2 ctdb_call protocol code
4 Copyright (C) Andrew Tridgell 2006
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.
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.
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/>.
20 see http://wiki.samba.org/index.php/Samba_%26_Clustering for
21 protocol design and packet details
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"
32 find the ctdb_db from a db index
34 struct ctdb_db_context *find_ctdb_db(struct ctdb_context *ctdb, uint32_t id)
36 struct ctdb_db_context *ctdb_db;
38 for (ctdb_db=ctdb->db_list; ctdb_db; ctdb_db=ctdb_db->next) {
39 if (ctdb_db->db_id == id) {
48 a varient of input packet that can be used in lock requeue
50 static void ctdb_call_input_pkt(void *p, struct ctdb_req_header *hdr)
52 struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
53 ctdb_input_pkt(ctdb, hdr);
60 static void ctdb_send_error(struct ctdb_context *ctdb,
61 struct ctdb_req_header *hdr, uint32_t status,
62 const char *fmt, ...) PRINTF_ATTRIBUTE(4,5);
63 static void ctdb_send_error(struct ctdb_context *ctdb,
64 struct ctdb_req_header *hdr, uint32_t status,
68 struct ctdb_reply_error *r;
72 if (ctdb->methods == NULL) {
73 DEBUG(DEBUG_INFO,(__location__ " Failed to send error. Transport is DOWN\n"));
78 msg = talloc_vasprintf(ctdb, fmt, ap);
80 ctdb_fatal(ctdb, "Unable to allocate error in ctdb_send_error\n");
84 msglen = strlen(msg)+1;
85 len = offsetof(struct ctdb_reply_error, msg);
86 r = ctdb_transport_allocate(ctdb, msg, CTDB_REPLY_ERROR, len + msglen,
87 struct ctdb_reply_error);
88 CTDB_NO_MEMORY_FATAL(ctdb, r);
90 r->hdr.destnode = hdr->srcnode;
91 r->hdr.reqid = hdr->reqid;
94 memcpy(&r->msg[0], msg, msglen);
96 ctdb_queue_packet(ctdb, &r->hdr);
103 * send a redirect reply
105 * The logic behind this function is this:
107 * A client wants to grab a record and sends a CTDB_REQ_CALL packet
108 * to its local ctdb (ctdb_request_call). If the node is not itself
109 * the record's DMASTER, it first redirects the packet to the
110 * record's LMASTER. The LMASTER then redirects the call packet to
111 * the current DMASTER. But there is a race: The record may have
112 * been migrated off the DMASTER while the redirected packet is
113 * on the wire (or in the local queue). So in case the record has
114 * migrated off the new destinaton of the call packet, instead of
115 * going back to the LMASTER to get the new DMASTER, we try to
116 * reduce rountrips by fist chasing the record a couple of times
117 * before giving up the direct chase and finally going back to the
118 * LMASTER (again). Note that this works because auf this: When
119 * a record is migrated off a node, then the new DMASTER is stored
120 * in the record's copy on the former DMASTER.
122 * The maxiumum number of attempts for direct chase to make before
123 * going back to the LMASTER is configurable by the tunable
124 * "MaxRedirectCount".
126 static void ctdb_call_send_redirect(struct ctdb_context *ctdb,
128 struct ctdb_req_call *c,
129 struct ctdb_ltdb_header *header)
132 uint32_t lmaster = ctdb_lmaster(ctdb, &key);
133 if (ctdb->pnn == lmaster) {
134 c->hdr.destnode = header->dmaster;
135 } else if ((c->hopcount % ctdb->tunable.max_redirect_count) == 0) {
136 c->hdr.destnode = lmaster;
138 c->hdr.destnode = header->dmaster;
141 ctdb_queue_packet(ctdb, &c->hdr);
148 caller must have the chainlock before calling this routine. Caller must be
151 static void ctdb_send_dmaster_reply(struct ctdb_db_context *ctdb_db,
152 struct ctdb_ltdb_header *header,
153 TDB_DATA key, TDB_DATA data,
154 uint32_t new_dmaster,
157 struct ctdb_context *ctdb = ctdb_db->ctdb;
158 struct ctdb_reply_dmaster *r;
162 if (ctdb->pnn != ctdb_lmaster(ctdb, &key)) {
163 DEBUG(DEBUG_ALERT,(__location__ " Caller is not lmaster!\n"));
167 header->dmaster = new_dmaster;
168 ret = ctdb_ltdb_store(ctdb_db, key, header, data);
170 ctdb_fatal(ctdb, "ctdb_send_dmaster_reply unable to update dmaster");
174 if (ctdb->methods == NULL) {
175 ctdb_fatal(ctdb, "ctdb_send_dmaster_reply cant update dmaster since transport is down");
179 /* put the packet on a temporary context, allowing us to safely free
180 it below even if ctdb_reply_dmaster() has freed it already */
181 tmp_ctx = talloc_new(ctdb);
183 /* send the CTDB_REPLY_DMASTER */
184 len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize + sizeof(uint32_t);
185 r = ctdb_transport_allocate(ctdb, tmp_ctx, CTDB_REPLY_DMASTER, len,
186 struct ctdb_reply_dmaster);
187 CTDB_NO_MEMORY_FATAL(ctdb, r);
189 r->hdr.destnode = new_dmaster;
190 r->hdr.reqid = reqid;
191 r->rsn = header->rsn;
192 r->keylen = key.dsize;
193 r->datalen = data.dsize;
194 r->db_id = ctdb_db->db_id;
195 memcpy(&r->data[0], key.dptr, key.dsize);
196 memcpy(&r->data[key.dsize], data.dptr, data.dsize);
197 memcpy(&r->data[key.dsize+data.dsize], &header->flags, sizeof(uint32_t));
199 ctdb_queue_packet(ctdb, &r->hdr);
201 talloc_free(tmp_ctx);
205 send a dmaster request (give another node the dmaster for a record)
207 This is always sent to the lmaster, which ensures that the lmaster
208 always knows who the dmaster is. The lmaster will then send a
209 CTDB_REPLY_DMASTER to the new dmaster
211 static void ctdb_call_send_dmaster(struct ctdb_db_context *ctdb_db,
212 struct ctdb_req_call *c,
213 struct ctdb_ltdb_header *header,
214 TDB_DATA *key, TDB_DATA *data)
216 struct ctdb_req_dmaster *r;
217 struct ctdb_context *ctdb = ctdb_db->ctdb;
219 uint32_t lmaster = ctdb_lmaster(ctdb, key);
221 if (ctdb->methods == NULL) {
222 ctdb_fatal(ctdb, "Failed ctdb_call_send_dmaster since transport is down");
226 if (data->dsize != 0) {
227 header->flags |= CTDB_REC_FLAG_MIGRATED_WITH_DATA;
230 if (lmaster == ctdb->pnn) {
231 ctdb_send_dmaster_reply(ctdb_db, header, *key, *data,
232 c->hdr.srcnode, c->hdr.reqid);
236 len = offsetof(struct ctdb_req_dmaster, data) + key->dsize + data->dsize
238 r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_DMASTER, len,
239 struct ctdb_req_dmaster);
240 CTDB_NO_MEMORY_FATAL(ctdb, r);
241 r->hdr.destnode = lmaster;
242 r->hdr.reqid = c->hdr.reqid;
244 r->rsn = header->rsn;
245 r->dmaster = c->hdr.srcnode;
246 r->keylen = key->dsize;
247 r->datalen = data->dsize;
248 memcpy(&r->data[0], key->dptr, key->dsize);
249 memcpy(&r->data[key->dsize], data->dptr, data->dsize);
250 memcpy(&r->data[key->dsize + data->dsize], &header->flags, sizeof(uint32_t));
252 header->dmaster = c->hdr.srcnode;
253 if (ctdb_ltdb_store(ctdb_db, *key, header, *data) != 0) {
254 ctdb_fatal(ctdb, "Failed to store record in ctdb_call_send_dmaster");
257 ctdb_queue_packet(ctdb, &r->hdr);
263 called when a CTDB_REPLY_DMASTER packet comes in, or when the lmaster
264 gets a CTDB_REQUEST_DMASTER for itself. We become the dmaster.
266 must be called with the chainlock held. This function releases the chainlock
268 static void ctdb_become_dmaster(struct ctdb_db_context *ctdb_db,
269 struct ctdb_req_header *hdr,
270 TDB_DATA key, TDB_DATA data,
271 uint64_t rsn, uint32_t record_flags)
273 struct ctdb_call_state *state;
274 struct ctdb_context *ctdb = ctdb_db->ctdb;
275 struct ctdb_ltdb_header header;
278 DEBUG(DEBUG_DEBUG,("pnn %u dmaster response %08x\n", ctdb->pnn, ctdb_hash(&key)));
281 header.rsn = rsn + 1;
282 header.dmaster = ctdb->pnn;
283 header.flags = record_flags;
285 state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
288 if (state->call->flags & CTDB_CALL_FLAG_VACUUM_MIGRATION) {
290 * We temporarily add the VACUUM_MIGRATED flag to
291 * the record flags, so that ctdb_ltdb_store can
292 * decide whether the record should be stored or
295 header.flags |= CTDB_REC_FLAG_VACUUM_MIGRATED;
299 if (ctdb_ltdb_store(ctdb_db, key, &header, data) != 0) {
300 ctdb_fatal(ctdb, "ctdb_reply_dmaster store failed\n");
302 ret = ctdb_ltdb_unlock(ctdb_db, key);
304 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
311 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_become_dmaster from node %u\n",
312 ctdb->pnn, hdr->reqid, hdr->srcnode));
314 ret = ctdb_ltdb_unlock(ctdb_db, key);
316 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
321 if (key.dsize != state->call->key.dsize || memcmp(key.dptr, state->call->key.dptr, key.dsize)) {
322 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));
324 ret = ctdb_ltdb_unlock(ctdb_db, key);
326 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
331 if (hdr->reqid != state->reqid) {
332 /* we found a record but it was the wrong one */
333 DEBUG(DEBUG_ERR, ("Dropped orphan in ctdb_become_dmaster with reqid:%u\n from node %u", hdr->reqid, hdr->srcnode));
335 ret = ctdb_ltdb_unlock(ctdb_db, key);
337 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
342 ctdb_call_local(ctdb_db, state->call, &header, state, &data, true);
344 ret = ctdb_ltdb_unlock(ctdb_db, state->call->key);
346 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
349 state->state = CTDB_CALL_DONE;
350 if (state->async.fn) {
351 state->async.fn(state);
358 called when a CTDB_REQ_DMASTER packet comes in
360 this comes into the lmaster for a record when the current dmaster
361 wants to give up the dmaster role and give it to someone else
363 void ctdb_request_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
365 struct ctdb_req_dmaster *c = (struct ctdb_req_dmaster *)hdr;
366 TDB_DATA key, data, data2;
367 struct ctdb_ltdb_header header;
368 struct ctdb_db_context *ctdb_db;
369 uint32_t record_flags = 0;
374 key.dsize = c->keylen;
375 data.dptr = c->data + c->keylen;
376 data.dsize = c->datalen;
377 len = offsetof(struct ctdb_req_dmaster, data) + key.dsize + data.dsize
379 if (len <= c->hdr.length) {
380 record_flags = *(uint32_t *)&c->data[c->keylen + c->datalen];
383 ctdb_db = find_ctdb_db(ctdb, c->db_id);
385 ctdb_send_error(ctdb, hdr, -1,
386 "Unknown database in request. db_id==0x%08x",
391 /* fetch the current record */
392 ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, key, &header, hdr, &data2,
393 ctdb_call_input_pkt, ctdb, False);
395 ctdb_fatal(ctdb, "ctdb_req_dmaster failed to fetch record");
399 DEBUG(DEBUG_INFO,(__location__ " deferring ctdb_request_dmaster\n"));
403 if (ctdb_lmaster(ctdb, &key) != ctdb->pnn) {
404 DEBUG(DEBUG_ALERT,("pnn %u dmaster request to non-lmaster lmaster=%u gen=%u curgen=%u\n",
405 ctdb->pnn, ctdb_lmaster(ctdb, &key),
406 hdr->generation, ctdb->vnn_map->generation));
407 ctdb_fatal(ctdb, "ctdb_req_dmaster to non-lmaster");
410 DEBUG(DEBUG_DEBUG,("pnn %u dmaster request on %08x for %u from %u\n",
411 ctdb->pnn, ctdb_hash(&key), c->dmaster, c->hdr.srcnode));
413 /* its a protocol error if the sending node is not the current dmaster */
414 if (header.dmaster != hdr->srcnode) {
415 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",
416 ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
417 ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation,
418 (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid,
419 (key.dsize >= 4)?(*(uint32_t *)key.dptr):0));
420 if (header.rsn != 0 || header.dmaster != ctdb->pnn) {
421 DEBUG(DEBUG_ERR,("ctdb_req_dmaster from non-master. Force a recovery.\n"));
423 ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
424 ctdb_ltdb_unlock(ctdb_db, key);
429 if (header.rsn > c->rsn) {
430 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",
431 ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
432 ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation,
433 (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid));
436 /* use the rsn from the sending node */
439 /* store the record flags from the sending node */
440 header.flags = record_flags;
442 /* check if the new dmaster is the lmaster, in which case we
443 skip the dmaster reply */
444 if (c->dmaster == ctdb->pnn) {
445 ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn, record_flags);
447 ctdb_send_dmaster_reply(ctdb_db, &header, key, data, c->dmaster, hdr->reqid);
449 ret = ctdb_ltdb_unlock(ctdb_db, key);
451 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
458 called when a CTDB_REQ_CALL packet comes in
460 void ctdb_request_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
462 struct ctdb_req_call *c = (struct ctdb_req_call *)hdr;
464 struct ctdb_reply_call *r;
466 struct ctdb_ltdb_header header;
467 struct ctdb_call *call;
468 struct ctdb_db_context *ctdb_db;
470 if (ctdb->methods == NULL) {
471 DEBUG(DEBUG_INFO,(__location__ " Failed ctdb_request_call. Transport is DOWN\n"));
476 ctdb_db = find_ctdb_db(ctdb, c->db_id);
478 ctdb_send_error(ctdb, hdr, -1,
479 "Unknown database in request. db_id==0x%08x",
484 call = talloc(hdr, struct ctdb_call);
485 CTDB_NO_MEMORY_FATAL(ctdb, call);
487 call->call_id = c->callid;
488 call->key.dptr = c->data;
489 call->key.dsize = c->keylen;
490 call->call_data.dptr = c->data + c->keylen;
491 call->call_data.dsize = c->calldatalen;
493 /* determine if we are the dmaster for this key. This also
494 fetches the record data (if any), thus avoiding a 2nd fetch of the data
495 if the call will be answered locally */
497 ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, call->key, &header, hdr, &data,
498 ctdb_call_input_pkt, ctdb, False);
500 ctdb_send_error(ctdb, hdr, ret, "ltdb fetch failed in ctdb_request_call");
504 DEBUG(DEBUG_INFO,(__location__ " deferred ctdb_request_call\n"));
508 /* if we are not the dmaster, then send a redirect to the
510 if (header.dmaster != ctdb->pnn) {
511 talloc_free(data.dptr);
512 ctdb_call_send_redirect(ctdb, call->key, c, &header);
514 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
516 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
521 CTDB_UPDATE_STAT(ctdb, max_hop_count, c->hopcount);
523 /* Try if possible to migrate the record off to the caller node.
524 * From the clients perspective a fetch of the data is just as
525 * expensive as a migration.
527 if (c->hdr.srcnode != ctdb->pnn) {
528 if (ctdb_db->transaction_active) {
529 DEBUG(DEBUG_INFO, (__location__ " refusing migration"
530 " of key %s while transaction is active\n",
531 (char *)call->key.dptr));
533 DEBUG(DEBUG_DEBUG,("pnn %u starting migration of %08x to %u\n",
534 ctdb->pnn, ctdb_hash(&(call->key)), c->hdr.srcnode));
535 ctdb_call_send_dmaster(ctdb_db, c, &header, &(call->key), &data);
536 talloc_free(data.dptr);
538 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
540 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
546 ret = ctdb_call_local(ctdb_db, call, &header, hdr, &data, true);
548 DEBUG(DEBUG_ERR,(__location__ " ctdb_call_local failed\n"));
552 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
554 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
557 len = offsetof(struct ctdb_reply_call, data) + call->reply_data.dsize;
558 r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len,
559 struct ctdb_reply_call);
560 CTDB_NO_MEMORY_FATAL(ctdb, r);
561 r->hdr.destnode = hdr->srcnode;
562 r->hdr.reqid = hdr->reqid;
563 r->status = call->status;
564 r->datalen = call->reply_data.dsize;
565 if (call->reply_data.dsize) {
566 memcpy(&r->data[0], call->reply_data.dptr, call->reply_data.dsize);
569 ctdb_queue_packet(ctdb, &r->hdr);
575 called when a CTDB_REPLY_CALL packet comes in
577 This packet comes in response to a CTDB_REQ_CALL request packet. It
578 contains any reply data from the call
580 void ctdb_reply_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
582 struct ctdb_reply_call *c = (struct ctdb_reply_call *)hdr;
583 struct ctdb_call_state *state;
585 state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
587 DEBUG(DEBUG_ERR, (__location__ " reqid %u not found\n", hdr->reqid));
591 if (hdr->reqid != state->reqid) {
592 /* we found a record but it was the wrong one */
593 DEBUG(DEBUG_ERR, ("Dropped orphaned call reply with reqid:%u\n",hdr->reqid));
597 state->call->reply_data.dptr = c->data;
598 state->call->reply_data.dsize = c->datalen;
599 state->call->status = c->status;
601 talloc_steal(state, c);
603 state->state = CTDB_CALL_DONE;
604 if (state->async.fn) {
605 state->async.fn(state);
611 called when a CTDB_REPLY_DMASTER packet comes in
613 This packet comes in from the lmaster response to a CTDB_REQ_CALL
614 request packet. It means that the current dmaster wants to give us
617 void ctdb_reply_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
619 struct ctdb_reply_dmaster *c = (struct ctdb_reply_dmaster *)hdr;
620 struct ctdb_db_context *ctdb_db;
622 uint32_t record_flags = 0;
626 ctdb_db = find_ctdb_db(ctdb, c->db_id);
627 if (ctdb_db == NULL) {
628 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_reply_dmaster\n", c->db_id));
633 key.dsize = c->keylen;
634 data.dptr = &c->data[key.dsize];
635 data.dsize = c->datalen;
636 len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize
638 if (len <= c->hdr.length) {
639 record_flags = *(uint32_t *)&c->data[c->keylen + c->datalen];
642 ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
643 ctdb_call_input_pkt, ctdb, False);
648 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_dmaster\n"));
652 ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn, record_flags);
657 called when a CTDB_REPLY_ERROR packet comes in
659 void ctdb_reply_error(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
661 struct ctdb_reply_error *c = (struct ctdb_reply_error *)hdr;
662 struct ctdb_call_state *state;
664 state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
666 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_reply_error\n",
667 ctdb->pnn, hdr->reqid));
671 if (hdr->reqid != state->reqid) {
672 /* we found a record but it was the wrong one */
673 DEBUG(DEBUG_ERR, ("Dropped orphaned error reply with reqid:%u\n",hdr->reqid));
677 talloc_steal(state, c);
679 state->state = CTDB_CALL_ERROR;
680 state->errmsg = (char *)c->msg;
681 if (state->async.fn) {
682 state->async.fn(state);
690 static int ctdb_call_destructor(struct ctdb_call_state *state)
692 DLIST_REMOVE(state->ctdb_db->ctdb->pending_calls, state);
693 ctdb_reqid_remove(state->ctdb_db->ctdb, state->reqid);
699 called when a ctdb_call needs to be resent after a reconfigure event
701 static void ctdb_call_resend(struct ctdb_call_state *state)
703 struct ctdb_context *ctdb = state->ctdb_db->ctdb;
705 state->generation = ctdb->vnn_map->generation;
707 /* use a new reqid, in case the old reply does eventually come in */
708 ctdb_reqid_remove(ctdb, state->reqid);
709 state->reqid = ctdb_reqid_new(ctdb, state);
710 state->c->hdr.reqid = state->reqid;
712 /* update the generation count for this request, so its valid with the new vnn_map */
713 state->c->hdr.generation = state->generation;
715 /* send the packet to ourselves, it will be redirected appropriately */
716 state->c->hdr.destnode = ctdb->pnn;
718 ctdb_queue_packet(ctdb, &state->c->hdr);
719 DEBUG(DEBUG_NOTICE,("resent ctdb_call\n"));
723 resend all pending calls on recovery
725 void ctdb_call_resend_all(struct ctdb_context *ctdb)
727 struct ctdb_call_state *state, *next;
728 for (state=ctdb->pending_calls;state;state=next) {
730 ctdb_call_resend(state);
735 this allows the caller to setup a async.fn
737 static void call_local_trigger(struct event_context *ev, struct timed_event *te,
738 struct timeval t, void *private_data)
740 struct ctdb_call_state *state = talloc_get_type(private_data, struct ctdb_call_state);
741 if (state->async.fn) {
742 state->async.fn(state);
748 construct an event driven local ctdb_call
750 this is used so that locally processed ctdb_call requests are processed
751 in an event driven manner
753 struct ctdb_call_state *ctdb_call_local_send(struct ctdb_db_context *ctdb_db,
754 struct ctdb_call *call,
755 struct ctdb_ltdb_header *header,
758 struct ctdb_call_state *state;
759 struct ctdb_context *ctdb = ctdb_db->ctdb;
762 state = talloc_zero(ctdb_db, struct ctdb_call_state);
763 CTDB_NO_MEMORY_NULL(ctdb, state);
765 talloc_steal(state, data->dptr);
767 state->state = CTDB_CALL_DONE;
768 state->call = talloc(state, struct ctdb_call);
769 CTDB_NO_MEMORY_NULL(ctdb, state->call);
770 *(state->call) = *call;
771 state->ctdb_db = ctdb_db;
773 ret = ctdb_call_local(ctdb_db, state->call, header, state, data, true);
775 event_add_timed(ctdb->ev, state, timeval_zero(), call_local_trigger, state);
782 make a remote ctdb call - async send. Called in daemon context.
784 This constructs a ctdb_call request and queues it for processing.
785 This call never blocks.
787 struct ctdb_call_state *ctdb_daemon_call_send_remote(struct ctdb_db_context *ctdb_db,
788 struct ctdb_call *call,
789 struct ctdb_ltdb_header *header)
792 struct ctdb_call_state *state;
793 struct ctdb_context *ctdb = ctdb_db->ctdb;
795 if (ctdb->methods == NULL) {
796 DEBUG(DEBUG_INFO,(__location__ " Failed send packet. Transport is down\n"));
800 state = talloc_zero(ctdb_db, struct ctdb_call_state);
801 CTDB_NO_MEMORY_NULL(ctdb, state);
802 state->call = talloc(state, struct ctdb_call);
803 CTDB_NO_MEMORY_NULL(ctdb, state->call);
805 state->reqid = ctdb_reqid_new(ctdb, state);
806 state->ctdb_db = ctdb_db;
807 talloc_set_destructor(state, ctdb_call_destructor);
809 len = offsetof(struct ctdb_req_call, data) + call->key.dsize + call->call_data.dsize;
810 state->c = ctdb_transport_allocate(ctdb, state, CTDB_REQ_CALL, len,
811 struct ctdb_req_call);
812 CTDB_NO_MEMORY_NULL(ctdb, state->c);
813 state->c->hdr.destnode = header->dmaster;
815 /* this limits us to 16k outstanding messages - not unreasonable */
816 state->c->hdr.reqid = state->reqid;
817 state->c->flags = call->flags;
818 state->c->db_id = ctdb_db->db_id;
819 state->c->callid = call->call_id;
820 state->c->hopcount = 0;
821 state->c->keylen = call->key.dsize;
822 state->c->calldatalen = call->call_data.dsize;
823 memcpy(&state->c->data[0], call->key.dptr, call->key.dsize);
824 memcpy(&state->c->data[call->key.dsize],
825 call->call_data.dptr, call->call_data.dsize);
826 *(state->call) = *call;
827 state->call->call_data.dptr = &state->c->data[call->key.dsize];
828 state->call->key.dptr = &state->c->data[0];
830 state->state = CTDB_CALL_WAIT;
831 state->generation = ctdb->vnn_map->generation;
833 DLIST_ADD(ctdb->pending_calls, state);
835 ctdb_queue_packet(ctdb, &state->c->hdr);
841 make a remote ctdb call - async recv - called in daemon context
843 This is called when the program wants to wait for a ctdb_call to complete and get the
844 results. This call will block unless the call has already completed.
846 int ctdb_daemon_call_recv(struct ctdb_call_state *state, struct ctdb_call *call)
848 while (state->state < CTDB_CALL_DONE) {
849 event_loop_once(state->ctdb_db->ctdb->ev);
851 if (state->state != CTDB_CALL_DONE) {
852 ctdb_set_error(state->ctdb_db->ctdb, "%s", state->errmsg);
857 if (state->call->reply_data.dsize) {
858 call->reply_data.dptr = talloc_memdup(call,
859 state->call->reply_data.dptr,
860 state->call->reply_data.dsize);
861 call->reply_data.dsize = state->call->reply_data.dsize;
863 call->reply_data.dptr = NULL;
864 call->reply_data.dsize = 0;
866 call->status = state->call->status;
873 send a keepalive packet to the other node
875 void ctdb_send_keepalive(struct ctdb_context *ctdb, uint32_t destnode)
877 struct ctdb_req_keepalive *r;
879 if (ctdb->methods == NULL) {
880 DEBUG(DEBUG_INFO,(__location__ " Failed to send keepalive. Transport is DOWN\n"));
884 r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_KEEPALIVE,
885 sizeof(struct ctdb_req_keepalive),
886 struct ctdb_req_keepalive);
887 CTDB_NO_MEMORY_FATAL(ctdb, r);
888 r->hdr.destnode = destnode;
891 CTDB_INCREMENT_STAT(ctdb, keepalive_packets_sent);
893 ctdb_queue_packet(ctdb, &r->hdr);
900 struct revokechild_deferred_call {
901 struct ctdb_context *ctdb;
902 struct ctdb_req_header *hdr;
903 deferred_requeue_fn fn;
907 struct revokechild_handle {
908 struct revokechild_handle *next, *prev;
909 struct ctdb_context *ctdb;
910 struct ctdb_db_context *ctdb_db;
911 struct fd_event *fde;
918 struct revokechild_requeue_handle {
919 struct ctdb_context *ctdb;
920 struct ctdb_req_header *hdr;
921 deferred_requeue_fn fn;
925 static void deferred_call_requeue(struct event_context *ev, struct timed_event *te,
926 struct timeval t, void *private_data)
928 struct revokechild_requeue_handle *requeue_handle = talloc_get_type(private_data, struct revokechild_requeue_handle);
930 requeue_handle->fn(requeue_handle->ctx, requeue_handle->hdr);
931 talloc_free(requeue_handle);
934 static int deferred_call_destructor(struct revokechild_deferred_call *deferred_call)
936 struct ctdb_context *ctdb = deferred_call->ctdb;
937 struct revokechild_requeue_handle *requeue_handle = talloc(ctdb, struct revokechild_requeue_handle);
938 struct ctdb_req_call *c = (struct ctdb_req_call *)deferred_call->hdr;
940 requeue_handle->ctdb = ctdb;
941 requeue_handle->hdr = deferred_call->hdr;
942 requeue_handle->fn = deferred_call->fn;
943 requeue_handle->ctx = deferred_call->ctx;
944 talloc_steal(requeue_handle, requeue_handle->hdr);
946 /* when revoking, any READONLY requests have 1 second grace to let read/write finish first */
947 event_add_timed(ctdb->ev, requeue_handle, timeval_current_ofs(c->flags & CTDB_WANT_READONLY ? 1 : 0, 0), deferred_call_requeue, requeue_handle);
953 static int revokechild_destructor(struct revokechild_handle *rc)
955 if (rc->fde != NULL) {
956 talloc_free(rc->fde);
959 if (rc->fd[0] != -1) {
962 if (rc->fd[1] != -1) {
965 kill(rc->child, SIGKILL);
967 DLIST_REMOVE(rc->ctdb_db->revokechild_active, rc);
971 static void revokechild_handler(struct event_context *ev, struct fd_event *fde,
972 uint16_t flags, void *private_data)
974 struct revokechild_handle *rc = talloc_get_type(private_data,
975 struct revokechild_handle);
979 ret = read(rc->fd[0], &c, 1);
981 DEBUG(DEBUG_ERR,("Failed to read status from revokechild. errno:%d\n", errno));
987 DEBUG(DEBUG_ERR,("revokechild returned failure. status:%d\n", c));
996 struct ctdb_revoke_state {
997 struct ctdb_db_context *ctdb_db;
999 struct ctdb_ltdb_header *header;
1006 static void update_record_cb(struct ctdb_client_control_state *state)
1008 struct ctdb_revoke_state *revoke_state;
1012 if (state == NULL) {
1015 revoke_state = state->async.private_data;
1017 state->async.fn = NULL;
1018 ret = ctdb_control_recv(state->ctdb, state, state, NULL, &res, NULL);
1019 if ((ret != 0) || (res != 0)) {
1020 DEBUG(DEBUG_ERR,("Recv for revoke update record failed ret:%d res:%d\n", ret, res));
1021 revoke_state->status = -1;
1024 revoke_state->count--;
1025 if (revoke_state->count <= 0) {
1026 revoke_state->finished = 1;
1030 static void revoke_send_cb(struct ctdb_context *ctdb, uint32_t pnn, void *private_data)
1032 struct ctdb_revoke_state *revoke_state = private_data;
1033 struct ctdb_client_control_state *state;
1035 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);
1036 if (state == NULL) {
1037 DEBUG(DEBUG_ERR,("Failure to send update record to revoke readonly delegation\n"));
1038 revoke_state->status = -1;
1041 state->async.fn = update_record_cb;
1042 state->async.private_data = revoke_state;
1044 revoke_state->count++;
1048 static void ctdb_revoke_timeout_handler(struct event_context *ev, struct timed_event *te,
1049 struct timeval yt, void *private_data)
1051 struct ctdb_revoke_state *state = private_data;
1053 DEBUG(DEBUG_ERR,("Timed out waiting for revoke to finish\n"));
1054 state->finished = 1;
1058 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)
1060 struct ctdb_revoke_state *state = talloc_zero(ctdb, struct ctdb_revoke_state);
1063 state->ctdb_db = ctdb_db;
1065 state->header = header;
1068 ctdb_trackingdb_traverse(ctdb, tdata, revoke_send_cb, state);
1070 event_add_timed(ctdb->ev, state, timeval_current_ofs(5, 0), ctdb_revoke_timeout_handler, state);
1072 while (state->finished == 0) {
1073 event_loop_once(ctdb->ev);
1076 status = state->status;
1079 struct ctdb_ltdb_header new_header;
1082 if (ctdb_ltdb_lock(ctdb_db, key) != 0) {
1083 DEBUG(DEBUG_ERR,("Failed to chainlock the database in revokechild\n"));
1087 if (ctdb_ltdb_fetch(ctdb_db, key, &new_header, state, &new_data) != 0) {
1088 ctdb_ltdb_unlock(ctdb_db, key);
1089 DEBUG(DEBUG_ERR,("Failed for fetch tdb record in revokechild\n"));
1094 if (new_header.rsn > header->rsn) {
1095 ctdb_ltdb_unlock(ctdb_db, key);
1096 DEBUG(DEBUG_ERR,("RSN too high in tdb record in revokechild\n"));
1100 if ( (new_header.flags & (CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS)) != (CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS) ) {
1101 ctdb_ltdb_unlock(ctdb_db, key);
1102 DEBUG(DEBUG_ERR,("Flags are wrong in tdb record in revokechild\n"));
1107 new_header.flags |= CTDB_REC_RO_REVOKE_COMPLETE;
1108 if (ctdb_ltdb_store(ctdb_db, key, &new_header, new_data) != 0) {
1109 ctdb_ltdb_unlock(ctdb_db, key);
1110 DEBUG(DEBUG_ERR,("Failed to write new record in revokechild\n"));
1114 ctdb_ltdb_unlock(ctdb_db, key);
1122 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)
1125 struct revokechild_handle *rc;
1126 pid_t parent = getpid();
1129 header->flags &= ~(CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY);
1132 if ((rc = talloc_zero(ctdb_db, struct revokechild_handle)) == NULL) {
1133 DEBUG(DEBUG_ERR,("Failed to allocate revokechild_handle\n"));
1137 tdata = tdb_fetch(ctdb_db->rottdb, key);
1138 if (tdata.dsize > 0) {
1142 tdata.dptr = talloc_memdup(rc, tdata.dptr, tdata.dsize);
1148 rc->ctdb_db = ctdb_db;
1152 talloc_set_destructor(rc, revokechild_destructor);
1154 rc->key.dsize = key.dsize;
1155 rc->key.dptr = talloc_memdup(rc, key.dptr, key.dsize);
1156 if (rc->key.dptr == NULL) {
1157 DEBUG(DEBUG_ERR,("Failed to allocate key for revokechild_handle\n"));
1164 DEBUG(DEBUG_ERR,("Failed to allocate key for revokechild_handle\n"));
1170 rc->child = ctdb_fork(ctdb);
1171 if (rc->child == (pid_t)-1) {
1172 DEBUG(DEBUG_ERR,("Failed to fork child for revokechild\n"));
1177 if (rc->child == 0) {
1180 debug_extra = talloc_asprintf(NULL, "revokechild-%s:", ctdb_db->db_name);
1182 if (switch_from_server_to_client(ctdb, "revokechild-%s", ctdb_db->db_name) != 0) {
1183 DEBUG(DEBUG_ERR,("Failed to switch from server to client for revokechild process\n"));
1185 goto child_finished;
1188 c = ctdb_revoke_all_delegations(ctdb, ctdb_db, tdata, key, header, data);
1191 write(rc->fd[1], &c, 1);
1192 /* make sure we die when our parent dies */
1193 while (kill(parent, 0) == 0 || errno != ESRCH) {
1201 set_close_on_exec(rc->fd[0]);
1203 /* This is an active revokechild child process */
1204 DLIST_ADD_END(ctdb_db->revokechild_active, rc, NULL);
1206 rc->fde = event_add_fd(ctdb->ev, rc, rc->fd[0],
1207 EVENT_FD_READ, revokechild_handler,
1209 if (rc->fde == NULL) {
1210 DEBUG(DEBUG_ERR,("Failed to set up fd event for revokechild process\n"));
1213 tevent_fd_set_auto_close(rc->fde);
1218 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)
1220 struct revokechild_handle *rc;
1221 struct revokechild_deferred_call *deferred_call;
1223 for (rc = ctdb_db->revokechild_active; rc; rc = rc->next) {
1224 if (rc->key.dsize == 0) {
1227 if (rc->key.dsize != key.dsize) {
1230 if (!memcmp(rc->key.dptr, key.dptr, key.dsize)) {
1236 DEBUG(DEBUG_ERR,("Failed to add deferred call to revoke list. revoke structure not found\n"));
1240 deferred_call = talloc(rc, struct revokechild_deferred_call);
1241 if (deferred_call == NULL) {
1242 DEBUG(DEBUG_ERR,("Failed to allocate deferred call structure for revoking record\n"));
1246 deferred_call->ctdb = ctdb;
1247 deferred_call->hdr = hdr;
1248 deferred_call->fn = fn;
1249 deferred_call->ctx = call_context;
1251 talloc_set_destructor(deferred_call, deferred_call_destructor);
1252 talloc_steal(deferred_call, hdr);