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 num_entries = 0;
  35         struct tdb_validation_status v_status;
  36
  37         v_status.tdb_error = False;
  38         v_status.bad_freelist = False;
  39         v_status.bad_entry = False;
  40         v_status.unknown_key = False;
  41         v_status.success = True;
  42
  43         if (!tdb) {
  44                 v_status.tdb_error = True;
  45                 v_status.success = False;
  46                 goto out;
  47         }
  48
  49         /*
  50          * we can simplify this by passing a check function,
  51          * but I don't want to change all the callers...
  52          */
  53         ret = tdb_check(tdb, NULL, NULL);
  54         if (ret != 0) {
  55                 v_status.tdb_error = True;
  56                 v_status.success = False;
  57                 goto out;
  58         }
  59
  60 #ifndef BUILD_TDB2
  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 #endif
  71
  72         /* Now traverse the tdb to validate it. */
  73         num_entries = tdb_traverse(tdb, validate_fn, (void *)&v_status);
  74         if (!v_status.success) {
  75                 goto out;
  76         } else if (num_entries < 0) {
  77                 v_status.tdb_error = True;
  78                 v_status.success = False;
  79                 goto out;
  80         }
  81
  82         DEBUG(10,("tdb_validate_child: tdb %s is good with %d entries\n",
  83                   tdb_name(tdb), num_entries));
  84         ret = 0; /* Cache is good. */
  85
  86 out:
  87         DEBUG(10,   ("tdb_validate_child: summary of validation status:\n"));
  88         DEBUGADD(10,(" * tdb error: %s\n", v_status.tdb_error ? "yes" : "no"));
  89         DEBUGADD(10,(" * bad freelist: %s\n",v_status.bad_freelist?"yes":"no"));
  90         DEBUGADD(10,(" * bad entry: %s\n", v_status.bad_entry ? "yes" : "no"));
  91         DEBUGADD(10,(" * unknown key: %s\n", v_status.unknown_key?"yes":"no"));
  92         DEBUGADD(10,(" => overall success: %s\n", v_status.success?"yes":"no"));
  93
  94         return ret;
  95 }
  96
  97 /*
  98  * tdb validation function.
  99  * returns 0 if tdb is ok, != 0 if it isn't.
 100  * this function expects an opened tdb.
 101  */
 102 int tdb_validate(struct tdb_context *tdb, tdb_validate_data_func validate_fn)
 103 {
 104         pid_t child_pid = -1;
 105         int child_status = 0;
 106         int wait_pid = 0;
 107         int ret = 1;
 108
 109         if (tdb == NULL) {
 110                 DEBUG(1, ("Error: tdb_validate called with tdb == NULL\n"));
 111                 return ret;
 112         }
 113
 114         DEBUG(5, ("tdb_validate called for tdb '%s'\n", tdb_name(tdb)));
 115
 116         /* fork and let the child do the validation.
 117          * benefit: no need to twist signal handlers and panic functions.
 118          * just let the child panic. we catch the signal. */
 119
 120         DEBUG(10, ("tdb_validate: forking to let child do validation.\n"));
 121         child_pid = sys_fork();
 122         if (child_pid == 0) {
 123                 /* child code */
 124                 DEBUG(10, ("tdb_validate (validation child): created\n"));
 125                 DEBUG(10, ("tdb_validate (validation child): "
 126                            "calling tdb_validate_child\n"));
 127                 exit(tdb_validate_child(tdb, validate_fn));
 128         }
 129         else if (child_pid < 0) {
 130                 DEBUG(1, ("tdb_validate: fork for validation failed.\n"));
 131                 goto done;
 132         }
 133
 134         /* parent */
 135
 136         DEBUG(10, ("tdb_validate: fork succeeded, child PID = %u\n",
 137                 (unsigned int)child_pid));
 138
 139         DEBUG(10, ("tdb_validate: waiting for child to finish...\n"));
 140         while  ((wait_pid = sys_waitpid(child_pid, &child_status, 0)) < 0) {
 141                 if (errno == EINTR) {
 142                         DEBUG(10, ("tdb_validate: got signal during waitpid, "
 143                                    "retrying\n"));
 144                         errno = 0;
 145                         continue;
 146                 }
 147                 DEBUG(1, ("tdb_validate: waitpid failed with error '%s'.\n",
 148                           strerror(errno)));
 149                 goto done;
 150         }
 151         if (wait_pid != child_pid) {
 152                 DEBUG(1, ("tdb_validate: waitpid returned pid %d, "
 153                           "but %u was expected\n", wait_pid, (unsigned int)child_pid));
 154                 goto done;
 155         }
 156
 157         DEBUG(10, ("tdb_validate: validating child returned.\n"));
 158         if (WIFEXITED(child_status)) {
 159                 DEBUG(10, ("tdb_validate: child exited, code %d.\n",
 160                            WEXITSTATUS(child_status)));
 161                 ret = WEXITSTATUS(child_status);
 162         }
 163         if (WIFSIGNALED(child_status)) {
 164                 DEBUG(10, ("tdb_validate: child terminated by signal %d\n",
 165                            WTERMSIG(child_status)));
 166 #ifdef WCOREDUMP
 167                 if (WCOREDUMP(child_status)) {
 168                         DEBUGADD(10, ("core dumped\n"));
 169                 }
 170 #endif
 171                 ret = WTERMSIG(child_status);
 172         }
 173         if (WIFSTOPPED(child_status)) {
 174                 DEBUG(10, ("tdb_validate: child was stopped by signal %d\n",
 175                            WSTOPSIG(child_status)));
 176                 ret = WSTOPSIG(child_status);
 177         }
 178
 179 done:
 180         DEBUG(5, ("tdb_validate returning code '%d' for tdb '%s'\n", ret,
 181                   tdb_name(tdb)));
 182
 183         return ret;
 184 }
 185
 186 /*
 187  * tdb validation function.
 188  * returns 0 if tdb is ok, != 0 if it isn't.
 189  * this is a wrapper around the actual validation function that opens and closes
 190  * the tdb.
 191  */
 192 int tdb_validate_open(const char *tdb_path, tdb_validate_data_func validate_fn)
 193 {
 194         TDB_CONTEXT *tdb = NULL;
 195         int ret = 1;
 196
 197         DEBUG(5, ("tdb_validate_open called for tdb '%s'\n", tdb_path));
 198
 199         tdb = tdb_open_log(tdb_path, 0, TDB_DEFAULT, O_RDWR, 0);
 200         if (!tdb) {
 201                 DEBUG(1, ("Error opening tdb %s\n", tdb_path));
 202                 return ret;
 203         }
 204
 205         ret = tdb_validate(tdb, validate_fn);
 206         tdb_close(tdb);
 207         return ret;
 208 }
 209
 210 /*
 211  * tdb backup function and helpers for tdb_validate wrapper with backup
 212  * handling.
 213  */
 214
 215 /* this structure eliminates the need for a global overall status for
 216  * the traverse-copy */
 217 struct tdb_copy_data {
 218         struct tdb_context *dst;
 219         bool success;
 220 };
 221
 222 static int traverse_copy_fn(struct tdb_context *tdb, TDB_DATA key,
 223                             TDB_DATA dbuf, void *private_data)
 224 {
 225         struct tdb_copy_data *data = (struct tdb_copy_data *)private_data;
 226
 227         if (tdb_store(data->dst, key, dbuf, TDB_INSERT) != 0) {
 228                 DEBUG(4, ("Failed to insert into %s: %s\n", tdb_name(data->dst),
 229                           strerror(errno)));
 230                 data->success = False;
 231                 return 1;
 232         }
 233         return 0;
 234 }
 235
 236 static int tdb_copy(struct tdb_context *src, struct tdb_context *dst)
 237 {
 238         struct tdb_copy_data data;
 239         int count;
 240
 241         data.dst = dst;
 242         data.success = True;
 243
 244         count = tdb_traverse(src, traverse_copy_fn, (void *)(&data));
 245         if ((count < 0) || (data.success == False)) {
 246                 return -1;
 247         }
 248         return count;
 249 }
 250
 251 static int tdb_verify_basic(struct tdb_context *tdb)
 252 {
 253         return tdb_traverse(tdb, NULL, NULL);
 254 }
 255
 256 /* this backup function is essentially taken from lib/tdb/tools/tdbbackup.tdb
 257  */
 258 static int tdb_backup(TALLOC_CTX *ctx, const char *src_path,
 259                       const char *dst_path, int hash_size)
 260 {
 261         struct tdb_context *src_tdb = NULL;
 262         struct tdb_context *dst_tdb = NULL;
 263         char *tmp_path = NULL;
 264         struct stat st;
 265         int count1, count2;
 266         int saved_errno = 0;
 267         int ret = -1;
 268
 269         if (stat(src_path, &st) != 0) {
 270                 DEBUG(3, ("Could not stat '%s': %s\n", src_path,
 271                           strerror(errno)));
 272                 goto done;
 273         }
 274
 275         /* open old tdb RDWR - so we can lock it */
 276         src_tdb = tdb_open_log(src_path, 0, TDB_DEFAULT, O_RDWR, 0);
 277         if (src_tdb == NULL) {
 278                 DEBUG(3, ("Failed to open tdb '%s'\n", src_path));
 279                 goto done;
 280         }
 281
 282         if (tdb_lockall(src_tdb) != 0) {
 283                 DEBUG(3, ("Failed to lock tdb '%s'\n", src_path));
 284                 goto done;
 285         }
 286
 287         tmp_path = talloc_asprintf(ctx, "%s%s", dst_path, ".tmp");
 288         if (!tmp_path) {
 289                 DEBUG(3, ("talloc fail\n"));
 290                 goto done;
 291         }
 292
 293         unlink(tmp_path);
 294
 295 #ifndef BUILD_TDB2
 296         if (!hash_size) {
 297                 hash_size = tdb_hash_size(src_tdb);
 298         }
 299 #endif
 300
 301         dst_tdb = tdb_open_log(tmp_path, hash_size,
 302                                TDB_DEFAULT, O_RDWR | O_CREAT | O_EXCL,
 303                                st.st_mode & 0777);
 304         if (dst_tdb == NULL) {
 305                 DEBUG(3, ("Error creating tdb '%s': %s\n", tmp_path,
 306                           strerror(errno)));
 307                 saved_errno = errno;
 308                 unlink(tmp_path);
 309                 goto done;
 310         }
 311
 312         count1 = tdb_copy(src_tdb, dst_tdb);
 313         if (count1 < 0) {
 314                 DEBUG(3, ("Failed to copy tdb '%s': %s\n", src_path,
 315                           strerror(errno)));
 316                 tdb_close(dst_tdb);
 317                 goto done;
 318         }
 319
 320         /* reopen ro and do basic verification */
 321         tdb_close(dst_tdb);
 322         dst_tdb = tdb_open_log(tmp_path, 0, TDB_DEFAULT, O_RDONLY, 0);
 323         if (!dst_tdb) {
 324                 DEBUG(3, ("Failed to reopen tdb '%s': %s\n", tmp_path,
 325                           strerror(errno)));
 326                 goto done;
 327         }
 328         count2 = tdb_verify_basic(dst_tdb);
 329         if (count2 != count1) {
 330                 DEBUG(3, ("Failed to verify result of copying tdb '%s'.\n",
 331                           src_path));
 332                 tdb_close(dst_tdb);
 333                 goto done;
 334         }
 335
 336         DEBUG(10, ("tdb_backup: successfully copied %d entries\n", count1));
 337
 338         /* make sure the new tdb has reached stable storage
 339          * then rename it to its destination */
 340         fsync(tdb_fd(dst_tdb));
 341         tdb_close(dst_tdb);
 342         unlink(dst_path);
 343         if (rename(tmp_path, dst_path) != 0) {
 344                 DEBUG(3, ("Failed to rename '%s' to '%s': %s\n",
 345                           tmp_path, dst_path, strerror(errno)));
 346                 goto done;
 347         }
 348
 349         /* success */
 350         ret = 0;
 351
 352 done:
 353         if (src_tdb != NULL) {
 354                 tdb_close(src_tdb);
 355         }
 356         if (tmp_path != NULL) {
 357                 unlink(tmp_path);
 358                 TALLOC_FREE(tmp_path);
 359         }
 360         if (saved_errno != 0) {
 361                 errno = saved_errno;
 362         }
 363         return ret;
 364 }
 365
 366 static int rename_file_with_suffix(TALLOC_CTX *ctx, const char *path,
 367                                    const char *suffix)
 368 {
 369         int ret = -1;
 370         char *dst_path;
 371
 372         dst_path = talloc_asprintf(ctx, "%s%s", path, suffix);
 373         if (dst_path == NULL) {
 374                 DEBUG(3, ("error out of memory\n"));
 375                 return ret;
 376         }
 377
 378         ret = (rename(path, dst_path) != 0);
 379
 380         if (ret == 0) {
 381                 DEBUG(5, ("moved '%s' to '%s'\n", path, dst_path));
 382         } else if (errno == ENOENT) {
 383                 DEBUG(3, ("file '%s' does not exist - so not moved\n", path));
 384                 ret = 0;
 385         } else {
 386                 DEBUG(3, ("error renaming %s to %s: %s\n", path, dst_path,
 387                           strerror(errno)));
 388         }
 389
 390         TALLOC_FREE(dst_path);
 391         return ret;
 392 }
 393
 394 /*
 395  * do a backup of a tdb, moving the destination out of the way first
 396  */
 397 static int tdb_backup_with_rotate(TALLOC_CTX *ctx, const char *src_path,
 398                                   const char *dst_path, int hash_size,
 399                                   const char *rotate_suffix,
 400                                   bool retry_norotate_if_nospc,
 401                                   bool rename_as_last_resort_if_nospc)
 402 {
 403         int ret;
 404
 405         rename_file_with_suffix(ctx, dst_path, rotate_suffix);
 406
 407         ret = tdb_backup(ctx, src_path, dst_path, hash_size);
 408
 409         if (ret != 0) {
 410                 DEBUG(10, ("backup of %s failed: %s\n", src_path, strerror(errno)));
 411         }
 412         if ((ret != 0) && (errno == ENOSPC) && retry_norotate_if_nospc)
 413         {
 414                 char *rotate_path = talloc_asprintf(ctx, "%s%s", dst_path,
 415                                                     rotate_suffix);
 416                 if (rotate_path == NULL) {
 417                         DEBUG(10, ("talloc fail\n"));
 418                         return -1;
 419                 }
 420                 DEBUG(10, ("backup of %s failed due to lack of space\n",
 421                            src_path));
 422                 DEBUGADD(10, ("trying to free some space by removing rotated "
 423                               "dst %s\n", rotate_path));
 424                 if (unlink(rotate_path) == -1) {
 425                         DEBUG(10, ("unlink of %s failed: %s\n", rotate_path,
 426                                    strerror(errno)));
 427                 } else {
 428                         ret = tdb_backup(ctx, src_path, dst_path, hash_size);
 429                 }
 430                 TALLOC_FREE(rotate_path);
 431         }
 432
 433         if ((ret != 0) && (errno == ENOSPC) && rename_as_last_resort_if_nospc)
 434         {
 435                 DEBUG(10, ("backup of %s failed due to lack of space\n",
 436                            src_path));
 437                 DEBUGADD(10, ("using 'rename' as a last resort\n"));
 438                 ret = rename(src_path, dst_path);
 439         }
 440
 441         return ret;
 442 }
 443
 444 /*
 445  * validation function with backup handling:
 446  *
 447  *  - calls tdb_validate
 448  *  - if the tdb is ok, create a backup "name.bak", possibly moving
 449  *    existing backup to name.bak.old,
 450  *    return 0 (success) even if the backup fails
 451  *  - if the tdb is corrupt:
 452  *    - move the tdb to "name.corrupt"
 453  *    - check if there is valid backup.
 454  *      if so, restore the backup.
 455  *      if restore is successful, return 0 (success),
 456  *    - otherwise return -1 (failure)
 457  */
 458 int tdb_validate_and_backup(const char *tdb_path,
 459                             tdb_validate_data_func validate_fn)
 460 {
 461         int ret = -1;
 462         const char *backup_suffix = ".bak";
 463         const char *corrupt_suffix = ".corrupt";
 464         const char *rotate_suffix = ".old";
 465         char *tdb_path_backup;
 466         struct stat st;
 467         TALLOC_CTX *ctx = NULL;
 468
 469         ctx = talloc_new(NULL);
 470         if (ctx == NULL) {
 471                 DEBUG(0, ("tdb_validate_and_backup: out of memory\n"));
 472                 goto done;
 473         }
 474
 475         tdb_path_backup = talloc_asprintf(ctx, "%s%s", tdb_path, backup_suffix);
 476         if (!tdb_path_backup) {
 477                 DEBUG(0, ("tdb_validate_and_backup: out of memory\n"));
 478                 goto done;
 479         }
 480
 481         ret = tdb_validate_open(tdb_path, validate_fn);
 482
 483         if (ret == 0) {
 484                 DEBUG(1, ("tdb '%s' is valid\n", tdb_path));
 485                 ret = tdb_backup_with_rotate(ctx, tdb_path, tdb_path_backup, 0,
 486                                              rotate_suffix, True, False);
 487                 if (ret != 0) {
 488                         DEBUG(1, ("Error creating backup of tdb '%s'\n",
 489                                   tdb_path));
 490                         /* the actual validation was successful: */
 491                         ret = 0;
 492                 } else {
 493                         DEBUG(1, ("Created backup '%s' of tdb '%s'\n",
 494                                   tdb_path_backup, tdb_path));
 495                 }
 496         } else {
 497                 DEBUG(1, ("tdb '%s' is invalid\n", tdb_path));
 498
 499                 ret =stat(tdb_path_backup, &st);
 500                 if (ret != 0) {
 501                         DEBUG(5, ("Could not stat '%s': %s\n", tdb_path_backup,
 502                                   strerror(errno)));
 503                         DEBUG(1, ("No backup found.\n"));
 504                 } else {
 505                         DEBUG(1, ("backup '%s' found.\n", tdb_path_backup));
 506                         ret = tdb_validate_open(tdb_path_backup, validate_fn);
 507                         if (ret != 0) {
 508                                 DEBUG(1, ("Backup '%s' is invalid.\n",
 509                                           tdb_path_backup));
 510                         }
 511                 }
 512
 513                 if (ret != 0) {
 514                         int renamed = rename_file_with_suffix(ctx, tdb_path,
 515                                                               corrupt_suffix);
 516                         if (renamed != 0) {
 517                                 DEBUG(1, ("Error moving tdb to '%s%s'\n",
 518                                           tdb_path, corrupt_suffix));
 519                         } else {
 520                                 DEBUG(1, ("Corrupt tdb stored as '%s%s'\n",
 521                                           tdb_path, corrupt_suffix));
 522                         }
 523                         goto done;
 524                 }
 525
 526                 DEBUG(1, ("valid backup '%s' found\n", tdb_path_backup));
 527                 ret = tdb_backup_with_rotate(ctx, tdb_path_backup, tdb_path, 0,
 528                                              corrupt_suffix, True, True);
 529                 if (ret != 0) {
 530                         DEBUG(1, ("Error restoring backup from '%s'\n",
 531                                   tdb_path_backup));
 532                 } else {
 533                         DEBUG(1, ("Restored tdb backup from '%s'\n",
 534                                   tdb_path_backup));
 535                 }
 536         }
 537
 538 done:
 539         TALLOC_FREE(ctx);
 540         return ret;
 541 }