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         int ret;
 249
 250         DEBUG(DEBUG_DEBUG,("pnn %u dmaster response %08x\n", ctdb->pnn, ctdb_hash(&key)));
 251
 252         ZERO_STRUCT(header);
 253         header.rsn = rsn + 1;
 254         header.dmaster = ctdb->pnn;
 255
 256         if (ctdb_ltdb_store(ctdb_db, key, &header, data) != 0) {
 257                 ctdb_fatal(ctdb, "ctdb_reply_dmaster store failed\n");
 258
 259                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 260                 if (ret != 0) {
 261                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 262                 }
 263                 return;
 264         }
 265
 266         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 267
 268         if (state == NULL) {
 269                 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_become_dmaster from node %u\n",
 270                          ctdb->pnn, hdr->reqid, hdr->srcnode));
 271
 272                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 273                 if (ret != 0) {
 274                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 275                 }
 276                 return;
 277         }
 278
 279         if (hdr->reqid != state->reqid) {
 280                 /* we found a record  but it was the wrong one */
 281                 DEBUG(DEBUG_ERR, ("Dropped orphan in ctdb_become_dmaster with reqid:%u\n from node %u", hdr->reqid, hdr->srcnode));
 282
 283                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 284                 if (ret != 0) {
 285                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 286                 }
 287                 return;
 288         }
 289
 290         ctdb_call_local(ctdb_db, state->call, &header, state, &data, ctdb->pnn);
 291
 292         ret = ctdb_ltdb_unlock(ctdb_db, key);
 293         if (ret != 0) {
 294                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 295         }
 296
 297         state->state = CTDB_CALL_DONE;
 298         if (state->async.fn) {
 299                 state->async.fn(state);
 300         }
 301 }
 302
 303
 304
 305 /*
 306   called when a CTDB_REQ_DMASTER packet comes in
 307
 308   this comes into the lmaster for a record when the current dmaster
 309   wants to give up the dmaster role and give it to someone else
 310 */
 311 void ctdb_request_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 312 {
 313         struct ctdb_req_dmaster *c = (struct ctdb_req_dmaster *)hdr;
 314         TDB_DATA key, data, data2;
 315         struct ctdb_ltdb_header header;
 316         struct ctdb_db_context *ctdb_db;
 317         int ret;
 318
 319         key.dptr = c->data;
 320         key.dsize = c->keylen;
 321         data.dptr = c->data + c->keylen;
 322         data.dsize = c->datalen;
 323
 324         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 325         if (!ctdb_db) {
 326                 ctdb_send_error(ctdb, hdr, -1,
 327                                 "Unknown database in request. db_id==0x%08x",
 328                                 c->db_id);
 329                 return;
 330         }
 331
 332         /* fetch the current record */
 333         ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, key, &header, hdr, &data2,
 334                                            ctdb_call_input_pkt, ctdb, False);
 335         if (ret == -1) {
 336                 ctdb_fatal(ctdb, "ctdb_req_dmaster failed to fetch record");
 337                 return;
 338         }
 339         if (ret == -2) {
 340                 DEBUG(DEBUG_INFO,(__location__ " deferring ctdb_request_dmaster\n"));
 341                 return;
 342         }
 343
 344         if (ctdb_lmaster(ctdb, &key) != ctdb->pnn) {
 345                 DEBUG(DEBUG_ALERT,("pnn %u dmaster request to non-lmaster lmaster=%u gen=%u curgen=%u\n",
 346                          ctdb->pnn, ctdb_lmaster(ctdb, &key),
 347                          hdr->generation, ctdb->vnn_map->generation));
 348                 ctdb_fatal(ctdb, "ctdb_req_dmaster to non-lmaster");
 349         }
 350
 351         DEBUG(DEBUG_DEBUG,("pnn %u dmaster request on %08x for %u from %u\n",
 352                  ctdb->pnn, ctdb_hash(&key), c->dmaster, c->hdr.srcnode));
 353
 354         /* its a protocol error if the sending node is not the current dmaster */
 355         if (header.dmaster != hdr->srcnode) {
 356                 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",
 357                          ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
 358                          ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation,
 359                          (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid,
 360                          (key.dsize >= 4)?(*(uint32_t *)key.dptr):0));
 361                 if (header.rsn != 0 || header.dmaster != ctdb->pnn) {
 362                         ctdb_fatal(ctdb, "ctdb_req_dmaster from non-master");
 363                         return;
 364                 }
 365         }
 366
 367         if (header.rsn > c->rsn) {
 368                 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",
 369                          ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
 370                          ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation,
 371                          (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid));
 372         }
 373
 374         /* use the rsn from the sending node */
 375         header.rsn = c->rsn;
 376
 377         /* check if the new dmaster is the lmaster, in which case we
 378            skip the dmaster reply */
 379         if (c->dmaster == ctdb->pnn) {
 380                 ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn);
 381         } else {
 382                 ctdb_send_dmaster_reply(ctdb_db, &header, key, data, c->dmaster, hdr->reqid);
 383
 384                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 385                 if (ret != 0) {
 386                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 387                 }
 388         }
 389 }
 390
 391
 392 /*
 393   called when a CTDB_REQ_CALL packet comes in
 394 */
 395 void ctdb_request_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 396 {
 397         struct ctdb_req_call *c = (struct ctdb_req_call *)hdr;
 398         TDB_DATA data;
 399         struct ctdb_reply_call *r;
 400         int ret, len;
 401         struct ctdb_ltdb_header header;
 402         struct ctdb_call *call;
 403         struct ctdb_db_context *ctdb_db;
 404
 405         if (ctdb->methods == NULL) {
 406                 DEBUG(DEBUG_ERR,(__location__ " Failed ctdb_request_call. Transport is DOWN\n"));
 407                 return;
 408         }
 409
 410
 411         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 412         if (!ctdb_db) {
 413                 ctdb_send_error(ctdb, hdr, -1,
 414                                 "Unknown database in request. db_id==0x%08x",
 415                                 c->db_id);
 416                 return;
 417         }
 418
 419         call = talloc(hdr, struct ctdb_call);
 420         CTDB_NO_MEMORY_FATAL(ctdb, call);
 421
 422         call->call_id  = c->callid;
 423         call->key.dptr = c->data;
 424         call->key.dsize = c->keylen;
 425         call->call_data.dptr = c->data + c->keylen;
 426         call->call_data.dsize = c->calldatalen;
 427
 428         /* determine if we are the dmaster for this key. This also
 429            fetches the record data (if any), thus avoiding a 2nd fetch of the data
 430            if the call will be answered locally */
 431
 432         ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, call->key, &header, hdr, &data,
 433                                            ctdb_call_input_pkt, ctdb, False);
 434         if (ret == -1) {
 435                 ctdb_send_error(ctdb, hdr, ret, "ltdb fetch failed in ctdb_request_call");
 436                 return;
 437         }
 438         if (ret == -2) {
 439                 DEBUG(DEBUG_INFO,(__location__ " deferred ctdb_request_call\n"));
 440                 return;
 441         }
 442
 443         /* if we are not the dmaster, then send a redirect to the
 444            requesting node */
 445         if (header.dmaster != ctdb->pnn) {
 446                 talloc_free(data.dptr);
 447                 ctdb_call_send_redirect(ctdb, call->key, c, &header);
 448
 449                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 450                 if (ret != 0) {
 451                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 452                 }
 453                 return;
 454         }
 455
 456         if (c->hopcount > ctdb->statistics.max_hop_count) {
 457                 ctdb->statistics.max_hop_count = c->hopcount;
 458         }
 459
 460         /* if this nodes has done enough consecutive calls on the same record
 461            then give them the record
 462            or if the node requested an immediate migration
 463         */
 464         if ( c->hdr.srcnode != ctdb->pnn &&
 465              ((header.laccessor == c->hdr.srcnode
 466                && header.lacount >= ctdb->tunable.max_lacount)
 467               || (c->flags & CTDB_IMMEDIATE_MIGRATION)) ) {
 468                 if (ctdb_db->transaction_active) {
 469                         DEBUG(DEBUG_INFO, (__location__ " refusing migration"
 470                               " of key %s while transaction is active\n",
 471                               (char *)call->key.dptr));
 472                 } else {
 473                         DEBUG(DEBUG_DEBUG,("pnn %u starting migration of %08x to %u\n",
 474                                  ctdb->pnn, ctdb_hash(&(call->key)), c->hdr.srcnode));
 475                         ctdb_call_send_dmaster(ctdb_db, c, &header, &(call->key), &data);
 476                         talloc_free(data.dptr);
 477
 478                         ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 479                         if (ret != 0) {
 480                                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 481                         }
 482                         return;
 483                 }
 484         }
 485
 486         ctdb_call_local(ctdb_db, call, &header, hdr, &data, c->hdr.srcnode);
 487
 488         ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 489         if (ret != 0) {
 490                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 491         }
 492
 493         len = offsetof(struct ctdb_reply_call, data) + call->reply_data.dsize;
 494         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len,
 495                                     struct ctdb_reply_call);
 496         CTDB_NO_MEMORY_FATAL(ctdb, r);
 497         r->hdr.destnode  = hdr->srcnode;
 498         r->hdr.reqid     = hdr->reqid;
 499         r->status        = call->status;
 500         r->datalen       = call->reply_data.dsize;
 501         if (call->reply_data.dsize) {
 502                 memcpy(&r->data[0], call->reply_data.dptr, call->reply_data.dsize);
 503         }
 504
 505         ctdb_queue_packet(ctdb, &r->hdr);
 506
 507         talloc_free(r);
 508 }
 509
 510 /*
 511   called when a CTDB_REPLY_CALL packet comes in
 512
 513   This packet comes in response to a CTDB_REQ_CALL request packet. It
 514   contains any reply data from the call
 515 */
 516 void ctdb_reply_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 517 {
 518         struct ctdb_reply_call *c = (struct ctdb_reply_call *)hdr;
 519         struct ctdb_call_state *state;
 520
 521         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 522         if (state == NULL) {
 523                 DEBUG(DEBUG_ERR, (__location__ " reqid %u not found\n", hdr->reqid));
 524                 return;
 525         }
 526
 527         if (hdr->reqid != state->reqid) {
 528                 /* we found a record  but it was the wrong one */
 529                 DEBUG(DEBUG_ERR, ("Dropped orphaned call reply with reqid:%u\n",hdr->reqid));
 530                 return;
 531         }
 532
 533         state->call->reply_data.dptr = c->data;
 534         state->call->reply_data.dsize = c->datalen;
 535         state->call->status = c->status;
 536
 537         talloc_steal(state, c);
 538
 539         state->state = CTDB_CALL_DONE;
 540         if (state->async.fn) {
 541                 state->async.fn(state);
 542         }
 543 }
 544
 545
 546 /*
 547   called when a CTDB_REPLY_DMASTER packet comes in
 548
 549   This packet comes in from the lmaster response to a CTDB_REQ_CALL
 550   request packet. It means that the current dmaster wants to give us
 551   the dmaster role
 552 */
 553 void ctdb_reply_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 554 {
 555         struct ctdb_reply_dmaster *c = (struct ctdb_reply_dmaster *)hdr;
 556         struct ctdb_db_context *ctdb_db;
 557         TDB_DATA key, data;
 558         int ret;
 559
 560         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 561         if (ctdb_db == NULL) {
 562                 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_reply_dmaster\n", c->db_id));
 563                 return;
 564         }
 565
 566         key.dptr = c->data;
 567         key.dsize = c->keylen;
 568         data.dptr = &c->data[key.dsize];
 569         data.dsize = c->datalen;
 570
 571         ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
 572                                      ctdb_call_input_pkt, ctdb, False);
 573         if (ret == -2) {
 574                 return;
 575         }
 576         if (ret != 0) {
 577                 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_dmaster\n"));
 578                 return;
 579         }
 580
 581         ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn);
 582 }
 583
 584
 585 /*
 586   called when a CTDB_REPLY_ERROR packet comes in
 587 */
 588 void ctdb_reply_error(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 589 {
 590         struct ctdb_reply_error *c = (struct ctdb_reply_error *)hdr;
 591         struct ctdb_call_state *state;
 592
 593         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 594         if (state == NULL) {
 595                 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_reply_error\n",
 596                          ctdb->pnn, hdr->reqid));
 597                 return;
 598         }
 599
 600         if (hdr->reqid != state->reqid) {
 601                 /* we found a record  but it was the wrong one */
 602                 DEBUG(DEBUG_ERR, ("Dropped orphaned error reply with reqid:%u\n",hdr->reqid));
 603                 return;
 604         }
 605
 606         talloc_steal(state, c);
 607
 608         state->state  = CTDB_CALL_ERROR;
 609         state->errmsg = (char *)c->msg;
 610         if (state->async.fn) {
 611                 state->async.fn(state);
 612         }
 613 }
 614
 615
 616 /*
 617   destroy a ctdb_call
 618 */
 619 static int ctdb_call_destructor(struct ctdb_call_state *state)
 620 {
 621         DLIST_REMOVE(state->ctdb_db->ctdb->pending_calls, state);
 622         ctdb_reqid_remove(state->ctdb_db->ctdb, state->reqid);
 623         return 0;
 624 }
 625
 626
 627 /*
 628   called when a ctdb_call needs to be resent after a reconfigure event
 629 */
 630 static void ctdb_call_resend(struct ctdb_call_state *state)
 631 {
 632         struct ctdb_context *ctdb = state->ctdb_db->ctdb;
 633
 634         state->generation = ctdb->vnn_map->generation;
 635
 636         /* use a new reqid, in case the old reply does eventually come in */
 637         ctdb_reqid_remove(ctdb, state->reqid);
 638         state->reqid = ctdb_reqid_new(ctdb, state);
 639         state->c->hdr.reqid = state->reqid;
 640
 641         /* update the generation count for this request, so its valid with the new vnn_map */
 642         state->c->hdr.generation = state->generation;
 643
 644         /* send the packet to ourselves, it will be redirected appropriately */
 645         state->c->hdr.destnode = ctdb->pnn;
 646
 647         ctdb_queue_packet(ctdb, &state->c->hdr);
 648         DEBUG(DEBUG_NOTICE,("resent ctdb_call\n"));
 649 }
 650
 651 /*
 652   resend all pending calls on recovery
 653  */
 654 void ctdb_call_resend_all(struct ctdb_context *ctdb)
 655 {
 656         struct ctdb_call_state *state, *next;
 657         for (state=ctdb->pending_calls;state;state=next) {
 658                 next = state->next;
 659                 ctdb_call_resend(state);
 660         }
 661 }
 662
 663 /*
 664   this allows the caller to setup a async.fn
 665 */
 666 static void call_local_trigger(struct event_context *ev, struct timed_event *te,
 667                        struct timeval t, void *private_data)
 668 {
 669         struct ctdb_call_state *state = talloc_get_type(private_data, struct ctdb_call_state);
 670         if (state->async.fn) {
 671                 state->async.fn(state);
 672         }
 673 }
 674
 675
 676 /*
 677   construct an event driven local ctdb_call
 678
 679   this is used so that locally processed ctdb_call requests are processed
 680   in an event driven manner
 681 */
 682 struct ctdb_call_state *ctdb_call_local_send(struct ctdb_db_context *ctdb_db,
 683                                              struct ctdb_call *call,
 684                                              struct ctdb_ltdb_header *header,
 685                                              TDB_DATA *data)
 686 {
 687         struct ctdb_call_state *state;
 688         struct ctdb_context *ctdb = ctdb_db->ctdb;
 689         int ret;
 690
 691         state = talloc_zero(ctdb_db, struct ctdb_call_state);
 692         CTDB_NO_MEMORY_NULL(ctdb, state);
 693
 694         talloc_steal(state, data->dptr);
 695
 696         state->state = CTDB_CALL_DONE;
 697         state->call  = talloc(state, struct ctdb_call);
 698         CTDB_NO_MEMORY_NULL(ctdb, state->call);
 699         *(state->call) = *call;
 700         state->ctdb_db = ctdb_db;
 701
 702         ret = ctdb_call_local(ctdb_db, state->call, header, state, data, ctdb->pnn);
 703
 704         event_add_timed(ctdb->ev, state, timeval_zero(), call_local_trigger, state);
 705
 706         return state;
 707 }
 708
 709
 710 /*
 711   make a remote ctdb call - async send. Called in daemon context.
 712
 713   This constructs a ctdb_call request and queues it for processing.
 714   This call never blocks.
 715 */
 716 struct ctdb_call_state *ctdb_daemon_call_send_remote(struct ctdb_db_context *ctdb_db,
 717                                                      struct ctdb_call *call,
 718                                                      struct ctdb_ltdb_header *header)
 719 {
 720         uint32_t len;
 721         struct ctdb_call_state *state;
 722         struct ctdb_context *ctdb = ctdb_db->ctdb;
 723
 724         if (ctdb->methods == NULL) {
 725                 DEBUG(DEBUG_ERR,(__location__ " Failed send packet. Transport is down\n"));
 726                 return NULL;
 727         }
 728
 729         state = talloc_zero(ctdb_db, struct ctdb_call_state);
 730         CTDB_NO_MEMORY_NULL(ctdb, state);
 731         state->call = talloc(state, struct ctdb_call);
 732         CTDB_NO_MEMORY_NULL(ctdb, state->call);
 733
 734         state->reqid = ctdb_reqid_new(ctdb, state);
 735         state->ctdb_db = ctdb_db;
 736         talloc_set_destructor(state, ctdb_call_destructor);
 737
 738         len = offsetof(struct ctdb_req_call, data) + call->key.dsize + call->call_data.dsize;
 739         state->c = ctdb_transport_allocate(ctdb, state, CTDB_REQ_CALL, len,
 740                                            struct ctdb_req_call);
 741         CTDB_NO_MEMORY_NULL(ctdb, state->c);
 742         state->c->hdr.destnode  = header->dmaster;
 743
 744         /* this limits us to 16k outstanding messages - not unreasonable */
 745         state->c->hdr.reqid     = state->reqid;
 746         state->c->flags         = call->flags;
 747         state->c->db_id         = ctdb_db->db_id;
 748         state->c->callid        = call->call_id;
 749         state->c->hopcount      = 0;
 750         state->c->keylen        = call->key.dsize;
 751         state->c->calldatalen   = call->call_data.dsize;
 752         memcpy(&state->c->data[0], call->key.dptr, call->key.dsize);
 753         memcpy(&state->c->data[call->key.dsize],
 754                call->call_data.dptr, call->call_data.dsize);
 755         *(state->call)              = *call;
 756         state->call->call_data.dptr = &state->c->data[call->key.dsize];
 757         state->call->key.dptr       = &state->c->data[0];
 758
 759         state->state  = CTDB_CALL_WAIT;
 760         state->generation = ctdb->vnn_map->generation;
 761
 762         DLIST_ADD(ctdb->pending_calls, state);
 763
 764         ctdb_queue_packet(ctdb, &state->c->hdr);
 765
 766         return state;
 767 }
 768
 769 /*
 770   make a remote ctdb call - async recv - called in daemon context
 771
 772   This is called when the program wants to wait for a ctdb_call to complete and get the
 773   results. This call will block unless the call has already completed.
 774 */
 775 int ctdb_daemon_call_recv(struct ctdb_call_state *state, struct ctdb_call *call)
 776 {
 777         while (state->state < CTDB_CALL_DONE) {
 778                 event_loop_once(state->ctdb_db->ctdb->ev);
 779         }
 780         if (state->state != CTDB_CALL_DONE) {
 781                 ctdb_set_error(state->ctdb_db->ctdb, "%s", state->errmsg);
 782                 talloc_free(state);
 783                 return -1;
 784         }
 785
 786         if (state->call->reply_data.dsize) {
 787                 call->reply_data.dptr = talloc_memdup(call,
 788                                                       state->call->reply_data.dptr,
 789                                                       state->call->reply_data.dsize);
 790                 call->reply_data.dsize = state->call->reply_data.dsize;
 791         } else {
 792                 call->reply_data.dptr = NULL;
 793                 call->reply_data.dsize = 0;
 794         }
 795         call->status = state->call->status;
 796         talloc_free(state);
 797         return 0;
 798 }
 799
 800
 801 /*
 802    send a keepalive packet to the other node
 803 */
 804 void ctdb_send_keepalive(struct ctdb_context *ctdb, uint32_t destnode)
 805 {
 806         struct ctdb_req_keepalive *r;
 807
 808         if (ctdb->methods == NULL) {
 809                 DEBUG(DEBUG_ERR,(__location__ " Failed to send keepalive. Transport is DOWN\n"));
 810                 return;
 811         }
 812
 813         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_KEEPALIVE,
 814                                     sizeof(struct ctdb_req_keepalive),
 815                                     struct ctdb_req_keepalive);
 816         CTDB_NO_MEMORY_FATAL(ctdb, r);
 817         r->hdr.destnode  = destnode;
 818         r->hdr.reqid     = 0;
 819
 820         ctdb->statistics.keepalive_packets_sent++;
 821
 822         ctdb_queue_packet(ctdb, &r->hdr);
 823
 824         talloc_free(r);
 825 }