Merge branch 'refs/heads/t/lopsub'
[osl.git] / fsck.c
1 /*
2 * Copyright (C) 2007-2009 Andre Noll <maan@tuebingen.mpg.de>
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 <lopsub.h>
13 #include <assert.h>
14 #include <pwd.h>
15
16 #include "log.h"
17 #include "osl.h"
18 #include "util.h"
19 #include "osl_core.h"
20 #include "oslfsck.lsg.h"
21
22 static struct lls_parse_result *lpr;
23 #define CMD_PTR (lls_cmd(0, oslfsck_suite))
24 #define OPT_RESULT(_name) \
25 (lls_opt_result(LSG_OSLFSCK_OSLFSCK_OPT_ ## _name, lpr))
26 #define OPT_GIVEN(_name) (lls_opt_given(OPT_RESULT(_name)))
27 #define OPT_STRING_VAL(_name) (lls_string_val(0, OPT_RESULT(_name)))
28 #define OPT_UINT32_VAL(_name) (lls_uint32_val(0, OPT_RESULT(_name)))
29
30 #define FSCK_ERRORS \
31 FSCK_ERROR(RANGE_VIOLATION, "range violation detected, very bad"), \
32 FSCK_ERROR(NOT_A_REGULAR_FILE, "not a regular file"), \
33 FSCK_ERROR(SYNTAX, "fsck syntax error"), \
34 FSCK_ERROR(ACCESS, "permission denied"), \
35 FSCK_ERROR(CHDIR, "could not change directory"), \
36 FSCK_ERROR(OPENDIR, "could not open directory"),
37
38 #define FSCK_ERROR_BIT 29
39 #define FSCK_ERROR(num, txt) E_FSCK_ ## num
40 enum {
41 FSCK_DUMMY = (1 << FSCK_ERROR_BIT) - 1,
42 FSCK_ERRORS
43 };
44 #undef FSCK_ERROR
45 #define FSCK_ERROR(num, txt) txt
46 static const char *fsck_errors[] = {
47 FSCK_ERRORS
48 };
49
50 static const char *fsck_strerror(int num)
51 {
52 if (num & (1 << FSCK_ERROR_BIT))
53 return fsck_errors[num & ((1 << FSCK_ERROR_BIT) - 1)];
54 return osl_strerror(num);
55 }
56
57 extern int loglevel;
58
59 /* taken from git */
60 static signed char hexval_table[256] = {
61 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
67 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
68 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
69 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
71 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
73 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
74 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
75 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
76 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
77 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
78 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
79 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
80 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
81 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
82 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
83 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
84 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
85 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
86 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
87 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
88 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
89 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
90 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
91 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
92 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
93 };
94
95 static int asc_to_hash(const char *asc_hash, int len, HASH_TYPE *hash)
96 {
97 int i = 0;
98 const unsigned char *asc = (const unsigned char *) asc_hash;
99
100 while (*asc && i++ < len) {
101 unsigned int val = (hexval_table[asc[0]] << 4) | hexval_table[asc[1]];
102 if (val & ~0xff)
103 return -1;
104 *hash++ = val;
105 asc += 2;
106
107 }
108 return 1;
109 }
110
111 static int _write_all(int fd, const char *buf, size_t len)
112 {
113 return write_all(fd, buf, &len);
114 }
115
116 /*
117 * Wrapper for malloc().
118 *
119 * \param size The desired new size.
120 *
121 * A wrapper for malloc(3) which exits on errors.
122 *
123 * \return A pointer to the allocated memory, which is suitably aligned for any
124 * kind of variable.
125 *
126 * \sa malloc(3).
127 */
128 __must_check __malloc static void *fsck_malloc(size_t size)
129 {
130 assert(size);
131 void *p = malloc(size);
132
133 if (!p) {
134 EMERG_LOG("malloc failed (size = %zu), aborting\n",
135 size);
136 exit(EXIT_FAILURE);
137 }
138 return p;
139 }
140
141 /**
142 * Allocate memory and fill with zeros.
143 *
144 * \param size The desired new size.
145 *
146 * A wrapper for calloc(3) which exits on errors.
147 *
148 * \return A pointer to the allocated and zeroed-out memory, which is suitably
149 * aligned for any kind of variable.
150 *
151 * \sa calloc(3)
152 */
153 __must_check __malloc static void *fsck_calloc(size_t size)
154 {
155 void *ret = fsck_malloc(size);
156
157 memset(ret, 0, size);
158 return ret;
159 }
160
161 /**
162 * Save version of strdup().
163 *
164 * \param s The string to be duplicated.
165 *
166 * A wrapper for strdup(3). It calls \p exit(EXIT_FAILURE) on errors, i.e.
167 * there is no need to check the return value in the caller.
168 *
169 * \return A pointer to the duplicated string. If \p s was the NULL pointer,
170 * an pointer to an empty string is returned.
171 *
172 * \sa strdup(3)
173 */
174 __must_check __malloc static char *fsck_strdup(const char *s)
175 {
176 char *ret;
177
178 if ((ret = strdup(s? s: "")))
179 return ret;
180 EMERG_LOG("strdup failed, aborting\n");
181 exit(EXIT_FAILURE);
182 }
183
184 /**
185 * Compare two osl objects pointing to unsigned integers of 32 bit size.
186 *
187 * \param obj1 Pointer to the first integer.
188 * \param obj2 Pointer to the second integer.
189 *
190 * \return The values required for an osl compare function.
191 *
192 * \sa osl_compare_func, osl_hash_compare().
193 */
194 static int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
195 {
196 uint32_t d1 = read_u32((const char *)obj1->data);
197 uint32_t d2 = read_u32((const char *)obj2->data);
198
199 if (d1 < d2)
200 return 1;
201 if (d1 > d2)
202 return -1;
203 return 0;
204 }
205
206 /**
207 * A wrapper for fchdir().
208 *
209 * \param fd An open file descriptor.
210 *
211 * \return Standard.
212 */
213 static inline int __fchdir(int fd)
214 {
215 if (fchdir(fd) >= 0)
216 return 1;
217 return errno == EACCES? -E_FSCK_ACCESS : -E_FSCK_CHDIR;
218 }
219
220 /**
221 * Wrapper for chdir(2).
222 *
223 * \param path The specified directory.
224 *
225 * \return Standard.
226 */
227 _static_inline_ int __chdir(const char *path)
228 {
229 if (chdir(path) >= 0)
230 return 1;
231 return errno == EACCES? -E_FSCK_ACCESS : -E_FSCK_CHDIR;
232 }
233
234 /**
235 * Save the cwd and open a given directory.
236 *
237 * \param dirname Path to the directory to open.
238 * \param dir Result pointer.
239 * \param cwd File descriptor of the current working directory.
240 *
241 * \return Standard.
242 *
243 * Opening the current directory (".") and calling fchdir() to return is
244 * usually faster and more reliable than saving cwd in some buffer and calling
245 * chdir() afterwards.
246 *
247 * If \a cwd is not \p NULL "." is opened and the resulting file descriptor is
248 * stored in \a cwd. If the function returns success, and \a cwd is not \p
249 * NULL, the caller must close this file descriptor (probably after calling
250 * fchdir(*cwd)).
251 *
252 * On errors, the function undos everything, so the caller needs neither close
253 * any files, nor change back to the original working directory.
254 *
255 * \sa getcwd(3).
256 *
257 */
258 static int fsck_opendir(const char *dirname, DIR **dir, int *cwd)
259 {
260 int ret;
261
262 if (cwd) {
263 ret = osl_open(".", O_RDONLY, 0);
264 if (ret < 0)
265 return ret;
266 *cwd = ret;
267 }
268 ret = __chdir(dirname);
269 if (ret < 0)
270 goto close_cwd;
271 *dir = opendir(".");
272 if (*dir)
273 return 1;
274 ret = errno == EACCES? -E_FSCK_ACCESS : -E_FSCK_OPENDIR;
275 /* Ignore return value of fchdir() and close(). We're busted anyway. */
276 if (cwd) {
277 int __a_unused ret2 = fchdir(*cwd); /* STFU, gcc */
278 }
279 close_cwd:
280 if (cwd)
281 close(*cwd);
282 return ret;
283 }
284
285
286
287 /**
288 * Traverse the given directory recursively.
289 *
290 * \param dirname The directory to traverse.
291 * \param func The function to call for each entry.
292 * \param private_data Pointer to an arbitrary data structure.
293 *
294 * For each regular file under \a dirname, the supplied function \a func is
295 * called. The full path of the regular file and the \a private_data pointer
296 * are passed to \a func. Directories for which the calling process has no
297 * permissions to change to are silently ignored.
298 *
299 * \return Standard.
300 */
301 static int for_each_file_in_dir(const char *dirname,
302 int (*func)(const char *, void *), void *private_data)
303 {
304 DIR *dir;
305 struct dirent *entry;
306 int cwd_fd, ret2, ret = fsck_opendir(dirname, &dir, &cwd_fd);
307
308 if (ret < 0)
309 return ret == -E_FSCK_ACCESS? 1 : ret;
310 /* scan cwd recursively */
311 while ((entry = readdir(dir))) {
312 mode_t m;
313 char *tmp;
314 struct stat s;
315
316 if (!strcmp(entry->d_name, "."))
317 continue;
318 if (!strcmp(entry->d_name, ".."))
319 continue;
320 if (lstat(entry->d_name, &s) == -1)
321 continue;
322 m = s.st_mode;
323 if (!S_ISREG(m) && !S_ISDIR(m))
324 continue;
325 tmp = make_message("%s/%s", dirname, entry->d_name);
326 if (!S_ISDIR(m)) {
327 ret = func(tmp, private_data);
328 free(tmp);
329 if (ret < 0)
330 goto out;
331 continue;
332 }
333 /* directory */
334 ret = for_each_file_in_dir(tmp, func, private_data);
335 free(tmp);
336 if (ret < 0)
337 goto out;
338 }
339 ret = 1;
340 out:
341 closedir(dir);
342 ret2 = __fchdir(cwd_fd);
343 if (ret2 < 0 && ret >= 0)
344 ret = ret2;
345 close(cwd_fd);
346 return ret;
347 }
348
349 /*
350 * check for object boundary violations
351 *
352 * test whether the range pointed to by the index entry for a given cell is
353 * contained in mapped data file. This should always be the case. Otherwise
354 * we are in real trouble.
355 */
356 static int check_range(struct osl_table *t, uint32_t row_num, uint32_t col_num)
357 {
358 char *index_entry;
359 struct osl_object obj;
360 struct osl_column *col;
361 int ret;
362 char *map_start, *obj_start;
363
364 ret = get_cell_index(t, row_num, col_num, &index_entry);
365 if (ret < 0)
366 return ret;
367 ret = get_mapped_object(t, col_num, row_num, &obj);
368 if (ret < 0)
369 return ret;
370 col = t->columns + col_num;
371 obj_start = obj.data;
372 map_start = col->data_map.data;
373 // INFO_LOG("obj: %p..%p\n", obj_start, obj_start + obj.size);
374 // INFO_LOG("map: %p..%p\n", map_start, map_start + col->data_map.size);
375 if (obj_start < map_start || obj_start + obj.size > map_start + col->data_map.size) {
376 CRIT_LOG("range violation in row %u, col %u\n", row_num,
377 col_num);
378 return -E_FSCK_RANGE_VIOLATION;
379 }
380 DEBUG_LOG("col %u: ok\n", col_num);
381 return 1;
382 }
383
384 static int fsck_mark_row_invalid(struct osl_table *t, int i)
385 {
386 if (OPT_GIVEN(DRY_RUN))
387 return 0;
388 return mark_row_invalid(t, i);
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 const struct osl_column_description *cd;
403
404 if (row_is_invalid(t, i)) {
405 t->num_invalid_rows++;
406 continue;
407 }
408 FOR_EACH_MAPPED_COLUMN(j, t, cd) {
409 ret = check_range(t, i, j);
410 if (ret < 0) {
411 if (ret != -E_FSCK_RANGE_VIOLATION)
412 goto err;
413 ret = fsck_mark_row_invalid(t, i);
414 if (ret < 0)
415 goto err;
416 t->num_invalid_rows++;
417 break;
418 }
419 }
420
421 }
422 if (t->num_invalid_rows)
423 NOTICE_LOG("ranges OK. %d invalid row(s) detected\n",
424 t->num_invalid_rows);
425 else
426 INFO_LOG("no invalid rows, no range violations, good\n");
427 return 1;
428 err:
429 return ret;
430 }
431
432 static int move_index_entry(struct osl_table *t, uint32_t dest, uint32_t src)
433 {
434 char *dest_ie, *src_ie;
435 int ret = get_row_index(t, dest, &dest_ie);
436
437 if (ret < 0)
438 return ret;
439 ret = get_row_index(t, src, &src_ie);
440 if (ret < 0)
441 return ret;
442 INFO_LOG("moving entry #%u to position %u\n", src, dest);
443 memcpy(dest_ie, src_ie, t->row_index_size);
444 return 1;
445 }
446
447 static int map_index(const struct osl_table_description *desc, struct osl_object *map)
448 {
449 char *filename = index_filename(desc);
450 int ret;
451
452 ret = mmap_full_file(filename, O_RDWR, &map->data, &map->size, NULL);
453 DEBUG_LOG("mapping index %s: ret: %d, size: %zu\n", filename, ret, map->size);
454 free(filename);
455 return ret;
456 }
457
458 static int prune_invalid_rows_from_index(struct osl_table *t)
459 {
460 uint32_t top = 0, bottom;
461 char *filename;
462 int ret;
463
464 if (!t->num_invalid_rows) {
465 INFO_LOG("all rows are valid, good\n");
466 return 1;
467 }
468 NOTICE_LOG("index contains %u invalid row(s) (%d bytes)\n",
469 t->num_invalid_rows, t->row_index_size * t->num_invalid_rows);
470 if (OPT_GIVEN(DRY_RUN))
471 return 0;
472 NOTICE_LOG("removing invalid rows from index\n");
473 bottom = t->num_rows - 1;
474 while (top < bottom) {
475 if (!row_is_invalid(t, top)) {
476 top++;
477 continue;
478 }
479 while (bottom > top) {
480 if (row_is_invalid(t, bottom)) {
481 bottom--;
482 continue;
483 }
484 /* move bottom index entry to top */
485 move_index_entry(t, top, bottom);
486 bottom--;
487 top++;
488 break;
489 }
490 }
491 DEBUG_LOG("unmapping index\n");
492 osl_munmap(t->index_map.data, t->index_map.size);
493 filename = index_filename(t->desc);
494 ret = truncate_file(filename, t->row_index_size
495 * t->num_invalid_rows);
496 free(filename);
497 if (ret < 0)
498 return ret;
499 ret = map_index(t->desc, &t->index_map);
500 if (ret < 0)
501 return ret;
502 t->num_rows = table_num_rows(t);
503 return 1;
504 }
505
506 static int check_for_invalid_objects(struct osl_table *t, uint32_t **lost_bytes)
507 {
508 int i, j, ret;
509 const struct osl_column_description *cd;
510 uint32_t *loss = fsck_malloc(sizeof(uint32_t) * t->desc->num_columns);
511
512 INFO_LOG("looking for mapped objects not contained in index\n");
513 /* first count used bytes */
514 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
515 loss[i] = t->columns[i].data_map.size;
516 DEBUG_LOG("column %i data map: %zu bytes\n", i,
517 t->columns[i].data_map.size);
518 for (j = 0; j < t->num_rows; j++) {
519 struct osl_object obj;
520 ret = get_mapped_object(t, i, j, &obj);
521 if (ret < 0)
522 goto err;
523 loss[i] -= obj.size;
524 }
525 }
526 ret = 0;
527 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
528 if (loss[i]) {
529 NOTICE_LOG("column %u contains %u lost bytes\n",
530 i, loss[i]);
531 ret = 1;
532 }
533 }
534 if (!ret)
535 INFO_LOG("all mapped objects are valid, good\n");
536 *lost_bytes = loss;
537 return ret;
538 err:
539 free(loss);
540 return ret;
541 }
542
543 /* prune_invalid_rows() must be run on the table before calling this */
544 static int prune_mapped_column(struct osl_table *t, uint32_t col_num, int fd)
545 {
546 int i, ret;
547 uint32_t written = 0;
548 struct osl_column *col = t->columns + col_num;
549
550 INFO_LOG("pruning col %u\n", col_num);
551 for (i = 0; i < t->num_rows; i++) {
552 struct osl_object obj;
553 char *index_entry;
554
555 DEBUG_LOG("checking row %u/%u\n", i, t->num_rows);
556 ret = get_mapped_object(t, col_num, i, &obj);
557 if (ret < 0)
558 return ret;
559 ret = _write_all(fd, (char *)(obj.data), obj.size);
560 if (ret < 0)
561 return ret;
562 written += obj.size;
563 ret = get_row_index(t, i, &index_entry);
564 if (ret < 0)
565 return ret;
566 update_cell_index(index_entry, col, written, obj.size);
567 }
568 return 1;
569 }
570
571 static int prune_objects(struct osl_table *t, uint32_t *lost_bytes)
572 {
573 int i, ret;
574 const struct osl_column_description *cd;
575 char **col_filenames = fsck_calloc(t->desc->num_columns * sizeof(char *));
576 char **new_col_filenames = fsck_calloc(t->desc->num_columns * sizeof(char *));
577 char *idx_filename = index_filename(t->desc);
578 char *old_idx_filename = make_message("%s.bak", idx_filename);
579 int fd;
580
581 NOTICE_LOG("removing unreferenced objects from data files\n");
582 /* first make a copy of the index */
583 DEBUG_LOG("opening %s\n", old_idx_filename);
584 ret = osl_open(old_idx_filename, O_WRONLY | O_CREAT | O_TRUNC, 0644);
585 if (ret < 0)
586 goto out_free;
587 fd = ret;
588 ret = _write_all(fd, t->index_map.data, t->index_map.size);
589 close(fd);
590 if (ret < 0)
591 goto out_free;
592 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
593 if (!lost_bytes[i])
594 continue;
595 col_filenames[i] = column_filename(t, i);
596 new_col_filenames[i] = make_message("%s.fsck", col_filenames[i]);
597 ret = osl_open(new_col_filenames[i], O_WRONLY | O_CREAT | O_EXCL, 0644);
598 if (ret < 0)
599 goto out_unlink_data;
600 fd = ret;
601 ret = prune_mapped_column(t, i, fd);
602 close(fd);
603 if (ret < 0)
604 goto out_unlink_data;
605 }
606 ret = unmap_table(t, OSL_MARK_CLEAN);
607 if (ret < 0)
608 goto out_unlink_data;
609 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
610 if (!lost_bytes[i])
611 continue;
612 ret = osl_rename(new_col_filenames[i], col_filenames[i]);
613 if (ret < 0) { /* we're kinda screwed here */
614 CRIT_LOG("rename of col %i failed: %s\n", i,
615 osl_strerror(errno));
616 goto out_free;
617 }
618 }
619 unlink(old_idx_filename);
620 ret = map_table(t, 0);
621 goto out_free;
622 out_unlink_data:
623 FOR_EACH_MAPPED_COLUMN(i, t, cd)
624 unlink(new_col_filenames[i]);
625 out_free:
626 free(old_idx_filename);
627 free(idx_filename);
628 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
629 free(col_filenames[i]);
630 free(new_col_filenames[i]);
631 }
632 free(col_filenames);
633 free(new_col_filenames);
634 return ret;
635 }
636
637 static struct osl_column_description hash_tree_table_cols[] = {
638 {
639 .storage_type = OSL_NO_STORAGE,
640 .storage_flags = OSL_RBTREE | OSL_FIXED_SIZE | OSL_UNIQUE,
641 .name = "hash",
642 .compare_function = uint32_compare,
643 .data_size = HASH_SIZE
644 },
645 };
646
647 static const struct osl_table_description hash_tree_table_desc = {
648 .dir = "/", /* irrelevant */
649 .name = "hash_tree",
650 .num_columns = 1,
651 .flags = 0,
652 .column_descriptions = hash_tree_table_cols
653 };
654
655 /**
656 * The hash_tree table contains all hashes of the disk storage name column.
657 * of each row. It is used for checking if a disk storage file has a reference
658 * in the table.
659 */
660 static struct osl_table *hash_tree_table;
661 static HASH_TYPE *hashes;
662
663 static int check_disk_storage_column(struct osl_table *t, int row_num,
664 int col_num, char *ds_name, unsigned *num_missing_objects)
665 {
666 int ret;
667 struct stat statbuf;
668 char *path = disk_storage_path(t, col_num, ds_name);
669 unsigned dsnc = t->disk_storage_name_column;
670 struct osl_object obj;
671
672 DEBUG_LOG("checking if %s is a regular file\n", path);
673 ret = stat(path, &statbuf);
674 if (ret < 0 && errno == ENOENT) {
675 struct osl_row *row;
676 (*num_missing_objects)++;
677 ERROR_LOG("row %d: object %s is missing\n", row_num, path);
678 ret = 0;
679 if (OPT_GIVEN(DRY_RUN))
680 goto out;
681 NOTICE_LOG("trying to delete row %d\n", row_num);
682 ret = osl_get_row(t, dsnc, &obj, &row);
683 if (ret < 0) {
684 CRIT_LOG("unable to get row %d\n", row_num);
685 fsck_mark_row_invalid(t, row_num);
686 CRIT_LOG("Please re-run fsck\n");
687 goto out;
688 }
689 ret = osl_del_row(t, row);
690 if (ret < 0)
691 goto out;
692 }
693 out:
694 free(path);
695 if (ret < 0)
696 return ret;
697 ret = -E_FSCK_NOT_A_REGULAR_FILE;
698 if (!(S_IFREG & statbuf.st_mode))
699 return ret;
700 return 1;
701 }
702
703 static int check_disk_storage_presence(struct osl_table *t)
704 {
705 int ret, i, j;
706 struct osl_object obj, hash_obj = {.size = HASH_SIZE};
707 char *ds_name;
708 const struct osl_column_description *cd;
709 unsigned dsnc = t->disk_storage_name_column, missing_objects = 0;
710
711 if (!t->num_rows)
712 return 1;
713 hashes = fsck_malloc(t->num_rows * HASH_SIZE);
714 INFO_LOG("looking for missing disk storage objects\n");
715 for (i = 0; i < t->num_rows; i++) {
716 if (row_is_invalid(t, i))
717 continue;
718 ret = get_mapped_object(t, dsnc, i, &obj);
719 if (ret < 0)
720 return ret;
721 hash_object(&obj, hashes + i * HASH_SIZE);
722 hash_obj.data = hashes + i * HASH_SIZE;
723 osl_add_row(hash_tree_table, &hash_obj);
724 ds_name = disk_storage_name_of_hash(t, hashes + i * HASH_SIZE);
725 FOR_EACH_DISK_STORAGE_COLUMN(j, t, cd) {
726 ret = check_disk_storage_column(t, i, j, ds_name,
727 &missing_objects);
728 if (ret < 0)
729 goto err;
730 }
731 free(ds_name);
732 }
733 if (!missing_objects)
734 INFO_LOG("all referenced disk storage objects exist, good\n");
735 else
736 NOTICE_LOG("%d missing object(s)\n", missing_objects);
737 return missing_objects;
738 err:
739 free(ds_name);
740 return ret;
741 }
742
743 static int dummy_compare(const struct osl_object *obj1, const struct osl_object *obj2)
744 {
745 if (obj1 < obj2)
746 return -1;
747 if (obj1 > obj2)
748 return 1;
749 return 0;
750 }
751
752 static unsigned files_pruned;
753
754 static int prune_disk_storage_file(const char *path, void *private_data)
755 {
756 HASH_TYPE hash[HASH_SIZE];
757 uint8_t flags = *(uint8_t *)private_data;
758 struct osl_object obj = {.data = hash, .size = HASH_SIZE};
759 struct osl_row *row;
760 int ret = -1;
761 size_t len = strlen(path);
762
763
764 DEBUG_LOG("path: %s\n", path);
765 if (flags & OSL_LARGE_TABLE) {
766 if (len < HASH_SIZE * 2 + 2)
767 goto invalid;
768 // NOTICE_LOG("p: %s\n", path + len - 2 * HASH_SIZE - 1);
769 ret = asc_to_hash(path + len - 2 * HASH_SIZE - 1, 1, hash);
770 if (ret < 0)
771 goto invalid;
772 ret = asc_to_hash(path + len - 2 * HASH_SIZE + 2, HASH_SIZE - 1,
773 hash + 1);
774 if (ret < 0)
775 goto invalid;
776 // INFO_LOG("high: %x, low: %x, hash: %x\n", high, low, hash);
777 } else {
778 if (len < 2 * HASH_SIZE + 1)
779 goto invalid;
780 ret = asc_to_hash(path + len - 2 * HASH_SIZE, 2 * HASH_SIZE, hash);
781 if (ret < 0)
782 goto invalid;
783 // INFO_LOG("hash: %x\n", hash);
784 }
785 #if 0
786 {
787 char asc[2 * HASH_SIZE + 1];
788 hash_to_asc(hash, asc);
789 NOTICE_LOG("before: %s\nafter: %s\n", path, asc);
790 }
791 #endif
792 ret = osl_get_row(hash_tree_table, 0, &obj, &row);
793 if (ret >= 0)
794 return 1;
795 NOTICE_LOG("unreferenced file in hash dir: %s\n", path);
796 goto remove;
797 invalid:
798 ERROR_LOG("could not read hash value of %s\n", path);
799 remove:
800 if (OPT_GIVEN(DRY_RUN))
801 return 0;
802 NOTICE_LOG("removing %s\n", path);
803 unlink(path);
804 files_pruned++;
805 return 1;
806 }
807
808 static int prune_disk_storage_files(struct osl_table *t)
809 {
810 int i, ret = 1;
811 const struct osl_column_description *cd;
812
813 INFO_LOG("looking for unreferenced disk storage files\n");
814 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
815 char *dirname = column_filename(t, i);
816 uint8_t flags = t->desc->flags;
817 ret = for_each_file_in_dir(dirname, prune_disk_storage_file,
818 &flags);
819 free(dirname);
820 }
821 if (files_pruned)
822 NOTICE_LOG("%u disk storage files deleted\n",
823 files_pruned);
824 else
825 INFO_LOG("all files are are referenced, good\n");
826 return ret;
827 }
828
829 static int check_disk_storage_columns(struct osl_table *t)
830 {
831 int ret, i;
832 const struct osl_column_description *cd;
833
834 if (!t->num_disk_storage_columns) {
835 INFO_LOG("no disk storage columns in table '%s', "
836 "skipping checks\n", t->desc->name);
837 return 1;
838 }
839 FOR_EACH_COLUMN(i, t->desc, cd)
840 t->desc->column_descriptions[i].compare_function = dummy_compare;
841 ret = init_rbtrees(t);
842 if (ret < 0)
843 return ret;
844 INFO_LOG("creating rbtree for disk storage hash values\n");
845 ret = osl_open_table(&hash_tree_table_desc, &hash_tree_table);
846 if (ret < 0)
847 goto out;
848 ret = check_disk_storage_presence(t);
849 if (ret < 0)
850 goto out_close_hash_tree;
851 ret = prune_disk_storage_files(t);
852 out_close_hash_tree:
853 osl_close_table(hash_tree_table, 0);
854 free(hashes);
855 hashes = NULL;
856 out:
857 clear_rbtrees(t); /* TODO why are we doing that here? Seems odd */
858 return ret;
859 }
860
861 static void set_dummy_contents(struct osl_table_description *desc)
862 {
863 int i;
864 struct osl_column_description *cd;
865
866 for (i = 0; i < desc->num_columns; i++) {
867 cd = get_column_description(desc, i);
868 cd->compare_function = dummy_compare;
869 }
870 }
871
872 static int fsck_init(struct osl_table_description *desc, struct osl_table **t)
873 {
874 struct osl_object map;
875 int ret = map_index(desc, &map);
876
877 if (ret < 0)
878 goto out;
879 ret = read_table_desc(&map, desc);
880 if (ret < 0) {
881 osl_munmap(map.data, map.size);
882 goto out;
883 }
884 set_dummy_contents(desc);
885 ret = init_table_structure(desc, t);
886 if (ret < 0) {
887 osl_munmap(map.data, map.size);
888 goto out;
889 }
890 DEBUG_LOG("unmapping index\n");
891 osl_munmap(map.data, map.size);
892 if (OPT_GIVEN(FORCE))
893 ret = map_table(*t, (MAP_TBL_FL_IGNORE_DIRTY));
894 else
895 ret = map_table(*t, 0);
896 if (ret >= 0) {
897 (*t)->num_rows = table_num_rows(*t);
898 DEBUG_LOG("index header size: %d\n", (*t)->index_header_size);
899 DEBUG_LOG("row index size: %d\n", (*t)->row_index_size);
900 }
901 out:
902 return ret;
903 }
904
905 static void fsck_cleanup(struct osl_table *t)
906 {
907 int i;
908
909 if (!t)
910 return;
911 if (t->desc->column_descriptions) {
912 struct osl_column_description *cd;
913 for (i = 0; i < t->desc->num_columns; i++) {
914 cd = get_column_description(t->desc, i);
915 free((char*)cd->name);
916 }
917 free(t->desc->column_descriptions);
918 }
919 free(t->columns);
920 free(t);
921
922 }
923
924 #define ST_CASE(st) case st: return #st
925
926 static const char *get_asc_storage_type(enum osl_storage_type st)
927 {
928 switch (st) {
929 ST_CASE(OSL_MAPPED_STORAGE);
930 ST_CASE(OSL_DISK_STORAGE);
931 ST_CASE(OSL_NO_STORAGE);
932 }
933 return NULL;
934 }
935
936 #define APPEND_ASC_SF(sf, flag, str) do { if (sf & flag) { \
937 if (str) str = make_message("%s%s", str, " | " # flag); \
938 else str = fsck_strdup(#flag); }} while (0)
939
940
941 static char *get_asc_storage_flags(enum osl_storage_type sf)
942 {
943 char *asc_sf = NULL;
944
945 APPEND_ASC_SF(sf, OSL_RBTREE, asc_sf);
946 APPEND_ASC_SF(sf, OSL_FIXED_SIZE, asc_sf);
947 APPEND_ASC_SF(sf, OSL_UNIQUE, asc_sf);
948 return asc_sf;
949 }
950
951 static int dump_table_desc(struct osl_table *t, int fd)
952 {
953 const struct osl_table_description *desc = t->desc;
954 int ret, i;
955 struct osl_column_description *cd;
956 char *msg = make_message("static struct osl_column_description cols[] = {\n");
957 ret = _write_all(fd, msg, strlen(msg));
958 if (ret < 0)
959 return ret;
960 free(msg);
961 FOR_EACH_COLUMN(i, desc, cd) {
962 const char *asc_st;
963 msg = make_message("\t[%d] = {\n", i);
964 ret = _write_all(fd, msg, strlen(msg));
965 if (ret < 0)
966 return ret;
967 free(msg);
968 asc_st = get_asc_storage_type(cd->storage_type);
969 msg = make_message("\t\t.storage_type = %s,\n", asc_st);
970 ret = _write_all(fd, msg, strlen(msg));
971 if (ret < 0)
972 return ret;
973 free(msg);
974 if (cd->storage_flags) {
975 char *asc_sf = get_asc_storage_flags(cd->storage_flags);
976 msg = make_message("\t\t,storage_flags = %s,\n", asc_sf);
977 free(asc_sf);
978 ret = _write_all(fd, msg, strlen(msg));
979 if (ret < 0)
980 return ret;
981 free(msg);
982 }
983 if (cd->storage_flags & OSL_FIXED_SIZE) {
984 msg = make_message("\t\t.data_size = %u,\n", cd->data_size);
985 ret = _write_all(fd, msg, strlen(msg));
986 if (ret < 0)
987 return ret;
988 free(msg);
989 }
990 msg = make_message("\t\t.name = \"%s\",\n", cd->name);
991 ret = _write_all(fd, msg, strlen(msg));
992 if (ret < 0)
993 return ret;
994 free(msg);
995 if (cd->storage_flags & OSL_RBTREE) {
996 msg = make_message("\t\t.compare_function = compare_func,\n");
997 ret = _write_all(fd, msg, strlen(msg));
998 if (ret < 0)
999 return ret;
1000 free(msg);
1001 }
1002 msg = make_message("\t},\n");
1003 ret = _write_all(fd, msg, strlen(msg));
1004 if (ret < 0)
1005 return ret;
1006 free(msg);
1007 }
1008 msg = make_message("};\n");
1009 ret = _write_all(fd, msg, strlen(msg));
1010 if (ret < 0)
1011 return ret;
1012 free(msg);
1013 return 1;
1014 }
1015
1016 static int dump_row(struct osl_table *t, unsigned row_num, const char *row_dir)
1017 {
1018 int ret, i;
1019 const struct osl_column_description *cd;
1020 unsigned dsnc;
1021 struct osl_object obj;
1022 char *ds_name;
1023 HASH_TYPE hash[HASH_SIZE];
1024 char *filename;
1025
1026 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
1027 ret = get_mapped_object(t, i, row_num, &obj);
1028 if (ret < 0)
1029 return ret;
1030 filename = make_message("%s/col_%03u", row_dir, i);
1031 ret = write_file(filename, obj.data, obj.size);
1032 free(filename);
1033 if (ret < 0)
1034 return ret;
1035 }
1036 if (!t->num_disk_storage_columns)
1037 return 1;
1038 dsnc = t->disk_storage_name_column;
1039 ret = get_mapped_object(t, dsnc, row_num, &obj);
1040 if (ret < 0)
1041 return ret;
1042 hash_object(&obj, hash);
1043 ds_name = disk_storage_name_of_hash(t, hash);
1044 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1045 filename = disk_storage_path(t, i, ds_name);
1046 ret = mmap_full_file(filename, O_RDONLY, &obj.data, &obj.size, NULL);
1047 free(filename);
1048 if (ret < 0)
1049 goto out;
1050 filename = make_message("%s/col_%03u", row_dir, i);
1051 ret = write_file(filename, obj.data, obj.size);
1052 free(filename);
1053 if (ret < 0)
1054 goto out;
1055 }
1056 ret = 1;
1057 out:
1058 free(ds_name);
1059 return ret;
1060 }
1061
1062 static int dump_rows(char *dump_dir, struct osl_table *t)
1063 {
1064 unsigned i;
1065 char *current_dir = NULL;
1066 int ret = 0;
1067
1068 for (i = 0; i < t->num_rows; i++) {
1069 char *row_dir;
1070 if (row_is_invalid(t, i))
1071 continue;
1072 if (!(i % 1000)) {
1073 free(current_dir);
1074 current_dir = make_message("%s/rows_%u-%u", dump_dir, i, i + 999);
1075 NOTICE_LOG("dumping rows %u - %u\n", i, i + 999);
1076 ret = osl_mkdir(current_dir, 0777);
1077 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1078 goto out;
1079 }
1080 row_dir = make_message("%s/row_%03u", current_dir, i);
1081 ret = osl_mkdir(row_dir, 0777);
1082 if (ret < 0 && ret != -E_OSL_DIR_EXISTS) {
1083 free(row_dir);
1084 goto out;
1085 }
1086 ret = dump_row(t, i, row_dir);
1087 free(row_dir);
1088 if (ret < 0)
1089 goto out;
1090 }
1091 out:
1092 free(current_dir);
1093 return ret;
1094 }
1095
1096 static int dump_table(const char *dump_dir, struct osl_table_description *desc)
1097 {
1098 struct osl_table *t = NULL;
1099 int fd, ret = fsck_init(desc, &t);
1100 char *desc_file;
1101 char *table_dump_dir = NULL;
1102
1103 if (ret < 0)
1104 goto out;
1105 ret = osl_mkdir(dump_dir, 0777);
1106 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1107 goto out;
1108 table_dump_dir = make_message("%s/%s", dump_dir, desc->name);
1109 ret = osl_mkdir(table_dump_dir, 0777);
1110 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1111 goto out;
1112 desc_file = make_message("%s/table_description.c", table_dump_dir);
1113 ret = osl_open(desc_file, O_WRONLY | O_CREAT | O_EXCL, 0644);
1114 free(desc_file);
1115 if (ret < 0)
1116 goto out;
1117 fd = ret;
1118 ret = dump_table_desc(t, fd);
1119 close(fd);
1120 if (ret < 0)
1121 goto out;
1122 ret = dump_rows(table_dump_dir, t);
1123 out:
1124 free(table_dump_dir);
1125 fsck_cleanup(t);
1126 return ret;
1127 }
1128
1129 static int fsck(struct osl_table_description *desc)
1130 {
1131 int ret;
1132 struct osl_table *t = NULL;
1133 uint32_t *lost_bytes = NULL;
1134
1135 ret = fsck_init(desc, &t);
1136 if (ret < 0)
1137 goto out;
1138 ret = check_index_ranges(t);
1139 if (ret < 0)
1140 goto out_unmap;
1141 ret = check_disk_storage_columns(t);
1142 if (ret < 0)
1143 goto out_unmap;
1144 ret = prune_invalid_rows_from_index(t);
1145 if (ret < 0)
1146 goto out_unmap;
1147 ret = check_for_invalid_objects(t, &lost_bytes);
1148 if (ret < 0)
1149 goto out_unmap;
1150 if (ret > 0 && !OPT_GIVEN(DRY_RUN)) {
1151 /* at least one mapped data file needs pruning */
1152 ret = prune_objects(t, lost_bytes);
1153 if (ret < 0)
1154 goto out_unmap;
1155 }
1156 free(lost_bytes);
1157 out_unmap:
1158 unmap_table(t, OSL_MARK_CLEAN);
1159 out:
1160 fsck_cleanup(t);
1161 return ret;
1162 }
1163
1164 static int check_table(const char *db_dir, const char *table_name)
1165 {
1166 struct osl_table_description desc = {
1167 .column_descriptions = NULL,
1168 .dir = db_dir,
1169 .name = table_name
1170 };
1171 int ret;
1172
1173 INFO_LOG("checking table %s\n", table_name);
1174 if (!OPT_GIVEN(NO_FSCK)) {
1175 ret = fsck(&desc);
1176 if (ret < 0)
1177 goto out;
1178 }
1179 ret = 1;
1180 if (!OPT_GIVEN(DUMP_DIR) || !*OPT_STRING_VAL(DUMP_DIR))
1181 goto out;
1182 ret = dump_table(OPT_STRING_VAL(DUMP_DIR), &desc);
1183 out:
1184 if (ret < 0)
1185 ERROR_LOG("failed to check table %s: %s\n", table_name,
1186 fsck_strerror(-ret));
1187 else
1188 NOTICE_LOG("successfully checked table %s\n", table_name);
1189 return ret;
1190 }
1191
1192 static int check_all_tables(const char *db_dir)
1193 {
1194 DIR *dir;
1195 struct dirent *entry;
1196 int cwd_fd, ret2, ret = fsck_opendir(db_dir, &dir, &cwd_fd);
1197
1198 if (ret < 0)
1199 return ret;
1200 while ((entry = readdir(dir))) {
1201 mode_t m;
1202 struct stat s;
1203 if (!strcmp(entry->d_name, "."))
1204 continue;
1205 if (!strcmp(entry->d_name, ".."))
1206 continue;
1207 if (lstat(entry->d_name, &s) == -1)
1208 continue;
1209 m = s.st_mode;
1210 if (!S_ISDIR(m))
1211 continue;
1212 ret = check_table(db_dir, entry->d_name);
1213 if (ret < 0)
1214 break;
1215 }
1216 closedir(dir);
1217 ret2 = __fchdir(cwd_fd);
1218 if (ret2 < 0 && ret >= 0)
1219 ret = ret2;
1220 close(cwd_fd);
1221 return ret;
1222 }
1223
1224 int main(int argc, char **argv)
1225 {
1226 int i, ret;
1227 char *errctx = NULL;
1228 const char *dd;
1229
1230 loglevel = 0;
1231 ret = lls_parse(argc, argv, CMD_PTR, &lpr, &errctx);
1232 if (ret < 0) {
1233 if (errctx)
1234 ERROR_LOG("%s\n", errctx);
1235 EMERG_LOG("%s\n", lls_strerror(-ret));
1236 exit(EXIT_FAILURE);
1237 }
1238 loglevel = OPT_UINT32_VAL(LOGLEVEL);
1239 if (OPT_GIVEN(DETAILED_HELP)) {
1240 printf("%s\n", lls_long_help(CMD_PTR));
1241 exit(EXIT_SUCCESS);
1242 }
1243 if (OPT_GIVEN(HELP)) {
1244 printf("%s\n", lls_short_help(CMD_PTR));
1245 exit(EXIT_SUCCESS);
1246 }
1247 if (OPT_GIVEN(VERSION)) {
1248 printf(
1249 "oslfsck " OSL_VERSION "\n"
1250 "Copyright (C) 2008-2009 Andre Noll\n"
1251 "This is free software with ABSOLUTELY NO WARRANTY."
1252 " See COPYING for details.\n"
1253 "Written by Andre Noll.\n"
1254 "Report bugs to <maan@tuebingen.mpg.de>.\n"
1255 );
1256 exit(EXIT_SUCCESS);
1257 }
1258 dd = OPT_STRING_VAL(DATABASE_DIR);
1259 if (!dd) {
1260 EMERG_LOG("--database-dir: mandatory option not given\n");
1261 exit(EXIT_FAILURE);
1262 }
1263 INFO_LOG("database dir: %s\n", dd);
1264 if (lls_num_inputs(lpr) == 0) {
1265 ret = check_all_tables(dd);
1266 goto out;
1267 }
1268 for (i = 0; i < lls_num_inputs(lpr); i++) {
1269 ret = check_table(dd, lls_input(i, lpr));
1270 if (ret < 0)
1271 break;
1272 }
1273 out:
1274 if (ret < 0) {
1275 ERROR_LOG("%s\n", fsck_strerror(-ret));
1276 if (loglevel > 1)
1277 EMERG_LOG("re-run with \"--loglevel %d\" to increase verbosity\n",
1278 loglevel - 1);
1279 } else
1280 NOTICE_LOG("success\n");
1281 return ret < 0? EXIT_FAILURE : EXIT_SUCCESS;
1282 }