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                 if (snprintf(helper, size, "%s/%s", dir, file) >= size) {
 115                         DEBUG(DEBUG_ERR,
 116                               ("Unable to set %s - path too long\n", type));
 117                         return false;
 118                 }
 119         }
 120
 121         if (stat(helper, &st) != 0) {
 122                 DEBUG(DEBUG_ERR,
 123                       ("Unable to set %s \"%s\" - %s\n",
 124                        type, helper, strerror(errno)));
 125                 return false;
 126         }
 127         if (!(st.st_mode & S_IXUSR)) {
 128                 DEBUG(DEBUG_ERR,
 129                       ("Unable to set %s \"%s\" - not executable\n",
 130                        type, helper));
 131                 return false;
 132         }
 133
 134         DEBUG(DEBUG_NOTICE,
 135               ("Set %s to \"%s\"\n", type, helper));
 136         return true;
 137 }
 138
 139 /*
 140   parse a IP:port pair
 141 */
 142 int ctdb_parse_address(TALLOC_CTX *mem_ctx, const char *str,
 143                        ctdb_sock_addr *address)
 144 {
 145         struct servent *se;
 146         int port;
 147         int ret;
 148
 149         setservent(0);
 150         se = getservbyname("ctdb", "tcp");
 151         endservent();
 152
 153         if (se == NULL) {
 154                 port = CTDB_PORT;
 155         } else {
 156                 port = ntohs(se->s_port);
 157         }
 158
 159         ret = ctdb_sock_addr_from_string(str, address, false);
 160         if (ret != 0) {
 161                 return -1;
 162         }
 163         ctdb_sock_addr_set_port(address, port);
 164
 165         return 0;
 166 }
 167
 168
 169 /*
 170   check if two addresses are the same
 171 */
 172 bool ctdb_same_address(ctdb_sock_addr *a1, ctdb_sock_addr *a2)
 173 {
 174         return ctdb_same_ip(a1, a2) &&
 175                 ctdb_addr_to_port(a1) == ctdb_addr_to_port(a2);
 176 }
 177
 178
 179 /*
 180   hash function for mapping data to a VNN - taken from tdb
 181 */
 182 uint32_t ctdb_hash(const TDB_DATA *key)
 183 {
 184         return tdb_jenkins_hash(discard_const(key));
 185 }
 186
 187
 188 static uint32_t ctdb_marshall_record_size(TDB_DATA key,
 189                                           struct ctdb_ltdb_header *header,
 190                                           TDB_DATA data)
 191 {
 192         return offsetof(struct ctdb_rec_data_old, data) + key.dsize +
 193                data.dsize + (header ? sizeof(*header) : 0);
 194 }
 195
 196 static void ctdb_marshall_record_copy(struct ctdb_rec_data_old *rec,
 197                                       uint32_t reqid,
 198                                       TDB_DATA key,
 199                                       struct ctdb_ltdb_header *header,
 200                                       TDB_DATA data,
 201                                       uint32_t length)
 202 {
 203         uint32_t offset;
 204
 205         rec->length = length;
 206         rec->reqid = reqid;
 207         rec->keylen = key.dsize;
 208         memcpy(&rec->data[0], key.dptr, key.dsize);
 209         offset = key.dsize;
 210
 211         if (header) {
 212                 rec->datalen = data.dsize + sizeof(*header);
 213                 memcpy(&rec->data[offset], header, sizeof(*header));
 214                 offset += sizeof(*header);
 215         } else {
 216                 rec->datalen = data.dsize;
 217         }
 218         memcpy(&rec->data[offset], data.dptr, data.dsize);
 219 }
 220
 221 /*
 222   form a ctdb_rec_data record from a key/data pair
 223
 224   note that header may be NULL. If not NULL then it is included in the data portion
 225   of the record
 226  */
 227 struct ctdb_rec_data_old *ctdb_marshall_record(TALLOC_CTX *mem_ctx,
 228                                                uint32_t reqid,
 229                                                TDB_DATA key,
 230                                                struct ctdb_ltdb_header *header,
 231                                                TDB_DATA data)
 232 {
 233         size_t length;
 234         struct ctdb_rec_data_old *d;
 235
 236         length = ctdb_marshall_record_size(key, header, data);
 237
 238         d = (struct ctdb_rec_data_old *)talloc_size(mem_ctx, length);
 239         if (d == NULL) {
 240                 return NULL;
 241         }
 242
 243         ctdb_marshall_record_copy(d, reqid, key, header, data, length);
 244         return d;
 245 }
 246
 247
 248 /* helper function for marshalling multiple records */
 249 struct ctdb_marshall_buffer *ctdb_marshall_add(TALLOC_CTX *mem_ctx,
 250                                                struct ctdb_marshall_buffer *m,
 251                                                uint32_t db_id,
 252                                                uint32_t reqid,
 253                                                TDB_DATA key,
 254                                                struct ctdb_ltdb_header *header,
 255                                                TDB_DATA data)
 256 {
 257         struct ctdb_rec_data_old *r;
 258         struct ctdb_marshall_buffer *m2;
 259         uint32_t length, offset;
 260
 261         length = ctdb_marshall_record_size(key, header, data);
 262
 263         if (m == NULL) {
 264                 offset = offsetof(struct ctdb_marshall_buffer, data);
 265                 m2 = talloc_zero_size(mem_ctx, offset + length);
 266         } else {
 267                 offset = talloc_get_size(m);
 268                 m2 = talloc_realloc_size(mem_ctx, m, offset + length);
 269         }
 270         if (m2 == NULL) {
 271                 TALLOC_FREE(m);
 272                 return NULL;
 273         }
 274
 275         if (m == NULL) {
 276                 m2->db_id = db_id;
 277         }
 278
 279         r = (struct ctdb_rec_data_old *)((uint8_t *)m2 + offset);
 280         ctdb_marshall_record_copy(r, reqid, key, header, data, length);
 281         m2->count++;
 282
 283         return m2;
 284 }
 285
 286 /* we've finished marshalling, return a data blob with the marshalled records */
 287 TDB_DATA ctdb_marshall_finish(struct ctdb_marshall_buffer *m)
 288 {
 289         TDB_DATA data;
 290         data.dptr = (uint8_t *)m;
 291         data.dsize = talloc_get_size(m);
 292         return data;
 293 }
 294
 295 /*
 296    loop over a marshalling buffer
 297
 298      - pass r==NULL to start
 299      - loop the number of times indicated by m->count
 300 */
 301 struct ctdb_rec_data_old *ctdb_marshall_loop_next(
 302                                 struct ctdb_marshall_buffer *m,
 303                                 struct ctdb_rec_data_old *r,
 304                                 uint32_t *reqid,
 305                                 struct ctdb_ltdb_header *header,
 306                                 TDB_DATA *key, TDB_DATA *data)
 307 {
 308         if (r == NULL) {
 309                 r = (struct ctdb_rec_data_old *)&m->data[0];
 310         } else {
 311                 r = (struct ctdb_rec_data_old *)(r->length + (uint8_t *)r);
 312         }
 313
 314         if (reqid != NULL) {
 315                 *reqid = r->reqid;
 316         }
 317
 318         if (key != NULL) {
 319                 key->dptr   = &r->data[0];
 320                 key->dsize  = r->keylen;
 321         }
 322         if (data != NULL) {
 323                 data->dptr  = &r->data[r->keylen];
 324                 data->dsize = r->datalen;
 325                 if (header != NULL) {
 326                         data->dptr += sizeof(*header);
 327                         data->dsize -= sizeof(*header);
 328                 }
 329         }
 330
 331         if (header != NULL) {
 332                 if (r->datalen < sizeof(*header)) {
 333                         return NULL;
 334                 }
 335                 memcpy(header, &r->data[r->keylen], sizeof(*header));
 336         }
 337
 338         return r;
 339 }
 340
 341 /*
 342    This is used to canonicalize a ctdb_sock_addr structure.
 343 */
 344 void ctdb_canonicalize_ip(const ctdb_sock_addr *ip, ctdb_sock_addr *cip)
 345 {
 346         ZERO_STRUCTP(cip);
 347
 348         if (ip->sa.sa_family == AF_INET6) {
 349                 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
 350                 if (memcmp(&ip->ip6.sin6_addr, prefix, sizeof(prefix)) == 0) {
 351                         /* Copy IPv4-mapped IPv6 addresses as IPv4 */
 352                         cip->ip.sin_family = AF_INET;
 353 #ifdef HAVE_SOCK_SIN_LEN
 354                         cip->ip.sin_len = sizeof(ctdb_sock_addr);
 355 #endif
 356                         cip->ip.sin_port   = ip->ip6.sin6_port;
 357                         memcpy(&cip->ip.sin_addr,
 358                                &ip->ip6.sin6_addr.s6_addr[12],
 359                                sizeof(cip->ip.sin_addr));
 360                 } else {
 361                         cip->ip6.sin6_family = AF_INET6;
 362 #ifdef HAVE_SOCK_SIN6_LEN
 363                         cip->ip6.sin6_len = sizeof(ctdb_sock_addr);
 364 #endif
 365                         cip->ip6.sin6_port   = ip->ip6.sin6_port;
 366                         memcpy(&cip->ip6.sin6_addr,
 367                                &ip->ip6.sin6_addr,
 368                                sizeof(cip->ip6.sin6_addr));
 369                 }
 370
 371                 return;
 372         }
 373
 374         if (ip->sa.sa_family == AF_INET) {
 375                 cip->ip.sin_family = AF_INET;
 376 #ifdef HAVE_SOCK_SIN_LEN
 377                 cip->ip.sin_len = sizeof(ctdb_sock_addr);
 378 #endif
 379                 cip->ip.sin_port = ip->ip.sin_port;
 380                 memcpy(&cip->ip.sin_addr,
 381                        &ip->ip.sin_addr,
 382                        sizeof(ip->ip.sin_addr));
 383
 384                 return;
 385         }
 386 }
 387
 388 bool ctdb_same_ip(const ctdb_sock_addr *tip1, const ctdb_sock_addr *tip2)
 389 {
 390         ctdb_sock_addr ip1, ip2;
 391
 392         ctdb_canonicalize_ip(tip1, &ip1);
 393         ctdb_canonicalize_ip(tip2, &ip2);
 394
 395         if (ip1.sa.sa_family != ip2.sa.sa_family) {
 396                 return false;
 397         }
 398
 399         switch (ip1.sa.sa_family) {
 400         case AF_INET:
 401                 return ip1.ip.sin_addr.s_addr == ip2.ip.sin_addr.s_addr;
 402         case AF_INET6:
 403                 return !memcmp(&ip1.ip6.sin6_addr.s6_addr[0],
 404                                 &ip2.ip6.sin6_addr.s6_addr[0],
 405                                 16);
 406         default:
 407                 DEBUG(DEBUG_ERR, (__location__ " CRITICAL Can not compare sockaddr structures of type %u\n", ip1.sa.sa_family));
 408                 return false;
 409         }
 410
 411         return true;
 412 }
 413
 414 /*
 415   compare two ctdb_sock_addr structures
 416  */
 417 bool ctdb_same_sockaddr(const ctdb_sock_addr *ip1, const ctdb_sock_addr *ip2)
 418 {
 419         return ctdb_same_ip(ip1, ip2) && ip1->ip.sin_port == ip2->ip.sin_port;
 420 }
 421
 422 char *ctdb_addr_to_str(ctdb_sock_addr *addr)
 423 {
 424         static char cip[128] = "";
 425
 426         switch (addr->sa.sa_family) {
 427         case AF_INET:
 428                 inet_ntop(addr->ip.sin_family, &addr->ip.sin_addr, cip, sizeof(cip));
 429                 break;
 430         case AF_INET6:
 431                 inet_ntop(addr->ip6.sin6_family, &addr->ip6.sin6_addr, cip, sizeof(cip));
 432                 break;
 433         default:
 434                 DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family %u\n", addr->sa.sa_family));
 435         }
 436
 437         return cip;
 438 }
 439
 440 unsigned ctdb_addr_to_port(ctdb_sock_addr *addr)
 441 {
 442         switch (addr->sa.sa_family) {
 443         case AF_INET:
 444                 return ntohs(addr->ip.sin_port);
 445                 break;
 446         case AF_INET6:
 447                 return ntohs(addr->ip6.sin6_port);
 448                 break;
 449         default:
 450                 DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family %u\n", addr->sa.sa_family));
 451         }
 452
 453         return 0;
 454 }
 455
 456 /* Add a node to a node map with given address and flags */
 457 static bool node_map_add(TALLOC_CTX *mem_ctx,
 458                          const char *nstr, uint32_t flags,
 459                          struct ctdb_node_map_old **node_map)
 460 {
 461         ctdb_sock_addr addr;
 462         uint32_t num;
 463         size_t s;
 464         struct ctdb_node_and_flags *n;
 465
 466         /* Might as well do this before trying to allocate memory */
 467         if (ctdb_parse_address(mem_ctx, nstr, &addr) == -1) {
 468                 return false;
 469         }
 470
 471         num = (*node_map)->num + 1;
 472         s = offsetof(struct ctdb_node_map_old, nodes) +
 473                 num * sizeof(struct ctdb_node_and_flags);
 474         *node_map = talloc_realloc_size(mem_ctx, *node_map, s);
 475         if (*node_map == NULL) {
 476                 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
 477                 return false;
 478         }
 479
 480         n = &(*node_map)->nodes[(*node_map)->num];
 481         n->addr = addr;
 482         n->pnn = (*node_map)->num;
 483         n->flags = flags;
 484
 485         (*node_map)->num++;
 486
 487         return true;
 488 }
 489
 490 /* Read a nodes file into a node map */
 491 struct ctdb_node_map_old *ctdb_read_nodes_file(TALLOC_CTX *mem_ctx,
 492                                            const char *nlist)
 493 {
 494         char **lines;
 495         int nlines;
 496         int i;
 497         struct ctdb_node_map_old *ret;
 498
 499         /* Allocate node map header */
 500         ret = talloc_zero_size(mem_ctx, offsetof(struct ctdb_node_map_old, nodes));
 501         if (ret == NULL) {
 502                 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
 503                 return false;
 504         }
 505
 506         lines = file_lines_load(nlist, &nlines, 0, mem_ctx);
 507         if (lines == NULL) {
 508                 DEBUG(DEBUG_ERR, ("Failed to read nodes file \"%s\"\n", nlist));
 509                 return false;
 510         }
 511         while (nlines > 0 && strcmp(lines[nlines-1], "") == 0) {
 512                 nlines--;
 513         }
 514
 515         for (i=0; i < nlines; i++) {
 516                 char *node;
 517                 uint32_t flags;
 518                 size_t len;
 519
 520                 node = lines[i];
 521                 /* strip leading spaces */
 522                 while((*node == ' ') || (*node == '\t')) {
 523                         node++;
 524                 }
 525
 526                 len = strlen(node);
 527
 528                 while ((len > 1) &&
 529                        ((node[len-1] == ' ') || (node[len-1] == '\t')))
 530                 {
 531                         node[len-1] = '\0';
 532                         len--;
 533                 }
 534
 535                 if (len == 0) {
 536                         continue;
 537                 }
 538                 if (*node == '#') {
 539                         /* A "deleted" node is a node that is
 540                            commented out in the nodes file.  This is
 541                            used instead of removing a line, which
 542                            would cause subsequent nodes to change
 543                            their PNN. */
 544                         flags = NODE_FLAGS_DELETED;
 545                         node = discard_const("0.0.0.0");
 546                 } else {
 547                         flags = 0;
 548                 }
 549                 if (!node_map_add(mem_ctx, node, flags, &ret)) {
 550                         talloc_free(lines);
 551                         TALLOC_FREE(ret);
 552                         return NULL;
 553                 }
 554         }
 555
 556         talloc_free(lines);
 557         return ret;
 558 }
 559
 560 struct ctdb_node_map_old *
 561 ctdb_node_list_to_map(struct ctdb_node **nodes, uint32_t num_nodes,
 562                       TALLOC_CTX *mem_ctx)
 563 {
 564         uint32_t i;
 565         size_t size;
 566         struct ctdb_node_map_old *node_map;
 567
 568         size = offsetof(struct ctdb_node_map_old, nodes) +
 569                 num_nodes * sizeof(struct ctdb_node_and_flags);
 570         node_map  = (struct ctdb_node_map_old *)talloc_zero_size(mem_ctx, size);
 571         if (node_map == NULL) {
 572                 DEBUG(DEBUG_ERR,
 573                       (__location__ " Failed to allocate nodemap array\n"));
 574                 return NULL;
 575         }
 576
 577         node_map->num = num_nodes;
 578         for (i=0; i<num_nodes; i++) {
 579                 node_map->nodes[i].addr  = nodes[i]->address;
 580                 node_map->nodes[i].pnn   = nodes[i]->pnn;
 581                 node_map->nodes[i].flags = nodes[i]->flags;
 582         }
 583
 584         return node_map;
 585 }
 586
 587 const char *ctdb_eventscript_call_names[] = {
 588         "init",
 589         "setup",
 590         "startup",
 591         "startrecovery",
 592         "recovered",
 593         "takeip",
 594         "releaseip",
 595         "stopped",
 596         "monitor",
 597         "status",
 598         "shutdown",
 599         "reload",
 600         "updateip",
 601         "ipreallocated"
 602 };
 603
 604 /* Runstate handling */
 605 static struct {
 606         enum ctdb_runstate runstate;
 607         const char * label;
 608 } runstate_map[] = {
 609         { CTDB_RUNSTATE_UNKNOWN, "UNKNOWN" },
 610         { CTDB_RUNSTATE_INIT, "INIT" },
 611         { CTDB_RUNSTATE_SETUP, "SETUP" },
 612         { CTDB_RUNSTATE_FIRST_RECOVERY, "FIRST_RECOVERY" },
 613         { CTDB_RUNSTATE_STARTUP, "STARTUP" },
 614         { CTDB_RUNSTATE_RUNNING, "RUNNING" },
 615         { CTDB_RUNSTATE_SHUTDOWN, "SHUTDOWN" },
 616         { -1, NULL },
 617 };
 618
 619 const char *runstate_to_string(enum ctdb_runstate runstate)
 620 {
 621         int i;
 622         for (i=0; runstate_map[i].label != NULL ; i++) {
 623                 if (runstate_map[i].runstate == runstate) {
 624                         return runstate_map[i].label;
 625                 }
 626         }
 627
 628         return runstate_map[0].label;
 629 }
 630
 631 enum ctdb_runstate runstate_from_string(const char *label)
 632 {
 633         int i;
 634         for (i=0; runstate_map[i].label != NULL; i++) {
 635                 if (strcasecmp(runstate_map[i].label, label) == 0) {
 636                         return runstate_map[i].runstate;
 637                 }
 638         }
 639
 640         return CTDB_RUNSTATE_UNKNOWN;
 641 }
 642
 643 void ctdb_set_runstate(struct ctdb_context *ctdb, enum ctdb_runstate runstate)
 644 {
 645         DEBUG(DEBUG_NOTICE,("Set runstate to %s (%d)\n",
 646                             runstate_to_string(runstate), runstate));
 647
 648         if (runstate <= ctdb->runstate) {
 649                 ctdb_fatal(ctdb, "runstate must always increase");
 650         }
 651
 652         ctdb->runstate = runstate;
 653 }
 654
 655 /* Convert arbitrary data to 4-byte boundary padded uint32 array */
 656 uint32_t *ctdb_key_to_idkey(TALLOC_CTX *mem_ctx, TDB_DATA key)
 657 {
 658         uint32_t idkey_size, *k;
 659
 660         idkey_size = 1 + (key.dsize + sizeof(uint32_t)-1) / sizeof(uint32_t);
 661
 662         k = talloc_zero_array(mem_ctx, uint32_t, idkey_size);
 663         if (k == NULL) {
 664                 return NULL;
 665         }
 666
 667         k[0] = idkey_size;
 668         memcpy(&k[1], key.dptr, key.dsize);
 669
 670         return k;
 671 }