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                 CTDB_INCREMENT_DB_STAT(ctdb_db, db_ro_revokes);
 518                 if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
 519                         ctdb_fatal(ctdb, "Failed to write header with cleared REVOKE flag");
 520                 }
 521                 /* and clear out the tracking data */
 522                 if (tdb_delete(ctdb_db->rottdb, call->key) != 0) {
 523                         DEBUG(DEBUG_ERR,(__location__ " Failed to clear out trackingdb record\n"));
 524                 }
 525         }
 526
 527         /* if we are revoking, we must defer all other calls until the revoke
 528          * had completed.
 529          */
 530         if (header.flags & CTDB_REC_RO_REVOKING_READONLY) {
 531                 talloc_free(data.dptr);
 532                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 533
 534                 if (ctdb_add_revoke_deferred_call(ctdb, ctdb_db, call->key, hdr, ctdb_call_input_pkt, ctdb) != 0) {
 535                         ctdb_fatal(ctdb, "Failed to add deferred call for revoke child");
 536                 }
 537                 talloc_free(call);
 538                 return;
 539         }
 540
 541         /* if we are not the dmaster and are not hosting any delegations,
 542            then send a redirect to the requesting node */
 543         if ((header.dmaster != ctdb->pnn)
 544             && (!(header.flags & CTDB_REC_RO_HAVE_DELEGATIONS)) ) {
 545                 talloc_free(data.dptr);
 546                 ctdb_call_send_redirect(ctdb, call->key, c, &header);
 547
 548                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 549                 if (ret != 0) {
 550                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 551                 }
 552                 return;
 553         }
 554
 555         if ( (!(c->flags & CTDB_WANT_READONLY))
 556         && (header.flags & (CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY)) ) {
 557                 header.flags   |= CTDB_REC_RO_REVOKING_READONLY;
 558                 if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
 559                         ctdb_fatal(ctdb, "Failed to store record with HAVE_DELEGATIONS set");
 560                 }
 561                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 562
 563                 if (ctdb_start_revoke_ro_record(ctdb, ctdb_db, call->key, &header, data) != 0) {
 564                         ctdb_fatal(ctdb, "Failed to start record revoke");
 565                 }
 566                 talloc_free(data.dptr);
 567
 568                 if (ctdb_add_revoke_deferred_call(ctdb, ctdb_db, call->key, hdr, ctdb_call_input_pkt, ctdb) != 0) {
 569                         ctdb_fatal(ctdb, "Failed to add deferred call for revoke child");
 570                 }
 571                 talloc_free(call);
 572
 573                 return;
 574         }
 575
 576         /* If this is the first request for delegation. bump rsn and set
 577          * the delegations flag
 578          */
 579         if ((c->flags & CTDB_WANT_READONLY)
 580         &&  (c->callid == CTDB_FETCH_WITH_HEADER_FUNC)
 581         &&  (!(header.flags & CTDB_REC_RO_HAVE_DELEGATIONS))) {
 582                 header.rsn     += 3;
 583                 header.flags   |= CTDB_REC_RO_HAVE_DELEGATIONS;
 584                 if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
 585                         ctdb_fatal(ctdb, "Failed to store record with HAVE_DELEGATIONS set");
 586                 }
 587         }
 588         if ((c->flags & CTDB_WANT_READONLY)
 589         &&  (call->call_id == CTDB_FETCH_WITH_HEADER_FUNC)) {
 590                 TDB_DATA tdata;
 591
 592                 tdata = tdb_fetch(ctdb_db->rottdb, call->key);
 593                 if (ctdb_trackingdb_add_pnn(ctdb, &tdata, c->hdr.srcnode) != 0) {
 594                         ctdb_fatal(ctdb, "Failed to add node to trackingdb");
 595                 }
 596                 if (tdb_store(ctdb_db->rottdb, call->key, tdata, TDB_REPLACE) != 0) {
 597                         ctdb_fatal(ctdb, "Failed to store trackingdb data");
 598                 }
 599                 free(tdata.dptr);
 600
 601                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 602                 if (ret != 0) {
 603                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 604                 }
 605
 606                 len = offsetof(struct ctdb_reply_call, data) + data.dsize + sizeof(struct ctdb_ltdb_header);
 607                 r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len,
 608                                             struct ctdb_reply_call);
 609                 CTDB_NO_MEMORY_FATAL(ctdb, r);
 610                 r->hdr.destnode  = c->hdr.srcnode;
 611                 r->hdr.reqid     = c->hdr.reqid;
 612                 r->status        = 0;
 613                 r->datalen       = data.dsize + sizeof(struct ctdb_ltdb_header);
 614                 header.rsn      -= 2;
 615                 header.flags   |= CTDB_REC_RO_HAVE_READONLY;
 616                 header.flags   &= ~CTDB_REC_RO_HAVE_DELEGATIONS;
 617                 memcpy(&r->data[0], &header, sizeof(struct ctdb_ltdb_header));
 618
 619                 if (data.dsize) {
 620                         memcpy(&r->data[sizeof(struct ctdb_ltdb_header)], data.dptr, data.dsize);
 621                 }
 622
 623                 ctdb_queue_packet(ctdb, &r->hdr);
 624                 CTDB_INCREMENT_STAT(ctdb, total_ro_delegations);
 625                 CTDB_INCREMENT_DB_STAT(ctdb_db, db_ro_delegations);
 626
 627                 talloc_free(r);
 628                 return;
 629         }
 630
 631         CTDB_UPDATE_STAT(ctdb, max_hop_count, c->hopcount);
 632
 633         /* Try if possible to migrate the record off to the caller node.
 634          * From the clients perspective a fetch of the data is just as
 635          * expensive as a migration.
 636          */
 637         if (c->hdr.srcnode != ctdb->pnn) {
 638                 if (ctdb_db->transaction_active) {
 639                         DEBUG(DEBUG_INFO, (__location__ " refusing migration"
 640                               " of key %s while transaction is active\n",
 641                               (char *)call->key.dptr));
 642                 } else {
 643                         DEBUG(DEBUG_DEBUG,("pnn %u starting migration of %08x to %u\n",
 644                                  ctdb->pnn, ctdb_hash(&(call->key)), c->hdr.srcnode));
 645                         ctdb_call_send_dmaster(ctdb_db, c, &header, &(call->key), &data);
 646                         talloc_free(data.dptr);
 647
 648                         ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 649                         if (ret != 0) {
 650                                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 651                         }
 652                         return;
 653                 }
 654         }
 655
 656         ret = ctdb_call_local(ctdb_db, call, &header, hdr, &data, true);
 657         if (ret != 0) {
 658                 DEBUG(DEBUG_ERR,(__location__ " ctdb_call_local failed\n"));
 659                 call->status = -1;
 660         }
 661
 662         ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 663         if (ret != 0) {
 664                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 665         }
 666
 667         len = offsetof(struct ctdb_reply_call, data) + call->reply_data.dsize;
 668         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len,
 669                                     struct ctdb_reply_call);
 670         CTDB_NO_MEMORY_FATAL(ctdb, r);
 671         r->hdr.destnode  = hdr->srcnode;
 672         r->hdr.reqid     = hdr->reqid;
 673         r->status        = call->status;
 674         r->datalen       = call->reply_data.dsize;
 675         if (call->reply_data.dsize) {
 676                 memcpy(&r->data[0], call->reply_data.dptr, call->reply_data.dsize);
 677         }
 678
 679         ctdb_queue_packet(ctdb, &r->hdr);
 680
 681         talloc_free(r);
 682 }
 683
 684 /*
 685   called when a CTDB_REPLY_CALL packet comes in
 686
 687   This packet comes in response to a CTDB_REQ_CALL request packet. It
 688   contains any reply data from the call
 689 */
 690 void ctdb_reply_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 691 {
 692         struct ctdb_reply_call *c = (struct ctdb_reply_call *)hdr;
 693         struct ctdb_call_state *state;
 694
 695         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 696         if (state == NULL) {
 697                 DEBUG(DEBUG_ERR, (__location__ " reqid %u not found\n", hdr->reqid));
 698                 return;
 699         }
 700
 701         if (hdr->reqid != state->reqid) {
 702                 /* we found a record  but it was the wrong one */
 703                 DEBUG(DEBUG_ERR, ("Dropped orphaned call reply with reqid:%u\n",hdr->reqid));
 704                 return;
 705         }
 706
 707
 708         /* read only delegation processing */
 709         /* If we got a FETCH_WITH_HEADER we should check if this is a ro
 710          * delegation since we may need to update the record header
 711          */
 712         if (state->c->callid == CTDB_FETCH_WITH_HEADER_FUNC) {
 713                 struct ctdb_db_context *ctdb_db = state->ctdb_db;
 714                 struct ctdb_ltdb_header *header = (struct ctdb_ltdb_header *)&c->data[0];
 715                 struct ctdb_ltdb_header oldheader;
 716                 TDB_DATA key, data, olddata;
 717                 int ret;
 718
 719                 if (!(header->flags & CTDB_REC_RO_HAVE_READONLY)) {
 720                         goto finished_ro;
 721                         return;
 722                 }
 723
 724                 key.dsize = state->c->keylen;
 725                 key.dptr  = state->c->data;
 726                 ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
 727                                      ctdb_call_input_pkt, ctdb, False);
 728                 if (ret == -2) {
 729                         return;
 730                 }
 731                 if (ret != 0) {
 732                         DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_call\n"));
 733                         return;
 734                 }
 735
 736                 ret = ctdb_ltdb_fetch(ctdb_db, key, &oldheader, state, &olddata);
 737                 if (ret != 0) {
 738                         DEBUG(DEBUG_ERR, ("Failed to fetch old record in ctdb_reply_call\n"));
 739                         ctdb_ltdb_unlock(ctdb_db, key);
 740                         goto finished_ro;
 741                 }
 742
 743                 if (header->rsn <= oldheader.rsn) {
 744                         ctdb_ltdb_unlock(ctdb_db, key);
 745                         goto finished_ro;
 746                 }
 747
 748                 if (c->datalen < sizeof(struct ctdb_ltdb_header)) {
 749                         DEBUG(DEBUG_ERR,(__location__ " Got FETCH_WITH_HEADER reply with too little data: %d bytes\n", c->datalen));
 750                         ctdb_ltdb_unlock(ctdb_db, key);
 751                         goto finished_ro;
 752                 }
 753
 754                 data.dsize = c->datalen - sizeof(struct ctdb_ltdb_header);
 755                 data.dptr  = &c->data[sizeof(struct ctdb_ltdb_header)];
 756                 ret = ctdb_ltdb_store(ctdb_db, key, header, data);
 757                 if (ret != 0) {
 758                         DEBUG(DEBUG_ERR, ("Failed to store new record in ctdb_reply_call\n"));
 759                         ctdb_ltdb_unlock(ctdb_db, key);
 760                         goto finished_ro;
 761                 }
 762
 763                 ctdb_ltdb_unlock(ctdb_db, key);
 764         }
 765 finished_ro:
 766
 767         state->call->reply_data.dptr = c->data;
 768         state->call->reply_data.dsize = c->datalen;
 769         state->call->status = c->status;
 770
 771         talloc_steal(state, c);
 772
 773         state->state = CTDB_CALL_DONE;
 774         if (state->async.fn) {
 775                 state->async.fn(state);
 776         }
 777 }
 778
 779
 780 /*
 781   called when a CTDB_REPLY_DMASTER packet comes in
 782
 783   This packet comes in from the lmaster response to a CTDB_REQ_CALL
 784   request packet. It means that the current dmaster wants to give us
 785   the dmaster role
 786 */
 787 void ctdb_reply_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 788 {
 789         struct ctdb_reply_dmaster *c = (struct ctdb_reply_dmaster *)hdr;
 790         struct ctdb_db_context *ctdb_db;
 791         TDB_DATA key, data;
 792         uint32_t record_flags = 0;
 793         size_t len;
 794         int ret;
 795
 796         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 797         if (ctdb_db == NULL) {
 798                 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_reply_dmaster\n", c->db_id));
 799                 return;
 800         }
 801
 802         key.dptr = c->data;
 803         key.dsize = c->keylen;
 804         data.dptr = &c->data[key.dsize];
 805         data.dsize = c->datalen;
 806         len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize
 807                 + sizeof(uint32_t);
 808         if (len <= c->hdr.length) {
 809                 record_flags = *(uint32_t *)&c->data[c->keylen + c->datalen];
 810         }
 811
 812         ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
 813                                      ctdb_call_input_pkt, ctdb, False);
 814         if (ret == -2) {
 815                 return;
 816         }
 817         if (ret != 0) {
 818                 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_dmaster\n"));
 819                 return;
 820         }
 821
 822         ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn, record_flags);
 823 }
 824
 825
 826 /*
 827   called when a CTDB_REPLY_ERROR packet comes in
 828 */
 829 void ctdb_reply_error(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 830 {
 831         struct ctdb_reply_error *c = (struct ctdb_reply_error *)hdr;
 832         struct ctdb_call_state *state;
 833
 834         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 835         if (state == NULL) {
 836                 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_reply_error\n",
 837                          ctdb->pnn, hdr->reqid));
 838                 return;
 839         }
 840
 841         if (hdr->reqid != state->reqid) {
 842                 /* we found a record  but it was the wrong one */
 843                 DEBUG(DEBUG_ERR, ("Dropped orphaned error reply with reqid:%u\n",hdr->reqid));
 844                 return;
 845         }
 846
 847         talloc_steal(state, c);
 848
 849         state->state  = CTDB_CALL_ERROR;
 850         state->errmsg = (char *)c->msg;
 851         if (state->async.fn) {
 852                 state->async.fn(state);
 853         }
 854 }
 855
 856
 857 /*
 858   destroy a ctdb_call
 859 */
 860 static int ctdb_call_destructor(struct ctdb_call_state *state)
 861 {
 862         DLIST_REMOVE(state->ctdb_db->ctdb->pending_calls, state);
 863         ctdb_reqid_remove(state->ctdb_db->ctdb, state->reqid);
 864         return 0;
 865 }
 866
 867
 868 /*
 869   called when a ctdb_call needs to be resent after a reconfigure event
 870 */
 871 static void ctdb_call_resend(struct ctdb_call_state *state)
 872 {
 873         struct ctdb_context *ctdb = state->ctdb_db->ctdb;
 874
 875         state->generation = ctdb->vnn_map->generation;
 876
 877         /* use a new reqid, in case the old reply does eventually come in */
 878         ctdb_reqid_remove(ctdb, state->reqid);
 879         state->reqid = ctdb_reqid_new(ctdb, state);
 880         state->c->hdr.reqid = state->reqid;
 881
 882         /* update the generation count for this request, so its valid with the new vnn_map */
 883         state->c->hdr.generation = state->generation;
 884
 885         /* send the packet to ourselves, it will be redirected appropriately */
 886         state->c->hdr.destnode = ctdb->pnn;
 887
 888         ctdb_queue_packet(ctdb, &state->c->hdr);
 889         DEBUG(DEBUG_NOTICE,("resent ctdb_call\n"));
 890 }
 891
 892 /*
 893   resend all pending calls on recovery
 894  */
 895 void ctdb_call_resend_all(struct ctdb_context *ctdb)
 896 {
 897         struct ctdb_call_state *state, *next;
 898         for (state=ctdb->pending_calls;state;state=next) {
 899                 next = state->next;
 900                 ctdb_call_resend(state);
 901         }
 902 }
 903
 904 /*
 905   this allows the caller to setup a async.fn
 906 */
 907 static void call_local_trigger(struct event_context *ev, struct timed_event *te,
 908                        struct timeval t, void *private_data)
 909 {
 910         struct ctdb_call_state *state = talloc_get_type(private_data, struct ctdb_call_state);
 911         if (state->async.fn) {
 912                 state->async.fn(state);
 913         }
 914 }
 915
 916
 917 /*
 918   construct an event driven local ctdb_call
 919
 920   this is used so that locally processed ctdb_call requests are processed
 921   in an event driven manner
 922 */
 923 struct ctdb_call_state *ctdb_call_local_send(struct ctdb_db_context *ctdb_db,
 924                                              struct ctdb_call *call,
 925                                              struct ctdb_ltdb_header *header,
 926                                              TDB_DATA *data)
 927 {
 928         struct ctdb_call_state *state;
 929         struct ctdb_context *ctdb = ctdb_db->ctdb;
 930         int ret;
 931
 932         state = talloc_zero(ctdb_db, struct ctdb_call_state);
 933         CTDB_NO_MEMORY_NULL(ctdb, state);
 934
 935         talloc_steal(state, data->dptr);
 936
 937         state->state = CTDB_CALL_DONE;
 938         state->call  = talloc(state, struct ctdb_call);
 939         CTDB_NO_MEMORY_NULL(ctdb, state->call);
 940         *(state->call) = *call;
 941         state->ctdb_db = ctdb_db;
 942
 943         ret = ctdb_call_local(ctdb_db, state->call, header, state, data, true);
 944         if (ret != 0) {
 945                 DEBUG(DEBUG_DEBUG,("ctdb_call_local() failed, ignoring return code %d\n", ret));
 946         }
 947
 948         event_add_timed(ctdb->ev, state, timeval_zero(), call_local_trigger, state);
 949
 950         return state;
 951 }
 952
 953
 954 /*
 955   make a remote ctdb call - async send. Called in daemon context.
 956
 957   This constructs a ctdb_call request and queues it for processing.
 958   This call never blocks.
 959 */
 960 struct ctdb_call_state *ctdb_daemon_call_send_remote(struct ctdb_db_context *ctdb_db,
 961                                                      struct ctdb_call *call,
 962                                                      struct ctdb_ltdb_header *header)
 963 {
 964         uint32_t len;
 965         struct ctdb_call_state *state;
 966         struct ctdb_context *ctdb = ctdb_db->ctdb;
 967
 968         if (ctdb->methods == NULL) {
 969                 DEBUG(DEBUG_INFO,(__location__ " Failed send packet. Transport is down\n"));
 970                 return NULL;
 971         }
 972
 973         state = talloc_zero(ctdb_db, struct ctdb_call_state);
 974         CTDB_NO_MEMORY_NULL(ctdb, state);
 975         state->call = talloc(state, struct ctdb_call);
 976         CTDB_NO_MEMORY_NULL(ctdb, state->call);
 977
 978         state->reqid = ctdb_reqid_new(ctdb, state);
 979         state->ctdb_db = ctdb_db;
 980         talloc_set_destructor(state, ctdb_call_destructor);
 981
 982         len = offsetof(struct ctdb_req_call, data) + call->key.dsize + call->call_data.dsize;
 983         state->c = ctdb_transport_allocate(ctdb, state, CTDB_REQ_CALL, len,
 984                                            struct ctdb_req_call);
 985         CTDB_NO_MEMORY_NULL(ctdb, state->c);
 986         state->c->hdr.destnode  = header->dmaster;
 987
 988         /* this limits us to 16k outstanding messages - not unreasonable */
 989         state->c->hdr.reqid     = state->reqid;
 990         state->c->flags         = call->flags;
 991         state->c->db_id         = ctdb_db->db_id;
 992         state->c->callid        = call->call_id;
 993         state->c->hopcount      = 0;
 994         state->c->keylen        = call->key.dsize;
 995         state->c->calldatalen   = call->call_data.dsize;
 996         memcpy(&state->c->data[0], call->key.dptr, call->key.dsize);
 997         memcpy(&state->c->data[call->key.dsize],
 998                call->call_data.dptr, call->call_data.dsize);
 999         *(state->call)              = *call;
1000         state->call->call_data.dptr = &state->c->data[call->key.dsize];
1001         state->call->key.dptr       = &state->c->data[0];
1002
1003         state->state  = CTDB_CALL_WAIT;
1004         state->generation = ctdb->vnn_map->generation;
1005
1006         DLIST_ADD(ctdb->pending_calls, state);
1007
1008         ctdb_queue_packet(ctdb, &state->c->hdr);
1009
1010         return state;
1011 }
1012
1013 /*
1014   make a remote ctdb call - async recv - called in daemon context
1015
1016   This is called when the program wants to wait for a ctdb_call to complete and get the
1017   results. This call will block unless the call has already completed.
1018 */
1019 int ctdb_daemon_call_recv(struct ctdb_call_state *state, struct ctdb_call *call)
1020 {
1021         while (state->state < CTDB_CALL_DONE) {
1022                 event_loop_once(state->ctdb_db->ctdb->ev);
1023         }
1024         if (state->state != CTDB_CALL_DONE) {
1025                 ctdb_set_error(state->ctdb_db->ctdb, "%s", state->errmsg);
1026                 talloc_free(state);
1027                 return -1;
1028         }
1029
1030         if (state->call->reply_data.dsize) {
1031                 call->reply_data.dptr = talloc_memdup(call,
1032                                                       state->call->reply_data.dptr,
1033                                                       state->call->reply_data.dsize);
1034                 call->reply_data.dsize = state->call->reply_data.dsize;
1035         } else {
1036                 call->reply_data.dptr = NULL;
1037                 call->reply_data.dsize = 0;
1038         }
1039         call->status = state->call->status;
1040         talloc_free(state);
1041         return 0;
1042 }
1043
1044
1045 /*
1046    send a keepalive packet to the other node
1047 */
1048 void ctdb_send_keepalive(struct ctdb_context *ctdb, uint32_t destnode)
1049 {
1050         struct ctdb_req_keepalive *r;
1051
1052         if (ctdb->methods == NULL) {
1053                 DEBUG(DEBUG_INFO,(__location__ " Failed to send keepalive. Transport is DOWN\n"));
1054                 return;
1055         }
1056
1057         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_KEEPALIVE,
1058                                     sizeof(struct ctdb_req_keepalive),
1059                                     struct ctdb_req_keepalive);
1060         CTDB_NO_MEMORY_FATAL(ctdb, r);
1061         r->hdr.destnode  = destnode;
1062         r->hdr.reqid     = 0;
1063
1064         CTDB_INCREMENT_STAT(ctdb, keepalive_packets_sent);
1065
1066         ctdb_queue_packet(ctdb, &r->hdr);
1067
1068         talloc_free(r);
1069 }
1070
1071
1072
1073 struct revokechild_deferred_call {
1074         struct ctdb_context *ctdb;
1075         struct ctdb_req_header *hdr;
1076         deferred_requeue_fn fn;
1077         void *ctx;
1078 };
1079
1080 struct revokechild_handle {
1081         struct revokechild_handle *next, *prev;
1082         struct ctdb_context *ctdb;
1083         struct ctdb_db_context *ctdb_db;
1084         struct fd_event *fde;
1085         int status;
1086         int fd[2];
1087         pid_t child;
1088         TDB_DATA key;
1089 };
1090
1091 struct revokechild_requeue_handle {
1092         struct ctdb_context *ctdb;
1093         struct ctdb_req_header *hdr;
1094         deferred_requeue_fn fn;
1095         void *ctx;
1096 };
1097
1098 static void deferred_call_requeue(struct event_context *ev, struct timed_event *te,
1099                        struct timeval t, void *private_data)
1100 {
1101         struct revokechild_requeue_handle *requeue_handle = talloc_get_type(private_data, struct revokechild_requeue_handle);
1102
1103         requeue_handle->fn(requeue_handle->ctx, requeue_handle->hdr);
1104         talloc_free(requeue_handle);
1105 }
1106
1107 static int deferred_call_destructor(struct revokechild_deferred_call *deferred_call)
1108 {
1109         struct ctdb_context *ctdb = deferred_call->ctdb;
1110         struct revokechild_requeue_handle *requeue_handle = talloc(ctdb, struct revokechild_requeue_handle);
1111         struct ctdb_req_call *c = (struct ctdb_req_call *)deferred_call->hdr;
1112
1113         requeue_handle->ctdb = ctdb;
1114         requeue_handle->hdr  = deferred_call->hdr;
1115         requeue_handle->fn   = deferred_call->fn;
1116         requeue_handle->ctx  = deferred_call->ctx;
1117         talloc_steal(requeue_handle, requeue_handle->hdr);
1118
1119         /* when revoking, any READONLY requests have 1 second grace to let read/write finish first */
1120         event_add_timed(ctdb->ev, requeue_handle, timeval_current_ofs(c->flags & CTDB_WANT_READONLY ? 1 : 0, 0), deferred_call_requeue, requeue_handle);
1121
1122         return 0;
1123 }
1124
1125
1126 static int revokechild_destructor(struct revokechild_handle *rc)
1127 {
1128         if (rc->fde != NULL) {
1129                 talloc_free(rc->fde);
1130         }
1131
1132         if (rc->fd[0] != -1) {
1133                 close(rc->fd[0]);
1134         }
1135         if (rc->fd[1] != -1) {
1136                 close(rc->fd[1]);
1137         }
1138         kill(rc->child, SIGKILL);
1139
1140         DLIST_REMOVE(rc->ctdb_db->revokechild_active, rc);
1141         return 0;
1142 }
1143
1144 static void revokechild_handler(struct event_context *ev, struct fd_event *fde,
1145                              uint16_t flags, void *private_data)
1146 {
1147         struct revokechild_handle *rc = talloc_get_type(private_data,
1148                                                      struct revokechild_handle);
1149         int ret;
1150         char c;
1151
1152         ret = read(rc->fd[0], &c, 1);
1153         if (ret != 1) {
1154                 DEBUG(DEBUG_ERR,("Failed to read status from revokechild. errno:%d\n", errno));
1155                 rc->status = -1;
1156                 talloc_free(rc);
1157                 return;
1158         }
1159         if (c != 0) {
1160                 DEBUG(DEBUG_ERR,("revokechild returned failure. status:%d\n", c));
1161                 rc->status = -1;
1162                 talloc_free(rc);
1163                 return;
1164         }
1165
1166         talloc_free(rc);
1167 }
1168
1169 struct ctdb_revoke_state {
1170         struct ctdb_db_context *ctdb_db;
1171         TDB_DATA key;
1172         struct ctdb_ltdb_header *header;
1173         TDB_DATA data;
1174         int count;
1175         int status;
1176         int finished;
1177 };
1178
1179 static void update_record_cb(struct ctdb_client_control_state *state)
1180 {
1181         struct ctdb_revoke_state *revoke_state;
1182         int ret;
1183         int32_t res;
1184
1185         if (state == NULL) {
1186                 return;
1187         }
1188         revoke_state = state->async.private_data;
1189
1190         state->async.fn = NULL;
1191         ret = ctdb_control_recv(state->ctdb, state, state, NULL, &res, NULL);
1192         if ((ret != 0) || (res != 0)) {
1193                 DEBUG(DEBUG_ERR,("Recv for revoke update record failed ret:%d res:%d\n", ret, res));
1194                 revoke_state->status = -1;
1195         }
1196
1197         revoke_state->count--;
1198         if (revoke_state->count <= 0) {
1199                 revoke_state->finished = 1;
1200         }
1201 }
1202
1203 static void revoke_send_cb(struct ctdb_context *ctdb, uint32_t pnn, void *private_data)
1204 {
1205         struct ctdb_revoke_state *revoke_state = private_data;
1206         struct ctdb_client_control_state *state;
1207
1208         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);
1209         if (state == NULL) {
1210                 DEBUG(DEBUG_ERR,("Failure to send update record to revoke readonly delegation\n"));
1211                 revoke_state->status = -1;
1212                 return;
1213         }
1214         state->async.fn           = update_record_cb;
1215         state->async.private_data = revoke_state;
1216
1217         revoke_state->count++;
1218
1219 }
1220
1221 static void ctdb_revoke_timeout_handler(struct event_context *ev, struct timed_event *te,
1222                               struct timeval yt, void *private_data)
1223 {
1224         struct ctdb_revoke_state *state = private_data;
1225
1226         DEBUG(DEBUG_ERR,("Timed out waiting for revoke to finish\n"));
1227         state->finished = 1;
1228         state->status   = -1;
1229 }
1230
1231 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)
1232 {
1233         struct ctdb_revoke_state *state = talloc_zero(ctdb, struct ctdb_revoke_state);
1234         int status;
1235
1236         state->ctdb_db = ctdb_db;
1237         state->key     = key;
1238         state->header  = header;
1239         state->data    = data;
1240
1241         ctdb_trackingdb_traverse(ctdb, tdata, revoke_send_cb, state);
1242
1243         event_add_timed(ctdb->ev, state, timeval_current_ofs(5, 0), ctdb_revoke_timeout_handler, state);
1244
1245         while (state->finished == 0) {
1246                 event_loop_once(ctdb->ev);
1247         }
1248
1249         status = state->status;
1250
1251         if (status == 0) {
1252                 struct ctdb_ltdb_header new_header;
1253                 TDB_DATA new_data;
1254
1255                 if (ctdb_ltdb_lock(ctdb_db, key) != 0) {
1256                         DEBUG(DEBUG_ERR,("Failed to chainlock the database in revokechild\n"));
1257                         talloc_free(state);
1258                         return -1;
1259                 }
1260                 if (ctdb_ltdb_fetch(ctdb_db, key, &new_header, state, &new_data) != 0) {
1261                         ctdb_ltdb_unlock(ctdb_db, key);
1262                         DEBUG(DEBUG_ERR,("Failed for fetch tdb record in revokechild\n"));
1263                         talloc_free(state);
1264                         return -1;
1265                 }
1266                 header->rsn++;
1267                 if (new_header.rsn > header->rsn) {
1268                         ctdb_ltdb_unlock(ctdb_db, key);
1269                         DEBUG(DEBUG_ERR,("RSN too high in tdb record in revokechild\n"));
1270                         talloc_free(state);
1271                         return -1;
1272                 }
1273                 if ( (new_header.flags & (CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS)) != (CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS) ) {
1274                         ctdb_ltdb_unlock(ctdb_db, key);
1275                         DEBUG(DEBUG_ERR,("Flags are wrong in tdb record in revokechild\n"));
1276                         talloc_free(state);
1277                         return -1;
1278                 }
1279                 new_header.rsn++;
1280                 new_header.flags |= CTDB_REC_RO_REVOKE_COMPLETE;
1281                 if (ctdb_ltdb_store(ctdb_db, key, &new_header, new_data) != 0) {
1282                         ctdb_ltdb_unlock(ctdb_db, key);
1283                         DEBUG(DEBUG_ERR,("Failed to write new record in revokechild\n"));
1284                         talloc_free(state);
1285                         return -1;
1286                 }
1287                 ctdb_ltdb_unlock(ctdb_db, key);
1288         }
1289
1290         talloc_free(state);
1291         return status;
1292 }
1293
1294
1295 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)
1296 {
1297         TDB_DATA tdata;
1298         struct revokechild_handle *rc;
1299         pid_t parent = getpid();
1300         int ret;
1301
1302         header->flags &= ~(CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY);
1303         header->flags |= CTDB_REC_FLAG_MIGRATED_WITH_DATA;
1304         header->rsn   -= 1;
1305
1306         if ((rc = talloc_zero(ctdb_db, struct revokechild_handle)) == NULL) {
1307                 DEBUG(DEBUG_ERR,("Failed to allocate revokechild_handle\n"));
1308                 return -1;
1309         }
1310
1311         tdata = tdb_fetch(ctdb_db->rottdb, key);
1312         if (tdata.dsize > 0) {
1313                 uint8_t *tmp;
1314
1315                 tmp = tdata.dptr;
1316                 tdata.dptr = talloc_memdup(rc, tdata.dptr, tdata.dsize);
1317                 free(tmp);
1318         }
1319
1320         rc->status    = 0;
1321         rc->ctdb      = ctdb;
1322         rc->ctdb_db   = ctdb_db;
1323         rc->fd[0]     = -1;
1324         rc->fd[1]     = -1;
1325
1326         talloc_set_destructor(rc, revokechild_destructor);
1327
1328         rc->key.dsize = key.dsize;
1329         rc->key.dptr  = talloc_memdup(rc, key.dptr, key.dsize);
1330         if (rc->key.dptr == NULL) {
1331                 DEBUG(DEBUG_ERR,("Failed to allocate key for revokechild_handle\n"));
1332                 talloc_free(rc);
1333                 return -1;
1334         }
1335
1336         ret = pipe(rc->fd);
1337         if (ret != 0) {
1338                 DEBUG(DEBUG_ERR,("Failed to allocate key for revokechild_handle\n"));
1339                 talloc_free(rc);
1340                 return -1;
1341         }
1342
1343
1344         rc->child = ctdb_fork(ctdb);
1345         if (rc->child == (pid_t)-1) {
1346                 DEBUG(DEBUG_ERR,("Failed to fork child for revokechild\n"));
1347                 talloc_free(rc);
1348                 return -1;
1349         }
1350
1351         if (rc->child == 0) {
1352                 char c = 0;
1353                 close(rc->fd[0]);
1354                 debug_extra = talloc_asprintf(NULL, "revokechild-%s:", ctdb_db->db_name);
1355
1356                 if (switch_from_server_to_client(ctdb, "revokechild-%s", ctdb_db->db_name) != 0) {
1357                         DEBUG(DEBUG_ERR,("Failed to switch from server to client for revokechild process\n"));
1358                         c = 1;
1359                         goto child_finished;
1360                 }
1361
1362                 c = ctdb_revoke_all_delegations(ctdb, ctdb_db, tdata, key, header, data);
1363
1364 child_finished:
1365                 write(rc->fd[1], &c, 1);
1366                 /* make sure we die when our parent dies */
1367                 while (kill(parent, 0) == 0 || errno != ESRCH) {
1368                         sleep(5);
1369                 }
1370                 _exit(0);
1371         }
1372
1373         close(rc->fd[1]);
1374         rc->fd[1] = -1;
1375         set_close_on_exec(rc->fd[0]);
1376
1377         /* This is an active revokechild child process */
1378         DLIST_ADD_END(ctdb_db->revokechild_active, rc, NULL);
1379
1380         rc->fde = event_add_fd(ctdb->ev, rc, rc->fd[0],
1381                                    EVENT_FD_READ, revokechild_handler,
1382                                    (void *)rc);
1383         if (rc->fde == NULL) {
1384                 DEBUG(DEBUG_ERR,("Failed to set up fd event for revokechild process\n"));
1385                 talloc_free(rc);
1386         }
1387         tevent_fd_set_auto_close(rc->fde);
1388
1389         return 0;
1390 }
1391
1392 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)
1393 {
1394         struct revokechild_handle *rc;
1395         struct revokechild_deferred_call *deferred_call;
1396
1397         for (rc = ctdb_db->revokechild_active; rc; rc = rc->next) {
1398                 if (rc->key.dsize == 0) {
1399                         continue;
1400                 }
1401                 if (rc->key.dsize != key.dsize) {
1402                         continue;
1403                 }
1404                 if (!memcmp(rc->key.dptr, key.dptr, key.dsize)) {
1405                         break;
1406                 }
1407         }
1408
1409         if (rc == NULL) {
1410                 DEBUG(DEBUG_ERR,("Failed to add deferred call to revoke list. revoke structure not found\n"));
1411                 return -1;
1412         }
1413
1414         deferred_call = talloc(rc, struct revokechild_deferred_call);
1415         if (deferred_call == NULL) {
1416                 DEBUG(DEBUG_ERR,("Failed to allocate deferred call structure for revoking record\n"));
1417                 return -1;
1418         }
1419
1420         deferred_call->ctdb = ctdb;
1421         deferred_call->hdr  = hdr;
1422         deferred_call->fn   = fn;
1423         deferred_call->ctx  = call_context;
1424
1425         talloc_set_destructor(deferred_call, deferred_call_destructor);
1426         talloc_steal(deferred_call, hdr);
1427
1428         return 0;
1429 }