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 "util_tdb.h"
  25 #include "tdb_validate.h"
  26
  27 /*
  28  * internal validation function, executed by the child.
  29  */
  30 static int tdb_validate_child(struct tdb_context *tdb,
  31                               tdb_validate_data_func validate_fn)
  32 {
  33         int ret = 1;
  34         int check_rc;
  35         int num_entries = 0;
  36         struct tdb_validation_status v_status;
  37
  38         v_status.tdb_error = False;
  39         v_status.bad_freelist = False;
  40         v_status.bad_entry = False;
  41         v_status.unknown_key = False;
  42         v_status.success = True;
  43
  44         if (!tdb) {
  45                 v_status.tdb_error = True;
  46                 v_status.success = False;
  47                 goto out;
  48         }
  49
  50         /*
  51          * we can simplify this by passing a check function,
  52          * but I don't want to change all the callers...
  53          */
  54         check_rc = tdb_check(tdb, NULL, NULL);
  55         if (check_rc != 0) {
  56                 v_status.tdb_error = True;
  57                 v_status.success = False;
  58                 goto out;
  59         }
  60
  61         /* Check if the tdb's freelist is good. */
  62         if (tdb_validate_freelist(tdb, &num_entries) == -1) {
  63                 v_status.bad_freelist = True;
  64                 v_status.success = False;
  65                 goto out;
  66         }
  67
  68         DEBUG(10,("tdb_validate_child: tdb %s freelist has %d entries\n",
  69                   tdb_name(tdb), num_entries));
  70
  71         /* Now traverse the tdb to validate it. */
  72         num_entries = tdb_traverse(tdb, validate_fn, (void *)&v_status);
  73         if (!v_status.success) {
  74                 goto out;
  75         } else if (num_entries < 0) {
  76                 v_status.tdb_error = True;
  77                 v_status.success = False;
  78                 goto out;
  79         }
  80
  81         DEBUG(10,("tdb_validate_child: tdb %s is good with %d entries\n",
  82                   tdb_name(tdb), num_entries));
  83         ret = 0; /* Cache is good. */
  84
  85 out:
  86         DBG_DEBUG("summary of validation status:\n"
  87                   " * tdb error: %s\n"
  88                   " * bad freelist: %s\n"
  89                   " * bad entry: %s\n"
  90                   " * unknown key: %s\n"
  91                   " => overall success: %s\n",
  92                   v_status.tdb_error ? "yes" : "no",
  93                   v_status.bad_freelist ? "yes" : "no",
  94                   v_status.bad_entry ? "yes" : "no",
  95                   v_status.unknown_key ? "yes" : "no",
  96                   v_status.success ? "yes" : "no");
  97
  98         return ret;
  99 }
 100
 101 /*
 102  * tdb validation function.
 103  * returns 0 if tdb is ok, != 0 if it isn't.
 104  * this function expects an opened tdb.
 105  */
 106 int tdb_validate(struct tdb_context *tdb, tdb_validate_data_func validate_fn)
 107 {
 108         pid_t child_pid = -1;
 109         int child_status = 0;
 110         int wait_pid = 0;
 111         int ret = 1;
 112
 113         if (tdb == NULL) {
 114                 DEBUG(1, ("Error: tdb_validate called with tdb == NULL\n"));
 115                 return ret;
 116         }
 117
 118         DEBUG(5, ("tdb_validate called for tdb '%s'\n", tdb_name(tdb)));
 119
 120         /* fork and let the child do the validation.
 121          * benefit: no need to twist signal handlers and panic functions.
 122          * just let the child panic. we catch the signal. */
 123
 124         DEBUG(10, ("tdb_validate: forking to let child do validation.\n"));
 125         child_pid = fork();
 126         if (child_pid == 0) {
 127                 /* child code */
 128                 DEBUG(10, ("tdb_validate (validation child): created\n"));
 129                 DEBUG(10, ("tdb_validate (validation child): "
 130                            "calling tdb_validate_child\n"));
 131                 exit(tdb_validate_child(tdb, validate_fn));
 132         }
 133         else if (child_pid < 0) {
 134                 DEBUG(1, ("tdb_validate: fork for validation failed.\n"));
 135                 goto done;
 136         }
 137
 138         /* parent */
 139
 140         DEBUG(10, ("tdb_validate: fork succeeded, child PID = %u\n",
 141                 (unsigned int)child_pid));
 142
 143         DEBUG(10, ("tdb_validate: waiting for child to finish...\n"));
 144         while  ((wait_pid = waitpid(child_pid, &child_status, 0)) < 0) {
 145                 if (errno == EINTR) {
 146                         DEBUG(10, ("tdb_validate: got signal during waitpid, "
 147                                    "retrying\n"));
 148                         errno = 0;
 149                         continue;
 150                 }
 151                 DEBUG(1, ("tdb_validate: waitpid failed with error '%s'.\n",
 152                           strerror(errno)));
 153                 goto done;
 154         }
 155         if (wait_pid != child_pid) {
 156                 DEBUG(1, ("tdb_validate: waitpid returned pid %d, "
 157                           "but %u was expected\n", wait_pid, (unsigned int)child_pid));
 158                 goto done;
 159         }
 160
 161         DEBUG(10, ("tdb_validate: validating child returned.\n"));
 162         if (WIFEXITED(child_status)) {
 163                 DEBUG(10, ("tdb_validate: child exited, code %d.\n",
 164                            WEXITSTATUS(child_status)));
 165                 ret = WEXITSTATUS(child_status);
 166         }
 167         if (WIFSIGNALED(child_status)) {
 168                 DEBUG(10, ("tdb_validate: child terminated by signal %d\n",
 169                            WTERMSIG(child_status)));
 170 #ifdef WCOREDUMP
 171                 if (WCOREDUMP(child_status)) {
 172                         DEBUGADD(10, ("core dumped\n"));
 173                 }
 174 #endif
 175                 ret = WTERMSIG(child_status);
 176         }
 177         if (WIFSTOPPED(child_status)) {
 178                 DEBUG(10, ("tdb_validate: child was stopped by signal %d\n",
 179                            WSTOPSIG(child_status)));
 180                 ret = WSTOPSIG(child_status);
 181         }
 182
 183 done:
 184         DEBUG(5, ("tdb_validate returning code '%d' for tdb '%s'\n", ret,
 185                   tdb_name(tdb)));
 186
 187         return ret;
 188 }
 189
 190 /*
 191  * tdb validation function.
 192  * returns 0 if tdb is ok, != 0 if it isn't.
 193  * this is a wrapper around the actual validation function that opens and closes
 194  * the tdb.
 195  */
 196 int tdb_validate_open(const char *tdb_path, tdb_validate_data_func validate_fn)
 197 {
 198         TDB_CONTEXT *tdb = NULL;
 199         int ret = 1;
 200
 201         DEBUG(5, ("tdb_validate_open called for tdb '%s'\n", tdb_path));
 202
 203         tdb = tdb_open_log(tdb_path, 0, TDB_DEFAULT, O_RDWR, 0);
 204         if (!tdb) {
 205                 DEBUG(1, ("Error opening tdb %s\n", tdb_path));
 206                 return ret;
 207         }
 208
 209         ret = tdb_validate(tdb, validate_fn);
 210         tdb_close(tdb);
 211         return ret;
 212 }
 213
 214 /*
 215  * tdb backup function and helpers for tdb_validate wrapper with backup
 216  * handling.
 217  */
 218
 219 /* this structure eliminates the need for a global overall status for
 220  * the traverse-copy */
 221 struct tdb_copy_data {
 222         struct tdb_context *dst;
 223         bool success;
 224 };
 225
 226 static int traverse_copy_fn(struct tdb_context *tdb, TDB_DATA key,
 227                             TDB_DATA dbuf, void *private_data)
 228 {
 229         struct tdb_copy_data *data = (struct tdb_copy_data *)private_data;
 230
 231         if (tdb_store(data->dst, key, dbuf, TDB_INSERT) != 0) {
 232                 DEBUG(4, ("Failed to insert into %s: %s\n", tdb_name(data->dst),
 233                           strerror(errno)));
 234                 data->success = False;
 235                 return 1;
 236         }
 237         return 0;
 238 }
 239
 240 static int tdb_copy(struct tdb_context *src, struct tdb_context *dst)
 241 {
 242         struct tdb_copy_data data;
 243         int count;
 244
 245         data.dst = dst;
 246         data.success = True;
 247
 248         count = tdb_traverse(src, traverse_copy_fn, (void *)(&data));
 249         if ((count < 0) || (data.success == False)) {
 250                 return -1;
 251         }
 252         return count;
 253 }
 254
 255 static int tdb_verify_basic(struct tdb_context *tdb)
 256 {
 257         return tdb_traverse(tdb, NULL, NULL);
 258 }
 259
 260 /* this backup function is essentially taken from lib/tdb/tools/tdbbackup.tdb
 261  */
 262 static int tdb_backup(TALLOC_CTX *ctx, const char *src_path,
 263                       const char *dst_path, int hash_size)
 264 {
 265         struct tdb_context *src_tdb = NULL;
 266         struct tdb_context *dst_tdb = NULL;
 267         char *tmp_path = NULL;
 268         struct stat st;
 269         int count1, count2;
 270         int saved_errno = 0;
 271         int ret = -1;
 272
 273         if (stat(src_path, &st) != 0) {
 274                 DEBUG(3, ("Could not stat '%s': %s\n", src_path,
 275                           strerror(errno)));
 276                 goto done;
 277         }
 278
 279         /* open old tdb RDWR - so we can lock it */
 280         src_tdb = tdb_open_log(src_path, 0, TDB_DEFAULT, O_RDWR, 0);
 281         if (src_tdb == NULL) {
 282                 DEBUG(3, ("Failed to open tdb '%s'\n", src_path));
 283                 goto done;
 284         }
 285
 286         if (tdb_lockall(src_tdb) != 0) {
 287                 DEBUG(3, ("Failed to lock tdb '%s'\n", src_path));
 288                 goto done;
 289         }
 290
 291         tmp_path = talloc_asprintf(ctx, "%s%s", dst_path, ".tmp");
 292         if (!tmp_path) {
 293                 DEBUG(3, ("talloc fail\n"));
 294                 goto done;
 295         }
 296
 297         unlink(tmp_path);
 298
 299         if (!hash_size) {
 300                 hash_size = tdb_hash_size(src_tdb);
 301         }
 302
 303         dst_tdb = tdb_open_log(tmp_path, hash_size,
 304                                TDB_DEFAULT, O_RDWR | O_CREAT | O_EXCL,
 305                                st.st_mode & 0777);
 306         if (dst_tdb == NULL) {
 307                 DEBUG(3, ("Error creating tdb '%s': %s\n", tmp_path,
 308                           strerror(errno)));
 309                 saved_errno = errno;
 310                 unlink(tmp_path);
 311                 goto done;
 312         }
 313
 314         count1 = tdb_copy(src_tdb, dst_tdb);
 315         if (count1 < 0) {
 316                 DEBUG(3, ("Failed to copy tdb '%s': %s\n", src_path,
 317                           strerror(errno)));
 318                 tdb_close(dst_tdb);
 319                 goto done;
 320         }
 321
 322         /* reopen ro and do basic verification */
 323         tdb_close(dst_tdb);
 324         dst_tdb = tdb_open_log(tmp_path, 0, TDB_DEFAULT, O_RDONLY, 0);
 325         if (!dst_tdb) {
 326                 DEBUG(3, ("Failed to reopen tdb '%s': %s\n", tmp_path,
 327                           strerror(errno)));
 328                 goto done;
 329         }
 330         count2 = tdb_verify_basic(dst_tdb);
 331         if (count2 != count1) {
 332                 DEBUG(3, ("Failed to verify result of copying tdb '%s'.\n",
 333                           src_path));
 334                 tdb_close(dst_tdb);
 335                 goto done;
 336         }
 337
 338         DEBUG(10, ("tdb_backup: successfully copied %d entries\n", count1));
 339
 340         /* make sure the new tdb has reached stable storage
 341          * then rename it to its destination */
 342         fsync(tdb_fd(dst_tdb));
 343         tdb_close(dst_tdb);
 344         unlink(dst_path);
 345         if (rename(tmp_path, dst_path) != 0) {
 346                 DEBUG(3, ("Failed to rename '%s' to '%s': %s\n",
 347                           tmp_path, dst_path, strerror(errno)));
 348                 goto done;
 349         }
 350
 351         /* success */
 352         ret = 0;
 353
 354 done:
 355         if (src_tdb != NULL) {
 356                 tdb_close(src_tdb);
 357         }
 358         if (tmp_path != NULL) {
 359                 unlink(tmp_path);
 360                 TALLOC_FREE(tmp_path);
 361         }
 362         if (saved_errno != 0) {
 363                 errno = saved_errno;
 364         }
 365         return ret;
 366 }
 367
 368 static int rename_file_with_suffix(TALLOC_CTX *ctx, const char *path,
 369                                    const char *suffix)
 370 {
 371         int ret = -1;
 372         char *dst_path;
 373
 374         dst_path = talloc_asprintf(ctx, "%s%s", path, suffix);
 375         if (dst_path == NULL) {
 376                 DEBUG(3, ("error out of memory\n"));
 377                 return ret;
 378         }
 379
 380         ret = (rename(path, dst_path) != 0);
 381
 382         if (ret == 0) {
 383                 DEBUG(5, ("moved '%s' to '%s'\n", path, dst_path));
 384         } else if (errno == ENOENT) {
 385                 DEBUG(3, ("file '%s' does not exist - so not moved\n", path));
 386                 ret = 0;
 387         } else {
 388                 DEBUG(3, ("error renaming %s to %s: %s\n", path, dst_path,
 389                           strerror(errno)));
 390         }
 391
 392         TALLOC_FREE(dst_path);
 393         return ret;
 394 }
 395
 396 /*
 397  * do a backup of a tdb, moving the destination out of the way first
 398  */
 399 static int tdb_backup_with_rotate(TALLOC_CTX *ctx, const char *src_path,
 400                                   const char *dst_path, int hash_size,
 401                                   const char *rotate_suffix,
 402                                   bool retry_norotate_if_nospc,
 403                                   bool rename_as_last_resort_if_nospc)
 404 {
 405         int ret;
 406
 407         rename_file_with_suffix(ctx, dst_path, rotate_suffix);
 408
 409         ret = tdb_backup(ctx, src_path, dst_path, hash_size);
 410
 411         if (ret != 0) {
 412                 DEBUG(10, ("backup of %s failed: %s\n", src_path, strerror(errno)));
 413         }
 414         if ((ret != 0) && (errno == ENOSPC) && retry_norotate_if_nospc)
 415         {
 416                 char *rotate_path = talloc_asprintf(ctx, "%s%s", dst_path,
 417                                                     rotate_suffix);
 418                 if (rotate_path == NULL) {
 419                         DEBUG(10, ("talloc fail\n"));
 420                         return -1;
 421                 }
 422                 DEBUG(10, ("backup of %s failed due to lack of space\n",
 423                            src_path));
 424                 DEBUGADD(10, ("trying to free some space by removing rotated "
 425                               "dst %s\n", rotate_path));
 426                 if (unlink(rotate_path) == -1) {
 427                         DEBUG(10, ("unlink of %s failed: %s\n", rotate_path,
 428                                    strerror(errno)));
 429                 } else {
 430                         ret = tdb_backup(ctx, src_path, dst_path, hash_size);
 431                 }
 432                 TALLOC_FREE(rotate_path);
 433         }
 434
 435         if ((ret != 0) && (errno == ENOSPC) && rename_as_last_resort_if_nospc)
 436         {
 437                 DEBUG(10, ("backup of %s failed due to lack of space\n",
 438                            src_path));
 439                 DEBUGADD(10, ("using 'rename' as a last resort\n"));
 440                 ret = rename(src_path, dst_path);
 441         }
 442
 443         return ret;
 444 }
 445
 446 /*
 447  * validation function with backup handling:
 448  *
 449  *  - calls tdb_validate
 450  *  - if the tdb is ok, create a backup "name.bak", possibly moving
 451  *    existing backup to name.bak.old,
 452  *    return 0 (success) even if the backup fails
 453  *  - if the tdb is corrupt:
 454  *    - move the tdb to "name.corrupt"
 455  *    - check if there is valid backup.
 456  *      if so, restore the backup.
 457  *      if restore is successful, return 0 (success),
 458  *    - otherwise return -1 (failure)
 459  */
 460 int tdb_validate_and_backup(const char *tdb_path,
 461                             tdb_validate_data_func validate_fn)
 462 {
 463         int ret = -1;
 464         const char *backup_suffix = ".bak";
 465         const char *corrupt_suffix = ".corrupt";
 466         const char *rotate_suffix = ".old";
 467         char *tdb_path_backup;
 468         struct stat st;
 469         TALLOC_CTX *ctx = NULL;
 470
 471         ctx = talloc_new(NULL);
 472         if (ctx == NULL) {
 473                 DEBUG(0, ("tdb_validate_and_backup: out of memory\n"));
 474                 goto done;
 475         }
 476
 477         tdb_path_backup = talloc_asprintf(ctx, "%s%s", tdb_path, backup_suffix);
 478         if (!tdb_path_backup) {
 479                 DEBUG(0, ("tdb_validate_and_backup: out of memory\n"));
 480                 goto done;
 481         }
 482
 483         ret = tdb_validate_open(tdb_path, validate_fn);
 484
 485         if (ret == 0) {
 486                 DEBUG(1, ("tdb '%s' is valid\n", tdb_path));
 487                 ret = tdb_backup_with_rotate(ctx, tdb_path, tdb_path_backup, 0,
 488                                              rotate_suffix, True, False);
 489                 if (ret != 0) {
 490                         DEBUG(1, ("Error creating backup of tdb '%s'\n",
 491                                   tdb_path));
 492                         /* the actual validation was successful: */
 493                         ret = 0;
 494                 } else {
 495                         DEBUG(1, ("Created backup '%s' of tdb '%s'\n",
 496                                   tdb_path_backup, tdb_path));
 497                 }
 498         } else {
 499                 DEBUG(1, ("tdb '%s' is invalid\n", tdb_path));
 500
 501                 ret =stat(tdb_path_backup, &st);
 502                 if (ret != 0) {
 503                         DEBUG(5, ("Could not stat '%s': %s\n", tdb_path_backup,
 504                                   strerror(errno)));
 505                         DEBUG(1, ("No backup found.\n"));
 506                 } else {
 507                         DEBUG(1, ("backup '%s' found.\n", tdb_path_backup));
 508                         ret = tdb_validate_open(tdb_path_backup, validate_fn);
 509                         if (ret != 0) {
 510                                 DEBUG(1, ("Backup '%s' is invalid.\n",
 511                                           tdb_path_backup));
 512                         }
 513                 }
 514
 515                 if (ret != 0) {
 516                         int renamed = rename_file_with_suffix(ctx, tdb_path,
 517                                                               corrupt_suffix);
 518                         if (renamed != 0) {
 519                                 DEBUG(1, ("Error moving tdb to '%s%s'\n",
 520                                           tdb_path, corrupt_suffix));
 521                         } else {
 522                                 DEBUG(1, ("Corrupt tdb stored as '%s%s'\n",
 523                                           tdb_path, corrupt_suffix));
 524                         }
 525                         goto done;
 526                 }
 527
 528                 DEBUG(1, ("valid backup '%s' found\n", tdb_path_backup));
 529                 ret = tdb_backup_with_rotate(ctx, tdb_path_backup, tdb_path, 0,
 530                                              corrupt_suffix, True, True);
 531                 if (ret != 0) {
 532                         DEBUG(1, ("Error restoring backup from '%s'\n",
 533                                   tdb_path_backup));
 534                 } else {
 535                         DEBUG(1, ("Restored tdb backup from '%s'\n",
 536                                   tdb_path_backup));
 537                 }
 538         }
 539
 540 done:
 541         TALLOC_FREE(ctx);
 542         return ret;
 543 }