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 "tdb.h"
  25 #include "lib/util/dlinklist.h"
  26 #include "system/network.h"
  27 #include "system/filesys.h"
  28 #include "../include/ctdb_private.h"
  29 #include "../common/rb_tree.h"
  30 #include "common/reqid.h"
  31 #include "common/system.h"
  32
  33 struct ctdb_sticky_record {
  34         struct ctdb_context *ctdb;
  35         struct ctdb_db_context *ctdb_db;
  36         TDB_CONTEXT *pindown;
  37 };
  38
  39 /*
  40   find the ctdb_db from a db index
  41  */
  42  struct ctdb_db_context *find_ctdb_db(struct ctdb_context *ctdb, uint32_t id)
  43 {
  44         struct ctdb_db_context *ctdb_db;
  45
  46         for (ctdb_db=ctdb->db_list; ctdb_db; ctdb_db=ctdb_db->next) {
  47                 if (ctdb_db->db_id == id) {
  48                         break;
  49                 }
  50         }
  51         return ctdb_db;
  52 }
  53
  54 /*
  55   a varient of input packet that can be used in lock requeue
  56 */
  57 static void ctdb_call_input_pkt(void *p, struct ctdb_req_header *hdr)
  58 {
  59         struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
  60         ctdb_input_pkt(ctdb, hdr);
  61 }
  62
  63
  64 /*
  65   send an error reply
  66 */
  67 static void ctdb_send_error(struct ctdb_context *ctdb,
  68                             struct ctdb_req_header *hdr, uint32_t status,
  69                             const char *fmt, ...) PRINTF_ATTRIBUTE(4,5);
  70 static void ctdb_send_error(struct ctdb_context *ctdb,
  71                             struct ctdb_req_header *hdr, uint32_t status,
  72                             const char *fmt, ...)
  73 {
  74         va_list ap;
  75         struct ctdb_reply_error *r;
  76         char *msg;
  77         int msglen, len;
  78
  79         if (ctdb->methods == NULL) {
  80                 DEBUG(DEBUG_INFO,(__location__ " Failed to send error. Transport is DOWN\n"));
  81                 return;
  82         }
  83
  84         va_start(ap, fmt);
  85         msg = talloc_vasprintf(ctdb, fmt, ap);
  86         if (msg == NULL) {
  87                 ctdb_fatal(ctdb, "Unable to allocate error in ctdb_send_error\n");
  88         }
  89         va_end(ap);
  90
  91         msglen = strlen(msg)+1;
  92         len = offsetof(struct ctdb_reply_error, msg);
  93         r = ctdb_transport_allocate(ctdb, msg, CTDB_REPLY_ERROR, len + msglen,
  94                                     struct ctdb_reply_error);
  95         CTDB_NO_MEMORY_FATAL(ctdb, r);
  96
  97         r->hdr.destnode  = hdr->srcnode;
  98         r->hdr.reqid     = hdr->reqid;
  99         r->status        = status;
 100         r->msglen        = msglen;
 101         memcpy(&r->msg[0], msg, msglen);
 102
 103         ctdb_queue_packet(ctdb, &r->hdr);
 104
 105         talloc_free(msg);
 106 }
 107
 108
 109 /**
 110  * send a redirect reply
 111  *
 112  * The logic behind this function is this:
 113  *
 114  * A client wants to grab a record and sends a CTDB_REQ_CALL packet
 115  * to its local ctdb (ctdb_request_call). If the node is not itself
 116  * the record's DMASTER, it first redirects the packet to  the
 117  * record's LMASTER. The LMASTER then redirects the call packet to
 118  * the current DMASTER. Note that this works because of this: When
 119  * a record is migrated off a node, then the new DMASTER is stored
 120  * in the record's copy on the former DMASTER.
 121  */
 122 static void ctdb_call_send_redirect(struct ctdb_context *ctdb,
 123                                     struct ctdb_db_context *ctdb_db,
 124                                     TDB_DATA key,
 125                                     struct ctdb_req_call *c,
 126                                     struct ctdb_ltdb_header *header)
 127 {
 128         uint32_t lmaster = ctdb_lmaster(ctdb, &key);
 129
 130         c->hdr.destnode = lmaster;
 131         if (ctdb->pnn == lmaster) {
 132                 c->hdr.destnode = header->dmaster;
 133         }
 134         c->hopcount++;
 135
 136         if (c->hopcount%100 > 95) {
 137                 DEBUG(DEBUG_WARNING,("High hopcount %d dbid:%s "
 138                         "key:0x%08x reqid=%08x pnn:%d src:%d lmaster:%d "
 139                         "header->dmaster:%d dst:%d\n",
 140                         c->hopcount, ctdb_db->db_name, ctdb_hash(&key),
 141                         c->hdr.reqid, ctdb->pnn, c->hdr.srcnode, lmaster,
 142                         header->dmaster, c->hdr.destnode));
 143         }
 144
 145         ctdb_queue_packet(ctdb, &c->hdr);
 146 }
 147
 148
 149 /*
 150   send a dmaster reply
 151
 152   caller must have the chainlock before calling this routine. Caller must be
 153   the lmaster
 154 */
 155 static void ctdb_send_dmaster_reply(struct ctdb_db_context *ctdb_db,
 156                                     struct ctdb_ltdb_header *header,
 157                                     TDB_DATA key, TDB_DATA data,
 158                                     uint32_t new_dmaster,
 159                                     uint32_t reqid)
 160 {
 161         struct ctdb_context *ctdb = ctdb_db->ctdb;
 162         struct ctdb_reply_dmaster *r;
 163         int ret, len;
 164         TALLOC_CTX *tmp_ctx;
 165
 166         if (ctdb->pnn != ctdb_lmaster(ctdb, &key)) {
 167                 DEBUG(DEBUG_ALERT,(__location__ " Caller is not lmaster!\n"));
 168                 return;
 169         }
 170
 171         header->dmaster = new_dmaster;
 172         ret = ctdb_ltdb_store(ctdb_db, key, header, data);
 173         if (ret != 0) {
 174                 ctdb_fatal(ctdb, "ctdb_send_dmaster_reply unable to update dmaster");
 175                 return;
 176         }
 177
 178         if (ctdb->methods == NULL) {
 179                 ctdb_fatal(ctdb, "ctdb_send_dmaster_reply cant update dmaster since transport is down");
 180                 return;
 181         }
 182
 183         /* put the packet on a temporary context, allowing us to safely free
 184            it below even if ctdb_reply_dmaster() has freed it already */
 185         tmp_ctx = talloc_new(ctdb);
 186
 187         /* send the CTDB_REPLY_DMASTER */
 188         len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize + sizeof(uint32_t);
 189         r = ctdb_transport_allocate(ctdb, tmp_ctx, CTDB_REPLY_DMASTER, len,
 190                                     struct ctdb_reply_dmaster);
 191         CTDB_NO_MEMORY_FATAL(ctdb, r);
 192
 193         r->hdr.destnode  = new_dmaster;
 194         r->hdr.reqid     = reqid;
 195         r->hdr.generation = ctdb_db->generation;
 196         r->rsn           = header->rsn;
 197         r->keylen        = key.dsize;
 198         r->datalen       = data.dsize;
 199         r->db_id         = ctdb_db->db_id;
 200         memcpy(&r->data[0], key.dptr, key.dsize);
 201         memcpy(&r->data[key.dsize], data.dptr, data.dsize);
 202         memcpy(&r->data[key.dsize+data.dsize], &header->flags, sizeof(uint32_t));
 203
 204         ctdb_queue_packet(ctdb, &r->hdr);
 205
 206         talloc_free(tmp_ctx);
 207 }
 208
 209 /*
 210   send a dmaster request (give another node the dmaster for a record)
 211
 212   This is always sent to the lmaster, which ensures that the lmaster
 213   always knows who the dmaster is. The lmaster will then send a
 214   CTDB_REPLY_DMASTER to the new dmaster
 215 */
 216 static void ctdb_call_send_dmaster(struct ctdb_db_context *ctdb_db,
 217                                    struct ctdb_req_call *c,
 218                                    struct ctdb_ltdb_header *header,
 219                                    TDB_DATA *key, TDB_DATA *data)
 220 {
 221         struct ctdb_req_dmaster *r;
 222         struct ctdb_context *ctdb = ctdb_db->ctdb;
 223         int len;
 224         uint32_t lmaster = ctdb_lmaster(ctdb, key);
 225
 226         if (ctdb->methods == NULL) {
 227                 ctdb_fatal(ctdb, "Failed ctdb_call_send_dmaster since transport is down");
 228                 return;
 229         }
 230
 231         if (data->dsize != 0) {
 232                 header->flags |= CTDB_REC_FLAG_MIGRATED_WITH_DATA;
 233         }
 234
 235         if (lmaster == ctdb->pnn) {
 236                 ctdb_send_dmaster_reply(ctdb_db, header, *key, *data,
 237                                         c->hdr.srcnode, c->hdr.reqid);
 238                 return;
 239         }
 240
 241         len = offsetof(struct ctdb_req_dmaster, data) + key->dsize + data->dsize
 242                         + sizeof(uint32_t);
 243         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_DMASTER, len,
 244                                     struct ctdb_req_dmaster);
 245         CTDB_NO_MEMORY_FATAL(ctdb, r);
 246         r->hdr.destnode  = lmaster;
 247         r->hdr.reqid     = c->hdr.reqid;
 248         r->hdr.generation = ctdb_db->generation;
 249         r->db_id         = c->db_id;
 250         r->rsn           = header->rsn;
 251         r->dmaster       = c->hdr.srcnode;
 252         r->keylen        = key->dsize;
 253         r->datalen       = data->dsize;
 254         memcpy(&r->data[0], key->dptr, key->dsize);
 255         memcpy(&r->data[key->dsize], data->dptr, data->dsize);
 256         memcpy(&r->data[key->dsize + data->dsize], &header->flags, sizeof(uint32_t));
 257
 258         header->dmaster = c->hdr.srcnode;
 259         if (ctdb_ltdb_store(ctdb_db, *key, header, *data) != 0) {
 260                 ctdb_fatal(ctdb, "Failed to store record in ctdb_call_send_dmaster");
 261         }
 262
 263         ctdb_queue_packet(ctdb, &r->hdr);
 264
 265         talloc_free(r);
 266 }
 267
 268 static void ctdb_sticky_pindown_timeout(struct tevent_context *ev,
 269                                         struct tevent_timer *te,
 270                                         struct timeval t, void *private_data)
 271 {
 272         struct ctdb_sticky_record *sr = talloc_get_type(private_data,
 273                                                        struct ctdb_sticky_record);
 274
 275         DEBUG(DEBUG_ERR,("Pindown timeout db:%s  unstick record\n", sr->ctdb_db->db_name));
 276         if (sr->pindown != NULL) {
 277                 talloc_free(sr->pindown);
 278                 sr->pindown = NULL;
 279         }
 280 }
 281
 282 static int
 283 ctdb_set_sticky_pindown(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, TDB_DATA key)
 284 {
 285         TALLOC_CTX *tmp_ctx = talloc_new(NULL);
 286         uint32_t *k;
 287         struct ctdb_sticky_record *sr;
 288
 289         k = ctdb_key_to_idkey(tmp_ctx, key);
 290         if (k == NULL) {
 291                 DEBUG(DEBUG_ERR,("Failed to allocate key for sticky record\n"));
 292                 talloc_free(tmp_ctx);
 293                 return -1;
 294         }
 295
 296         sr = trbt_lookuparray32(ctdb_db->sticky_records, k[0], &k[0]);
 297         if (sr == NULL) {
 298                 talloc_free(tmp_ctx);
 299                 return 0;
 300         }
 301
 302         talloc_free(tmp_ctx);
 303
 304         if (sr->pindown == NULL) {
 305                 DEBUG(DEBUG_ERR,("Pinning down record in %s for %d ms\n", ctdb_db->db_name, ctdb->tunable.sticky_pindown));
 306                 sr->pindown = talloc_new(sr);
 307                 if (sr->pindown == NULL) {
 308                         DEBUG(DEBUG_ERR,("Failed to allocate pindown context for sticky record\n"));
 309                         return -1;
 310                 }
 311                 tevent_add_timer(ctdb->ev, sr->pindown,
 312                                  timeval_current_ofs(ctdb->tunable.sticky_pindown / 1000,
 313                                                      (ctdb->tunable.sticky_pindown * 1000) % 1000000),
 314                                  ctdb_sticky_pindown_timeout, sr);
 315         }
 316
 317         return 0;
 318 }
 319
 320 /*
 321   called when a CTDB_REPLY_DMASTER packet comes in, or when the lmaster
 322   gets a CTDB_REQUEST_DMASTER for itself. We become the dmaster.
 323
 324   must be called with the chainlock held. This function releases the chainlock
 325 */
 326 static void ctdb_become_dmaster(struct ctdb_db_context *ctdb_db,
 327                                 struct ctdb_req_header *hdr,
 328                                 TDB_DATA key, TDB_DATA data,
 329                                 uint64_t rsn, uint32_t record_flags)
 330 {
 331         struct ctdb_call_state *state;
 332         struct ctdb_context *ctdb = ctdb_db->ctdb;
 333         struct ctdb_ltdb_header header;
 334         int ret;
 335
 336         DEBUG(DEBUG_DEBUG,("pnn %u dmaster response %08x\n", ctdb->pnn, ctdb_hash(&key)));
 337
 338         ZERO_STRUCT(header);
 339         header.rsn = rsn;
 340         header.dmaster = ctdb->pnn;
 341         header.flags = record_flags;
 342
 343         state = reqid_find(ctdb->idr, hdr->reqid, struct ctdb_call_state);
 344
 345         if (state) {
 346                 if (state->call->flags & CTDB_CALL_FLAG_VACUUM_MIGRATION) {
 347                         /*
 348                          * We temporarily add the VACUUM_MIGRATED flag to
 349                          * the record flags, so that ctdb_ltdb_store can
 350                          * decide whether the record should be stored or
 351                          * deleted.
 352                          */
 353                         header.flags |= CTDB_REC_FLAG_VACUUM_MIGRATED;
 354                 }
 355         }
 356
 357         if (ctdb_ltdb_store(ctdb_db, key, &header, data) != 0) {
 358                 ctdb_fatal(ctdb, "ctdb_reply_dmaster store failed\n");
 359
 360                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 361                 if (ret != 0) {
 362                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 363                 }
 364                 return;
 365         }
 366
 367         /* we just became DMASTER and this database is "sticky",
 368            see if the record is flagged as "hot" and set up a pin-down
 369            context to stop migrations for a little while if so
 370         */
 371         if (ctdb_db->sticky) {
 372                 ctdb_set_sticky_pindown(ctdb, ctdb_db, key);
 373         }
 374
 375         if (state == NULL) {
 376                 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_become_dmaster from node %u\n",
 377                          ctdb->pnn, hdr->reqid, hdr->srcnode));
 378
 379                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 380                 if (ret != 0) {
 381                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 382                 }
 383                 return;
 384         }
 385
 386         if (key.dsize != state->call->key.dsize || memcmp(key.dptr, state->call->key.dptr, key.dsize)) {
 387                 DEBUG(DEBUG_ERR, ("Got bogus DMASTER packet reqid:%u from node %u. Key does not match key held in matching idr.\n", hdr->reqid, hdr->srcnode));
 388
 389                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 390                 if (ret != 0) {
 391                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 392                 }
 393                 return;
 394         }
 395
 396         if (hdr->reqid != state->reqid) {
 397                 /* we found a record  but it was the wrong one */
 398                 DEBUG(DEBUG_ERR, ("Dropped orphan in ctdb_become_dmaster with reqid:%u\n from node %u", hdr->reqid, hdr->srcnode));
 399
 400                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 401                 if (ret != 0) {
 402                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 403                 }
 404                 return;
 405         }
 406
 407         ctdb_call_local(ctdb_db, state->call, &header, state, &data, true);
 408
 409         ret = ctdb_ltdb_unlock(ctdb_db, state->call->key);
 410         if (ret != 0) {
 411                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 412         }
 413
 414         state->state = CTDB_CALL_DONE;
 415         if (state->async.fn) {
 416                 state->async.fn(state);
 417         }
 418 }
 419
 420 struct dmaster_defer_call {
 421         struct dmaster_defer_call *next, *prev;
 422         struct ctdb_context *ctdb;
 423         struct ctdb_req_header *hdr;
 424 };
 425
 426 struct dmaster_defer_queue {
 427         struct ctdb_db_context *ctdb_db;
 428         uint32_t generation;
 429         struct dmaster_defer_call *deferred_calls;
 430 };
 431
 432 static void dmaster_defer_reprocess(struct tevent_context *ev,
 433                                     struct tevent_timer *te,
 434                                     struct timeval t,
 435                                     void *private_data)
 436 {
 437         struct dmaster_defer_call *call = talloc_get_type(
 438                 private_data, struct dmaster_defer_call);
 439
 440         ctdb_input_pkt(call->ctdb, call->hdr);
 441         talloc_free(call);
 442 }
 443
 444 static int dmaster_defer_queue_destructor(struct dmaster_defer_queue *ddq)
 445 {
 446         /* Ignore requests, if database recovery happens in-between. */
 447         if (ddq->generation != ddq->ctdb_db->generation) {
 448                 return 0;
 449         }
 450
 451         while (ddq->deferred_calls != NULL) {
 452                 struct dmaster_defer_call *call = ddq->deferred_calls;
 453
 454                 DLIST_REMOVE(ddq->deferred_calls, call);
 455
 456                 talloc_steal(call->ctdb, call);
 457                 tevent_add_timer(call->ctdb->ev, call, timeval_zero(),
 458                                  dmaster_defer_reprocess, call);
 459         }
 460         return 0;
 461 }
 462
 463 static void *insert_ddq_callback(void *parm, void *data)
 464 {
 465         if (data) {
 466                 talloc_free(data);
 467         }
 468         return parm;
 469 }
 470
 471 /**
 472  * This function is used to reigster a key in database that needs to be updated.
 473  * Any requests for that key should get deferred till this is completed.
 474  */
 475 static int dmaster_defer_setup(struct ctdb_db_context *ctdb_db,
 476                                struct ctdb_req_header *hdr,
 477                                TDB_DATA key)
 478 {
 479         uint32_t *k;
 480         struct dmaster_defer_queue *ddq;
 481
 482         k = ctdb_key_to_idkey(hdr, key);
 483         if (k == NULL) {
 484                 DEBUG(DEBUG_ERR, ("Failed to allocate key for dmaster defer setup\n"));
 485                 return -1;
 486         }
 487
 488         /* Already exists */
 489         ddq = trbt_lookuparray32(ctdb_db->defer_dmaster, k[0], k);
 490         if (ddq != NULL) {
 491                 if (ddq->generation == ctdb_db->generation) {
 492                         talloc_free(k);
 493                         return 0;
 494                 }
 495
 496                 /* Recovery ocurred - get rid of old queue. All the deferred
 497                  * requests will be resent anyway from ctdb_call_resend_db.
 498                  */
 499                 talloc_free(ddq);
 500         }
 501
 502         ddq = talloc(hdr, struct dmaster_defer_queue);
 503         if (ddq == NULL) {
 504                 DEBUG(DEBUG_ERR, ("Failed to allocate dmaster defer queue\n"));
 505                 talloc_free(k);
 506                 return -1;
 507         }
 508         ddq->ctdb_db = ctdb_db;
 509         ddq->generation = hdr->generation;
 510         ddq->deferred_calls = NULL;
 511
 512         trbt_insertarray32_callback(ctdb_db->defer_dmaster, k[0], k,
 513                                     insert_ddq_callback, ddq);
 514         talloc_set_destructor(ddq, dmaster_defer_queue_destructor);
 515
 516         talloc_free(k);
 517         return 0;
 518 }
 519
 520 static int dmaster_defer_add(struct ctdb_db_context *ctdb_db,
 521                              struct ctdb_req_header *hdr,
 522                              TDB_DATA key)
 523 {
 524         struct dmaster_defer_queue *ddq;
 525         struct dmaster_defer_call *call;
 526         uint32_t *k;
 527
 528         k = ctdb_key_to_idkey(hdr, key);
 529         if (k == NULL) {
 530                 DEBUG(DEBUG_ERR, ("Failed to allocate key for dmaster defer add\n"));
 531                 return -1;
 532         }
 533
 534         ddq = trbt_lookuparray32(ctdb_db->defer_dmaster, k[0], k);
 535         if (ddq == NULL) {
 536                 talloc_free(k);
 537                 return -1;
 538         }
 539
 540         talloc_free(k);
 541
 542         if (ddq->generation != hdr->generation) {
 543                 talloc_set_destructor(ddq, NULL);
 544                 talloc_free(ddq);
 545                 return -1;
 546         }
 547
 548         call = talloc(ddq, struct dmaster_defer_call);
 549         if (call == NULL) {
 550                 DEBUG(DEBUG_ERR, ("Failed to allocate dmaster defer call\n"));
 551                 return -1;
 552         }
 553
 554         call->ctdb = ctdb_db->ctdb;
 555         call->hdr = talloc_steal(call, hdr);
 556
 557         DLIST_ADD_END(ddq->deferred_calls, call, NULL);
 558
 559         return 0;
 560 }
 561
 562 /*
 563   called when a CTDB_REQ_DMASTER packet comes in
 564
 565   this comes into the lmaster for a record when the current dmaster
 566   wants to give up the dmaster role and give it to someone else
 567 */
 568 void ctdb_request_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 569 {
 570         struct ctdb_req_dmaster *c = (struct ctdb_req_dmaster *)hdr;
 571         TDB_DATA key, data, data2;
 572         struct ctdb_ltdb_header header;
 573         struct ctdb_db_context *ctdb_db;
 574         uint32_t record_flags = 0;
 575         size_t len;
 576         int ret;
 577
 578         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 579         if (!ctdb_db) {
 580                 ctdb_send_error(ctdb, hdr, -1,
 581                                 "Unknown database in request. db_id==0x%08x",
 582                                 c->db_id);
 583                 return;
 584         }
 585
 586         if (hdr->generation != ctdb_db->generation) {
 587                 DEBUG(DEBUG_DEBUG,
 588                       ("ctdb operation %u request %u from node %u to %u had an"
 589                        " invalid generation:%u while our generation is:%u\n",
 590                        hdr->operation, hdr->reqid, hdr->srcnode, hdr->destnode,
 591                        hdr->generation, ctdb_db->generation));
 592                 return;
 593         }
 594
 595         key.dptr = c->data;
 596         key.dsize = c->keylen;
 597         data.dptr = c->data + c->keylen;
 598         data.dsize = c->datalen;
 599         len = offsetof(struct ctdb_req_dmaster, data) + key.dsize + data.dsize
 600                         + sizeof(uint32_t);
 601         if (len <= c->hdr.length) {
 602                 memcpy(&record_flags, &c->data[c->keylen + c->datalen],
 603                        sizeof(record_flags));
 604         }
 605
 606         dmaster_defer_setup(ctdb_db, hdr, key);
 607
 608         /* fetch the current record */
 609         ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, key, &header, hdr, &data2,
 610                                            ctdb_call_input_pkt, ctdb, false);
 611         if (ret == -1) {
 612                 ctdb_fatal(ctdb, "ctdb_req_dmaster failed to fetch record");
 613                 return;
 614         }
 615         if (ret == -2) {
 616                 DEBUG(DEBUG_INFO,(__location__ " deferring ctdb_request_dmaster\n"));
 617                 return;
 618         }
 619
 620         if (ctdb_lmaster(ctdb, &key) != ctdb->pnn) {
 621                 DEBUG(DEBUG_ALERT,("pnn %u dmaster request to non-lmaster lmaster=%u gen=%u curgen=%u\n",
 622                          ctdb->pnn, ctdb_lmaster(ctdb, &key),
 623                          hdr->generation, ctdb->vnn_map->generation));
 624                 ctdb_fatal(ctdb, "ctdb_req_dmaster to non-lmaster");
 625         }
 626
 627         DEBUG(DEBUG_DEBUG,("pnn %u dmaster request on %08x for %u from %u\n",
 628                  ctdb->pnn, ctdb_hash(&key), c->dmaster, c->hdr.srcnode));
 629
 630         /* its a protocol error if the sending node is not the current dmaster */
 631         if (header.dmaster != hdr->srcnode) {
 632                 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",
 633                          ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
 634                          ctdb_db->db_id, hdr->generation, ctdb_db->generation,
 635                          (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid,
 636                          (key.dsize >= 4)?(*(uint32_t *)key.dptr):0));
 637                 if (header.rsn != 0 || header.dmaster != ctdb->pnn) {
 638                         DEBUG(DEBUG_ERR,("ctdb_req_dmaster from non-master. Force a recovery.\n"));
 639
 640                         ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
 641                         ctdb_ltdb_unlock(ctdb_db, key);
 642                         return;
 643                 }
 644         }
 645
 646         if (header.rsn > c->rsn) {
 647                 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",
 648                          ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
 649                          ctdb_db->db_id, hdr->generation, ctdb_db->generation,
 650                          (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid));
 651         }
 652
 653         /* use the rsn from the sending node */
 654         header.rsn = c->rsn;
 655
 656         /* store the record flags from the sending node */
 657         header.flags = record_flags;
 658
 659         /* check if the new dmaster is the lmaster, in which case we
 660            skip the dmaster reply */
 661         if (c->dmaster == ctdb->pnn) {
 662                 ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn, record_flags);
 663         } else {
 664                 ctdb_send_dmaster_reply(ctdb_db, &header, key, data, c->dmaster, hdr->reqid);
 665
 666                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 667                 if (ret != 0) {
 668                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 669                 }
 670         }
 671 }
 672
 673 static void ctdb_sticky_record_timeout(struct tevent_context *ev,
 674                                        struct tevent_timer *te,
 675                                        struct timeval t, void *private_data)
 676 {
 677         struct ctdb_sticky_record *sr = talloc_get_type(private_data,
 678                                                        struct ctdb_sticky_record);
 679         talloc_free(sr);
 680 }
 681
 682 static void *ctdb_make_sticky_record_callback(void *parm, void *data)
 683 {
 684         if (data) {
 685                 DEBUG(DEBUG_ERR,("Already have sticky record registered. Free old %p and create new %p\n", data, parm));
 686                 talloc_free(data);
 687         }
 688         return parm;
 689 }
 690
 691 static int
 692 ctdb_make_record_sticky(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, TDB_DATA key)
 693 {
 694         TALLOC_CTX *tmp_ctx = talloc_new(NULL);
 695         uint32_t *k;
 696         struct ctdb_sticky_record *sr;
 697
 698         k = ctdb_key_to_idkey(tmp_ctx, key);
 699         if (k == NULL) {
 700                 DEBUG(DEBUG_ERR,("Failed to allocate key for sticky record\n"));
 701                 talloc_free(tmp_ctx);
 702                 return -1;
 703         }
 704
 705         sr = trbt_lookuparray32(ctdb_db->sticky_records, k[0], &k[0]);
 706         if (sr != NULL) {
 707                 talloc_free(tmp_ctx);
 708                 return 0;
 709         }
 710
 711         sr = talloc(ctdb_db->sticky_records, struct ctdb_sticky_record);
 712         if (sr == NULL) {
 713                 talloc_free(tmp_ctx);
 714                 DEBUG(DEBUG_ERR,("Failed to allocate sticky record structure\n"));
 715                 return -1;
 716         }
 717
 718         sr->ctdb    = ctdb;
 719         sr->ctdb_db = ctdb_db;
 720         sr->pindown = NULL;
 721
 722         DEBUG(DEBUG_ERR,("Make record sticky for %d seconds in db %s key:0x%08x.\n",
 723                          ctdb->tunable.sticky_duration,
 724                          ctdb_db->db_name, ctdb_hash(&key)));
 725
 726         trbt_insertarray32_callback(ctdb_db->sticky_records, k[0], &k[0], ctdb_make_sticky_record_callback, sr);
 727
 728         tevent_add_timer(ctdb->ev, sr,
 729                          timeval_current_ofs(ctdb->tunable.sticky_duration, 0),
 730                          ctdb_sticky_record_timeout, sr);
 731
 732         talloc_free(tmp_ctx);
 733         return 0;
 734 }
 735
 736 struct pinned_down_requeue_handle {
 737         struct ctdb_context *ctdb;
 738         struct ctdb_req_header *hdr;
 739 };
 740
 741 struct pinned_down_deferred_call {
 742         struct ctdb_context *ctdb;
 743         struct ctdb_req_header *hdr;
 744 };
 745
 746 static void pinned_down_requeue(struct tevent_context *ev,
 747                                 struct tevent_timer *te,
 748                                 struct timeval t, void *private_data)
 749 {
 750         struct pinned_down_requeue_handle *handle = talloc_get_type(private_data, struct pinned_down_requeue_handle);
 751         struct ctdb_context *ctdb = handle->ctdb;
 752
 753         talloc_steal(ctdb, handle->hdr);
 754         ctdb_call_input_pkt(ctdb, handle->hdr);
 755
 756         talloc_free(handle);
 757 }
 758
 759 static int pinned_down_destructor(struct pinned_down_deferred_call *pinned_down)
 760 {
 761         struct ctdb_context *ctdb = pinned_down->ctdb;
 762         struct pinned_down_requeue_handle *handle = talloc(ctdb, struct pinned_down_requeue_handle);
 763
 764         handle->ctdb = pinned_down->ctdb;
 765         handle->hdr  = pinned_down->hdr;
 766         talloc_steal(handle, handle->hdr);
 767
 768         tevent_add_timer(ctdb->ev, handle, timeval_zero(),
 769                          pinned_down_requeue, handle);
 770
 771         return 0;
 772 }
 773
 774 static int
 775 ctdb_defer_pinned_down_request(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, TDB_DATA key, struct ctdb_req_header *hdr)
 776 {
 777         TALLOC_CTX *tmp_ctx = talloc_new(NULL);
 778         uint32_t *k;
 779         struct ctdb_sticky_record *sr;
 780         struct pinned_down_deferred_call *pinned_down;
 781
 782         k = ctdb_key_to_idkey(tmp_ctx, key);
 783         if (k == NULL) {
 784                 DEBUG(DEBUG_ERR,("Failed to allocate key for sticky record\n"));
 785                 talloc_free(tmp_ctx);
 786                 return -1;
 787         }
 788
 789         sr = trbt_lookuparray32(ctdb_db->sticky_records, k[0], &k[0]);
 790         if (sr == NULL) {
 791                 talloc_free(tmp_ctx);
 792                 return -1;
 793         }
 794
 795         talloc_free(tmp_ctx);
 796
 797         if (sr->pindown == NULL) {
 798                 return -1;
 799         }
 800
 801         pinned_down = talloc(sr->pindown, struct pinned_down_deferred_call);
 802         if (pinned_down == NULL) {
 803                 DEBUG(DEBUG_ERR,("Failed to allocate structure for deferred pinned down request\n"));
 804                 return -1;
 805         }
 806
 807         pinned_down->ctdb = ctdb;
 808         pinned_down->hdr  = hdr;
 809
 810         talloc_set_destructor(pinned_down, pinned_down_destructor);
 811         talloc_steal(pinned_down, hdr);
 812
 813         return 0;
 814 }
 815
 816 static void
 817 ctdb_update_db_stat_hot_keys(struct ctdb_db_context *ctdb_db, TDB_DATA key, int hopcount)
 818 {
 819         int i, id;
 820
 821         /* smallest value is always at index 0 */
 822         if (hopcount <= ctdb_db->statistics.hot_keys[0].count) {
 823                 return;
 824         }
 825
 826         /* see if we already know this key */
 827         for (i = 0; i < MAX_HOT_KEYS; i++) {
 828                 if (key.dsize != ctdb_db->statistics.hot_keys[i].key.dsize) {
 829                         continue;
 830                 }
 831                 if (memcmp(key.dptr, ctdb_db->statistics.hot_keys[i].key.dptr, key.dsize)) {
 832                         continue;
 833                 }
 834                 /* found an entry for this key */
 835                 if (hopcount <= ctdb_db->statistics.hot_keys[i].count) {
 836                         return;
 837                 }
 838                 ctdb_db->statistics.hot_keys[i].count = hopcount;
 839                 goto sort_keys;
 840         }
 841
 842         if (ctdb_db->statistics.num_hot_keys < MAX_HOT_KEYS) {
 843                 id = ctdb_db->statistics.num_hot_keys;
 844                 ctdb_db->statistics.num_hot_keys++;
 845         } else {
 846                 id = 0;
 847         }
 848
 849         if (ctdb_db->statistics.hot_keys[id].key.dptr != NULL) {
 850                 talloc_free(ctdb_db->statistics.hot_keys[id].key.dptr);
 851         }
 852         ctdb_db->statistics.hot_keys[id].key.dsize = key.dsize;
 853         ctdb_db->statistics.hot_keys[id].key.dptr  = talloc_memdup(ctdb_db, key.dptr, key.dsize);
 854         ctdb_db->statistics.hot_keys[id].count = hopcount;
 855         DEBUG(DEBUG_NOTICE,("Updated hot key database=%s key=0x%08x id=%d hop_count=%d\n",
 856                             ctdb_db->db_name, ctdb_hash(&key), id, hopcount));
 857
 858 sort_keys:
 859         for (i = 1; i < MAX_HOT_KEYS; i++) {
 860                 if (ctdb_db->statistics.hot_keys[i].count == 0) {
 861                         continue;
 862                 }
 863                 if (ctdb_db->statistics.hot_keys[i].count < ctdb_db->statistics.hot_keys[0].count) {
 864                         hopcount = ctdb_db->statistics.hot_keys[i].count;
 865                         ctdb_db->statistics.hot_keys[i].count = ctdb_db->statistics.hot_keys[0].count;
 866                         ctdb_db->statistics.hot_keys[0].count = hopcount;
 867
 868                         key = ctdb_db->statistics.hot_keys[i].key;
 869                         ctdb_db->statistics.hot_keys[i].key = ctdb_db->statistics.hot_keys[0].key;
 870                         ctdb_db->statistics.hot_keys[0].key = key;
 871                 }
 872         }
 873 }
 874
 875 /*
 876   called when a CTDB_REQ_CALL packet comes in
 877 */
 878 void ctdb_request_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 879 {
 880         struct ctdb_req_call *c = (struct ctdb_req_call *)hdr;
 881         TDB_DATA data;
 882         struct ctdb_reply_call *r;
 883         int ret, len;
 884         struct ctdb_ltdb_header header;
 885         struct ctdb_call *call;
 886         struct ctdb_db_context *ctdb_db;
 887         int tmp_count, bucket;
 888
 889         if (ctdb->methods == NULL) {
 890                 DEBUG(DEBUG_INFO,(__location__ " Failed ctdb_request_call. Transport is DOWN\n"));
 891                 return;
 892         }
 893
 894         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 895         if (!ctdb_db) {
 896                 ctdb_send_error(ctdb, hdr, -1,
 897                                 "Unknown database in request. db_id==0x%08x",
 898                                 c->db_id);
 899                 return;
 900         }
 901
 902         if (hdr->generation != ctdb_db->generation) {
 903                 DEBUG(DEBUG_DEBUG,
 904                       ("ctdb operation %u request %u from node %u to %u had an"
 905                        " invalid generation:%u while our generation is:%u\n",
 906                        hdr->operation, hdr->reqid, hdr->srcnode, hdr->destnode,
 907                        hdr->generation, ctdb_db->generation));
 908                 return;
 909         }
 910
 911         call = talloc(hdr, struct ctdb_call);
 912         CTDB_NO_MEMORY_FATAL(ctdb, call);
 913
 914         call->call_id  = c->callid;
 915         call->key.dptr = c->data;
 916         call->key.dsize = c->keylen;
 917         call->call_data.dptr = c->data + c->keylen;
 918         call->call_data.dsize = c->calldatalen;
 919         call->reply_data.dptr  = NULL;
 920         call->reply_data.dsize = 0;
 921
 922
 923         /* If this record is pinned down we should defer the
 924            request until the pindown times out
 925         */
 926         if (ctdb_db->sticky) {
 927                 if (ctdb_defer_pinned_down_request(ctdb, ctdb_db, call->key, hdr) == 0) {
 928                         DEBUG(DEBUG_WARNING,
 929                               ("Defer request for pinned down record in %s\n", ctdb_db->db_name));
 930                         talloc_free(call);
 931                         return;
 932                 }
 933         }
 934
 935         if (dmaster_defer_add(ctdb_db, hdr, call->key) == 0) {
 936                 talloc_free(call);
 937                 return;
 938         }
 939
 940         /* determine if we are the dmaster for this key. This also
 941            fetches the record data (if any), thus avoiding a 2nd fetch of the data
 942            if the call will be answered locally */
 943
 944         ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, call->key, &header, hdr, &data,
 945                                            ctdb_call_input_pkt, ctdb, false);
 946         if (ret == -1) {
 947                 ctdb_send_error(ctdb, hdr, ret, "ltdb fetch failed in ctdb_request_call");
 948                 talloc_free(call);
 949                 return;
 950         }
 951         if (ret == -2) {
 952                 DEBUG(DEBUG_INFO,(__location__ " deferred ctdb_request_call\n"));
 953                 talloc_free(call);
 954                 return;
 955         }
 956
 957         /* Dont do READONLY if we dont have a tracking database */
 958         if ((c->flags & CTDB_WANT_READONLY) && !ctdb_db->readonly) {
 959                 c->flags &= ~CTDB_WANT_READONLY;
 960         }
 961
 962         if (header.flags & CTDB_REC_RO_REVOKE_COMPLETE) {
 963                 header.flags &= ~CTDB_REC_RO_FLAGS;
 964                 CTDB_INCREMENT_STAT(ctdb, total_ro_revokes);
 965                 CTDB_INCREMENT_DB_STAT(ctdb_db, db_ro_revokes);
 966                 if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
 967                         ctdb_fatal(ctdb, "Failed to write header with cleared REVOKE flag");
 968                 }
 969                 /* and clear out the tracking data */
 970                 if (tdb_delete(ctdb_db->rottdb, call->key) != 0) {
 971                         DEBUG(DEBUG_ERR,(__location__ " Failed to clear out trackingdb record\n"));
 972                 }
 973         }
 974
 975         /* if we are revoking, we must defer all other calls until the revoke
 976          * had completed.
 977          */
 978         if (header.flags & CTDB_REC_RO_REVOKING_READONLY) {
 979                 talloc_free(data.dptr);
 980                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 981
 982                 if (ctdb_add_revoke_deferred_call(ctdb, ctdb_db, call->key, hdr, ctdb_call_input_pkt, ctdb) != 0) {
 983                         ctdb_fatal(ctdb, "Failed to add deferred call for revoke child");
 984                 }
 985                 talloc_free(call);
 986                 return;
 987         }
 988
 989         /*
 990          * If we are not the dmaster and are not hosting any delegations,
 991          * then we redirect the request to the node than can answer it
 992          * (the lmaster or the dmaster).
 993          */
 994         if ((header.dmaster != ctdb->pnn)
 995             && (!(header.flags & CTDB_REC_RO_HAVE_DELEGATIONS)) ) {
 996                 talloc_free(data.dptr);
 997                 ctdb_call_send_redirect(ctdb, ctdb_db, call->key, c, &header);
 998
 999                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
1000                 if (ret != 0) {
1001                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
1002                 }
1003                 talloc_free(call);
1004                 return;
1005         }
1006
1007         if ( (!(c->flags & CTDB_WANT_READONLY))
1008         && (header.flags & (CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY)) ) {
1009                 header.flags   |= CTDB_REC_RO_REVOKING_READONLY;
1010                 if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
1011                         ctdb_fatal(ctdb, "Failed to store record with HAVE_DELEGATIONS set");
1012                 }
1013                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
1014
1015                 if (ctdb_start_revoke_ro_record(ctdb, ctdb_db, call->key, &header, data) != 0) {
1016                         ctdb_fatal(ctdb, "Failed to start record revoke");
1017                 }
1018                 talloc_free(data.dptr);
1019
1020                 if (ctdb_add_revoke_deferred_call(ctdb, ctdb_db, call->key, hdr, ctdb_call_input_pkt, ctdb) != 0) {
1021                         ctdb_fatal(ctdb, "Failed to add deferred call for revoke child");
1022                 }
1023                 talloc_free(call);
1024
1025                 return;
1026         }
1027
1028         /* If this is the first request for delegation. bump rsn and set
1029          * the delegations flag
1030          */
1031         if ((c->flags & CTDB_WANT_READONLY)
1032         &&  (c->callid == CTDB_FETCH_WITH_HEADER_FUNC)
1033         &&  (!(header.flags & CTDB_REC_RO_HAVE_DELEGATIONS))) {
1034                 header.rsn     += 3;
1035                 header.flags   |= CTDB_REC_RO_HAVE_DELEGATIONS;
1036                 if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
1037                         ctdb_fatal(ctdb, "Failed to store record with HAVE_DELEGATIONS set");
1038                 }
1039         }
1040         if ((c->flags & CTDB_WANT_READONLY)
1041         &&  (call->call_id == CTDB_FETCH_WITH_HEADER_FUNC)) {
1042                 TDB_DATA tdata;
1043
1044                 tdata = tdb_fetch(ctdb_db->rottdb, call->key);
1045                 if (ctdb_trackingdb_add_pnn(ctdb, &tdata, c->hdr.srcnode) != 0) {
1046                         ctdb_fatal(ctdb, "Failed to add node to trackingdb");
1047                 }
1048                 if (tdb_store(ctdb_db->rottdb, call->key, tdata, TDB_REPLACE) != 0) {
1049                         ctdb_fatal(ctdb, "Failed to store trackingdb data");
1050                 }
1051                 free(tdata.dptr);
1052
1053                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
1054                 if (ret != 0) {
1055                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
1056                 }
1057
1058                 len = offsetof(struct ctdb_reply_call, data) + data.dsize + sizeof(struct ctdb_ltdb_header);
1059                 r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len,
1060                                             struct ctdb_reply_call);
1061                 CTDB_NO_MEMORY_FATAL(ctdb, r);
1062                 r->hdr.destnode  = c->hdr.srcnode;
1063                 r->hdr.reqid     = c->hdr.reqid;
1064                 r->hdr.generation = ctdb_db->generation;
1065                 r->status        = 0;
1066                 r->datalen       = data.dsize + sizeof(struct ctdb_ltdb_header);
1067                 header.rsn      -= 2;
1068                 header.flags   |= CTDB_REC_RO_HAVE_READONLY;
1069                 header.flags   &= ~CTDB_REC_RO_HAVE_DELEGATIONS;
1070                 memcpy(&r->data[0], &header, sizeof(struct ctdb_ltdb_header));
1071
1072                 if (data.dsize) {
1073                         memcpy(&r->data[sizeof(struct ctdb_ltdb_header)], data.dptr, data.dsize);
1074                 }
1075
1076                 ctdb_queue_packet(ctdb, &r->hdr);
1077                 CTDB_INCREMENT_STAT(ctdb, total_ro_delegations);
1078                 CTDB_INCREMENT_DB_STAT(ctdb_db, db_ro_delegations);
1079
1080                 talloc_free(r);
1081                 talloc_free(call);
1082                 return;
1083         }
1084
1085         CTDB_UPDATE_STAT(ctdb, max_hop_count, c->hopcount);
1086         tmp_count = c->hopcount;
1087         bucket = 0;
1088         while (tmp_count) {
1089                 tmp_count >>= 2;
1090                 bucket++;
1091         }
1092         if (bucket >= MAX_COUNT_BUCKETS) {
1093                 bucket = MAX_COUNT_BUCKETS - 1;
1094         }
1095         CTDB_INCREMENT_STAT(ctdb, hop_count_bucket[bucket]);
1096         CTDB_INCREMENT_DB_STAT(ctdb_db, hop_count_bucket[bucket]);
1097         ctdb_update_db_stat_hot_keys(ctdb_db, call->key, c->hopcount);
1098
1099         /* If this database supports sticky records, then check if the
1100            hopcount is big. If it is it means the record is hot and we
1101            should make it sticky.
1102         */
1103         if (ctdb_db->sticky && c->hopcount >= ctdb->tunable.hopcount_make_sticky) {
1104                 ctdb_make_record_sticky(ctdb, ctdb_db, call->key);
1105         }
1106
1107
1108         /* Try if possible to migrate the record off to the caller node.
1109          * From the clients perspective a fetch of the data is just as
1110          * expensive as a migration.
1111          */
1112         if (c->hdr.srcnode != ctdb->pnn) {
1113                 if (ctdb_db->persistent_state) {
1114                         DEBUG(DEBUG_INFO, (__location__ " refusing migration"
1115                               " of key %s while transaction is active\n",
1116                               (char *)call->key.dptr));
1117                 } else {
1118                         DEBUG(DEBUG_DEBUG,("pnn %u starting migration of %08x to %u\n",
1119                                  ctdb->pnn, ctdb_hash(&(call->key)), c->hdr.srcnode));
1120                         ctdb_call_send_dmaster(ctdb_db, c, &header, &(call->key), &data);
1121                         talloc_free(data.dptr);
1122
1123                         ret = ctdb_ltdb_unlock(ctdb_db, call->key);
1124                         if (ret != 0) {
1125                                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
1126                         }
1127                 }
1128                 talloc_free(call);
1129                 return;
1130         }
1131
1132         ret = ctdb_call_local(ctdb_db, call, &header, hdr, &data, true);
1133         if (ret != 0) {
1134                 DEBUG(DEBUG_ERR,(__location__ " ctdb_call_local failed\n"));
1135                 call->status = -1;
1136         }
1137
1138         ret = ctdb_ltdb_unlock(ctdb_db, call->key);
1139         if (ret != 0) {
1140                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
1141         }
1142
1143         len = offsetof(struct ctdb_reply_call, data) + call->reply_data.dsize;
1144         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len,
1145                                     struct ctdb_reply_call);
1146         CTDB_NO_MEMORY_FATAL(ctdb, r);
1147         r->hdr.destnode  = hdr->srcnode;
1148         r->hdr.reqid     = hdr->reqid;
1149         r->hdr.generation = ctdb_db->generation;
1150         r->status        = call->status;
1151         r->datalen       = call->reply_data.dsize;
1152         if (call->reply_data.dsize) {
1153                 memcpy(&r->data[0], call->reply_data.dptr, call->reply_data.dsize);
1154         }
1155
1156         ctdb_queue_packet(ctdb, &r->hdr);
1157
1158         talloc_free(r);
1159         talloc_free(call);
1160 }
1161
1162 /**
1163  * called when a CTDB_REPLY_CALL packet comes in
1164  *
1165  * This packet comes in response to a CTDB_REQ_CALL request packet. It
1166  * contains any reply data from the call
1167  */
1168 void ctdb_reply_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
1169 {
1170         struct ctdb_reply_call *c = (struct ctdb_reply_call *)hdr;
1171         struct ctdb_call_state *state;
1172
1173         state = reqid_find(ctdb->idr, hdr->reqid, struct ctdb_call_state);
1174         if (state == NULL) {
1175                 DEBUG(DEBUG_ERR, (__location__ " reqid %u not found\n", hdr->reqid));
1176                 return;
1177         }
1178
1179         if (hdr->reqid != state->reqid) {
1180                 /* we found a record  but it was the wrong one */
1181                 DEBUG(DEBUG_ERR, ("Dropped orphaned call reply with reqid:%u\n",hdr->reqid));
1182                 return;
1183         }
1184
1185         if (hdr->generation != state->generation) {
1186                 DEBUG(DEBUG_DEBUG,
1187                       ("ctdb operation %u request %u from node %u to %u had an"
1188                        " invalid generation:%u while our generation is:%u\n",
1189                        hdr->operation, hdr->reqid, hdr->srcnode, hdr->destnode,
1190                        hdr->generation, state->generation));
1191                 return;
1192         }
1193
1194
1195         /* read only delegation processing */
1196         /* If we got a FETCH_WITH_HEADER we should check if this is a ro
1197          * delegation since we may need to update the record header
1198          */
1199         if (state->c->callid == CTDB_FETCH_WITH_HEADER_FUNC) {
1200                 struct ctdb_db_context *ctdb_db = state->ctdb_db;
1201                 struct ctdb_ltdb_header *header = (struct ctdb_ltdb_header *)&c->data[0];
1202                 struct ctdb_ltdb_header oldheader;
1203                 TDB_DATA key, data, olddata;
1204                 int ret;
1205
1206                 if (!(header->flags & CTDB_REC_RO_HAVE_READONLY)) {
1207                         goto finished_ro;
1208                         return;
1209                 }
1210
1211                 key.dsize = state->c->keylen;
1212                 key.dptr  = state->c->data;
1213                 ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
1214                                      ctdb_call_input_pkt, ctdb, false);
1215                 if (ret == -2) {
1216                         return;
1217                 }
1218                 if (ret != 0) {
1219                         DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_call\n"));
1220                         return;
1221                 }
1222
1223                 ret = ctdb_ltdb_fetch(ctdb_db, key, &oldheader, state, &olddata);
1224                 if (ret != 0) {
1225                         DEBUG(DEBUG_ERR, ("Failed to fetch old record in ctdb_reply_call\n"));
1226                         ctdb_ltdb_unlock(ctdb_db, key);
1227                         goto finished_ro;
1228                 }
1229
1230                 if (header->rsn <= oldheader.rsn) {
1231                         ctdb_ltdb_unlock(ctdb_db, key);
1232                         goto finished_ro;
1233                 }
1234
1235                 if (c->datalen < sizeof(struct ctdb_ltdb_header)) {
1236                         DEBUG(DEBUG_ERR,(__location__ " Got FETCH_WITH_HEADER reply with too little data: %d bytes\n", c->datalen));
1237                         ctdb_ltdb_unlock(ctdb_db, key);
1238                         goto finished_ro;
1239                 }
1240
1241                 data.dsize = c->datalen - sizeof(struct ctdb_ltdb_header);
1242                 data.dptr  = &c->data[sizeof(struct ctdb_ltdb_header)];
1243                 ret = ctdb_ltdb_store(ctdb_db, key, header, data);
1244                 if (ret != 0) {
1245                         DEBUG(DEBUG_ERR, ("Failed to store new record in ctdb_reply_call\n"));
1246                         ctdb_ltdb_unlock(ctdb_db, key);
1247                         goto finished_ro;
1248                 }
1249
1250                 ctdb_ltdb_unlock(ctdb_db, key);
1251         }
1252 finished_ro:
1253
1254         state->call->reply_data.dptr = c->data;
1255         state->call->reply_data.dsize = c->datalen;
1256         state->call->status = c->status;
1257
1258         talloc_steal(state, c);
1259
1260         state->state = CTDB_CALL_DONE;
1261         if (state->async.fn) {
1262                 state->async.fn(state);
1263         }
1264 }
1265
1266
1267 /**
1268  * called when a CTDB_REPLY_DMASTER packet comes in
1269  *
1270  * This packet comes in from the lmaster in response to a CTDB_REQ_CALL
1271  * request packet. It means that the current dmaster wants to give us
1272  * the dmaster role.
1273  */
1274 void ctdb_reply_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
1275 {
1276         struct ctdb_reply_dmaster *c = (struct ctdb_reply_dmaster *)hdr;
1277         struct ctdb_db_context *ctdb_db;
1278         TDB_DATA key, data;
1279         uint32_t record_flags = 0;
1280         size_t len;
1281         int ret;
1282
1283         ctdb_db = find_ctdb_db(ctdb, c->db_id);
1284         if (ctdb_db == NULL) {
1285                 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_reply_dmaster\n", c->db_id));
1286                 return;
1287         }
1288
1289         if (hdr->generation != ctdb_db->generation) {
1290                 DEBUG(DEBUG_DEBUG,
1291                       ("ctdb operation %u request %u from node %u to %u had an"
1292                        " invalid generation:%u while our generation is:%u\n",
1293                        hdr->operation, hdr->reqid, hdr->srcnode, hdr->destnode,
1294                        hdr->generation, ctdb_db->generation));
1295                 return;
1296         }
1297
1298         key.dptr = c->data;
1299         key.dsize = c->keylen;
1300         data.dptr = &c->data[key.dsize];
1301         data.dsize = c->datalen;
1302         len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize
1303                 + sizeof(uint32_t);
1304         if (len <= c->hdr.length) {
1305                 memcpy(&record_flags, &c->data[c->keylen + c->datalen],
1306                        sizeof(record_flags));
1307         }
1308
1309         dmaster_defer_setup(ctdb_db, hdr, key);
1310
1311         ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
1312                                      ctdb_call_input_pkt, ctdb, false);
1313         if (ret == -2) {
1314                 return;
1315         }
1316         if (ret != 0) {
1317                 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_dmaster\n"));
1318                 return;
1319         }
1320
1321         ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn, record_flags);
1322 }
1323
1324
1325 /*
1326   called when a CTDB_REPLY_ERROR packet comes in
1327 */
1328 void ctdb_reply_error(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
1329 {
1330         struct ctdb_reply_error *c = (struct ctdb_reply_error *)hdr;
1331         struct ctdb_call_state *state;
1332
1333         state = reqid_find(ctdb->idr, hdr->reqid, struct ctdb_call_state);
1334         if (state == NULL) {
1335                 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_reply_error\n",
1336                          ctdb->pnn, hdr->reqid));
1337                 return;
1338         }
1339
1340         if (hdr->reqid != state->reqid) {
1341                 /* we found a record  but it was the wrong one */
1342                 DEBUG(DEBUG_ERR, ("Dropped orphaned error reply with reqid:%u\n",hdr->reqid));
1343                 return;
1344         }
1345
1346         talloc_steal(state, c);
1347
1348         state->state  = CTDB_CALL_ERROR;
1349         state->errmsg = (char *)c->msg;
1350         if (state->async.fn) {
1351                 state->async.fn(state);
1352         }
1353 }
1354
1355
1356 /*
1357   destroy a ctdb_call
1358 */
1359 static int ctdb_call_destructor(struct ctdb_call_state *state)
1360 {
1361         DLIST_REMOVE(state->ctdb_db->pending_calls, state);
1362         reqid_remove(state->ctdb_db->ctdb->idr, state->reqid);
1363         return 0;
1364 }
1365
1366
1367 /*
1368   called when a ctdb_call needs to be resent after a reconfigure event
1369 */
1370 static void ctdb_call_resend(struct ctdb_call_state *state)
1371 {
1372         struct ctdb_context *ctdb = state->ctdb_db->ctdb;
1373
1374         state->generation = state->ctdb_db->generation;
1375
1376         /* use a new reqid, in case the old reply does eventually come in */
1377         reqid_remove(ctdb->idr, state->reqid);
1378         state->reqid = reqid_new(ctdb->idr, state);
1379         state->c->hdr.reqid = state->reqid;
1380
1381         /* update the generation count for this request, so its valid with the new vnn_map */
1382         state->c->hdr.generation = state->generation;
1383
1384         /* send the packet to ourselves, it will be redirected appropriately */
1385         state->c->hdr.destnode = ctdb->pnn;
1386
1387         ctdb_queue_packet(ctdb, &state->c->hdr);
1388         DEBUG(DEBUG_NOTICE,("resent ctdb_call for db %s reqid %u generation %u\n",
1389                             state->ctdb_db->db_name, state->reqid, state->generation));
1390 }
1391
1392 /*
1393   resend all pending calls on recovery
1394  */
1395 void ctdb_call_resend_db(struct ctdb_db_context *ctdb_db)
1396 {
1397         struct ctdb_call_state *state, *next;
1398
1399         for (state = ctdb_db->pending_calls; state; state = next) {
1400                 next = state->next;
1401                 ctdb_call_resend(state);
1402         }
1403 }
1404
1405 void ctdb_call_resend_all(struct ctdb_context *ctdb)
1406 {
1407         struct ctdb_db_context *ctdb_db;
1408
1409         for (ctdb_db = ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
1410                 ctdb_call_resend_db(ctdb_db);
1411         }
1412 }
1413
1414 /*
1415   this allows the caller to setup a async.fn
1416 */
1417 static void call_local_trigger(struct tevent_context *ev,
1418                                struct tevent_timer *te,
1419                                struct timeval t, void *private_data)
1420 {
1421         struct ctdb_call_state *state = talloc_get_type(private_data, struct ctdb_call_state);
1422         if (state->async.fn) {
1423                 state->async.fn(state);
1424         }
1425 }
1426
1427
1428 /*
1429   construct an event driven local ctdb_call
1430
1431   this is used so that locally processed ctdb_call requests are processed
1432   in an event driven manner
1433 */
1434 struct ctdb_call_state *ctdb_call_local_send(struct ctdb_db_context *ctdb_db,
1435                                              struct ctdb_call *call,
1436                                              struct ctdb_ltdb_header *header,
1437                                              TDB_DATA *data)
1438 {
1439         struct ctdb_call_state *state;
1440         struct ctdb_context *ctdb = ctdb_db->ctdb;
1441         int ret;
1442
1443         state = talloc_zero(ctdb_db, struct ctdb_call_state);
1444         CTDB_NO_MEMORY_NULL(ctdb, state);
1445
1446         talloc_steal(state, data->dptr);
1447
1448         state->state = CTDB_CALL_DONE;
1449         state->call  = talloc(state, struct ctdb_call);
1450         CTDB_NO_MEMORY_NULL(ctdb, state->call);
1451         *(state->call) = *call;
1452         state->ctdb_db = ctdb_db;
1453
1454         ret = ctdb_call_local(ctdb_db, state->call, header, state, data, true);
1455         if (ret != 0) {
1456                 DEBUG(DEBUG_DEBUG,("ctdb_call_local() failed, ignoring return code %d\n", ret));
1457         }
1458
1459         tevent_add_timer(ctdb->ev, state, timeval_zero(),
1460                          call_local_trigger, state);
1461
1462         return state;
1463 }
1464
1465
1466 /*
1467   make a remote ctdb call - async send. Called in daemon context.
1468
1469   This constructs a ctdb_call request and queues it for processing.
1470   This call never blocks.
1471 */
1472 struct ctdb_call_state *ctdb_daemon_call_send_remote(struct ctdb_db_context *ctdb_db,
1473                                                      struct ctdb_call *call,
1474                                                      struct ctdb_ltdb_header *header)
1475 {
1476         uint32_t len;
1477         struct ctdb_call_state *state;
1478         struct ctdb_context *ctdb = ctdb_db->ctdb;
1479
1480         if (ctdb->methods == NULL) {
1481                 DEBUG(DEBUG_INFO,(__location__ " Failed send packet. Transport is down\n"));
1482                 return NULL;
1483         }
1484
1485         state = talloc_zero(ctdb_db, struct ctdb_call_state);
1486         CTDB_NO_MEMORY_NULL(ctdb, state);
1487         state->call = talloc(state, struct ctdb_call);
1488         CTDB_NO_MEMORY_NULL(ctdb, state->call);
1489
1490         state->reqid = reqid_new(ctdb->idr, state);
1491         state->ctdb_db = ctdb_db;
1492         talloc_set_destructor(state, ctdb_call_destructor);
1493
1494         len = offsetof(struct ctdb_req_call, data) + call->key.dsize + call->call_data.dsize;
1495         state->c = ctdb_transport_allocate(ctdb, state, CTDB_REQ_CALL, len,
1496                                            struct ctdb_req_call);
1497         CTDB_NO_MEMORY_NULL(ctdb, state->c);
1498         state->c->hdr.destnode  = header->dmaster;
1499
1500         /* this limits us to 16k outstanding messages - not unreasonable */
1501         state->c->hdr.reqid     = state->reqid;
1502         state->c->hdr.generation = ctdb_db->generation;
1503         state->c->flags         = call->flags;
1504         state->c->db_id         = ctdb_db->db_id;
1505         state->c->callid        = call->call_id;
1506         state->c->hopcount      = 0;
1507         state->c->keylen        = call->key.dsize;
1508         state->c->calldatalen   = call->call_data.dsize;
1509         memcpy(&state->c->data[0], call->key.dptr, call->key.dsize);
1510         memcpy(&state->c->data[call->key.dsize],
1511                call->call_data.dptr, call->call_data.dsize);
1512         *(state->call)              = *call;
1513         state->call->call_data.dptr = &state->c->data[call->key.dsize];
1514         state->call->key.dptr       = &state->c->data[0];
1515
1516         state->state  = CTDB_CALL_WAIT;
1517         state->generation = ctdb_db->generation;
1518
1519         DLIST_ADD(ctdb_db->pending_calls, state);
1520
1521         ctdb_queue_packet(ctdb, &state->c->hdr);
1522
1523         return state;
1524 }
1525
1526 /*
1527   make a remote ctdb call - async recv - called in daemon context
1528
1529   This is called when the program wants to wait for a ctdb_call to complete and get the
1530   results. This call will block unless the call has already completed.
1531 */
1532 int ctdb_daemon_call_recv(struct ctdb_call_state *state, struct ctdb_call *call)
1533 {
1534         while (state->state < CTDB_CALL_DONE) {
1535                 tevent_loop_once(state->ctdb_db->ctdb->ev);
1536         }
1537         if (state->state != CTDB_CALL_DONE) {
1538                 ctdb_set_error(state->ctdb_db->ctdb, "%s", state->errmsg);
1539                 talloc_free(state);
1540                 return -1;
1541         }
1542
1543         if (state->call->reply_data.dsize) {
1544                 call->reply_data.dptr = talloc_memdup(call,
1545                                                       state->call->reply_data.dptr,
1546                                                       state->call->reply_data.dsize);
1547                 call->reply_data.dsize = state->call->reply_data.dsize;
1548         } else {
1549                 call->reply_data.dptr = NULL;
1550                 call->reply_data.dsize = 0;
1551         }
1552         call->status = state->call->status;
1553         talloc_free(state);
1554         return 0;
1555 }
1556
1557
1558 /*
1559    send a keepalive packet to the other node
1560 */
1561 void ctdb_send_keepalive(struct ctdb_context *ctdb, uint32_t destnode)
1562 {
1563         struct ctdb_req_keepalive *r;
1564
1565         if (ctdb->methods == NULL) {
1566                 DEBUG(DEBUG_INFO,(__location__ " Failed to send keepalive. Transport is DOWN\n"));
1567                 return;
1568         }
1569
1570         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_KEEPALIVE,
1571                                     sizeof(struct ctdb_req_keepalive),
1572                                     struct ctdb_req_keepalive);
1573         CTDB_NO_MEMORY_FATAL(ctdb, r);
1574         r->hdr.destnode  = destnode;
1575         r->hdr.reqid     = 0;
1576
1577         CTDB_INCREMENT_STAT(ctdb, keepalive_packets_sent);
1578
1579         ctdb_queue_packet(ctdb, &r->hdr);
1580
1581         talloc_free(r);
1582 }
1583
1584
1585
1586 struct revokechild_deferred_call {
1587         struct ctdb_context *ctdb;
1588         struct ctdb_req_header *hdr;
1589         deferred_requeue_fn fn;
1590         void *ctx;
1591 };
1592
1593 struct revokechild_handle {
1594         struct revokechild_handle *next, *prev;
1595         struct ctdb_context *ctdb;
1596         struct ctdb_db_context *ctdb_db;
1597         struct tevent_fd *fde;
1598         int status;
1599         int fd[2];
1600         pid_t child;
1601         TDB_DATA key;
1602 };
1603
1604 struct revokechild_requeue_handle {
1605         struct ctdb_context *ctdb;
1606         struct ctdb_req_header *hdr;
1607         deferred_requeue_fn fn;
1608         void *ctx;
1609 };
1610
1611 static void deferred_call_requeue(struct tevent_context *ev,
1612                                   struct tevent_timer *te,
1613                                   struct timeval t, void *private_data)
1614 {
1615         struct revokechild_requeue_handle *requeue_handle = talloc_get_type(private_data, struct revokechild_requeue_handle);
1616
1617         requeue_handle->fn(requeue_handle->ctx, requeue_handle->hdr);
1618         talloc_free(requeue_handle);
1619 }
1620
1621 static int deferred_call_destructor(struct revokechild_deferred_call *deferred_call)
1622 {
1623         struct ctdb_context *ctdb = deferred_call->ctdb;
1624         struct revokechild_requeue_handle *requeue_handle = talloc(ctdb, struct revokechild_requeue_handle);
1625         struct ctdb_req_call *c = (struct ctdb_req_call *)deferred_call->hdr;
1626
1627         requeue_handle->ctdb = ctdb;
1628         requeue_handle->hdr  = deferred_call->hdr;
1629         requeue_handle->fn   = deferred_call->fn;
1630         requeue_handle->ctx  = deferred_call->ctx;
1631         talloc_steal(requeue_handle, requeue_handle->hdr);
1632
1633         /* when revoking, any READONLY requests have 1 second grace to let read/write finish first */
1634         tevent_add_timer(ctdb->ev, requeue_handle,
1635                          timeval_current_ofs(c->flags & CTDB_WANT_READONLY ? 1 : 0, 0),
1636                          deferred_call_requeue, requeue_handle);
1637
1638         return 0;
1639 }
1640
1641
1642 static int revokechild_destructor(struct revokechild_handle *rc)
1643 {
1644         if (rc->fde != NULL) {
1645                 talloc_free(rc->fde);
1646         }
1647
1648         if (rc->fd[0] != -1) {
1649                 close(rc->fd[0]);
1650         }
1651         if (rc->fd[1] != -1) {
1652                 close(rc->fd[1]);
1653         }
1654         ctdb_kill(rc->ctdb, rc->child, SIGKILL);
1655
1656         DLIST_REMOVE(rc->ctdb_db->revokechild_active, rc);
1657         return 0;
1658 }
1659
1660 static void revokechild_handler(struct tevent_context *ev,
1661                                 struct tevent_fd *fde,
1662                                 uint16_t flags, void *private_data)
1663 {
1664         struct revokechild_handle *rc = talloc_get_type(private_data,
1665                                                      struct revokechild_handle);
1666         int ret;
1667         char c;
1668
1669         ret = sys_read(rc->fd[0], &c, 1);
1670         if (ret != 1) {
1671                 DEBUG(DEBUG_ERR,("Failed to read status from revokechild. errno:%d\n", errno));
1672                 rc->status = -1;
1673                 talloc_free(rc);
1674                 return;
1675         }
1676         if (c != 0) {
1677                 DEBUG(DEBUG_ERR,("revokechild returned failure. status:%d\n", c));
1678                 rc->status = -1;
1679                 talloc_free(rc);
1680                 return;
1681         }
1682
1683         talloc_free(rc);
1684 }
1685
1686 struct ctdb_revoke_state {
1687         struct ctdb_db_context *ctdb_db;
1688         TDB_DATA key;
1689         struct ctdb_ltdb_header *header;
1690         TDB_DATA data;
1691         int count;
1692         int status;
1693         int finished;
1694 };
1695
1696 static void update_record_cb(struct ctdb_client_control_state *state)
1697 {
1698         struct ctdb_revoke_state *revoke_state;
1699         int ret;
1700         int32_t res;
1701
1702         if (state == NULL) {
1703                 return;
1704         }
1705         revoke_state = state->async.private_data;
1706
1707         state->async.fn = NULL;
1708         ret = ctdb_control_recv(state->ctdb, state, state, NULL, &res, NULL);
1709         if ((ret != 0) || (res != 0)) {
1710                 DEBUG(DEBUG_ERR,("Recv for revoke update record failed ret:%d res:%d\n", ret, res));
1711                 revoke_state->status = -1;
1712         }
1713
1714         revoke_state->count--;
1715         if (revoke_state->count <= 0) {
1716                 revoke_state->finished = 1;
1717         }
1718 }
1719
1720 static void revoke_send_cb(struct ctdb_context *ctdb, uint32_t pnn, void *private_data)
1721 {
1722         struct ctdb_revoke_state *revoke_state = private_data;
1723         struct ctdb_client_control_state *state;
1724
1725         state = ctdb_ctrl_updaterecord_send(ctdb, revoke_state, timeval_current_ofs(ctdb->tunable.control_timeout,0), pnn, revoke_state->ctdb_db, revoke_state->key, revoke_state->header, revoke_state->data);
1726         if (state == NULL) {
1727                 DEBUG(DEBUG_ERR,("Failure to send update record to revoke readonly delegation\n"));
1728                 revoke_state->status = -1;
1729                 return;
1730         }
1731         state->async.fn           = update_record_cb;
1732         state->async.private_data = revoke_state;
1733
1734         revoke_state->count++;
1735
1736 }
1737
1738 static void ctdb_revoke_timeout_handler(struct tevent_context *ev,
1739                                         struct tevent_timer *te,
1740                                         struct timeval yt, void *private_data)
1741 {
1742         struct ctdb_revoke_state *state = private_data;
1743
1744         DEBUG(DEBUG_ERR,("Timed out waiting for revoke to finish\n"));
1745         state->finished = 1;
1746         state->status   = -1;
1747 }
1748
1749 static int ctdb_revoke_all_delegations(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, TDB_DATA tdata, TDB_DATA key, struct ctdb_ltdb_header *header, TDB_DATA data)
1750 {
1751         struct ctdb_revoke_state *state = talloc_zero(ctdb, struct ctdb_revoke_state);
1752         struct ctdb_ltdb_header new_header;
1753         TDB_DATA new_data;
1754
1755         state->ctdb_db = ctdb_db;
1756         state->key     = key;
1757         state->header  = header;
1758         state->data    = data;
1759
1760         ctdb_trackingdb_traverse(ctdb, tdata, revoke_send_cb, state);
1761
1762         tevent_add_timer(ctdb->ev, state,
1763                          timeval_current_ofs(ctdb->tunable.control_timeout, 0),
1764                          ctdb_revoke_timeout_handler, state);
1765
1766         while (state->finished == 0) {
1767                 tevent_loop_once(ctdb->ev);
1768         }
1769
1770         if (ctdb_ltdb_lock(ctdb_db, key) != 0) {
1771                 DEBUG(DEBUG_ERR,("Failed to chainlock the database in revokechild\n"));
1772                 talloc_free(state);
1773                 return -1;
1774         }
1775         if (ctdb_ltdb_fetch(ctdb_db, key, &new_header, state, &new_data) != 0) {
1776                 ctdb_ltdb_unlock(ctdb_db, key);
1777                 DEBUG(DEBUG_ERR,("Failed for fetch tdb record in revokechild\n"));
1778                 talloc_free(state);
1779                 return -1;
1780         }
1781         header->rsn++;
1782         if (new_header.rsn > header->rsn) {
1783                 ctdb_ltdb_unlock(ctdb_db, key);
1784                 DEBUG(DEBUG_ERR,("RSN too high in tdb record in revokechild\n"));
1785                 talloc_free(state);
1786                 return -1;
1787         }
1788         if ( (new_header.flags & (CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS)) != (CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS) ) {
1789                 ctdb_ltdb_unlock(ctdb_db, key);
1790                 DEBUG(DEBUG_ERR,("Flags are wrong in tdb record in revokechild\n"));
1791                 talloc_free(state);
1792                 return -1;
1793         }
1794
1795         /*
1796          * If revoke on all nodes succeed, revoke is complete.  Otherwise,
1797          * remove CTDB_REC_RO_REVOKING_READONLY flag and retry.
1798          */
1799         if (state->status == 0) {
1800                 new_header.rsn++;
1801                 new_header.flags |= CTDB_REC_RO_REVOKE_COMPLETE;
1802         } else {
1803                 DEBUG(DEBUG_NOTICE, ("Revoke all delegations failed, retrying.\n"));
1804                 new_header.flags &= ~CTDB_REC_RO_REVOKING_READONLY;
1805         }
1806         if (ctdb_ltdb_store(ctdb_db, key, &new_header, new_data) != 0) {
1807                 ctdb_ltdb_unlock(ctdb_db, key);
1808                 DEBUG(DEBUG_ERR,("Failed to write new record in revokechild\n"));
1809                 talloc_free(state);
1810                 return -1;
1811         }
1812         ctdb_ltdb_unlock(ctdb_db, key);
1813
1814         talloc_free(state);
1815         return 0;
1816 }
1817
1818
1819 int ctdb_start_revoke_ro_record(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, TDB_DATA key, struct ctdb_ltdb_header *header, TDB_DATA data)
1820 {
1821         TDB_DATA tdata;
1822         struct revokechild_handle *rc;
1823         pid_t parent = getpid();
1824         int ret;
1825
1826         header->flags &= ~(CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY);
1827         header->flags |= CTDB_REC_FLAG_MIGRATED_WITH_DATA;
1828         header->rsn   -= 1;
1829
1830         if ((rc = talloc_zero(ctdb_db, struct revokechild_handle)) == NULL) {
1831                 DEBUG(DEBUG_ERR,("Failed to allocate revokechild_handle\n"));
1832                 return -1;
1833         }
1834
1835         tdata = tdb_fetch(ctdb_db->rottdb, key);
1836         if (tdata.dsize > 0) {
1837                 uint8_t *tmp;
1838
1839                 tmp = tdata.dptr;
1840                 tdata.dptr = talloc_memdup(rc, tdata.dptr, tdata.dsize);
1841                 free(tmp);
1842         }
1843
1844         rc->status    = 0;
1845         rc->ctdb      = ctdb;
1846         rc->ctdb_db   = ctdb_db;
1847         rc->fd[0]     = -1;
1848         rc->fd[1]     = -1;
1849
1850         talloc_set_destructor(rc, revokechild_destructor);
1851
1852         rc->key.dsize = key.dsize;
1853         rc->key.dptr  = talloc_memdup(rc, key.dptr, key.dsize);
1854         if (rc->key.dptr == NULL) {
1855                 DEBUG(DEBUG_ERR,("Failed to allocate key for revokechild_handle\n"));
1856                 talloc_free(rc);
1857                 return -1;
1858         }
1859
1860         ret = pipe(rc->fd);
1861         if (ret != 0) {
1862                 DEBUG(DEBUG_ERR,("Failed to allocate key for revokechild_handle\n"));
1863                 talloc_free(rc);
1864                 return -1;
1865         }
1866
1867
1868         rc->child = ctdb_fork(ctdb);
1869         if (rc->child == (pid_t)-1) {
1870                 DEBUG(DEBUG_ERR,("Failed to fork child for revokechild\n"));
1871                 talloc_free(rc);
1872                 return -1;
1873         }
1874
1875         if (rc->child == 0) {
1876                 char c = 0;
1877                 close(rc->fd[0]);
1878                 debug_extra = talloc_asprintf(NULL, "revokechild-%s:", ctdb_db->db_name);
1879
1880                 ctdb_set_process_name("ctdb_revokechild");
1881                 if (switch_from_server_to_client(ctdb, "revokechild-%s", ctdb_db->db_name) != 0) {
1882                         DEBUG(DEBUG_ERR,("Failed to switch from server to client for revokechild process\n"));
1883                         c = 1;
1884                         goto child_finished;
1885                 }
1886
1887                 c = ctdb_revoke_all_delegations(ctdb, ctdb_db, tdata, key, header, data);
1888
1889 child_finished:
1890                 sys_write(rc->fd[1], &c, 1);
1891                 /* make sure we die when our parent dies */
1892                 while (ctdb_kill(ctdb, parent, 0) == 0 || errno != ESRCH) {
1893                         sleep(5);
1894                 }
1895                 _exit(0);
1896         }
1897
1898         close(rc->fd[1]);
1899         rc->fd[1] = -1;
1900         set_close_on_exec(rc->fd[0]);
1901
1902         /* This is an active revokechild child process */
1903         DLIST_ADD_END(ctdb_db->revokechild_active, rc, NULL);
1904
1905         rc->fde = tevent_add_fd(ctdb->ev, rc, rc->fd[0], TEVENT_FD_READ,
1906                                 revokechild_handler, (void *)rc);
1907         if (rc->fde == NULL) {
1908                 DEBUG(DEBUG_ERR,("Failed to set up fd event for revokechild process\n"));
1909                 talloc_free(rc);
1910         }
1911         tevent_fd_set_auto_close(rc->fde);
1912
1913         return 0;
1914 }
1915
1916 int ctdb_add_revoke_deferred_call(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, TDB_DATA key, struct ctdb_req_header *hdr, deferred_requeue_fn fn, void *call_context)
1917 {
1918         struct revokechild_handle *rc;
1919         struct revokechild_deferred_call *deferred_call;
1920
1921         for (rc = ctdb_db->revokechild_active; rc; rc = rc->next) {
1922                 if (rc->key.dsize == 0) {
1923                         continue;
1924                 }
1925                 if (rc->key.dsize != key.dsize) {
1926                         continue;
1927                 }
1928                 if (!memcmp(rc->key.dptr, key.dptr, key.dsize)) {
1929                         break;
1930                 }
1931         }
1932
1933         if (rc == NULL) {
1934                 DEBUG(DEBUG_ERR,("Failed to add deferred call to revoke list. revoke structure not found\n"));
1935                 return -1;
1936         }
1937
1938         deferred_call = talloc(rc, struct revokechild_deferred_call);
1939         if (deferred_call == NULL) {
1940                 DEBUG(DEBUG_ERR,("Failed to allocate deferred call structure for revoking record\n"));
1941                 return -1;
1942         }
1943
1944         deferred_call->ctdb = ctdb;
1945         deferred_call->hdr  = hdr;
1946         deferred_call->fn   = fn;
1947         deferred_call->ctx  = call_context;
1948
1949         talloc_set_destructor(deferred_call, deferred_call_destructor);
1950         talloc_steal(deferred_call, hdr);
1951
1952         return 0;
1953 }