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