source3/lib/tdb_validate.c

   1 /*
   2  * Unix SMB/CIFS implementation.
   3  *
   4  * A general tdb content validation mechanism
   5  *
   6  * Copyright (C) Michael Adam      2007
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; either version 3 of the License, or
  11  * (at your option) any later version.
  12  *
  13  * This program is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  20  */
  21
  22 #include "includes.h"
  23 #include "system/filesys.h"
  24 #include "tdb_validate.h"
  25
  26 /*
  27  * internal validation function, executed by the child.
  28  */
  29 static int tdb_validate_child(struct tdb_context *tdb,
  30                               tdb_validate_data_func validate_fn)
  31 {
  32         int ret = 1;
  33         int num_entries = 0;
  34         struct tdb_validation_status v_status;
  35
  36         v_status.tdb_error = False;
  37         v_status.bad_freelist = False;
  38         v_status.bad_entry = False;
  39         v_status.unknown_key = False;
  40         v_status.success = True;
  41
  42         if (!tdb) {
  43                 v_status.tdb_error = True;
  44                 v_status.success = False;
  45                 goto out;
  46         }
  47
  48         /*
  49          * we can simplify this by passing a check function,
  50          * but I don't want to change all the callers...
  51          */
  52         ret = tdb_check(tdb, NULL, NULL);
  53         if (ret == -1) {
  54                 v_status.tdb_error = True;
  55                 v_status.success = False;
  56                 goto out;
  57         }
  58
  59         /* Check if the tdb's freelist is good. */
  60         if (tdb_validate_freelist(tdb, &num_entries) == -1) {
  61                 v_status.bad_freelist = True;
  62                 v_status.success = False;
  63                 goto out;
  64         }
  65
  66         DEBUG(10,("tdb_validate_child: tdb %s freelist has %d entries\n",
  67                   tdb_name(tdb), num_entries));
  68
  69         /* Now traverse the tdb to validate it. */
  70         num_entries = tdb_traverse(tdb, validate_fn, (void *)&v_status);
  71         if (!v_status.success) {
  72                 goto out;
  73         } else if (num_entries == -1) {
  74                 v_status.tdb_error = True;
  75                 v_status.success = False;
  76                 goto out;
  77         }
  78
  79         DEBUG(10,("tdb_validate_child: tdb %s is good with %d entries\n",
  80                   tdb_name(tdb), num_entries));
  81         ret = 0; /* Cache is good. */
  82
  83 out:
  84         DEBUG(10,   ("tdb_validate_child: summary of validation status:\n"));
  85         DEBUGADD(10,(" * tdb error: %s\n", v_status.tdb_error ? "yes" : "no"));
  86         DEBUGADD(10,(" * bad freelist: %s\n",v_status.bad_freelist?"yes":"no"));
  87         DEBUGADD(10,(" * bad entry: %s\n", v_status.bad_entry ? "yes" : "no"));
  88         DEBUGADD(10,(" * unknown key: %s\n", v_status.unknown_key?"yes":"no"));
  89         DEBUGADD(10,(" => overall success: %s\n", v_status.success?"yes":"no"));
  90
  91         return ret;
  92 }
  93
  94 /*
  95  * tdb validation function.
  96  * returns 0 if tdb is ok, != 0 if it isn't.
  97  * this function expects an opened tdb.
  98  */
  99 int tdb_validate(struct tdb_context *tdb, tdb_validate_data_func validate_fn)
 100 {
 101         pid_t child_pid = -1;
 102         int child_status = 0;
 103         int wait_pid = 0;
 104         int ret = 1;
 105
 106         if (tdb == NULL) {
 107                 DEBUG(1, ("Error: tdb_validate called with tdb == NULL\n"));
 108                 return ret;
 109         }
 110
 111         DEBUG(5, ("tdb_validate called for tdb '%s'\n", tdb_name(tdb)));
 112
 113         /* fork and let the child do the validation.
 114          * benefit: no need to twist signal handlers and panic functions.
 115          * just let the child panic. we catch the signal. */
 116
 117         DEBUG(10, ("tdb_validate: forking to let child do validation.\n"));
 118         child_pid = sys_fork();
 119         if (child_pid == 0) {
 120                 /* child code */
 121                 DEBUG(10, ("tdb_validate (validation child): created\n"));
 122                 DEBUG(10, ("tdb_validate (validation child): "
 123                            "calling tdb_validate_child\n"));
 124                 exit(tdb_validate_child(tdb, validate_fn));
 125         }
 126         else if (child_pid < 0) {
 127                 DEBUG(1, ("tdb_validate: fork for validation failed.\n"));
 128                 goto done;
 129         }
 130
 131         /* parent */
 132
 133         DEBUG(10, ("tdb_validate: fork succeeded, child PID = %u\n",
 134                 (unsigned int)child_pid));
 135
 136         DEBUG(10, ("tdb_validate: waiting for child to finish...\n"));
 137         while  ((wait_pid = sys_waitpid(child_pid, &child_status, 0)) < 0) {
 138                 if (errno == EINTR) {
 139                         DEBUG(10, ("tdb_validate: got signal during waitpid, "
 140                                    "retrying\n"));
 141                         errno = 0;
 142                         continue;
 143                 }
 144                 DEBUG(1, ("tdb_validate: waitpid failed with error '%s'.\n",
 145                           strerror(errno)));
 146                 goto done;
 147         }
 148         if (wait_pid != child_pid) {
 149                 DEBUG(1, ("tdb_validate: waitpid returned pid %d, "
 150                           "but %u was expected\n", wait_pid, (unsigned int)child_pid));
 151                 goto done;
 152         }
 153
 154         DEBUG(10, ("tdb_validate: validating child returned.\n"));
 155         if (WIFEXITED(child_status)) {
 156                 DEBUG(10, ("tdb_validate: child exited, code %d.\n",
 157                            WEXITSTATUS(child_status)));
 158                 ret = WEXITSTATUS(child_status);
 159         }
 160         if (WIFSIGNALED(child_status)) {
 161                 DEBUG(10, ("tdb_validate: child terminated by signal %d\n",
 162                            WTERMSIG(child_status)));
 163 #ifdef WCOREDUMP
 164                 if (WCOREDUMP(child_status)) {
 165                         DEBUGADD(10, ("core dumped\n"));
 166                 }
 167 #endif
 168                 ret = WTERMSIG(child_status);
 169         }
 170         if (WIFSTOPPED(child_status)) {
 171                 DEBUG(10, ("tdb_validate: child was stopped by signal %d\n",
 172                            WSTOPSIG(child_status)));
 173                 ret = WSTOPSIG(child_status);
 174         }
 175
 176 done:
 177         DEBUG(5, ("tdb_validate returning code '%d' for tdb '%s'\n", ret,
 178                   tdb_name(tdb)));
 179
 180         return ret;
 181 }
 182
 183 /*
 184  * tdb validation function.
 185  * returns 0 if tdb is ok, != 0 if it isn't.
 186  * this is a wrapper around the actual validation function that opens and closes
 187  * the tdb.
 188  */
 189 int tdb_validate_open(const char *tdb_path, tdb_validate_data_func validate_fn)
 190 {
 191         TDB_CONTEXT *tdb = NULL;
 192         int ret = 1;
 193
 194         DEBUG(5, ("tdb_validate_open called for tdb '%s'\n", tdb_path));
 195
 196         tdb = tdb_open_log(tdb_path, 0, TDB_DEFAULT, O_RDWR, 0);
 197         if (!tdb) {
 198                 DEBUG(1, ("Error opening tdb %s\n", tdb_path));
 199                 return ret;
 200         }
 201
 202         ret = tdb_validate(tdb, validate_fn);
 203         tdb_close(tdb);
 204         return ret;
 205 }
 206
 207 /*
 208  * tdb backup function and helpers for tdb_validate wrapper with backup
 209  * handling.
 210  */
 211
 212 /* this structure eliminates the need for a global overall status for
 213  * the traverse-copy */
 214 struct tdb_copy_data {
 215         struct tdb_context *dst;
 216         bool success;
 217 };
 218
 219 static int traverse_copy_fn(struct tdb_context *tdb, TDB_DATA key,
 220                             TDB_DATA dbuf, void *private_data)
 221 {
 222         struct tdb_copy_data *data = (struct tdb_copy_data *)private_data;
 223
 224         if (tdb_store(data->dst, key, dbuf, TDB_INSERT) != 0) {
 225                 DEBUG(4, ("Failed to insert into %s: %s\n", tdb_name(data->dst),
 226                           strerror(errno)));
 227                 data->success = False;
 228                 return 1;
 229         }
 230         return 0;
 231 }
 232
 233 static int tdb_copy(struct tdb_context *src, struct tdb_context *dst)
 234 {
 235         struct tdb_copy_data data;
 236         int count;
 237
 238         data.dst = dst;
 239         data.success = True;
 240
 241         count = tdb_traverse(src, traverse_copy_fn, (void *)(&data));
 242         if ((count < 0) || (data.success == False)) {
 243                 return -1;
 244         }
 245         return count;
 246 }
 247
 248 static int tdb_verify_basic(struct tdb_context *tdb)
 249 {
 250         return tdb_traverse(tdb, NULL, NULL);
 251 }
 252
 253 /* this backup function is essentially taken from lib/tdb/tools/tdbbackup.tdb
 254  */
 255 static int tdb_backup(TALLOC_CTX *ctx, const char *src_path,
 256                       const char *dst_path, int hash_size)
 257 {
 258         struct tdb_context *src_tdb = NULL;
 259         struct tdb_context *dst_tdb = NULL;
 260         char *tmp_path = NULL;
 261         struct stat st;
 262         int count1, count2;
 263         int saved_errno = 0;
 264         int ret = -1;
 265
 266         if (stat(src_path, &st) != 0) {
 267                 DEBUG(3, ("Could not stat '%s': %s\n", src_path,
 268                           strerror(errno)));
 269                 goto done;
 270         }
 271
 272         /* open old tdb RDWR - so we can lock it */
 273         src_tdb = tdb_open_log(src_path, 0, TDB_DEFAULT, O_RDWR, 0);
 274         if (src_tdb == NULL) {
 275                 DEBUG(3, ("Failed to open tdb '%s'\n", src_path));
 276                 goto done;
 277         }
 278
 279         if (tdb_lockall(src_tdb) != 0) {
 280                 DEBUG(3, ("Failed to lock tdb '%s'\n", src_path));
 281                 goto done;
 282         }
 283
 284         tmp_path = talloc_asprintf(ctx, "%s%s", dst_path, ".tmp");
 285         if (!tmp_path) {
 286                 DEBUG(3, ("talloc fail\n"));
 287                 goto done;
 288         }
 289
 290         unlink(tmp_path);
 291         dst_tdb = tdb_open_log(tmp_path,
 292                                hash_size ? hash_size : tdb_hash_size(src_tdb),
 293                                TDB_DEFAULT, O_RDWR | O_CREAT | O_EXCL,
 294                                st.st_mode & 0777);
 295         if (dst_tdb == NULL) {
 296                 DEBUG(3, ("Error creating tdb '%s': %s\n", tmp_path,
 297                           strerror(errno)));
 298                 saved_errno = errno;
 299                 unlink(tmp_path);
 300                 goto done;
 301         }
 302
 303         count1 = tdb_copy(src_tdb, dst_tdb);
 304         if (count1 < 0) {
 305                 DEBUG(3, ("Failed to copy tdb '%s': %s\n", src_path,
 306                           strerror(errno)));
 307                 tdb_close(dst_tdb);
 308                 goto done;
 309         }
 310
 311         /* reopen ro and do basic verification */
 312         tdb_close(dst_tdb);
 313         dst_tdb = tdb_open_log(tmp_path, 0, TDB_DEFAULT, O_RDONLY, 0);
 314         if (!dst_tdb) {
 315                 DEBUG(3, ("Failed to reopen tdb '%s': %s\n", tmp_path,
 316                           strerror(errno)));
 317                 goto done;
 318         }
 319         count2 = tdb_verify_basic(dst_tdb);
 320         if (count2 != count1) {
 321                 DEBUG(3, ("Failed to verify result of copying tdb '%s'.\n",
 322                           src_path));
 323                 tdb_close(dst_tdb);
 324                 goto done;
 325         }
 326
 327         DEBUG(10, ("tdb_backup: successfully copied %d entries\n", count1));
 328
 329         /* make sure the new tdb has reached stable storage
 330          * then rename it to its destination */
 331         fsync(tdb_fd(dst_tdb));
 332         tdb_close(dst_tdb);
 333         unlink(dst_path);
 334         if (rename(tmp_path, dst_path) != 0) {
 335                 DEBUG(3, ("Failed to rename '%s' to '%s': %s\n",
 336                           tmp_path, dst_path, strerror(errno)));
 337                 goto done;
 338         }
 339
 340         /* success */
 341         ret = 0;
 342
 343 done:
 344         if (src_tdb != NULL) {
 345                 tdb_close(src_tdb);
 346         }
 347         if (tmp_path != NULL) {
 348                 unlink(tmp_path);
 349                 TALLOC_FREE(tmp_path);
 350         }
 351         if (saved_errno != 0) {
 352                 errno = saved_errno;
 353         }
 354         return ret;
 355 }
 356
 357 static int rename_file_with_suffix(TALLOC_CTX *ctx, const char *path,
 358                                    const char *suffix)
 359 {
 360         int ret = -1;
 361         char *dst_path;
 362
 363         dst_path = talloc_asprintf(ctx, "%s%s", path, suffix);
 364         if (dst_path == NULL) {
 365                 DEBUG(3, ("error out of memory\n"));
 366                 return ret;
 367         }
 368
 369         ret = (rename(path, dst_path) != 0);
 370
 371         if (ret == 0) {
 372                 DEBUG(5, ("moved '%s' to '%s'\n", path, dst_path));
 373         } else if (errno == ENOENT) {
 374                 DEBUG(3, ("file '%s' does not exist - so not moved\n", path));
 375                 ret = 0;
 376         } else {
 377                 DEBUG(3, ("error renaming %s to %s: %s\n", path, dst_path,
 378                           strerror(errno)));
 379         }
 380
 381         TALLOC_FREE(dst_path);
 382         return ret;
 383 }
 384
 385 /*
 386  * do a backup of a tdb, moving the destination out of the way first
 387  */
 388 static int tdb_backup_with_rotate(TALLOC_CTX *ctx, const char *src_path,
 389                                   const char *dst_path, int hash_size,
 390                                   const char *rotate_suffix,
 391                                   bool retry_norotate_if_nospc,
 392                                   bool rename_as_last_resort_if_nospc)
 393 {
 394         int ret;
 395
 396         rename_file_with_suffix(ctx, dst_path, rotate_suffix);
 397
 398         ret = tdb_backup(ctx, src_path, dst_path, hash_size);
 399
 400         if (ret != 0) {
 401                 DEBUG(10, ("backup of %s failed: %s\n", src_path, strerror(errno)));
 402         }
 403         if ((ret != 0) && (errno == ENOSPC) && retry_norotate_if_nospc)
 404         {
 405                 char *rotate_path = talloc_asprintf(ctx, "%s%s", dst_path,
 406                                                     rotate_suffix);
 407                 if (rotate_path == NULL) {
 408                         DEBUG(10, ("talloc fail\n"));
 409                         return -1;
 410                 }
 411                 DEBUG(10, ("backup of %s failed due to lack of space\n",
 412                            src_path));
 413                 DEBUGADD(10, ("trying to free some space by removing rotated "
 414                               "dst %s\n", rotate_path));
 415                 if (unlink(rotate_path) == -1) {
 416                         DEBUG(10, ("unlink of %s failed: %s\n", rotate_path,
 417                                    strerror(errno)));
 418                 } else {
 419                         ret = tdb_backup(ctx, src_path, dst_path, hash_size);
 420                 }
 421                 TALLOC_FREE(rotate_path);
 422         }
 423
 424         if ((ret != 0) && (errno == ENOSPC) && rename_as_last_resort_if_nospc)
 425         {
 426                 DEBUG(10, ("backup of %s failed due to lack of space\n",
 427                            src_path));
 428                 DEBUGADD(10, ("using 'rename' as a last resort\n"));
 429                 ret = rename(src_path, dst_path);
 430         }
 431
 432         return ret;
 433 }
 434
 435 /*
 436  * validation function with backup handling:
 437  *
 438  *  - calls tdb_validate
 439  *  - if the tdb is ok, create a backup "name.bak", possibly moving
 440  *    existing backup to name.bak.old,
 441  *    return 0 (success) even if the backup fails
 442  *  - if the tdb is corrupt:
 443  *    - move the tdb to "name.corrupt"
 444  *    - check if there is valid backup.
 445  *      if so, restore the backup.
 446  *      if restore is successful, return 0 (success),
 447  *    - otherwise return -1 (failure)
 448  */
 449 int tdb_validate_and_backup(const char *tdb_path,
 450                             tdb_validate_data_func validate_fn)
 451 {
 452         int ret = -1;
 453         const char *backup_suffix = ".bak";
 454         const char *corrupt_suffix = ".corrupt";
 455         const char *rotate_suffix = ".old";
 456         char *tdb_path_backup;
 457         struct stat st;
 458         TALLOC_CTX *ctx = NULL;
 459
 460         ctx = talloc_new(NULL);
 461         if (ctx == NULL) {
 462                 DEBUG(0, ("tdb_validate_and_backup: out of memory\n"));
 463                 goto done;
 464         }
 465
 466         tdb_path_backup = talloc_asprintf(ctx, "%s%s", tdb_path, backup_suffix);
 467         if (!tdb_path_backup) {
 468                 DEBUG(0, ("tdb_validate_and_backup: out of memory\n"));
 469                 goto done;
 470         }
 471
 472         ret = tdb_validate_open(tdb_path, validate_fn);
 473
 474         if (ret == 0) {
 475                 DEBUG(1, ("tdb '%s' is valid\n", tdb_path));
 476                 ret = tdb_backup_with_rotate(ctx, tdb_path, tdb_path_backup, 0,
 477                                              rotate_suffix, True, False);
 478                 if (ret != 0) {
 479                         DEBUG(1, ("Error creating backup of tdb '%s'\n",
 480                                   tdb_path));
 481                         /* the actual validation was successful: */
 482                         ret = 0;
 483                 } else {
 484                         DEBUG(1, ("Created backup '%s' of tdb '%s'\n",
 485                                   tdb_path_backup, tdb_path));
 486                 }
 487         } else {
 488                 DEBUG(1, ("tdb '%s' is invalid\n", tdb_path));
 489
 490                 ret =stat(tdb_path_backup, &st);
 491                 if (ret != 0) {
 492                         DEBUG(5, ("Could not stat '%s': %s\n", tdb_path_backup,
 493                                   strerror(errno)));
 494                         DEBUG(1, ("No backup found.\n"));
 495                 } else {
 496                         DEBUG(1, ("backup '%s' found.\n", tdb_path_backup));
 497                         ret = tdb_validate_open(tdb_path_backup, validate_fn);
 498                         if (ret != 0) {
 499                                 DEBUG(1, ("Backup '%s' is invalid.\n",
 500                                           tdb_path_backup));
 501                         }
 502                 }
 503
 504                 if (ret != 0) {
 505                         int renamed = rename_file_with_suffix(ctx, tdb_path,
 506                                                               corrupt_suffix);
 507                         if (renamed != 0) {
 508                                 DEBUG(1, ("Error moving tdb to '%s%s'\n",
 509                                           tdb_path, corrupt_suffix));
 510                         } else {
 511                                 DEBUG(1, ("Corrupt tdb stored as '%s%s'\n",
 512                                           tdb_path, corrupt_suffix));
 513                         }
 514                         goto done;
 515                 }
 516
 517                 DEBUG(1, ("valid backup '%s' found\n", tdb_path_backup));
 518                 ret = tdb_backup_with_rotate(ctx, tdb_path_backup, tdb_path, 0,
 519                                              corrupt_suffix, True, True);
 520                 if (ret != 0) {
 521                         DEBUG(1, ("Error restoring backup from '%s'\n",
 522                                   tdb_path_backup));
 523                 } else {
 524                         DEBUG(1, ("Restored tdb backup from '%s'\n",
 525                                   tdb_path_backup));
 526                 }
 527         }
 528
 529 done:
 530         TALLOC_FREE(ctx);
 531         return ret;
 532 }