ctdb/common/ctdb_util.c

   1 /*
   2    ctdb utility code
   3
   4    Copyright (C) Andrew Tridgell  2006
   5
   6    This program is free software; you can redistribute it and/or modify
   7    it under the terms of the GNU General Public License as published by
   8    the Free Software Foundation; either version 3 of the License, or
   9    (at your option) any later version.
  10
  11    This program is distributed in the hope that it will be useful,
  12    but WITHOUT ANY WARRANTY; without even the implied warranty of
  13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14    GNU General Public License for more details.
  15
  16    You should have received a copy of the GNU General Public License
  17    along with this program; if not, see <http://www.gnu.org/licenses/>.
  18 */
  19
  20 #include "replace.h"
  21 #include "system/network.h"
  22 #include "system/filesys.h"
  23 #include "system/wait.h"
  24
  25 #include <tdb.h>
  26
  27 #include "lib/util/debug.h"
  28 #include "lib/util/samba_util.h"
  29
  30 #include "ctdb_private.h"
  31
  32 #include "protocol/protocol_util.h"
  33
  34 #include "common/reqid.h"
  35 #include "common/system.h"
  36 #include "common/common.h"
  37 #include "common/logging.h"
  38
  39 /*
  40   return error string for last error
  41 */
  42 const char *ctdb_errstr(struct ctdb_context *ctdb)
  43 {
  44         return ctdb->err_msg;
  45 }
  46
  47
  48 /*
  49   remember an error message
  50 */
  51 void ctdb_set_error(struct ctdb_context *ctdb, const char *fmt, ...)
  52 {
  53         va_list ap;
  54         talloc_free(ctdb->err_msg);
  55         va_start(ap, fmt);
  56         ctdb->err_msg = talloc_vasprintf(ctdb, fmt, ap);
  57         DEBUG(DEBUG_ERR,("ctdb error: %s\n", ctdb->err_msg));
  58         va_end(ap);
  59 }
  60
  61 /*
  62   a fatal internal error occurred - no hope for recovery
  63 */
  64 void ctdb_fatal(struct ctdb_context *ctdb, const char *msg)
  65 {
  66         DEBUG(DEBUG_ALERT,("ctdb fatal error: %s\n", msg));
  67         abort();
  68 }
  69
  70 /*
  71   like ctdb_fatal() but a core/backtrace would not be useful
  72 */
  73 void ctdb_die(struct ctdb_context *ctdb, const char *msg)
  74 {
  75         DEBUG(DEBUG_ALERT,("ctdb exiting with error: %s\n", msg));
  76         exit(1);
  77 }
  78
  79 /* Set the path of a helper program from envvar, falling back to
  80  * dir/file if envvar unset. type is a string to print in log
  81  * messages.  helper is assumed to point to a statically allocated
  82  * array of size bytes, initialised to "".  If file is NULL don't fall
  83  * back if envvar is unset.  If dir is NULL and envvar is unset (but
  84  * file is not NULL) then this is an error.  Returns true if helper is
  85  * set, either previously or this time. */
  86 bool ctdb_set_helper(const char *type, char *helper, size_t size,
  87                      const char *envvar,
  88                      const char *dir, const char *file)
  89 {
  90         const char *t;
  91         struct stat st;
  92
  93         if (helper[0] != '\0') {
  94                 /* Already set */
  95                 return true;
  96         }
  97
  98         t = getenv(envvar);
  99         if (t != NULL) {
 100                 if (strlen(t) >= size) {
 101                         DEBUG(DEBUG_ERR,
 102                               ("Unable to set %s - path too long\n", type));
 103                         return false;
 104                 }
 105
 106                 strncpy(helper, t, size);
 107         } else if (file == NULL) {
 108                 return false;
 109         } else if (dir == NULL) {
 110                         DEBUG(DEBUG_ERR,
 111                               ("Unable to set %s - dir is NULL\n", type));
 112                 return false;
 113         } else {
 114                 int ret;
 115
 116                 ret = snprintf(helper, size, "%s/%s", dir, file);
 117                 if (ret < 0 || (size_t)ret >= size) {
 118                         DEBUG(DEBUG_ERR,
 119                               ("Unable to set %s - path too long\n", type));
 120                         return false;
 121                 }
 122         }
 123
 124         if (stat(helper, &st) != 0) {
 125                 DEBUG(DEBUG_ERR,
 126                       ("Unable to set %s \"%s\" - %s\n",
 127                        type, helper, strerror(errno)));
 128                 return false;
 129         }
 130         if (!(st.st_mode & S_IXUSR)) {
 131                 DEBUG(DEBUG_ERR,
 132                       ("Unable to set %s \"%s\" - not executable\n",
 133                        type, helper));
 134                 return false;
 135         }
 136
 137         DEBUG(DEBUG_NOTICE,
 138               ("Set %s to \"%s\"\n", type, helper));
 139         return true;
 140 }
 141
 142 /*
 143   parse a IP:port pair
 144 */
 145 int ctdb_parse_address(TALLOC_CTX *mem_ctx, const char *str,
 146                        ctdb_sock_addr *address)
 147 {
 148         struct servent *se;
 149         int port;
 150         int ret;
 151
 152         setservent(0);
 153         se = getservbyname("ctdb", "tcp");
 154         endservent();
 155
 156         if (se == NULL) {
 157                 port = CTDB_PORT;
 158         } else {
 159                 port = ntohs(se->s_port);
 160         }
 161
 162         ret = ctdb_sock_addr_from_string(str, address, false);
 163         if (ret != 0) {
 164                 return -1;
 165         }
 166         ctdb_sock_addr_set_port(address, port);
 167
 168         return 0;
 169 }
 170
 171
 172 /*
 173   check if two addresses are the same
 174 */
 175 bool ctdb_same_address(ctdb_sock_addr *a1, ctdb_sock_addr *a2)
 176 {
 177         return ctdb_same_ip(a1, a2) &&
 178                 ctdb_addr_to_port(a1) == ctdb_addr_to_port(a2);
 179 }
 180
 181
 182 /*
 183   hash function for mapping data to a VNN - taken from tdb
 184 */
 185 uint32_t ctdb_hash(const TDB_DATA *key)
 186 {
 187         return tdb_jenkins_hash(discard_const(key));
 188 }
 189
 190
 191 static uint32_t ctdb_marshall_record_size(TDB_DATA key,
 192                                           struct ctdb_ltdb_header *header,
 193                                           TDB_DATA data)
 194 {
 195         return offsetof(struct ctdb_rec_data_old, data) + key.dsize +
 196                data.dsize + (header ? sizeof(*header) : 0);
 197 }
 198
 199 static void ctdb_marshall_record_copy(struct ctdb_rec_data_old *rec,
 200                                       uint32_t reqid,
 201                                       TDB_DATA key,
 202                                       struct ctdb_ltdb_header *header,
 203                                       TDB_DATA data,
 204                                       uint32_t length)
 205 {
 206         uint32_t offset;
 207
 208         rec->length = length;
 209         rec->reqid = reqid;
 210         rec->keylen = key.dsize;
 211         memcpy(&rec->data[0], key.dptr, key.dsize);
 212         offset = key.dsize;
 213
 214         if (header) {
 215                 rec->datalen = data.dsize + sizeof(*header);
 216                 memcpy(&rec->data[offset], header, sizeof(*header));
 217                 offset += sizeof(*header);
 218         } else {
 219                 rec->datalen = data.dsize;
 220         }
 221         memcpy(&rec->data[offset], data.dptr, data.dsize);
 222 }
 223
 224 /*
 225   form a ctdb_rec_data record from a key/data pair
 226
 227   note that header may be NULL. If not NULL then it is included in the data portion
 228   of the record
 229  */
 230 struct ctdb_rec_data_old *ctdb_marshall_record(TALLOC_CTX *mem_ctx,
 231                                                uint32_t reqid,
 232                                                TDB_DATA key,
 233                                                struct ctdb_ltdb_header *header,
 234                                                TDB_DATA data)
 235 {
 236         size_t length;
 237         struct ctdb_rec_data_old *d;
 238
 239         length = ctdb_marshall_record_size(key, header, data);
 240
 241         d = (struct ctdb_rec_data_old *)talloc_size(mem_ctx, length);
 242         if (d == NULL) {
 243                 return NULL;
 244         }
 245
 246         ctdb_marshall_record_copy(d, reqid, key, header, data, length);
 247         return d;
 248 }
 249
 250
 251 /* helper function for marshalling multiple records */
 252 struct ctdb_marshall_buffer *ctdb_marshall_add(TALLOC_CTX *mem_ctx,
 253                                                struct ctdb_marshall_buffer *m,
 254                                                uint32_t db_id,
 255                                                uint32_t reqid,
 256                                                TDB_DATA key,
 257                                                struct ctdb_ltdb_header *header,
 258                                                TDB_DATA data)
 259 {
 260         struct ctdb_rec_data_old *r;
 261         struct ctdb_marshall_buffer *m2;
 262         uint32_t length, offset;
 263
 264         length = ctdb_marshall_record_size(key, header, data);
 265
 266         if (m == NULL) {
 267                 offset = offsetof(struct ctdb_marshall_buffer, data);
 268                 m2 = talloc_zero_size(mem_ctx, offset + length);
 269         } else {
 270                 offset = talloc_get_size(m);
 271                 m2 = talloc_realloc_size(mem_ctx, m, offset + length);
 272         }
 273         if (m2 == NULL) {
 274                 TALLOC_FREE(m);
 275                 return NULL;
 276         }
 277
 278         if (m == NULL) {
 279                 m2->db_id = db_id;
 280         }
 281
 282         r = (struct ctdb_rec_data_old *)((uint8_t *)m2 + offset);
 283         ctdb_marshall_record_copy(r, reqid, key, header, data, length);
 284         m2->count++;
 285
 286         return m2;
 287 }
 288
 289 /* we've finished marshalling, return a data blob with the marshalled records */
 290 TDB_DATA ctdb_marshall_finish(struct ctdb_marshall_buffer *m)
 291 {
 292         TDB_DATA data;
 293         data.dptr = (uint8_t *)m;
 294         data.dsize = talloc_get_size(m);
 295         return data;
 296 }
 297
 298 /*
 299    loop over a marshalling buffer
 300
 301      - pass r==NULL to start
 302      - loop the number of times indicated by m->count
 303 */
 304 struct ctdb_rec_data_old *ctdb_marshall_loop_next(
 305                                 struct ctdb_marshall_buffer *m,
 306                                 struct ctdb_rec_data_old *r,
 307                                 uint32_t *reqid,
 308                                 struct ctdb_ltdb_header *header,
 309                                 TDB_DATA *key, TDB_DATA *data)
 310 {
 311         if (r == NULL) {
 312                 r = (struct ctdb_rec_data_old *)&m->data[0];
 313         } else {
 314                 r = (struct ctdb_rec_data_old *)(r->length + (uint8_t *)r);
 315         }
 316
 317         if (reqid != NULL) {
 318                 *reqid = r->reqid;
 319         }
 320
 321         if (key != NULL) {
 322                 key->dptr   = &r->data[0];
 323                 key->dsize  = r->keylen;
 324         }
 325         if (data != NULL) {
 326                 data->dptr  = &r->data[r->keylen];
 327                 data->dsize = r->datalen;
 328                 if (header != NULL) {
 329                         data->dptr += sizeof(*header);
 330                         data->dsize -= sizeof(*header);
 331                 }
 332         }
 333
 334         if (header != NULL) {
 335                 if (r->datalen < sizeof(*header)) {
 336                         return NULL;
 337                 }
 338                 memcpy(header, &r->data[r->keylen], sizeof(*header));
 339         }
 340
 341         return r;
 342 }
 343
 344 /*
 345    This is used to canonicalize a ctdb_sock_addr structure.
 346 */
 347 void ctdb_canonicalize_ip(const ctdb_sock_addr *ip, ctdb_sock_addr *cip)
 348 {
 349         ZERO_STRUCTP(cip);
 350
 351         if (ip->sa.sa_family == AF_INET6) {
 352                 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
 353                 if (memcmp(&ip->ip6.sin6_addr, prefix, sizeof(prefix)) == 0) {
 354                         /* Copy IPv4-mapped IPv6 addresses as IPv4 */
 355                         cip->ip.sin_family = AF_INET;
 356 #ifdef HAVE_SOCK_SIN_LEN
 357                         cip->ip.sin_len = sizeof(ctdb_sock_addr);
 358 #endif
 359                         cip->ip.sin_port   = ip->ip6.sin6_port;
 360                         memcpy(&cip->ip.sin_addr,
 361                                &ip->ip6.sin6_addr.s6_addr[12],
 362                                sizeof(cip->ip.sin_addr));
 363                 } else {
 364                         cip->ip6.sin6_family = AF_INET6;
 365 #ifdef HAVE_SOCK_SIN6_LEN
 366                         cip->ip6.sin6_len = sizeof(ctdb_sock_addr);
 367 #endif
 368                         cip->ip6.sin6_port   = ip->ip6.sin6_port;
 369                         memcpy(&cip->ip6.sin6_addr,
 370                                &ip->ip6.sin6_addr,
 371                                sizeof(cip->ip6.sin6_addr));
 372                 }
 373
 374                 return;
 375         }
 376
 377         if (ip->sa.sa_family == AF_INET) {
 378                 cip->ip.sin_family = AF_INET;
 379 #ifdef HAVE_SOCK_SIN_LEN
 380                 cip->ip.sin_len = sizeof(ctdb_sock_addr);
 381 #endif
 382                 cip->ip.sin_port = ip->ip.sin_port;
 383                 memcpy(&cip->ip.sin_addr,
 384                        &ip->ip.sin_addr,
 385                        sizeof(ip->ip.sin_addr));
 386
 387                 return;
 388         }
 389 }
 390
 391 void ctdb_canonicalize_ip_inplace(ctdb_sock_addr *ip)
 392 {
 393         ctdb_sock_addr tmp;
 394         ctdb_canonicalize_ip(ip, &tmp);
 395         memcpy(ip, &tmp, sizeof(tmp));
 396 }
 397
 398 bool ctdb_same_ip(const ctdb_sock_addr *tip1, const ctdb_sock_addr *tip2)
 399 {
 400         ctdb_sock_addr ip1, ip2;
 401
 402         ctdb_canonicalize_ip(tip1, &ip1);
 403         ctdb_canonicalize_ip(tip2, &ip2);
 404
 405         if (ip1.sa.sa_family != ip2.sa.sa_family) {
 406                 return false;
 407         }
 408
 409         switch (ip1.sa.sa_family) {
 410         case AF_INET:
 411                 return ip1.ip.sin_addr.s_addr == ip2.ip.sin_addr.s_addr;
 412         case AF_INET6:
 413                 return !memcmp(&ip1.ip6.sin6_addr.s6_addr[0],
 414                                 &ip2.ip6.sin6_addr.s6_addr[0],
 415                                 16);
 416         default:
 417                 DEBUG(DEBUG_ERR, (__location__ " CRITICAL Can not compare sockaddr structures of type %u\n", ip1.sa.sa_family));
 418                 return false;
 419         }
 420
 421         return true;
 422 }
 423
 424 /*
 425   compare two ctdb_sock_addr structures
 426  */
 427 bool ctdb_same_sockaddr(const ctdb_sock_addr *ip1, const ctdb_sock_addr *ip2)
 428 {
 429         return ctdb_same_ip(ip1, ip2) && ip1->ip.sin_port == ip2->ip.sin_port;
 430 }
 431
 432 char *ctdb_addr_to_str(ctdb_sock_addr *addr)
 433 {
 434         static char cip[128] = "";
 435
 436         switch (addr->sa.sa_family) {
 437         case AF_INET:
 438                 inet_ntop(addr->ip.sin_family, &addr->ip.sin_addr, cip, sizeof(cip));
 439                 break;
 440         case AF_INET6:
 441                 inet_ntop(addr->ip6.sin6_family, &addr->ip6.sin6_addr, cip, sizeof(cip));
 442                 break;
 443         default:
 444                 DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family %u\n", addr->sa.sa_family));
 445         }
 446
 447         return cip;
 448 }
 449
 450 unsigned ctdb_addr_to_port(ctdb_sock_addr *addr)
 451 {
 452         switch (addr->sa.sa_family) {
 453         case AF_INET:
 454                 return ntohs(addr->ip.sin_port);
 455                 break;
 456         case AF_INET6:
 457                 return ntohs(addr->ip6.sin6_port);
 458                 break;
 459         default:
 460                 DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family %u\n", addr->sa.sa_family));
 461         }
 462
 463         return 0;
 464 }
 465
 466 /* Add a node to a node map with given address and flags */
 467 static bool node_map_add(TALLOC_CTX *mem_ctx,
 468                          const char *nstr, uint32_t flags,
 469                          struct ctdb_node_map_old **node_map)
 470 {
 471         ctdb_sock_addr addr;
 472         uint32_t num;
 473         size_t s;
 474         struct ctdb_node_and_flags *n;
 475
 476         /* Might as well do this before trying to allocate memory */
 477         if (ctdb_parse_address(mem_ctx, nstr, &addr) == -1) {
 478                 return false;
 479         }
 480
 481         num = (*node_map)->num + 1;
 482         s = offsetof(struct ctdb_node_map_old, nodes) +
 483                 num * sizeof(struct ctdb_node_and_flags);
 484         *node_map = talloc_realloc_size(mem_ctx, *node_map, s);
 485         if (*node_map == NULL) {
 486                 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
 487                 return false;
 488         }
 489
 490         n = &(*node_map)->nodes[(*node_map)->num];
 491         n->addr = addr;
 492         n->pnn = (*node_map)->num;
 493         n->flags = flags;
 494
 495         (*node_map)->num++;
 496
 497         return true;
 498 }
 499
 500 /* Read a nodes file into a node map */
 501 struct ctdb_node_map_old *ctdb_read_nodes_file(TALLOC_CTX *mem_ctx,
 502                                            const char *nlist)
 503 {
 504         char **lines;
 505         int nlines;
 506         int i;
 507         struct ctdb_node_map_old *ret;
 508
 509         /* Allocate node map header */
 510         ret = talloc_zero_size(mem_ctx, offsetof(struct ctdb_node_map_old, nodes));
 511         if (ret == NULL) {
 512                 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
 513                 return false;
 514         }
 515
 516         lines = file_lines_load(nlist, &nlines, 0, mem_ctx);
 517         if (lines == NULL) {
 518                 DEBUG(DEBUG_ERR, ("Failed to read nodes file \"%s\"\n", nlist));
 519                 return false;
 520         }
 521         while (nlines > 0 && strcmp(lines[nlines-1], "") == 0) {
 522                 nlines--;
 523         }
 524
 525         for (i=0; i < nlines; i++) {
 526                 char *node;
 527                 uint32_t flags;
 528                 size_t len;
 529
 530                 node = lines[i];
 531                 /* strip leading spaces */
 532                 while((*node == ' ') || (*node == '\t')) {
 533                         node++;
 534                 }
 535
 536                 len = strlen(node);
 537
 538                 while ((len > 1) &&
 539                        ((node[len-1] == ' ') || (node[len-1] == '\t')))
 540                 {
 541                         node[len-1] = '\0';
 542                         len--;
 543                 }
 544
 545                 if (len == 0) {
 546                         continue;
 547                 }
 548                 if (*node == '#') {
 549                         /* A "deleted" node is a node that is
 550                            commented out in the nodes file.  This is
 551                            used instead of removing a line, which
 552                            would cause subsequent nodes to change
 553                            their PNN. */
 554                         flags = NODE_FLAGS_DELETED;
 555                         node = discard_const("0.0.0.0");
 556                 } else {
 557                         flags = 0;
 558                 }
 559                 if (!node_map_add(mem_ctx, node, flags, &ret)) {
 560                         talloc_free(lines);
 561                         TALLOC_FREE(ret);
 562                         return NULL;
 563                 }
 564         }
 565
 566         talloc_free(lines);
 567         return ret;
 568 }
 569
 570 struct ctdb_node_map_old *
 571 ctdb_node_list_to_map(struct ctdb_node **nodes, uint32_t num_nodes,
 572                       TALLOC_CTX *mem_ctx)
 573 {
 574         uint32_t i;
 575         size_t size;
 576         struct ctdb_node_map_old *node_map;
 577
 578         size = offsetof(struct ctdb_node_map_old, nodes) +
 579                 num_nodes * sizeof(struct ctdb_node_and_flags);
 580         node_map  = (struct ctdb_node_map_old *)talloc_zero_size(mem_ctx, size);
 581         if (node_map == NULL) {
 582                 DEBUG(DEBUG_ERR,
 583                       (__location__ " Failed to allocate nodemap array\n"));
 584                 return NULL;
 585         }
 586
 587         node_map->num = num_nodes;
 588         for (i=0; i<num_nodes; i++) {
 589                 node_map->nodes[i].addr  = nodes[i]->address;
 590                 node_map->nodes[i].pnn   = nodes[i]->pnn;
 591                 node_map->nodes[i].flags = nodes[i]->flags;
 592         }
 593
 594         return node_map;
 595 }
 596
 597 const char *ctdb_eventscript_call_names[] = {
 598         "init",
 599         "setup",
 600         "startup",
 601         "startrecovery",
 602         "recovered",
 603         "takeip",
 604         "releaseip",
 605         "stopped",
 606         "monitor",
 607         "status",
 608         "shutdown",
 609         "reload",
 610         "updateip",
 611         "ipreallocated"
 612 };
 613
 614 /* Runstate handling */
 615 static struct {
 616         enum ctdb_runstate runstate;
 617         const char * label;
 618 } runstate_map[] = {
 619         { CTDB_RUNSTATE_UNKNOWN, "UNKNOWN" },
 620         { CTDB_RUNSTATE_INIT, "INIT" },
 621         { CTDB_RUNSTATE_SETUP, "SETUP" },
 622         { CTDB_RUNSTATE_FIRST_RECOVERY, "FIRST_RECOVERY" },
 623         { CTDB_RUNSTATE_STARTUP, "STARTUP" },
 624         { CTDB_RUNSTATE_RUNNING, "RUNNING" },
 625         { CTDB_RUNSTATE_SHUTDOWN, "SHUTDOWN" },
 626         { -1, NULL },
 627 };
 628
 629 const char *runstate_to_string(enum ctdb_runstate runstate)
 630 {
 631         int i;
 632         for (i=0; runstate_map[i].label != NULL ; i++) {
 633                 if (runstate_map[i].runstate == runstate) {
 634                         return runstate_map[i].label;
 635                 }
 636         }
 637
 638         return runstate_map[0].label;
 639 }
 640
 641 enum ctdb_runstate runstate_from_string(const char *label)
 642 {
 643         int i;
 644         for (i=0; runstate_map[i].label != NULL; i++) {
 645                 if (strcasecmp(runstate_map[i].label, label) == 0) {
 646                         return runstate_map[i].runstate;
 647                 }
 648         }
 649
 650         return CTDB_RUNSTATE_UNKNOWN;
 651 }
 652
 653 void ctdb_set_runstate(struct ctdb_context *ctdb, enum ctdb_runstate runstate)
 654 {
 655         DEBUG(DEBUG_NOTICE,("Set runstate to %s (%d)\n",
 656                             runstate_to_string(runstate), runstate));
 657
 658         if (runstate <= ctdb->runstate) {
 659                 ctdb_fatal(ctdb, "runstate must always increase");
 660         }
 661
 662         ctdb->runstate = runstate;
 663 }
 664
 665 /* Convert arbitrary data to 4-byte boundary padded uint32 array */
 666 uint32_t *ctdb_key_to_idkey(TALLOC_CTX *mem_ctx, TDB_DATA key)
 667 {
 668         uint32_t idkey_size, *k;
 669
 670         idkey_size = 1 + (key.dsize + sizeof(uint32_t)-1) / sizeof(uint32_t);
 671
 672         k = talloc_zero_array(mem_ctx, uint32_t, idkey_size);
 673         if (k == NULL) {
 674                 return NULL;
 675         }
 676
 677         k[0] = idkey_size;
 678         memcpy(&k[1], key.dptr, key.dsize);
 679
 680         return k;
 681 }