ctdb/server/ctdb_call.c

   1 /*
   2    ctdb_call protocol 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   see http://wiki.samba.org/index.php/Samba_%26_Clustering for
  21   protocol design and packet details
  22 */
  23 #include "includes.h"
  24 #include "lib/events/events.h"
  25 #include "lib/tdb/include/tdb.h"
  26 #include "lib/util/dlinklist.h"
  27 #include "system/network.h"
  28 #include "system/filesys.h"
  29 #include "../include/ctdb_private.h"
  30
  31 /*
  32   find the ctdb_db from a db index
  33  */
  34  struct ctdb_db_context *find_ctdb_db(struct ctdb_context *ctdb, uint32_t id)
  35 {
  36         struct ctdb_db_context *ctdb_db;
  37
  38         for (ctdb_db=ctdb->db_list; ctdb_db; ctdb_db=ctdb_db->next) {
  39                 if (ctdb_db->db_id == id) {
  40                         break;
  41                 }
  42         }
  43         return ctdb_db;
  44 }
  45
  46
  47 /*
  48   a varient of input packet that can be used in lock requeue
  49 */
  50 static void ctdb_call_input_pkt(void *p, struct ctdb_req_header *hdr)
  51 {
  52         struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
  53         ctdb_input_pkt(ctdb, hdr);
  54 }
  55
  56
  57 /*
  58   send an error reply
  59 */
  60 static void ctdb_send_error(struct ctdb_context *ctdb,
  61                             struct ctdb_req_header *hdr, uint32_t status,
  62                             const char *fmt, ...) PRINTF_ATTRIBUTE(4,5);
  63 static void ctdb_send_error(struct ctdb_context *ctdb,
  64                             struct ctdb_req_header *hdr, uint32_t status,
  65                             const char *fmt, ...)
  66 {
  67         va_list ap;
  68         struct ctdb_reply_error *r;
  69         char *msg;
  70         int msglen, len;
  71
  72         if (ctdb->methods == NULL) {
  73                 DEBUG(DEBUG_ERR,(__location__ " Failed to send error. Transport is DOWN\n"));
  74                 return;
  75         }
  76
  77         va_start(ap, fmt);
  78         msg = talloc_vasprintf(ctdb, fmt, ap);
  79         if (msg == NULL) {
  80                 ctdb_fatal(ctdb, "Unable to allocate error in ctdb_send_error\n");
  81         }
  82         va_end(ap);
  83
  84         msglen = strlen(msg)+1;
  85         len = offsetof(struct ctdb_reply_error, msg);
  86         r = ctdb_transport_allocate(ctdb, msg, CTDB_REPLY_ERROR, len + msglen,
  87                                     struct ctdb_reply_error);
  88         CTDB_NO_MEMORY_FATAL(ctdb, r);
  89
  90         r->hdr.destnode  = hdr->srcnode;
  91         r->hdr.reqid     = hdr->reqid;
  92         r->status        = status;
  93         r->msglen        = msglen;
  94         memcpy(&r->msg[0], msg, msglen);
  95
  96         ctdb_queue_packet(ctdb, &r->hdr);
  97
  98         talloc_free(msg);
  99 }
 100
 101
 102 /*
 103   send a redirect reply
 104 */
 105 static void ctdb_call_send_redirect(struct ctdb_context *ctdb,
 106                                     TDB_DATA key,
 107                                     struct ctdb_req_call *c,
 108                                     struct ctdb_ltdb_header *header)
 109 {
 110
 111         uint32_t lmaster = ctdb_lmaster(ctdb, &key);
 112         if (ctdb->pnn == lmaster) {
 113                 c->hdr.destnode = header->dmaster;
 114         } else if ((c->hopcount % ctdb->tunable.max_redirect_count) == 0) {
 115                 c->hdr.destnode = lmaster;
 116         } else {
 117                 c->hdr.destnode = header->dmaster;
 118         }
 119         c->hopcount++;
 120         ctdb_queue_packet(ctdb, &c->hdr);
 121 }
 122
 123
 124 /*
 125   send a dmaster reply
 126
 127   caller must have the chainlock before calling this routine. Caller must be
 128   the lmaster
 129 */
 130 static void ctdb_send_dmaster_reply(struct ctdb_db_context *ctdb_db,
 131                                     struct ctdb_ltdb_header *header,
 132                                     TDB_DATA key, TDB_DATA data,
 133                                     uint32_t new_dmaster,
 134                                     uint32_t reqid)
 135 {
 136         struct ctdb_context *ctdb = ctdb_db->ctdb;
 137         struct ctdb_reply_dmaster *r;
 138         int ret, len;
 139         TALLOC_CTX *tmp_ctx;
 140
 141         if (ctdb->pnn != ctdb_lmaster(ctdb, &key)) {
 142                 DEBUG(DEBUG_ALERT,(__location__ " Caller is not lmaster!\n"));
 143                 return;
 144         }
 145
 146         header->dmaster = new_dmaster;
 147         ret = ctdb_ltdb_store(ctdb_db, key, header, data);
 148         if (ret != 0) {
 149                 ctdb_fatal(ctdb, "ctdb_send_dmaster_reply unable to update dmaster");
 150                 return;
 151         }
 152
 153         if (ctdb->methods == NULL) {
 154                 ctdb_fatal(ctdb, "ctdb_send_dmaster_reply cant update dmaster since transport is down");
 155                 return;
 156         }
 157
 158         /* put the packet on a temporary context, allowing us to safely free
 159            it below even if ctdb_reply_dmaster() has freed it already */
 160         tmp_ctx = talloc_new(ctdb);
 161
 162         /* send the CTDB_REPLY_DMASTER */
 163         len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize;
 164         r = ctdb_transport_allocate(ctdb, tmp_ctx, CTDB_REPLY_DMASTER, len,
 165                                     struct ctdb_reply_dmaster);
 166         CTDB_NO_MEMORY_FATAL(ctdb, r);
 167
 168         r->hdr.destnode  = new_dmaster;
 169         r->hdr.reqid     = reqid;
 170         r->rsn           = header->rsn;
 171         r->keylen        = key.dsize;
 172         r->datalen       = data.dsize;
 173         r->db_id         = ctdb_db->db_id;
 174         memcpy(&r->data[0], key.dptr, key.dsize);
 175         memcpy(&r->data[key.dsize], data.dptr, data.dsize);
 176
 177         ctdb_queue_packet(ctdb, &r->hdr);
 178
 179         talloc_free(tmp_ctx);
 180 }
 181
 182 /*
 183   send a dmaster request (give another node the dmaster for a record)
 184
 185   This is always sent to the lmaster, which ensures that the lmaster
 186   always knows who the dmaster is. The lmaster will then send a
 187   CTDB_REPLY_DMASTER to the new dmaster
 188 */
 189 static void ctdb_call_send_dmaster(struct ctdb_db_context *ctdb_db,
 190                                    struct ctdb_req_call *c,
 191                                    struct ctdb_ltdb_header *header,
 192                                    TDB_DATA *key, TDB_DATA *data)
 193 {
 194         struct ctdb_req_dmaster *r;
 195         struct ctdb_context *ctdb = ctdb_db->ctdb;
 196         int len;
 197         uint32_t lmaster = ctdb_lmaster(ctdb, key);
 198
 199         if (ctdb->methods == NULL) {
 200                 ctdb_fatal(ctdb, "Failed ctdb_call_send_dmaster since transport is down");
 201                 return;
 202         }
 203
 204         if (lmaster == ctdb->pnn) {
 205                 ctdb_send_dmaster_reply(ctdb_db, header, *key, *data,
 206                                         c->hdr.srcnode, c->hdr.reqid);
 207                 return;
 208         }
 209
 210         len = offsetof(struct ctdb_req_dmaster, data) + key->dsize + data->dsize;
 211         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_DMASTER, len,
 212                                     struct ctdb_req_dmaster);
 213         CTDB_NO_MEMORY_FATAL(ctdb, r);
 214         r->hdr.destnode  = lmaster;
 215         r->hdr.reqid     = c->hdr.reqid;
 216         r->db_id         = c->db_id;
 217         r->rsn           = header->rsn;
 218         r->dmaster       = c->hdr.srcnode;
 219         r->keylen        = key->dsize;
 220         r->datalen       = data->dsize;
 221         memcpy(&r->data[0], key->dptr, key->dsize);
 222         memcpy(&r->data[key->dsize], data->dptr, data->dsize);
 223
 224         header->dmaster = c->hdr.srcnode;
 225         if (ctdb_ltdb_store(ctdb_db, *key, header, *data) != 0) {
 226                 ctdb_fatal(ctdb, "Failed to store record in ctdb_call_send_dmaster");
 227         }
 228
 229         ctdb_queue_packet(ctdb, &r->hdr);
 230
 231         talloc_free(r);
 232 }
 233
 234 /*
 235   called when a CTDB_REPLY_DMASTER packet comes in, or when the lmaster
 236   gets a CTDB_REQUEST_DMASTER for itself. We become the dmaster.
 237
 238   must be called with the chainlock held. This function releases the chainlock
 239 */
 240 static void ctdb_become_dmaster(struct ctdb_db_context *ctdb_db,
 241                                 struct ctdb_req_header *hdr,
 242                                 TDB_DATA key, TDB_DATA data,
 243                                 uint64_t rsn)
 244 {
 245         struct ctdb_call_state *state;
 246         struct ctdb_context *ctdb = ctdb_db->ctdb;
 247         struct ctdb_ltdb_header header;
 248
 249         DEBUG(DEBUG_DEBUG,("pnn %u dmaster response %08x\n", ctdb->pnn, ctdb_hash(&key)));
 250
 251         ZERO_STRUCT(header);
 252         header.rsn = rsn + 1;
 253         header.dmaster = ctdb->pnn;
 254
 255         if (ctdb_ltdb_store(ctdb_db, key, &header, data) != 0) {
 256                 ctdb_fatal(ctdb, "ctdb_reply_dmaster store failed\n");
 257                 ctdb_ltdb_unlock(ctdb_db, key);
 258                 return;
 259         }
 260
 261         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 262
 263         if (state == NULL) {
 264                 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_become_dmaster from node %u\n",
 265                          ctdb->pnn, hdr->reqid, hdr->srcnode));
 266                 ctdb_ltdb_unlock(ctdb_db, key);
 267                 return;
 268         }
 269
 270         if (hdr->reqid != state->reqid) {
 271                 /* we found a record  but it was the wrong one */
 272                 DEBUG(DEBUG_ERR, ("Dropped orphan in ctdb_become_dmaster with reqid:%u\n from node %u", hdr->reqid, hdr->srcnode));
 273                 ctdb_ltdb_unlock(ctdb_db, key);
 274                 return;
 275         }
 276
 277         ctdb_call_local(ctdb_db, state->call, &header, state, &data, ctdb->pnn);
 278
 279         ctdb_ltdb_unlock(ctdb_db, state->call->key);
 280
 281         state->state = CTDB_CALL_DONE;
 282         if (state->async.fn) {
 283                 state->async.fn(state);
 284         }
 285 }
 286
 287
 288
 289 /*
 290   called when a CTDB_REQ_DMASTER packet comes in
 291
 292   this comes into the lmaster for a record when the current dmaster
 293   wants to give up the dmaster role and give it to someone else
 294 */
 295 void ctdb_request_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 296 {
 297         struct ctdb_req_dmaster *c = (struct ctdb_req_dmaster *)hdr;
 298         TDB_DATA key, data, data2;
 299         struct ctdb_ltdb_header header;
 300         struct ctdb_db_context *ctdb_db;
 301         int ret;
 302
 303         key.dptr = c->data;
 304         key.dsize = c->keylen;
 305         data.dptr = c->data + c->keylen;
 306         data.dsize = c->datalen;
 307
 308         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 309         if (!ctdb_db) {
 310                 ctdb_send_error(ctdb, hdr, -1,
 311                                 "Unknown database in request. db_id==0x%08x",
 312                                 c->db_id);
 313                 return;
 314         }
 315
 316         /* fetch the current record */
 317         ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, key, &header, hdr, &data2,
 318                                            ctdb_call_input_pkt, ctdb, False);
 319         if (ret == -1) {
 320                 ctdb_fatal(ctdb, "ctdb_req_dmaster failed to fetch record");
 321                 return;
 322         }
 323         if (ret == -2) {
 324                 DEBUG(DEBUG_INFO,(__location__ " deferring ctdb_request_dmaster\n"));
 325                 return;
 326         }
 327
 328         if (ctdb_lmaster(ctdb, &key) != ctdb->pnn) {
 329                 DEBUG(DEBUG_ALERT,("pnn %u dmaster request to non-lmaster lmaster=%u gen=%u curgen=%u\n",
 330                          ctdb->pnn, ctdb_lmaster(ctdb, &key),
 331                          hdr->generation, ctdb->vnn_map->generation));
 332                 ctdb_fatal(ctdb, "ctdb_req_dmaster to non-lmaster");
 333         }
 334
 335         DEBUG(DEBUG_DEBUG,("pnn %u dmaster request on %08x for %u from %u\n",
 336                  ctdb->pnn, ctdb_hash(&key), c->dmaster, c->hdr.srcnode));
 337
 338         /* its a protocol error if the sending node is not the current dmaster */
 339         if (header.dmaster != hdr->srcnode) {
 340                 DEBUG(DEBUG_ALERT,("pnn %u dmaster request for new-dmaster %u from non-master %u real-dmaster=%u key %08x dbid 0x%08x gen=%u curgen=%u c->rsn=%llu header.rsn=%llu reqid=%u keyval=0x%08x\n",
 341                          ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
 342                          ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation,
 343                          (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid,
 344                          (key.dsize >= 4)?(*(uint32_t *)key.dptr):0));
 345                 if (header.rsn != 0 || header.dmaster != ctdb->pnn) {
 346                         ctdb_fatal(ctdb, "ctdb_req_dmaster from non-master");
 347                         return;
 348                 }
 349         }
 350
 351         if (header.rsn > c->rsn) {
 352                 DEBUG(DEBUG_ALERT,("pnn %u dmaster request with older RSN new-dmaster %u from %u real-dmaster=%u key %08x dbid 0x%08x gen=%u curgen=%u c->rsn=%llu header.rsn=%llu reqid=%u\n",
 353                          ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
 354                          ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation,
 355                          (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid));
 356         }
 357
 358         /* use the rsn from the sending node */
 359         header.rsn = c->rsn;
 360
 361         /* check if the new dmaster is the lmaster, in which case we
 362            skip the dmaster reply */
 363         if (c->dmaster == ctdb->pnn) {
 364                 ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn);
 365         } else {
 366                 ctdb_send_dmaster_reply(ctdb_db, &header, key, data, c->dmaster, hdr->reqid);
 367                 ctdb_ltdb_unlock(ctdb_db, key);
 368         }
 369 }
 370
 371
 372 /*
 373   called when a CTDB_REQ_CALL packet comes in
 374 */
 375 void ctdb_request_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 376 {
 377         struct ctdb_req_call *c = (struct ctdb_req_call *)hdr;
 378         TDB_DATA data;
 379         struct ctdb_reply_call *r;
 380         int ret, len;
 381         struct ctdb_ltdb_header header;
 382         struct ctdb_call *call;
 383         struct ctdb_db_context *ctdb_db;
 384
 385         if (ctdb->methods == NULL) {
 386                 DEBUG(DEBUG_ERR,(__location__ " Failed ctdb_request_call. Transport is DOWN\n"));
 387                 return;
 388         }
 389
 390
 391         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 392         if (!ctdb_db) {
 393                 ctdb_send_error(ctdb, hdr, -1,
 394                                 "Unknown database in request. db_id==0x%08x",
 395                                 c->db_id);
 396                 return;
 397         }
 398
 399         call = talloc(hdr, struct ctdb_call);
 400         CTDB_NO_MEMORY_FATAL(ctdb, call);
 401
 402         call->call_id  = c->callid;
 403         call->key.dptr = c->data;
 404         call->key.dsize = c->keylen;
 405         call->call_data.dptr = c->data + c->keylen;
 406         call->call_data.dsize = c->calldatalen;
 407
 408         /* determine if we are the dmaster for this key. This also
 409            fetches the record data (if any), thus avoiding a 2nd fetch of the data
 410            if the call will be answered locally */
 411
 412         ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, call->key, &header, hdr, &data,
 413                                            ctdb_call_input_pkt, ctdb, False);
 414         if (ret == -1) {
 415                 ctdb_send_error(ctdb, hdr, ret, "ltdb fetch failed in ctdb_request_call");
 416                 return;
 417         }
 418         if (ret == -2) {
 419                 DEBUG(DEBUG_INFO,(__location__ " deferred ctdb_request_call\n"));
 420                 return;
 421         }
 422
 423         /* if we are not the dmaster, then send a redirect to the
 424            requesting node */
 425         if (header.dmaster != ctdb->pnn) {
 426                 talloc_free(data.dptr);
 427                 ctdb_call_send_redirect(ctdb, call->key, c, &header);
 428                 ctdb_ltdb_unlock(ctdb_db, call->key);
 429                 return;
 430         }
 431
 432         if (c->hopcount > ctdb->statistics.max_hop_count) {
 433                 ctdb->statistics.max_hop_count = c->hopcount;
 434         }
 435
 436         /* if this nodes has done enough consecutive calls on the same record
 437            then give them the record
 438            or if the node requested an immediate migration
 439         */
 440         if ( c->hdr.srcnode != ctdb->pnn &&
 441              ((header.laccessor == c->hdr.srcnode
 442                && header.lacount >= ctdb->tunable.max_lacount)
 443               || (c->flags & CTDB_IMMEDIATE_MIGRATION)) ) {
 444                 if (ctdb_db->transaction_active) {
 445                         DEBUG(DEBUG_INFO, (__location__ " refusing migration"
 446                               " of key %s while transaction is active\n",
 447                               (char *)call->key.dptr));
 448                 } else {
 449                         DEBUG(DEBUG_DEBUG,("pnn %u starting migration of %08x to %u\n",
 450                                  ctdb->pnn, ctdb_hash(&(call->key)), c->hdr.srcnode));
 451                         ctdb_call_send_dmaster(ctdb_db, c, &header, &(call->key), &data);
 452                         talloc_free(data.dptr);
 453                         ctdb_ltdb_unlock(ctdb_db, call->key);
 454                         return;
 455                 }
 456         }
 457
 458         ctdb_call_local(ctdb_db, call, &header, hdr, &data, c->hdr.srcnode);
 459
 460         ctdb_ltdb_unlock(ctdb_db, call->key);
 461
 462         len = offsetof(struct ctdb_reply_call, data) + call->reply_data.dsize;
 463         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len,
 464                                     struct ctdb_reply_call);
 465         CTDB_NO_MEMORY_FATAL(ctdb, r);
 466         r->hdr.destnode  = hdr->srcnode;
 467         r->hdr.reqid     = hdr->reqid;
 468         r->status        = call->status;
 469         r->datalen       = call->reply_data.dsize;
 470         if (call->reply_data.dsize) {
 471                 memcpy(&r->data[0], call->reply_data.dptr, call->reply_data.dsize);
 472         }
 473
 474         ctdb_queue_packet(ctdb, &r->hdr);
 475
 476         talloc_free(r);
 477 }
 478
 479 /*
 480   called when a CTDB_REPLY_CALL packet comes in
 481
 482   This packet comes in response to a CTDB_REQ_CALL request packet. It
 483   contains any reply data from the call
 484 */
 485 void ctdb_reply_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 486 {
 487         struct ctdb_reply_call *c = (struct ctdb_reply_call *)hdr;
 488         struct ctdb_call_state *state;
 489
 490         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 491         if (state == NULL) {
 492                 DEBUG(DEBUG_ERR, (__location__ " reqid %u not found\n", hdr->reqid));
 493                 return;
 494         }
 495
 496         if (hdr->reqid != state->reqid) {
 497                 /* we found a record  but it was the wrong one */
 498                 DEBUG(DEBUG_ERR, ("Dropped orphaned call reply with reqid:%u\n",hdr->reqid));
 499                 return;
 500         }
 501
 502         state->call->reply_data.dptr = c->data;
 503         state->call->reply_data.dsize = c->datalen;
 504         state->call->status = c->status;
 505
 506         talloc_steal(state, c);
 507
 508         state->state = CTDB_CALL_DONE;
 509         if (state->async.fn) {
 510                 state->async.fn(state);
 511         }
 512 }
 513
 514
 515 /*
 516   called when a CTDB_REPLY_DMASTER packet comes in
 517
 518   This packet comes in from the lmaster response to a CTDB_REQ_CALL
 519   request packet. It means that the current dmaster wants to give us
 520   the dmaster role
 521 */
 522 void ctdb_reply_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 523 {
 524         struct ctdb_reply_dmaster *c = (struct ctdb_reply_dmaster *)hdr;
 525         struct ctdb_db_context *ctdb_db;
 526         TDB_DATA key, data;
 527         int ret;
 528
 529         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 530         if (ctdb_db == NULL) {
 531                 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_reply_dmaster\n", c->db_id));
 532                 return;
 533         }
 534
 535         key.dptr = c->data;
 536         key.dsize = c->keylen;
 537         data.dptr = &c->data[key.dsize];
 538         data.dsize = c->datalen;
 539
 540         ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
 541                                      ctdb_call_input_pkt, ctdb, False);
 542         if (ret == -2) {
 543                 return;
 544         }
 545         if (ret != 0) {
 546                 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_dmaster\n"));
 547                 return;
 548         }
 549
 550         ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn);
 551 }
 552
 553
 554 /*
 555   called when a CTDB_REPLY_ERROR packet comes in
 556 */
 557 void ctdb_reply_error(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 558 {
 559         struct ctdb_reply_error *c = (struct ctdb_reply_error *)hdr;
 560         struct ctdb_call_state *state;
 561
 562         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 563         if (state == NULL) {
 564                 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_reply_error\n",
 565                          ctdb->pnn, hdr->reqid));
 566                 return;
 567         }
 568
 569         if (hdr->reqid != state->reqid) {
 570                 /* we found a record  but it was the wrong one */
 571                 DEBUG(DEBUG_ERR, ("Dropped orphaned error reply with reqid:%u\n",hdr->reqid));
 572                 return;
 573         }
 574
 575         talloc_steal(state, c);
 576
 577         state->state  = CTDB_CALL_ERROR;
 578         state->errmsg = (char *)c->msg;
 579         if (state->async.fn) {
 580                 state->async.fn(state);
 581         }
 582 }
 583
 584
 585 /*
 586   destroy a ctdb_call
 587 */
 588 static int ctdb_call_destructor(struct ctdb_call_state *state)
 589 {
 590         DLIST_REMOVE(state->ctdb_db->ctdb->pending_calls, state);
 591         ctdb_reqid_remove(state->ctdb_db->ctdb, state->reqid);
 592         return 0;
 593 }
 594
 595
 596 /*
 597   called when a ctdb_call needs to be resent after a reconfigure event
 598 */
 599 static void ctdb_call_resend(struct ctdb_call_state *state)
 600 {
 601         struct ctdb_context *ctdb = state->ctdb_db->ctdb;
 602
 603         state->generation = ctdb->vnn_map->generation;
 604
 605         /* use a new reqid, in case the old reply does eventually come in */
 606         ctdb_reqid_remove(ctdb, state->reqid);
 607         state->reqid = ctdb_reqid_new(ctdb, state);
 608         state->c->hdr.reqid = state->reqid;
 609
 610         /* update the generation count for this request, so its valid with the new vnn_map */
 611         state->c->hdr.generation = state->generation;
 612
 613         /* send the packet to ourselves, it will be redirected appropriately */
 614         state->c->hdr.destnode = ctdb->pnn;
 615
 616         ctdb_queue_packet(ctdb, &state->c->hdr);
 617         DEBUG(DEBUG_NOTICE,("resent ctdb_call\n"));
 618 }
 619
 620 /*
 621   resend all pending calls on recovery
 622  */
 623 void ctdb_call_resend_all(struct ctdb_context *ctdb)
 624 {
 625         struct ctdb_call_state *state, *next;
 626         for (state=ctdb->pending_calls;state;state=next) {
 627                 next = state->next;
 628                 ctdb_call_resend(state);
 629         }
 630 }
 631
 632 /*
 633   this allows the caller to setup a async.fn
 634 */
 635 static void call_local_trigger(struct event_context *ev, struct timed_event *te,
 636                        struct timeval t, void *private_data)
 637 {
 638         struct ctdb_call_state *state = talloc_get_type(private_data, struct ctdb_call_state);
 639         if (state->async.fn) {
 640                 state->async.fn(state);
 641         }
 642 }
 643
 644
 645 /*
 646   construct an event driven local ctdb_call
 647
 648   this is used so that locally processed ctdb_call requests are processed
 649   in an event driven manner
 650 */
 651 struct ctdb_call_state *ctdb_call_local_send(struct ctdb_db_context *ctdb_db,
 652                                              struct ctdb_call *call,
 653                                              struct ctdb_ltdb_header *header,
 654                                              TDB_DATA *data)
 655 {
 656         struct ctdb_call_state *state;
 657         struct ctdb_context *ctdb = ctdb_db->ctdb;
 658         int ret;
 659
 660         state = talloc_zero(ctdb_db, struct ctdb_call_state);
 661         CTDB_NO_MEMORY_NULL(ctdb, state);
 662
 663         talloc_steal(state, data->dptr);
 664
 665         state->state = CTDB_CALL_DONE;
 666         state->call  = talloc(state, struct ctdb_call);
 667         CTDB_NO_MEMORY_NULL(ctdb, state->call);
 668         *(state->call) = *call;
 669         state->ctdb_db = ctdb_db;
 670
 671         ret = ctdb_call_local(ctdb_db, state->call, header, state, data, ctdb->pnn);
 672
 673         event_add_timed(ctdb->ev, state, timeval_zero(), call_local_trigger, state);
 674
 675         return state;
 676 }
 677
 678
 679 /*
 680   make a remote ctdb call - async send. Called in daemon context.
 681
 682   This constructs a ctdb_call request and queues it for processing.
 683   This call never blocks.
 684 */
 685 struct ctdb_call_state *ctdb_daemon_call_send_remote(struct ctdb_db_context *ctdb_db,
 686                                                      struct ctdb_call *call,
 687                                                      struct ctdb_ltdb_header *header)
 688 {
 689         uint32_t len;
 690         struct ctdb_call_state *state;
 691         struct ctdb_context *ctdb = ctdb_db->ctdb;
 692
 693         if (ctdb->methods == NULL) {
 694                 DEBUG(DEBUG_ERR,(__location__ " Failed send packet. Transport is down\n"));
 695                 return NULL;
 696         }
 697
 698         state = talloc_zero(ctdb_db, struct ctdb_call_state);
 699         CTDB_NO_MEMORY_NULL(ctdb, state);
 700         state->call = talloc(state, struct ctdb_call);
 701         CTDB_NO_MEMORY_NULL(ctdb, state->call);
 702
 703         state->reqid = ctdb_reqid_new(ctdb, state);
 704         state->ctdb_db = ctdb_db;
 705         talloc_set_destructor(state, ctdb_call_destructor);
 706
 707         len = offsetof(struct ctdb_req_call, data) + call->key.dsize + call->call_data.dsize;
 708         state->c = ctdb_transport_allocate(ctdb, state, CTDB_REQ_CALL, len,
 709                                            struct ctdb_req_call);
 710         CTDB_NO_MEMORY_NULL(ctdb, state->c);
 711         state->c->hdr.destnode  = header->dmaster;
 712
 713         /* this limits us to 16k outstanding messages - not unreasonable */
 714         state->c->hdr.reqid     = state->reqid;
 715         state->c->flags         = call->flags;
 716         state->c->db_id         = ctdb_db->db_id;
 717         state->c->callid        = call->call_id;
 718         state->c->hopcount      = 0;
 719         state->c->keylen        = call->key.dsize;
 720         state->c->calldatalen   = call->call_data.dsize;
 721         memcpy(&state->c->data[0], call->key.dptr, call->key.dsize);
 722         memcpy(&state->c->data[call->key.dsize],
 723                call->call_data.dptr, call->call_data.dsize);
 724         *(state->call)              = *call;
 725         state->call->call_data.dptr = &state->c->data[call->key.dsize];
 726         state->call->key.dptr       = &state->c->data[0];
 727
 728         state->state  = CTDB_CALL_WAIT;
 729         state->generation = ctdb->vnn_map->generation;
 730
 731         DLIST_ADD(ctdb->pending_calls, state);
 732
 733         ctdb_queue_packet(ctdb, &state->c->hdr);
 734
 735         return state;
 736 }
 737
 738 /*
 739   make a remote ctdb call - async recv - called in daemon context
 740
 741   This is called when the program wants to wait for a ctdb_call to complete and get the
 742   results. This call will block unless the call has already completed.
 743 */
 744 int ctdb_daemon_call_recv(struct ctdb_call_state *state, struct ctdb_call *call)
 745 {
 746         while (state->state < CTDB_CALL_DONE) {
 747                 event_loop_once(state->ctdb_db->ctdb->ev);
 748         }
 749         if (state->state != CTDB_CALL_DONE) {
 750                 ctdb_set_error(state->ctdb_db->ctdb, "%s", state->errmsg);
 751                 talloc_free(state);
 752                 return -1;
 753         }
 754
 755         if (state->call->reply_data.dsize) {
 756                 call->reply_data.dptr = talloc_memdup(call,
 757                                                       state->call->reply_data.dptr,
 758                                                       state->call->reply_data.dsize);
 759                 call->reply_data.dsize = state->call->reply_data.dsize;
 760         } else {
 761                 call->reply_data.dptr = NULL;
 762                 call->reply_data.dsize = 0;
 763         }
 764         call->status = state->call->status;
 765         talloc_free(state);
 766         return 0;
 767 }
 768
 769
 770 /*
 771    send a keepalive packet to the other node
 772 */
 773 void ctdb_send_keepalive(struct ctdb_context *ctdb, uint32_t destnode)
 774 {
 775         struct ctdb_req_keepalive *r;
 776
 777         if (ctdb->methods == NULL) {
 778                 DEBUG(DEBUG_ERR,(__location__ " Failed to send keepalive. Transport is DOWN\n"));
 779                 return;
 780         }
 781
 782         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_KEEPALIVE,
 783                                     sizeof(struct ctdb_req_keepalive),
 784                                     struct ctdb_req_keepalive);
 785         CTDB_NO_MEMORY_FATAL(ctdb, r);
 786         r->hdr.destnode  = destnode;
 787         r->hdr.reqid     = 0;
 788
 789         ctdb->statistics.keepalive_packets_sent++;
 790
 791         ctdb_queue_packet(ctdb, &r->hdr);
 792
 793         talloc_free(r);
 794 }