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