ctdb/server/ctdb_call.c

   1 /*
   2    ctdb_call protocol code
   3
   4    Copyright (C) Andrew Tridgell  2006
   5
   6    This program is free software; you can redistribute it and/or modify
   7    it under the terms of the GNU General Public License as published by
   8    the Free Software Foundation; either version 3 of the License, or
   9    (at your option) any later version.
  10
  11    This program is distributed in the hope that it will be useful,
  12    but WITHOUT ANY WARRANTY; without even the implied warranty of
  13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14    GNU General Public License for more details.
  15
  16    You should have received a copy of the GNU General Public License
  17    along with this program; if not, see <http://www.gnu.org/licenses/>.
  18 */
  19 /*
  20   see http://wiki.samba.org/index.php/Samba_%26_Clustering for
  21   protocol design and packet details
  22 */
  23 #include "includes.h"
  24 #include "lib/tevent/tevent.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   a varient of input packet that can be used in lock requeue
  48 */
  49 static void ctdb_call_input_pkt(void *p, struct ctdb_req_header *hdr)
  50 {
  51         struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
  52         ctdb_input_pkt(ctdb, hdr);
  53 }
  54
  55
  56 /*
  57   send an error reply
  58 */
  59 static void ctdb_send_error(struct ctdb_context *ctdb,
  60                             struct ctdb_req_header *hdr, uint32_t status,
  61                             const char *fmt, ...) PRINTF_ATTRIBUTE(4,5);
  62 static void ctdb_send_error(struct ctdb_context *ctdb,
  63                             struct ctdb_req_header *hdr, uint32_t status,
  64                             const char *fmt, ...)
  65 {
  66         va_list ap;
  67         struct ctdb_reply_error *r;
  68         char *msg;
  69         int msglen, len;
  70
  71         if (ctdb->methods == NULL) {
  72                 DEBUG(DEBUG_INFO,(__location__ " Failed to send error. Transport is DOWN\n"));
  73                 return;
  74         }
  75
  76         va_start(ap, fmt);
  77         msg = talloc_vasprintf(ctdb, fmt, ap);
  78         if (msg == NULL) {
  79                 ctdb_fatal(ctdb, "Unable to allocate error in ctdb_send_error\n");
  80         }
  81         va_end(ap);
  82
  83         msglen = strlen(msg)+1;
  84         len = offsetof(struct ctdb_reply_error, msg);
  85         r = ctdb_transport_allocate(ctdb, msg, CTDB_REPLY_ERROR, len + msglen,
  86                                     struct ctdb_reply_error);
  87         CTDB_NO_MEMORY_FATAL(ctdb, r);
  88
  89         r->hdr.destnode  = hdr->srcnode;
  90         r->hdr.reqid     = hdr->reqid;
  91         r->status        = status;
  92         r->msglen        = msglen;
  93         memcpy(&r->msg[0], msg, msglen);
  94
  95         ctdb_queue_packet(ctdb, &r->hdr);
  96
  97         talloc_free(msg);
  98 }
  99
 100
 101 /**
 102  * send a redirect reply
 103  *
 104  * The logic behind this function is this:
 105  *
 106  * A client wants to grab a record and sends a CTDB_REQ_CALL packet
 107  * to its local ctdb (ctdb_request_call). If the node is not itself
 108  * the record's DMASTER, it first redirects the packet to  the
 109  * record's LMASTER. The LMASTER then redirects the call packet to
 110  * the current DMASTER. But there is a race: The record may have
 111  * been migrated off the DMASTER while the redirected packet is
 112  * on the wire (or in the local queue). So in case the record has
 113  * migrated off the new destinaton of the call packet, instead of
 114  * going back to the LMASTER to get the new DMASTER, we try to
 115  * reduce round-trips by first chasing the record a couple of times
 116  * before giving up the direct chase and finally going back to the
 117  * LMASTER (again). Note that this works because of this: When
 118  * a record is migrated off a node, then the new DMASTER is stored
 119  * in the record's copy on the former DMASTER.
 120  *
 121  * The maximum number of attempts for direct chase to make before
 122  * going back to the LMASTER is configurable by the tunable
 123  * "MaxRedirectCount".
 124  */
 125 static void ctdb_call_send_redirect(struct ctdb_context *ctdb,
 126                                     TDB_DATA key,
 127                                     struct ctdb_req_call *c,
 128                                     struct ctdb_ltdb_header *header)
 129 {
 130
 131         uint32_t lmaster = ctdb_lmaster(ctdb, &key);
 132         if (ctdb->pnn == lmaster) {
 133                 c->hdr.destnode = header->dmaster;
 134         } else if ((c->hopcount % ctdb->tunable.max_redirect_count) == 0) {
 135                 c->hdr.destnode = lmaster;
 136         } else {
 137                 c->hdr.destnode = header->dmaster;
 138         }
 139         c->hopcount++;
 140         ctdb_queue_packet(ctdb, &c->hdr);
 141 }
 142
 143
 144 /*
 145   send a dmaster reply
 146
 147   caller must have the chainlock before calling this routine. Caller must be
 148   the lmaster
 149 */
 150 static void ctdb_send_dmaster_reply(struct ctdb_db_context *ctdb_db,
 151                                     struct ctdb_ltdb_header *header,
 152                                     TDB_DATA key, TDB_DATA data,
 153                                     uint32_t new_dmaster,
 154                                     uint32_t reqid)
 155 {
 156         struct ctdb_context *ctdb = ctdb_db->ctdb;
 157         struct ctdb_reply_dmaster *r;
 158         int ret, len;
 159         TALLOC_CTX *tmp_ctx;
 160
 161         if (ctdb->pnn != ctdb_lmaster(ctdb, &key)) {
 162                 DEBUG(DEBUG_ALERT,(__location__ " Caller is not lmaster!\n"));
 163                 return;
 164         }
 165
 166         header->dmaster = new_dmaster;
 167         ret = ctdb_ltdb_store(ctdb_db, key, header, data);
 168         if (ret != 0) {
 169                 ctdb_fatal(ctdb, "ctdb_send_dmaster_reply unable to update dmaster");
 170                 return;
 171         }
 172
 173         if (ctdb->methods == NULL) {
 174                 ctdb_fatal(ctdb, "ctdb_send_dmaster_reply cant update dmaster since transport is down");
 175                 return;
 176         }
 177
 178         /* put the packet on a temporary context, allowing us to safely free
 179            it below even if ctdb_reply_dmaster() has freed it already */
 180         tmp_ctx = talloc_new(ctdb);
 181
 182         /* send the CTDB_REPLY_DMASTER */
 183         len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize + sizeof(uint32_t);
 184         r = ctdb_transport_allocate(ctdb, tmp_ctx, CTDB_REPLY_DMASTER, len,
 185                                     struct ctdb_reply_dmaster);
 186         CTDB_NO_MEMORY_FATAL(ctdb, r);
 187
 188         r->hdr.destnode  = new_dmaster;
 189         r->hdr.reqid     = reqid;
 190         r->rsn           = header->rsn;
 191         r->keylen        = key.dsize;
 192         r->datalen       = data.dsize;
 193         r->db_id         = ctdb_db->db_id;
 194         memcpy(&r->data[0], key.dptr, key.dsize);
 195         memcpy(&r->data[key.dsize], data.dptr, data.dsize);
 196         memcpy(&r->data[key.dsize+data.dsize], &header->flags, sizeof(uint32_t));
 197
 198         ctdb_queue_packet(ctdb, &r->hdr);
 199
 200         talloc_free(tmp_ctx);
 201 }
 202
 203 /*
 204   send a dmaster request (give another node the dmaster for a record)
 205
 206   This is always sent to the lmaster, which ensures that the lmaster
 207   always knows who the dmaster is. The lmaster will then send a
 208   CTDB_REPLY_DMASTER to the new dmaster
 209 */
 210 static void ctdb_call_send_dmaster(struct ctdb_db_context *ctdb_db,
 211                                    struct ctdb_req_call *c,
 212                                    struct ctdb_ltdb_header *header,
 213                                    TDB_DATA *key, TDB_DATA *data)
 214 {
 215         struct ctdb_req_dmaster *r;
 216         struct ctdb_context *ctdb = ctdb_db->ctdb;
 217         int len;
 218         uint32_t lmaster = ctdb_lmaster(ctdb, key);
 219
 220         if (ctdb->methods == NULL) {
 221                 ctdb_fatal(ctdb, "Failed ctdb_call_send_dmaster since transport is down");
 222                 return;
 223         }
 224
 225         if (data->dsize != 0) {
 226                 header->flags |= CTDB_REC_FLAG_MIGRATED_WITH_DATA;
 227         }
 228
 229         if (lmaster == ctdb->pnn) {
 230                 ctdb_send_dmaster_reply(ctdb_db, header, *key, *data,
 231                                         c->hdr.srcnode, c->hdr.reqid);
 232                 return;
 233         }
 234
 235         len = offsetof(struct ctdb_req_dmaster, data) + key->dsize + data->dsize
 236                         + sizeof(uint32_t);
 237         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_DMASTER, len,
 238                                     struct ctdb_req_dmaster);
 239         CTDB_NO_MEMORY_FATAL(ctdb, r);
 240         r->hdr.destnode  = lmaster;
 241         r->hdr.reqid     = c->hdr.reqid;
 242         r->db_id         = c->db_id;
 243         r->rsn           = header->rsn;
 244         r->dmaster       = c->hdr.srcnode;
 245         r->keylen        = key->dsize;
 246         r->datalen       = data->dsize;
 247         memcpy(&r->data[0], key->dptr, key->dsize);
 248         memcpy(&r->data[key->dsize], data->dptr, data->dsize);
 249         memcpy(&r->data[key->dsize + data->dsize], &header->flags, sizeof(uint32_t));
 250
 251         header->dmaster = c->hdr.srcnode;
 252         if (ctdb_ltdb_store(ctdb_db, *key, header, *data) != 0) {
 253                 ctdb_fatal(ctdb, "Failed to store record in ctdb_call_send_dmaster");
 254         }
 255
 256         ctdb_queue_packet(ctdb, &r->hdr);
 257
 258         talloc_free(r);
 259 }
 260
 261 /*
 262   called when a CTDB_REPLY_DMASTER packet comes in, or when the lmaster
 263   gets a CTDB_REQUEST_DMASTER for itself. We become the dmaster.
 264
 265   must be called with the chainlock held. This function releases the chainlock
 266 */
 267 static void ctdb_become_dmaster(struct ctdb_db_context *ctdb_db,
 268                                 struct ctdb_req_header *hdr,
 269                                 TDB_DATA key, TDB_DATA data,
 270                                 uint64_t rsn, uint32_t record_flags)
 271 {
 272         struct ctdb_call_state *state;
 273         struct ctdb_context *ctdb = ctdb_db->ctdb;
 274         struct ctdb_ltdb_header header;
 275         int ret;
 276
 277         DEBUG(DEBUG_DEBUG,("pnn %u dmaster response %08x\n", ctdb->pnn, ctdb_hash(&key)));
 278
 279         ZERO_STRUCT(header);
 280         header.rsn = rsn + 1;
 281         header.dmaster = ctdb->pnn;
 282         header.flags = record_flags;
 283
 284         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 285
 286         if (state) {
 287                 if (state->call->flags & CTDB_CALL_FLAG_VACUUM_MIGRATION) {
 288                         /*
 289                          * We temporarily add the VACUUM_MIGRATED flag to
 290                          * the record flags, so that ctdb_ltdb_store can
 291                          * decide whether the record should be stored or
 292                          * deleted.
 293                          */
 294                         header.flags |= CTDB_REC_FLAG_VACUUM_MIGRATED;
 295                 }
 296         }
 297
 298         if (ctdb_ltdb_store(ctdb_db, key, &header, data) != 0) {
 299                 ctdb_fatal(ctdb, "ctdb_reply_dmaster store failed\n");
 300
 301                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 302                 if (ret != 0) {
 303                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 304                 }
 305                 return;
 306         }
 307
 308
 309         if (state == NULL) {
 310                 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_become_dmaster from node %u\n",
 311                          ctdb->pnn, hdr->reqid, hdr->srcnode));
 312
 313                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 314                 if (ret != 0) {
 315                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 316                 }
 317                 return;
 318         }
 319
 320         if (key.dsize != state->call->key.dsize || memcmp(key.dptr, state->call->key.dptr, key.dsize)) {
 321                 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));
 322
 323                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 324                 if (ret != 0) {
 325                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 326                 }
 327                 return;
 328         }
 329
 330         if (hdr->reqid != state->reqid) {
 331                 /* we found a record  but it was the wrong one */
 332                 DEBUG(DEBUG_ERR, ("Dropped orphan in ctdb_become_dmaster with reqid:%u\n from node %u", hdr->reqid, hdr->srcnode));
 333
 334                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 335                 if (ret != 0) {
 336                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 337                 }
 338                 return;
 339         }
 340
 341         ctdb_call_local(ctdb_db, state->call, &header, state, &data, true);
 342
 343         ret = ctdb_ltdb_unlock(ctdb_db, state->call->key);
 344         if (ret != 0) {
 345                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 346         }
 347
 348         state->state = CTDB_CALL_DONE;
 349         if (state->async.fn) {
 350                 state->async.fn(state);
 351         }
 352 }
 353
 354
 355
 356 /*
 357   called when a CTDB_REQ_DMASTER packet comes in
 358
 359   this comes into the lmaster for a record when the current dmaster
 360   wants to give up the dmaster role and give it to someone else
 361 */
 362 void ctdb_request_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 363 {
 364         struct ctdb_req_dmaster *c = (struct ctdb_req_dmaster *)hdr;
 365         TDB_DATA key, data, data2;
 366         struct ctdb_ltdb_header header;
 367         struct ctdb_db_context *ctdb_db;
 368         uint32_t record_flags = 0;
 369         size_t len;
 370         int ret;
 371
 372         key.dptr = c->data;
 373         key.dsize = c->keylen;
 374         data.dptr = c->data + c->keylen;
 375         data.dsize = c->datalen;
 376         len = offsetof(struct ctdb_req_dmaster, data) + key.dsize + data.dsize
 377                         + sizeof(uint32_t);
 378         if (len <= c->hdr.length) {
 379                 record_flags = *(uint32_t *)&c->data[c->keylen + c->datalen];
 380         }
 381
 382         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 383         if (!ctdb_db) {
 384                 ctdb_send_error(ctdb, hdr, -1,
 385                                 "Unknown database in request. db_id==0x%08x",
 386                                 c->db_id);
 387                 return;
 388         }
 389
 390         /* fetch the current record */
 391         ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, key, &header, hdr, &data2,
 392                                            ctdb_call_input_pkt, ctdb, False);
 393         if (ret == -1) {
 394                 ctdb_fatal(ctdb, "ctdb_req_dmaster failed to fetch record");
 395                 return;
 396         }
 397         if (ret == -2) {
 398                 DEBUG(DEBUG_INFO,(__location__ " deferring ctdb_request_dmaster\n"));
 399                 return;
 400         }
 401
 402         if (ctdb_lmaster(ctdb, &key) != ctdb->pnn) {
 403                 DEBUG(DEBUG_ALERT,("pnn %u dmaster request to non-lmaster lmaster=%u gen=%u curgen=%u\n",
 404                          ctdb->pnn, ctdb_lmaster(ctdb, &key),
 405                          hdr->generation, ctdb->vnn_map->generation));
 406                 ctdb_fatal(ctdb, "ctdb_req_dmaster to non-lmaster");
 407         }
 408
 409         DEBUG(DEBUG_DEBUG,("pnn %u dmaster request on %08x for %u from %u\n",
 410                  ctdb->pnn, ctdb_hash(&key), c->dmaster, c->hdr.srcnode));
 411
 412         /* its a protocol error if the sending node is not the current dmaster */
 413         if (header.dmaster != hdr->srcnode) {
 414                 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",
 415                          ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
 416                          ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation,
 417                          (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid,
 418                          (key.dsize >= 4)?(*(uint32_t *)key.dptr):0));
 419                 if (header.rsn != 0 || header.dmaster != ctdb->pnn) {
 420                         DEBUG(DEBUG_ERR,("ctdb_req_dmaster from non-master. Force a recovery.\n"));
 421
 422                         ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
 423                         ctdb_ltdb_unlock(ctdb_db, key);
 424                         return;
 425                 }
 426         }
 427
 428         if (header.rsn > c->rsn) {
 429                 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",
 430                          ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
 431                          ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation,
 432                          (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid));
 433         }
 434
 435         /* use the rsn from the sending node */
 436         header.rsn = c->rsn;
 437
 438         /* store the record flags from the sending node */
 439         header.flags = record_flags;
 440
 441         /* check if the new dmaster is the lmaster, in which case we
 442            skip the dmaster reply */
 443         if (c->dmaster == ctdb->pnn) {
 444                 ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn, record_flags);
 445         } else {
 446                 ctdb_send_dmaster_reply(ctdb_db, &header, key, data, c->dmaster, hdr->reqid);
 447
 448                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 449                 if (ret != 0) {
 450                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 451                 }
 452         }
 453 }
 454
 455
 456 /*
 457   called when a CTDB_REQ_CALL packet comes in
 458 */
 459 void ctdb_request_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 460 {
 461         struct ctdb_req_call *c = (struct ctdb_req_call *)hdr;
 462         TDB_DATA data;
 463         struct ctdb_reply_call *r;
 464         int ret, len;
 465         struct ctdb_ltdb_header header;
 466         struct ctdb_call *call;
 467         struct ctdb_db_context *ctdb_db;
 468
 469         if (ctdb->methods == NULL) {
 470                 DEBUG(DEBUG_INFO,(__location__ " Failed ctdb_request_call. Transport is DOWN\n"));
 471                 return;
 472         }
 473
 474
 475         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 476         if (!ctdb_db) {
 477                 ctdb_send_error(ctdb, hdr, -1,
 478                                 "Unknown database in request. db_id==0x%08x",
 479                                 c->db_id);
 480                 return;
 481         }
 482
 483         call = talloc(hdr, struct ctdb_call);
 484         CTDB_NO_MEMORY_FATAL(ctdb, call);
 485
 486         call->call_id  = c->callid;
 487         call->key.dptr = c->data;
 488         call->key.dsize = c->keylen;
 489         call->call_data.dptr = c->data + c->keylen;
 490         call->call_data.dsize = c->calldatalen;
 491         call->reply_data.dptr  = NULL;
 492         call->reply_data.dsize = 0;
 493
 494         /* determine if we are the dmaster for this key. This also
 495            fetches the record data (if any), thus avoiding a 2nd fetch of the data
 496            if the call will be answered locally */
 497
 498         ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, call->key, &header, hdr, &data,
 499                                            ctdb_call_input_pkt, ctdb, False);
 500         if (ret == -1) {
 501                 ctdb_send_error(ctdb, hdr, ret, "ltdb fetch failed in ctdb_request_call");
 502                 return;
 503         }
 504         if (ret == -2) {
 505                 DEBUG(DEBUG_INFO,(__location__ " deferred ctdb_request_call\n"));
 506                 return;
 507         }
 508
 509         /* Dont do READONLY if we dont have a tracking database */
 510         if ((c->flags & CTDB_WANT_READONLY) && !ctdb_db->readonly) {
 511                 c->flags &= ~CTDB_WANT_READONLY;
 512         }
 513
 514         if (header.flags & CTDB_REC_RO_REVOKE_COMPLETE) {
 515                 header.flags &= ~(CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY|CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_REVOKE_COMPLETE);
 516                 CTDB_INCREMENT_STAT(ctdb, total_ro_revokes);
 517                 if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
 518                         ctdb_fatal(ctdb, "Failed to write header with cleared REVOKE flag");
 519                 }
 520                 /* and clear out the tracking data */
 521                 if (tdb_delete(ctdb_db->rottdb, call->key) != 0) {
 522                         DEBUG(DEBUG_ERR,(__location__ " Failed to clear out trackingdb record\n"));
 523                 }
 524         }
 525
 526         /* if we are revoking, we must defer all other calls until the revoke
 527          * had completed.
 528          */
 529         if (header.flags & CTDB_REC_RO_REVOKING_READONLY) {
 530                 talloc_free(data.dptr);
 531                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 532
 533                 if (ctdb_add_revoke_deferred_call(ctdb, ctdb_db, call->key, hdr, ctdb_call_input_pkt, ctdb) != 0) {
 534                         ctdb_fatal(ctdb, "Failed to add deferred call for revoke child");
 535                 }
 536                 talloc_free(call);
 537                 return;
 538         }
 539
 540         /* if we are not the dmaster and are not hosting any delegations,
 541            then send a redirect to the requesting node */
 542         if ((header.dmaster != ctdb->pnn)
 543             && (!(header.flags & CTDB_REC_RO_HAVE_DELEGATIONS)) ) {
 544                 talloc_free(data.dptr);
 545                 ctdb_call_send_redirect(ctdb, call->key, c, &header);
 546
 547                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 548                 if (ret != 0) {
 549                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 550                 }
 551                 return;
 552         }
 553
 554         if ( (!(c->flags & CTDB_WANT_READONLY))
 555         && (header.flags & (CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY)) ) {
 556                 header.flags   |= CTDB_REC_RO_REVOKING_READONLY;
 557                 if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
 558                         ctdb_fatal(ctdb, "Failed to store record with HAVE_DELEGATIONS set");
 559                 }
 560                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 561
 562                 if (ctdb_start_revoke_ro_record(ctdb, ctdb_db, call->key, &header, data) != 0) {
 563                         ctdb_fatal(ctdb, "Failed to start record revoke");
 564                 }
 565                 talloc_free(data.dptr);
 566
 567                 if (ctdb_add_revoke_deferred_call(ctdb, ctdb_db, call->key, hdr, ctdb_call_input_pkt, ctdb) != 0) {
 568                         ctdb_fatal(ctdb, "Failed to add deferred call for revoke child");
 569                 }
 570                 talloc_free(call);
 571
 572                 return;
 573         }
 574
 575         /* If this is the first request for delegation. bump rsn and set
 576          * the delegations flag
 577          */
 578         if ((c->flags & CTDB_WANT_READONLY)
 579         &&  (c->callid == CTDB_FETCH_WITH_HEADER_FUNC)
 580         &&  (!(header.flags & CTDB_REC_RO_HAVE_DELEGATIONS))) {
 581                 header.rsn     += 3;
 582                 header.flags   |= CTDB_REC_RO_HAVE_DELEGATIONS;
 583                 if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
 584                         ctdb_fatal(ctdb, "Failed to store record with HAVE_DELEGATIONS set");
 585                 }
 586         }
 587         if ((c->flags & CTDB_WANT_READONLY)
 588         &&  (call->call_id == CTDB_FETCH_WITH_HEADER_FUNC)) {
 589                 TDB_DATA tdata;
 590
 591                 tdata = tdb_fetch(ctdb_db->rottdb, call->key);
 592                 if (ctdb_trackingdb_add_pnn(ctdb, &tdata, c->hdr.srcnode) != 0) {
 593                         ctdb_fatal(ctdb, "Failed to add node to trackingdb");
 594                 }
 595                 if (tdb_store(ctdb_db->rottdb, call->key, tdata, TDB_REPLACE) != 0) {
 596                         ctdb_fatal(ctdb, "Failed to store trackingdb data");
 597                 }
 598                 free(tdata.dptr);
 599
 600                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 601                 if (ret != 0) {
 602                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 603                 }
 604
 605                 len = offsetof(struct ctdb_reply_call, data) + data.dsize + sizeof(struct ctdb_ltdb_header);
 606                 r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len,
 607                                             struct ctdb_reply_call);
 608                 CTDB_NO_MEMORY_FATAL(ctdb, r);
 609                 r->hdr.destnode  = c->hdr.srcnode;
 610                 r->hdr.reqid     = c->hdr.reqid;
 611                 r->status        = 0;
 612                 r->datalen       = data.dsize + sizeof(struct ctdb_ltdb_header);
 613                 header.rsn      -= 2;
 614                 header.flags   |= CTDB_REC_RO_HAVE_READONLY;
 615                 header.flags   &= ~CTDB_REC_RO_HAVE_DELEGATIONS;
 616                 memcpy(&r->data[0], &header, sizeof(struct ctdb_ltdb_header));
 617
 618                 if (data.dsize) {
 619                         memcpy(&r->data[sizeof(struct ctdb_ltdb_header)], data.dptr, data.dsize);
 620                 }
 621
 622                 ctdb_queue_packet(ctdb, &r->hdr);
 623                 CTDB_INCREMENT_STAT(ctdb, total_ro_delegations);
 624
 625                 talloc_free(r);
 626                 return;
 627         }
 628
 629         CTDB_UPDATE_STAT(ctdb, max_hop_count, c->hopcount);
 630
 631         /* Try if possible to migrate the record off to the caller node.
 632          * From the clients perspective a fetch of the data is just as
 633          * expensive as a migration.
 634          */
 635         if (c->hdr.srcnode != ctdb->pnn) {
 636                 if (ctdb_db->transaction_active) {
 637                         DEBUG(DEBUG_INFO, (__location__ " refusing migration"
 638                               " of key %s while transaction is active\n",
 639                               (char *)call->key.dptr));
 640                 } else {
 641                         DEBUG(DEBUG_DEBUG,("pnn %u starting migration of %08x to %u\n",
 642                                  ctdb->pnn, ctdb_hash(&(call->key)), c->hdr.srcnode));
 643                         ctdb_call_send_dmaster(ctdb_db, c, &header, &(call->key), &data);
 644                         talloc_free(data.dptr);
 645
 646                         ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 647                         if (ret != 0) {
 648                                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 649                         }
 650                         return;
 651                 }
 652         }
 653
 654         ret = ctdb_call_local(ctdb_db, call, &header, hdr, &data, true);
 655         if (ret != 0) {
 656                 DEBUG(DEBUG_ERR,(__location__ " ctdb_call_local failed\n"));
 657                 call->status = -1;
 658         }
 659
 660         ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 661         if (ret != 0) {
 662                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 663         }
 664
 665         len = offsetof(struct ctdb_reply_call, data) + call->reply_data.dsize;
 666         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len,
 667                                     struct ctdb_reply_call);
 668         CTDB_NO_MEMORY_FATAL(ctdb, r);
 669         r->hdr.destnode  = hdr->srcnode;
 670         r->hdr.reqid     = hdr->reqid;
 671         r->status        = call->status;
 672         r->datalen       = call->reply_data.dsize;
 673         if (call->reply_data.dsize) {
 674                 memcpy(&r->data[0], call->reply_data.dptr, call->reply_data.dsize);
 675         }
 676
 677         ctdb_queue_packet(ctdb, &r->hdr);
 678
 679         talloc_free(r);
 680 }
 681
 682 /*
 683   called when a CTDB_REPLY_CALL packet comes in
 684
 685   This packet comes in response to a CTDB_REQ_CALL request packet. It
 686   contains any reply data from the call
 687 */
 688 void ctdb_reply_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 689 {
 690         struct ctdb_reply_call *c = (struct ctdb_reply_call *)hdr;
 691         struct ctdb_call_state *state;
 692
 693         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 694         if (state == NULL) {
 695                 DEBUG(DEBUG_ERR, (__location__ " reqid %u not found\n", hdr->reqid));
 696                 return;
 697         }
 698
 699         if (hdr->reqid != state->reqid) {
 700                 /* we found a record  but it was the wrong one */
 701                 DEBUG(DEBUG_ERR, ("Dropped orphaned call reply with reqid:%u\n",hdr->reqid));
 702                 return;
 703         }
 704
 705
 706         /* read only delegation processing */
 707         /* If we got a FETCH_WITH_HEADER we should check if this is a ro
 708          * delegation since we may need to update the record header
 709          */
 710         if (state->c->callid == CTDB_FETCH_WITH_HEADER_FUNC) {
 711                 struct ctdb_db_context *ctdb_db = state->ctdb_db;
 712                 struct ctdb_ltdb_header *header = (struct ctdb_ltdb_header *)&c->data[0];
 713                 struct ctdb_ltdb_header oldheader;
 714                 TDB_DATA key, data, olddata;
 715                 int ret;
 716
 717                 if (!(header->flags & CTDB_REC_RO_HAVE_READONLY)) {
 718                         goto finished_ro;
 719                         return;
 720                 }
 721
 722                 key.dsize = state->c->keylen;
 723                 key.dptr  = state->c->data;
 724                 ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
 725                                      ctdb_call_input_pkt, ctdb, False);
 726                 if (ret == -2) {
 727                         return;
 728                 }
 729                 if (ret != 0) {
 730                         DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_call\n"));
 731                         return;
 732                 }
 733
 734                 ret = ctdb_ltdb_fetch(ctdb_db, key, &oldheader, state, &olddata);
 735                 if (ret != 0) {
 736                         DEBUG(DEBUG_ERR, ("Failed to fetch old record in ctdb_reply_call\n"));
 737                         ctdb_ltdb_unlock(ctdb_db, key);
 738                         goto finished_ro;
 739                 }
 740
 741                 if (header->rsn <= oldheader.rsn) {
 742                         ctdb_ltdb_unlock(ctdb_db, key);
 743                         goto finished_ro;
 744                 }
 745
 746                 if (c->datalen < sizeof(struct ctdb_ltdb_header)) {
 747                         DEBUG(DEBUG_ERR,(__location__ " Got FETCH_WITH_HEADER reply with too little data: %d bytes\n", c->datalen));
 748                         ctdb_ltdb_unlock(ctdb_db, key);
 749                         goto finished_ro;
 750                 }
 751
 752                 data.dsize = c->datalen - sizeof(struct ctdb_ltdb_header);
 753                 data.dptr  = &c->data[sizeof(struct ctdb_ltdb_header)];
 754                 ret = ctdb_ltdb_store(ctdb_db, key, header, data);
 755                 if (ret != 0) {
 756                         DEBUG(DEBUG_ERR, ("Failed to store new record in ctdb_reply_call\n"));
 757                         ctdb_ltdb_unlock(ctdb_db, key);
 758                         goto finished_ro;
 759                 }
 760
 761                 ctdb_ltdb_unlock(ctdb_db, key);
 762         }
 763 finished_ro:
 764
 765         state->call->reply_data.dptr = c->data;
 766         state->call->reply_data.dsize = c->datalen;
 767         state->call->status = c->status;
 768
 769         talloc_steal(state, c);
 770
 771         state->state = CTDB_CALL_DONE;
 772         if (state->async.fn) {
 773                 state->async.fn(state);
 774         }
 775 }
 776
 777
 778 /*
 779   called when a CTDB_REPLY_DMASTER packet comes in
 780
 781   This packet comes in from the lmaster response to a CTDB_REQ_CALL
 782   request packet. It means that the current dmaster wants to give us
 783   the dmaster role
 784 */
 785 void ctdb_reply_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 786 {
 787         struct ctdb_reply_dmaster *c = (struct ctdb_reply_dmaster *)hdr;
 788         struct ctdb_db_context *ctdb_db;
 789         TDB_DATA key, data;
 790         uint32_t record_flags = 0;
 791         size_t len;
 792         int ret;
 793
 794         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 795         if (ctdb_db == NULL) {
 796                 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_reply_dmaster\n", c->db_id));
 797                 return;
 798         }
 799
 800         key.dptr = c->data;
 801         key.dsize = c->keylen;
 802         data.dptr = &c->data[key.dsize];
 803         data.dsize = c->datalen;
 804         len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize
 805                 + sizeof(uint32_t);
 806         if (len <= c->hdr.length) {
 807                 record_flags = *(uint32_t *)&c->data[c->keylen + c->datalen];
 808         }
 809
 810         ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
 811                                      ctdb_call_input_pkt, ctdb, False);
 812         if (ret == -2) {
 813                 return;
 814         }
 815         if (ret != 0) {
 816                 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_dmaster\n"));
 817                 return;
 818         }
 819
 820         ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn, record_flags);
 821 }
 822
 823
 824 /*
 825   called when a CTDB_REPLY_ERROR packet comes in
 826 */
 827 void ctdb_reply_error(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 828 {
 829         struct ctdb_reply_error *c = (struct ctdb_reply_error *)hdr;
 830         struct ctdb_call_state *state;
 831
 832         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 833         if (state == NULL) {
 834                 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_reply_error\n",
 835                          ctdb->pnn, hdr->reqid));
 836                 return;
 837         }
 838
 839         if (hdr->reqid != state->reqid) {
 840                 /* we found a record  but it was the wrong one */
 841                 DEBUG(DEBUG_ERR, ("Dropped orphaned error reply with reqid:%u\n",hdr->reqid));
 842                 return;
 843         }
 844
 845         talloc_steal(state, c);
 846
 847         state->state  = CTDB_CALL_ERROR;
 848         state->errmsg = (char *)c->msg;
 849         if (state->async.fn) {
 850                 state->async.fn(state);
 851         }
 852 }
 853
 854
 855 /*
 856   destroy a ctdb_call
 857 */
 858 static int ctdb_call_destructor(struct ctdb_call_state *state)
 859 {
 860         DLIST_REMOVE(state->ctdb_db->ctdb->pending_calls, state);
 861         ctdb_reqid_remove(state->ctdb_db->ctdb, state->reqid);
 862         return 0;
 863 }
 864
 865
 866 /*
 867   called when a ctdb_call needs to be resent after a reconfigure event
 868 */
 869 static void ctdb_call_resend(struct ctdb_call_state *state)
 870 {
 871         struct ctdb_context *ctdb = state->ctdb_db->ctdb;
 872
 873         state->generation = ctdb->vnn_map->generation;
 874
 875         /* use a new reqid, in case the old reply does eventually come in */
 876         ctdb_reqid_remove(ctdb, state->reqid);
 877         state->reqid = ctdb_reqid_new(ctdb, state);
 878         state->c->hdr.reqid = state->reqid;
 879
 880         /* update the generation count for this request, so its valid with the new vnn_map */
 881         state->c->hdr.generation = state->generation;
 882
 883         /* send the packet to ourselves, it will be redirected appropriately */
 884         state->c->hdr.destnode = ctdb->pnn;
 885
 886         ctdb_queue_packet(ctdb, &state->c->hdr);
 887         DEBUG(DEBUG_NOTICE,("resent ctdb_call\n"));
 888 }
 889
 890 /*
 891   resend all pending calls on recovery
 892  */
 893 void ctdb_call_resend_all(struct ctdb_context *ctdb)
 894 {
 895         struct ctdb_call_state *state, *next;
 896         for (state=ctdb->pending_calls;state;state=next) {
 897                 next = state->next;
 898                 ctdb_call_resend(state);
 899         }
 900 }
 901
 902 /*
 903   this allows the caller to setup a async.fn
 904 */
 905 static void call_local_trigger(struct event_context *ev, struct timed_event *te,
 906                        struct timeval t, void *private_data)
 907 {
 908         struct ctdb_call_state *state = talloc_get_type(private_data, struct ctdb_call_state);
 909         if (state->async.fn) {
 910                 state->async.fn(state);
 911         }
 912 }
 913
 914
 915 /*
 916   construct an event driven local ctdb_call
 917
 918   this is used so that locally processed ctdb_call requests are processed
 919   in an event driven manner
 920 */
 921 struct ctdb_call_state *ctdb_call_local_send(struct ctdb_db_context *ctdb_db,
 922                                              struct ctdb_call *call,
 923                                              struct ctdb_ltdb_header *header,
 924                                              TDB_DATA *data)
 925 {
 926         struct ctdb_call_state *state;
 927         struct ctdb_context *ctdb = ctdb_db->ctdb;
 928         int ret;
 929
 930         state = talloc_zero(ctdb_db, struct ctdb_call_state);
 931         CTDB_NO_MEMORY_NULL(ctdb, state);
 932
 933         talloc_steal(state, data->dptr);
 934
 935         state->state = CTDB_CALL_DONE;
 936         state->call  = talloc(state, struct ctdb_call);
 937         CTDB_NO_MEMORY_NULL(ctdb, state->call);
 938         *(state->call) = *call;
 939         state->ctdb_db = ctdb_db;
 940
 941         ret = ctdb_call_local(ctdb_db, state->call, header, state, data, true);
 942         if (ret != 0) {
 943                 DEBUG(DEBUG_DEBUG,("ctdb_call_local() failed, ignoring return code %d\n", ret));
 944         }
 945
 946         event_add_timed(ctdb->ev, state, timeval_zero(), call_local_trigger, state);
 947
 948         return state;
 949 }
 950
 951
 952 /*
 953   make a remote ctdb call - async send. Called in daemon context.
 954
 955   This constructs a ctdb_call request and queues it for processing.
 956   This call never blocks.
 957 */
 958 struct ctdb_call_state *ctdb_daemon_call_send_remote(struct ctdb_db_context *ctdb_db,
 959                                                      struct ctdb_call *call,
 960                                                      struct ctdb_ltdb_header *header)
 961 {
 962         uint32_t len;
 963         struct ctdb_call_state *state;
 964         struct ctdb_context *ctdb = ctdb_db->ctdb;
 965
 966         if (ctdb->methods == NULL) {
 967                 DEBUG(DEBUG_INFO,(__location__ " Failed send packet. Transport is down\n"));
 968                 return NULL;
 969         }
 970
 971         state = talloc_zero(ctdb_db, struct ctdb_call_state);
 972         CTDB_NO_MEMORY_NULL(ctdb, state);
 973         state->call = talloc(state, struct ctdb_call);
 974         CTDB_NO_MEMORY_NULL(ctdb, state->call);
 975
 976         state->reqid = ctdb_reqid_new(ctdb, state);
 977         state->ctdb_db = ctdb_db;
 978         talloc_set_destructor(state, ctdb_call_destructor);
 979
 980         len = offsetof(struct ctdb_req_call, data) + call->key.dsize + call->call_data.dsize;
 981         state->c = ctdb_transport_allocate(ctdb, state, CTDB_REQ_CALL, len,
 982                                            struct ctdb_req_call);
 983         CTDB_NO_MEMORY_NULL(ctdb, state->c);
 984         state->c->hdr.destnode  = header->dmaster;
 985
 986         /* this limits us to 16k outstanding messages - not unreasonable */
 987         state->c->hdr.reqid     = state->reqid;
 988         state->c->flags         = call->flags;
 989         state->c->db_id         = ctdb_db->db_id;
 990         state->c->callid        = call->call_id;
 991         state->c->hopcount      = 0;
 992         state->c->keylen        = call->key.dsize;
 993         state->c->calldatalen   = call->call_data.dsize;
 994         memcpy(&state->c->data[0], call->key.dptr, call->key.dsize);
 995         memcpy(&state->c->data[call->key.dsize],
 996                call->call_data.dptr, call->call_data.dsize);
 997         *(state->call)              = *call;
 998         state->call->call_data.dptr = &state->c->data[call->key.dsize];
 999         state->call->key.dptr       = &state->c->data[0];
1000
1001         state->state  = CTDB_CALL_WAIT;
1002         state->generation = ctdb->vnn_map->generation;
1003
1004         DLIST_ADD(ctdb->pending_calls, state);
1005
1006         ctdb_queue_packet(ctdb, &state->c->hdr);
1007
1008         return state;
1009 }
1010
1011 /*
1012   make a remote ctdb call - async recv - called in daemon context
1013
1014   This is called when the program wants to wait for a ctdb_call to complete and get the
1015   results. This call will block unless the call has already completed.
1016 */
1017 int ctdb_daemon_call_recv(struct ctdb_call_state *state, struct ctdb_call *call)
1018 {
1019         while (state->state < CTDB_CALL_DONE) {
1020                 event_loop_once(state->ctdb_db->ctdb->ev);
1021         }
1022         if (state->state != CTDB_CALL_DONE) {
1023                 ctdb_set_error(state->ctdb_db->ctdb, "%s", state->errmsg);
1024                 talloc_free(state);
1025                 return -1;
1026         }
1027
1028         if (state->call->reply_data.dsize) {
1029                 call->reply_data.dptr = talloc_memdup(call,
1030                                                       state->call->reply_data.dptr,
1031                                                       state->call->reply_data.dsize);
1032                 call->reply_data.dsize = state->call->reply_data.dsize;
1033         } else {
1034                 call->reply_data.dptr = NULL;
1035                 call->reply_data.dsize = 0;
1036         }
1037         call->status = state->call->status;
1038         talloc_free(state);
1039         return 0;
1040 }
1041
1042
1043 /*
1044    send a keepalive packet to the other node
1045 */
1046 void ctdb_send_keepalive(struct ctdb_context *ctdb, uint32_t destnode)
1047 {
1048         struct ctdb_req_keepalive *r;
1049
1050         if (ctdb->methods == NULL) {
1051                 DEBUG(DEBUG_INFO,(__location__ " Failed to send keepalive. Transport is DOWN\n"));
1052                 return;
1053         }
1054
1055         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_KEEPALIVE,
1056                                     sizeof(struct ctdb_req_keepalive),
1057                                     struct ctdb_req_keepalive);
1058         CTDB_NO_MEMORY_FATAL(ctdb, r);
1059         r->hdr.destnode  = destnode;
1060         r->hdr.reqid     = 0;
1061
1062         CTDB_INCREMENT_STAT(ctdb, keepalive_packets_sent);
1063
1064         ctdb_queue_packet(ctdb, &r->hdr);
1065
1066         talloc_free(r);
1067 }
1068
1069
1070
1071 struct revokechild_deferred_call {
1072         struct ctdb_context *ctdb;
1073         struct ctdb_req_header *hdr;
1074         deferred_requeue_fn fn;
1075         void *ctx;
1076 };
1077
1078 struct revokechild_handle {
1079         struct revokechild_handle *next, *prev;
1080         struct ctdb_context *ctdb;
1081         struct ctdb_db_context *ctdb_db;
1082         struct fd_event *fde;
1083         int status;
1084         int fd[2];
1085         pid_t child;
1086         TDB_DATA key;
1087 };
1088
1089 struct revokechild_requeue_handle {
1090         struct ctdb_context *ctdb;
1091         struct ctdb_req_header *hdr;
1092         deferred_requeue_fn fn;
1093         void *ctx;
1094 };
1095
1096 static void deferred_call_requeue(struct event_context *ev, struct timed_event *te,
1097                        struct timeval t, void *private_data)
1098 {
1099         struct revokechild_requeue_handle *requeue_handle = talloc_get_type(private_data, struct revokechild_requeue_handle);
1100
1101         requeue_handle->fn(requeue_handle->ctx, requeue_handle->hdr);
1102         talloc_free(requeue_handle);
1103 }
1104
1105 static int deferred_call_destructor(struct revokechild_deferred_call *deferred_call)
1106 {
1107         struct ctdb_context *ctdb = deferred_call->ctdb;
1108         struct revokechild_requeue_handle *requeue_handle = talloc(ctdb, struct revokechild_requeue_handle);
1109         struct ctdb_req_call *c = (struct ctdb_req_call *)deferred_call->hdr;
1110
1111         requeue_handle->ctdb = ctdb;
1112         requeue_handle->hdr  = deferred_call->hdr;
1113         requeue_handle->fn   = deferred_call->fn;
1114         requeue_handle->ctx  = deferred_call->ctx;
1115         talloc_steal(requeue_handle, requeue_handle->hdr);
1116
1117         /* when revoking, any READONLY requests have 1 second grace to let read/write finish first */
1118         event_add_timed(ctdb->ev, requeue_handle, timeval_current_ofs(c->flags & CTDB_WANT_READONLY ? 1 : 0, 0), deferred_call_requeue, requeue_handle);
1119
1120         return 0;
1121 }
1122
1123
1124 static int revokechild_destructor(struct revokechild_handle *rc)
1125 {
1126         if (rc->fde != NULL) {
1127                 talloc_free(rc->fde);
1128         }
1129
1130         if (rc->fd[0] != -1) {
1131                 close(rc->fd[0]);
1132         }
1133         if (rc->fd[1] != -1) {
1134                 close(rc->fd[1]);
1135         }
1136         kill(rc->child, SIGKILL);
1137
1138         DLIST_REMOVE(rc->ctdb_db->revokechild_active, rc);
1139         return 0;
1140 }
1141
1142 static void revokechild_handler(struct event_context *ev, struct fd_event *fde,
1143                              uint16_t flags, void *private_data)
1144 {
1145         struct revokechild_handle *rc = talloc_get_type(private_data,
1146                                                      struct revokechild_handle);
1147         int ret;
1148         char c;
1149
1150         ret = read(rc->fd[0], &c, 1);
1151         if (ret != 1) {
1152                 DEBUG(DEBUG_ERR,("Failed to read status from revokechild. errno:%d\n", errno));
1153                 rc->status = -1;
1154                 talloc_free(rc);
1155                 return;
1156         }
1157         if (c != 0) {
1158                 DEBUG(DEBUG_ERR,("revokechild returned failure. status:%d\n", c));
1159                 rc->status = -1;
1160                 talloc_free(rc);
1161                 return;
1162         }
1163
1164         talloc_free(rc);
1165 }
1166
1167 struct ctdb_revoke_state {
1168         struct ctdb_db_context *ctdb_db;
1169         TDB_DATA key;
1170         struct ctdb_ltdb_header *header;
1171         TDB_DATA data;
1172         int count;
1173         int status;
1174         int finished;
1175 };
1176
1177 static void update_record_cb(struct ctdb_client_control_state *state)
1178 {
1179         struct ctdb_revoke_state *revoke_state;
1180         int ret;
1181         int32_t res;
1182
1183         if (state == NULL) {
1184                 return;
1185         }
1186         revoke_state = state->async.private_data;
1187
1188         state->async.fn = NULL;
1189         ret = ctdb_control_recv(state->ctdb, state, state, NULL, &res, NULL);
1190         if ((ret != 0) || (res != 0)) {
1191                 DEBUG(DEBUG_ERR,("Recv for revoke update record failed ret:%d res:%d\n", ret, res));
1192                 revoke_state->status = -1;
1193         }
1194
1195         revoke_state->count--;
1196         if (revoke_state->count <= 0) {
1197                 revoke_state->finished = 1;
1198         }
1199 }
1200
1201 static void revoke_send_cb(struct ctdb_context *ctdb, uint32_t pnn, void *private_data)
1202 {
1203         struct ctdb_revoke_state *revoke_state = private_data;
1204         struct ctdb_client_control_state *state;
1205
1206         state = ctdb_ctrl_updaterecord_send(ctdb, revoke_state, timeval_current_ofs(5,0), pnn, revoke_state->ctdb_db, revoke_state->key, revoke_state->header, revoke_state->data);
1207         if (state == NULL) {
1208                 DEBUG(DEBUG_ERR,("Failure to send update record to revoke readonly delegation\n"));
1209                 revoke_state->status = -1;
1210                 return;
1211         }
1212         state->async.fn           = update_record_cb;
1213         state->async.private_data = revoke_state;
1214
1215         revoke_state->count++;
1216
1217 }
1218
1219 static void ctdb_revoke_timeout_handler(struct event_context *ev, struct timed_event *te,
1220                               struct timeval yt, void *private_data)
1221 {
1222         struct ctdb_revoke_state *state = private_data;
1223
1224         DEBUG(DEBUG_ERR,("Timed out waiting for revoke to finish\n"));
1225         state->finished = 1;
1226         state->status   = -1;
1227 }
1228
1229 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)
1230 {
1231         struct ctdb_revoke_state *state = talloc_zero(ctdb, struct ctdb_revoke_state);
1232         int status;
1233
1234         state->ctdb_db = ctdb_db;
1235         state->key     = key;
1236         state->header  = header;
1237         state->data    = data;
1238
1239         ctdb_trackingdb_traverse(ctdb, tdata, revoke_send_cb, state);
1240
1241         event_add_timed(ctdb->ev, state, timeval_current_ofs(5, 0), ctdb_revoke_timeout_handler, state);
1242
1243         while (state->finished == 0) {
1244                 event_loop_once(ctdb->ev);
1245         }
1246
1247         status = state->status;
1248
1249         if (status == 0) {
1250                 struct ctdb_ltdb_header new_header;
1251                 TDB_DATA new_data;
1252
1253                 if (ctdb_ltdb_lock(ctdb_db, key) != 0) {
1254                         DEBUG(DEBUG_ERR,("Failed to chainlock the database in revokechild\n"));
1255                         talloc_free(state);
1256                         return -1;
1257                 }
1258                 if (ctdb_ltdb_fetch(ctdb_db, key, &new_header, state, &new_data) != 0) {
1259                         ctdb_ltdb_unlock(ctdb_db, key);
1260                         DEBUG(DEBUG_ERR,("Failed for fetch tdb record in revokechild\n"));
1261                         talloc_free(state);
1262                         return -1;
1263                 }
1264                 header->rsn++;
1265                 if (new_header.rsn > header->rsn) {
1266                         ctdb_ltdb_unlock(ctdb_db, key);
1267                         DEBUG(DEBUG_ERR,("RSN too high in tdb record in revokechild\n"));
1268                         talloc_free(state);
1269                         return -1;
1270                 }
1271                 if ( (new_header.flags & (CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS)) != (CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS) ) {
1272                         ctdb_ltdb_unlock(ctdb_db, key);
1273                         DEBUG(DEBUG_ERR,("Flags are wrong in tdb record in revokechild\n"));
1274                         talloc_free(state);
1275                         return -1;
1276                 }
1277                 new_header.rsn++;
1278                 new_header.flags |= CTDB_REC_RO_REVOKE_COMPLETE;
1279                 if (ctdb_ltdb_store(ctdb_db, key, &new_header, new_data) != 0) {
1280                         ctdb_ltdb_unlock(ctdb_db, key);
1281                         DEBUG(DEBUG_ERR,("Failed to write new record in revokechild\n"));
1282                         talloc_free(state);
1283                         return -1;
1284                 }
1285                 ctdb_ltdb_unlock(ctdb_db, key);
1286         }
1287
1288         talloc_free(state);
1289         return status;
1290 }
1291
1292
1293 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)
1294 {
1295         TDB_DATA tdata;
1296         struct revokechild_handle *rc;
1297         pid_t parent = getpid();
1298         int ret;
1299
1300         header->flags &= ~(CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY);
1301         header->rsn   -= 1;
1302
1303         if ((rc = talloc_zero(ctdb_db, struct revokechild_handle)) == NULL) {
1304                 DEBUG(DEBUG_ERR,("Failed to allocate revokechild_handle\n"));
1305                 return -1;
1306         }
1307
1308         tdata = tdb_fetch(ctdb_db->rottdb, key);
1309         if (tdata.dsize > 0) {
1310                 uint8_t *tmp;
1311
1312                 tmp = tdata.dptr;
1313                 tdata.dptr = talloc_memdup(rc, tdata.dptr, tdata.dsize);
1314                 free(tmp);
1315         }
1316
1317         rc->status    = 0;
1318         rc->ctdb      = ctdb;
1319         rc->ctdb_db   = ctdb_db;
1320         rc->fd[0]     = -1;
1321         rc->fd[1]     = -1;
1322
1323         talloc_set_destructor(rc, revokechild_destructor);
1324
1325         rc->key.dsize = key.dsize;
1326         rc->key.dptr  = talloc_memdup(rc, key.dptr, key.dsize);
1327         if (rc->key.dptr == NULL) {
1328                 DEBUG(DEBUG_ERR,("Failed to allocate key for revokechild_handle\n"));
1329                 talloc_free(rc);
1330                 return -1;
1331         }
1332
1333         ret = pipe(rc->fd);
1334         if (ret != 0) {
1335                 DEBUG(DEBUG_ERR,("Failed to allocate key for revokechild_handle\n"));
1336                 talloc_free(rc);
1337                 return -1;
1338         }
1339
1340
1341         rc->child = ctdb_fork(ctdb);
1342         if (rc->child == (pid_t)-1) {
1343                 DEBUG(DEBUG_ERR,("Failed to fork child for revokechild\n"));
1344                 talloc_free(rc);
1345                 return -1;
1346         }
1347
1348         if (rc->child == 0) {
1349                 char c = 0;
1350                 close(rc->fd[0]);
1351                 debug_extra = talloc_asprintf(NULL, "revokechild-%s:", ctdb_db->db_name);
1352
1353                 if (switch_from_server_to_client(ctdb, "revokechild-%s", ctdb_db->db_name) != 0) {
1354                         DEBUG(DEBUG_ERR,("Failed to switch from server to client for revokechild process\n"));
1355                         c = 1;
1356                         goto child_finished;
1357                 }
1358
1359                 c = ctdb_revoke_all_delegations(ctdb, ctdb_db, tdata, key, header, data);
1360
1361 child_finished:
1362                 write(rc->fd[1], &c, 1);
1363                 /* make sure we die when our parent dies */
1364                 while (kill(parent, 0) == 0 || errno != ESRCH) {
1365                         sleep(5);
1366                 }
1367                 _exit(0);
1368         }
1369
1370         close(rc->fd[1]);
1371         rc->fd[1] = -1;
1372         set_close_on_exec(rc->fd[0]);
1373
1374         /* This is an active revokechild child process */
1375         DLIST_ADD_END(ctdb_db->revokechild_active, rc, NULL);
1376
1377         rc->fde = event_add_fd(ctdb->ev, rc, rc->fd[0],
1378                                    EVENT_FD_READ, revokechild_handler,
1379                                    (void *)rc);
1380         if (rc->fde == NULL) {
1381                 DEBUG(DEBUG_ERR,("Failed to set up fd event for revokechild process\n"));
1382                 talloc_free(rc);
1383         }
1384         tevent_fd_set_auto_close(rc->fde);
1385
1386         return 0;
1387 }
1388
1389 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)
1390 {
1391         struct revokechild_handle *rc;
1392         struct revokechild_deferred_call *deferred_call;
1393
1394         for (rc = ctdb_db->revokechild_active; rc; rc = rc->next) {
1395                 if (rc->key.dsize == 0) {
1396                         continue;
1397                 }
1398                 if (rc->key.dsize != key.dsize) {
1399                         continue;
1400                 }
1401                 if (!memcmp(rc->key.dptr, key.dptr, key.dsize)) {
1402                         break;
1403                 }
1404         }
1405
1406         if (rc == NULL) {
1407                 DEBUG(DEBUG_ERR,("Failed to add deferred call to revoke list. revoke structure not found\n"));
1408                 return -1;
1409         }
1410
1411         deferred_call = talloc(rc, struct revokechild_deferred_call);
1412         if (deferred_call == NULL) {
1413                 DEBUG(DEBUG_ERR,("Failed to allocate deferred call structure for revoking record\n"));
1414                 return -1;
1415         }
1416
1417         deferred_call->ctdb = ctdb;
1418         deferred_call->hdr  = hdr;
1419         deferred_call->fn   = fn;
1420         deferred_call->ctx  = call_context;
1421
1422         talloc_set_destructor(deferred_call, deferred_call_destructor);
1423         talloc_steal(deferred_call, hdr);
1424
1425         return 0;
1426 }