash(path + len - 2 * HASH_SIZE, 2 * HASH_SIZE, hash);
710 if (ret < 0)
711 goto invalid;
712 // INFO_LOG("hash: %x\n", hash);
713 }
714 #if 0
715 {
716 char asc[2 * HASH_SIZE + 1];
717 hash_to_asc(hash, asc);
718 NOTICE_LOG("before: %s\nafter: %s\n", path, asc);
719 }
720 #endif
721 ret = osl_get_row(hash_tree_table, 0, &obj, &row);
722 if (ret >= 0)
723 return 1;
724 NOTICE_LOG("unreferenced file in hash dir: %s\n", path);
725 goto remove;
726 invalid:
727 ERROR_LOG("could not read hash value of %s\n", path);
728 remove:
729 NOTICE_LOG("removing %s\n", path);
730 unlink(path);
731 files_pruned++;
732 return 1;
733 }
734
735 static int prune_disk_storage_files(struct osl_table *t)
736 {
737 int i, ret = 1;
738 const struct osl_column_description *cd;
739
740 INFO_LOG("looking for unreferenced disk storage files\n");
741 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
742 char *dirname = column_filename(t, i);
743 ret = for_each_file_in_dir(dirname, prune_disk_storage_file,
744 (unsigned *)&t->desc->flags);
745 free(dirname);
746 }
747 if (files_pruned)
748 NOTICE_LOG("%u disk storage files deleted\n",
749 files_pruned);
750 else
751 INFO_LOG("all files are are referenced, good\n");
752 return ret;
753 }
754
755 static int check_disk_storage_columns(struct osl_table *t)
756 {
757 int ret, i;
758 const struct osl_column_description *cd;
759
760 if (!t->num_disk_storage_columns) {
761 INFO_LOG("no disk storage columns in table '%s', "
762 "skipping checks\n", t->desc->name);
763 return 1;
764 }
765 FOR_EACH_COLUMN(i, t->desc, cd)
766 t->desc->column_descriptions[i].compare_function = dummy_compare;
767 ret = init_rbtrees(t);
768 if (ret < 0)
769 return ret;
770 INFO_LOG("creating rbtree for disk storage hash values\n");
771 ret = osl_open_table(&hash_tree_table_desc, &hash_tree_table);
772 if (ret < 0)
773 goto out;
774 ret = check_disk_storage_presence(t);
775 if (ret < 0)
776 goto out_close_hash_tree;
777 ret = prune_disk_storage_files(t);
778 out_close_hash_tree:
779 osl_close_table(hash_tree_table, 0);
780 free(hashes);
781 hashes = NULL;
782 out:
783 clear_rbtrees(t); /* TODO why are we doing that here? Seems odd */
784 return ret;
785 }
786
787 static void set_dummy_contents(struct osl_table_description *desc)
788 {
789 int i;
790 struct osl_column_description *cd;
791
792 for (i = 0; i < desc->num_columns; i++) {
793 cd = get_column_description(desc, i);
794 cd->compare_function = dummy_compare;
795 }
796 }
797
798 static int fsck_init(struct osl_table_description *desc, struct osl_table **t)
799 {
800 struct osl_object map;
801 int ret = map_index(desc, &map);
802
803 if (ret < 0)
804 goto out;
805 ret = read_table_desc(&map, desc);
806 if (ret < 0) {
807 osl_munmap(map.data, map.size);
808 goto out;
809 }
810 set_dummy_contents(desc);
811 ret = init_table_structure(desc, t);
812 if (ret < 0) {
813 osl_munmap(map.data, map.size);
814 goto out;
815 }
816 DEBUG_LOG("unmapping index\n");
817 osl_munmap(map.data, map.size);
818 if (conf.force_given)
819 ret = map_table(*t, (MAP_TBL_FL_IGNORE_DIRTY));
820 else
821 ret = map_table(*t, 0);
822 if (ret >= 0)
823 (*t)->num_rows = table_num_rows(*t);
824 out:
825 return ret;
826 }
827
828 static void fsck_cleanup(struct osl_table *t)
829 {
830 int i;
831
832 if (!t)
833 return;
834 if (t->desc->column_descriptions) {
835 struct osl_column_description *cd;
836 for (i = 0; i < t->desc->num_columns; i++) {
837 cd = get_column_description(t->desc, i);
838 free((char*)cd->name);
839 }
840 free(t->desc->column_descriptions);
841 }
842 free(t->columns);
843 free(t);
844
845 }
846
847 #define ST_CASE(st) case st: return #st
848
849 const char *get_asc_storage_type(enum osl_storage_type st)
850 {
851 switch (st) {
852 ST_CASE(OSL_MAPPED_STORAGE);
853 ST_CASE(OSL_DISK_STORAGE);
854 ST_CASE(OSL_NO_STORAGE);
855 }
856 return NULL;
857 }
858
859 #define APPEND_ASC_SF(sf, flag, str) do { if (sf & flag) { \
860 if (str) str = make_message("%s%s", str, " | " # flag); \
861 else str = para_strdup(#flag); }} while (0)
862
863
864 char *get_asc_storage_flags(enum osl_storage_type sf)
865 {
866 char *asc_sf = NULL;
867
868 APPEND_ASC_SF(sf, OSL_RBTREE, asc_sf);
869 APPEND_ASC_SF(sf, OSL_FIXED_SIZE, asc_sf);
870 APPEND_ASC_SF(sf, OSL_UNIQUE, asc_sf);
871 return asc_sf;
872 }
873
874 static int dump_table_desc(struct osl_table *t, int fd)
875 {
876 const struct osl_table_description *desc = t->desc;
877 int ret, i;
878 struct osl_column_description *cd;
879 char *msg = make_message("static struct osl_column_description cols[] = {\n");
880 ret = _write_all(fd, msg, strlen(msg));
881 if (ret < 0)
882 return ret;
883 free(msg);
884 FOR_EACH_COLUMN(i, desc, cd) {
885 const char *asc_st;
886 msg = make_message("\t[%d] = {\n", i);
887 ret = _write_all(fd, msg, strlen(msg));
888 if (ret < 0)
889 return ret;
890 free(msg);
891 asc_st = get_asc_storage_type(cd->storage_type);
892 msg = make_message("\t\t.storage_type = %s,\n", asc_st);
893 ret = _write_all(fd, msg, strlen(msg));
894 if (ret < 0)
895 return ret;
896 free(msg);
897 if (cd->storage_flags) {
898 char *asc_sf = get_asc_storage_flags(cd->storage_flags);
899 msg = make_message("\t\t,storage_flags = %s,\n", asc_sf);
900 free(asc_sf);
901 ret = _write_all(fd, msg, strlen(msg));
902 if (ret < 0)
903 return ret;
904 free(msg);
905 }
906 if (cd->storage_flags & OSL_FIXED_SIZE) {
907 msg = make_message("\t\t.data_size = %u,\n", cd->data_size);
908 ret = _write_all(fd, msg, strlen(msg));
909 if (ret < 0)
910 return ret;
911 free(msg);
912 }
913 msg = make_message("\t\t.name = \"%s\",\n", cd->name);
914 ret = _write_all(fd, msg, strlen(msg));
915 if (ret < 0)
916 return ret;
917 free(msg);
918 if (cd->storage_flags & OSL_RBTREE) {
919 msg = make_message("\t\t.compare_function = compare_func,\n");
920 ret = _write_all(fd, msg, strlen(msg));
921 if (ret < 0)
922 return ret;
923 free(msg);
924 }
925 msg = make_message("\t},\n");
926 ret = _write_all(fd, msg, strlen(msg));
927 if (ret < 0)
928 return ret;
929 free(msg);
930 }
931 msg = make_message("};\n");
932 ret = _write_all(fd, msg, strlen(msg));
933 if (ret < 0)
934 return ret;
935 free(msg);
936 return 1;
937 }
938
939 static int dump_row(struct osl_table *t, unsigned row_num, const char *row_dir)
940 {
941 int ret, i;
942 const struct osl_column_description *cd;
943 unsigned dsnc;
944 struct osl_object obj;
945 char *ds_name;
946 HASH_TYPE hash[HASH_SIZE];
947 char *filename;
948
949 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
950 ret = get_mapped_object(t, i, row_num, &obj);
951 if (ret < 0)
952 return ret;
953 filename = make_message("%s/col_%03u", row_dir, i);
954 ret = write_file(filename, obj.data, obj.size);
955 free(filename);
956 if (ret < 0)
957 return ret;
958 }
959 if (!t->num_disk_storage_columns)
960 return 1;
961 dsnc = t->disk_storage_name_column;
962 ret = get_mapped_object(t, dsnc, row_num, &obj);
963 if (ret < 0)
964 return ret;
965 hash_object(&obj, hash);
966 ds_name = disk_storage_name_of_hash(t, hash);
967 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
968 filename = disk_storage_path(t, i, ds_name);
969 ret = mmap_full_file(filename, O_RDONLY, &obj.data, &obj.size, NULL);
970 free(filename);
971 if (ret < 0)
972 goto out;
973 filename = make_message("%s/col_%03u", row_dir, i);
974 ret = write_file(filename, obj.data, obj.size);
975 free(filename);
976 if (ret < 0)
977 goto out;
978 }
979 ret = 1;
980 out:
981 free(ds_name);
982 return ret;
983 }
984
985 static int dump_rows(char *dump_dir, struct osl_table *t)
986 {
987 unsigned i;
988 char *current_dir = NULL;
989 int ret = 0;
990
991 for (i = 0; i < t->num_rows; i++) {
992 char *row_dir;
993 if (row_is_invalid(t, i))
994 continue;
995 if (!(i % 1000)) {
996 free(current_dir);
997 current_dir = make_message("%s/rows_%u-%u", dump_dir, i, i + 999);
998 NOTICE_LOG("dumping rows %u - %u\n", i, i + 999);
999 ret = osl_mkdir(current_dir, 0777);
1000 if (ret < 0 && !is_errno(-ret, EEXIST))
1001 goto out;
1002 }
1003 row_dir = make_message("%s/row_%03u", current_dir, i);
1004 ret = osl_mkdir(row_dir, 0777);
1005 if (ret < 0 && !is_errno(-ret, EEXIST)) {
1006 free(row_dir);
1007 goto out;
1008 }
1009 ret = dump_row(t, i, row_dir);
1010 free(row_dir);
1011 if (ret < 0)
1012 goto out;
1013 }
1014 out:
1015 free(current_dir);
1016 return ret;
1017 }
1018
1019 static int dump_table(char *dump_dir, struct osl_table_description *desc)
1020 {
1021 struct osl_table *t = NULL;
1022 int fd, ret = fsck_init(desc, &t);
1023 char *desc_file;
1024 char *table_dump_dir = NULL;
1025
1026 if (ret < 0)
1027 goto out;
1028 ret = osl_mkdir(dump_dir, 0777);
1029 if (ret < 0 && !is_errno(-ret, EEXIST))
1030 goto out;
1031 table_dump_dir = make_message("%s/%s", dump_dir, desc->name);
1032 ret = osl_mkdir(table_dump_dir, 0777);
1033 if (ret < 0 && !is_errno(-ret, EEXIST))
1034 goto out;
1035 desc_file = make_message("%s/table_description.c", table_dump_dir);
1036 ret = osl_open(desc_file, O_WRONLY | O_CREAT | O_EXCL, 0644);
1037 free(desc_file);
1038 if (ret < 0)
1039 goto out;
1040 fd = ret;
1041 ret = dump_table_desc(t, fd);
1042 close(fd);
1043 if (ret < 0)
1044 goto out;
1045 ret = dump_rows(table_dump_dir, t);
1046 out:
1047 free(table_dump_dir);
1048 fsck_cleanup(t);
1049 return ret;
1050 }
1051
1052 static int fsck(struct osl_table_description *desc)
1053 {
1054 int ret;
1055 struct osl_table *t = NULL;
1056 uint32_t *lost_bytes = NULL;
1057
1058 ret = fsck_init(desc, &t);
1059 if (ret < 0)
1060 goto out;
1061 ret = check_index_ranges(t);
1062 if (ret < 0)
1063 goto out_unmap;
1064 ret = check_disk_storage_columns(t);
1065 if (ret < 0)
1066 goto out_unmap;
1067 ret = prune_invalid_rows_from_index(t);
1068 if (ret < 0)
1069 goto out_unmap;
1070 ret = check_for_invalid_objects(t, &lost_bytes);
1071 if (ret < 0)
1072 goto out_unmap;
1073 if (ret > 0) { /* at least one mapped data file needs pruning */
1074 ret = prune_objects(t, lost_bytes);
1075 if (ret < 0)
1076 goto out_unmap;
1077 }
1078 free(lost_bytes);
1079 out_unmap:
1080 unmap_table(t, OSL_MARK_CLEAN);
1081 out:
1082 fsck_cleanup(t);
1083 return ret;
1084 }
1085
1086 static int check_table(char *base_dir, char *table_name)
1087 {
1088 struct osl_table_description desc = {
1089 .column_descriptions = NULL,
1090 .dir = base_dir,
1091 .name = table_name
1092 };
1093 int ret;
1094
1095 INFO_LOG("checking table %s\n", table_name);
1096 if (!conf.no_fsck_given) {
1097 ret = fsck(&desc);
1098 if (ret < 0)
1099 goto out;
1100 }
1101 ret = 1;
1102 if (!conf.dump_dir_given || !*conf.dump_dir_arg)
1103 goto out;
1104 ret = dump_table(conf.dump_dir_arg, &desc);
1105 out:
1106 if (ret < 0)
1107 ERROR_LOG("failed to check table %s\n", table_name);
1108 else
1109 NOTICE_LOG("successfully checked table %s\n", table_name);
1110 return ret;
1111 }
1112
1113 static int check_all_tables(char *base_dir)
1114 {
1115 DIR *dir;
1116 struct dirent *entry;
1117 int cwd_fd, ret2, ret = para_opendir(base_dir, &dir, &cwd_fd);
1118
1119 if (ret < 0)
1120 return ret;
1121 while ((entry = readdir(dir))) {
1122 mode_t m;
1123 struct stat s;
1124 if (!strcmp(entry->d_name, "."))
1125 continue;
1126 if (!strcmp(entry->d_name, ".."))
1127 continue;
1128 if (lstat(entry->d_name, &s) == -1)
1129 continue;
1130 m = s.st_mode;
1131 if (!S_ISDIR(m))
1132 continue;
1133 ret = check_table(base_dir, entry->d_name);
1134 if (ret < 0)
1135 break;
1136 }
1137 closedir(dir);
1138 ret2 = para_fchdir(cwd_fd);
1139 if (ret2 < 0 && ret >= 0)
1140 ret = ret2;
1141 close(cwd_fd);
1142 return ret;
1143 }
1144
1145 int main(int argc, char **argv)
1146 {
1147 int i, ret;
1148 char *base_dir = NULL;
1149
1150 ret = fsck_cmdline_parser(argc, argv, &conf);
1151 if (ret < 0) {
1152 ret = -E_FSCK_SYNTAX;
1153 goto out;
1154 }
1155 HANDLE_VERSION_FLAG("fsck", conf);
1156 if (conf.base_dir_given)
1157 base_dir = para_strdup(conf.base_dir_arg);
1158 else {
1159 char *home = para_homedir();
1160 base_dir = make_message("%s/.paraslash/afs_database", home);
1161 free(home);
1162 }
1163 if (!conf.inputs_num) {
1164 ret = check_all_tables(base_dir);
1165 goto out;
1166 }
1167 for (i = 0; i < conf.inputs_num; i++) {
1168 ret = check_table(base_dir, conf.inputs[i]);
1169 if (ret < 0)
1170 break;
1171 }
1172 out:
1173 if (ret < 0) {
1174 /* FIXME: osl_strerror() is BAD!!! */
1175 ERROR_LOG("%s%s: %s\n",
1176 base_dir? "base_dir: " : "",
1177 base_dir? base_dir : "",
1178 osl_strerror(-ret)
1179 );
1180 if (conf.loglevel_arg > 1)
1181 EMERG_LOG("re-run with \"--loglevel %d\" to increase verbosity\n",
1182 conf.loglevel_arg - 1);
1183 } else
1184 NOTICE_LOG("success\n");
1185 if (base_dir)
1186 free(base_dir);
1187 return ret < 0? EXIT_FAILURE : EXIT_SUCCESS;
1188 }