4 Copyright (C) Rusty Russell 2010
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/>.
25 #include <sys/socket.h>
27 #include "libctdb_private.h"
29 #include "local_tdb.h"
31 #include <dlinklist.h>
32 #include <ctdb_protocol.h>
34 /* Remove type-safety macros. */
35 #undef ctdb_attachdb_send
36 #undef ctdb_readrecordlock_async
40 struct ctdb_lock *next, *prev;
42 struct ctdb_db *ctdb_db;
45 /* This will always be true by the time user sees this. */
47 struct ctdb_ltdb_header *hdr;
49 /* For convenience, we stash original callback here. */
50 ctdb_rrl_callback_t callback;
53 static void remove_lock(struct ctdb_connection *ctdb, struct ctdb_lock *lock)
55 DLIST_REMOVE(ctdb->locks, lock);
58 /* FIXME: for thread safety, need tid info too. */
59 static bool holding_lock(struct ctdb_connection *ctdb)
61 /* For the moment, you can't ever hold more than 1 lock. */
62 return (ctdb->locks != NULL);
65 static void add_lock(struct ctdb_connection *ctdb, struct ctdb_lock *lock)
67 DLIST_ADD(ctdb->locks, lock);
70 /* FIXME: Could be in shared util code with rest of ctdb */
71 static void close_noerr(int fd)
78 /* FIXME: Could be in shared util code with rest of ctdb */
79 static void free_noerr(void *p)
86 /* FIXME: Could be in shared util code with rest of ctdb */
87 static void set_nonblocking(int fd)
90 v = fcntl(fd, F_GETFL, 0);
91 fcntl(fd, F_SETFL, v | O_NONBLOCK);
94 /* FIXME: Could be in shared util code with rest of ctdb */
95 static void set_close_on_exec(int fd)
98 v = fcntl(fd, F_GETFD, 0);
99 fcntl(fd, F_SETFD, v | FD_CLOEXEC);
102 static void set_pnn(struct ctdb_connection *ctdb,
103 struct ctdb_request *req,
106 if (!ctdb_getpnn_recv(ctdb, req, &ctdb->pnn)) {
107 DEBUG(ctdb, LOG_CRIT,
108 "ctdb_connect(async): failed to get pnn");
111 ctdb_request_free(ctdb, req);
114 struct ctdb_connection *ctdb_connect(const char *addr,
115 ctdb_log_fn_t log_fn, void *log_priv)
117 struct ctdb_connection *ctdb;
118 struct sockaddr_un sun;
120 ctdb = malloc(sizeof(*ctdb));
122 /* With no format string, we hope it doesn't use ap! */
124 memset(&ap, 0, sizeof(ap));
126 log_fn(log_priv, LOG_ERR, "ctdb_connect: no memory", ap);
132 ctdb->message_handlers = NULL;
134 ctdb->broken = false;
136 ctdb->log_priv = log_priv;
138 memset(&sun, 0, sizeof(sun));
139 sun.sun_family = AF_UNIX;
142 strncpy(sun.sun_path, addr, sizeof(sun.sun_path));
143 ctdb->fd = socket(AF_UNIX, SOCK_STREAM, 0);
147 set_nonblocking(ctdb->fd);
148 set_close_on_exec(ctdb->fd);
150 if (connect(ctdb->fd, (struct sockaddr *)&sun, sizeof(sun)) == -1)
153 /* Immediately queue a request to get our pnn. */
154 if (!ctdb_getpnn_send(ctdb, CTDB_CURRENT_NODE, set_pnn, NULL))
160 close_noerr(ctdb->fd);
167 int ctdb_get_fd(struct ctdb_connection *ctdb)
172 int ctdb_which_events(struct ctdb_connection *ctdb)
181 struct ctdb_request *new_ctdb_request(size_t len,
182 ctdb_callback_t cb, void *cbdata)
184 struct ctdb_request *req = malloc(sizeof(*req));
187 req->io = new_io_elem(len);
192 req->hdr.hdr = io_elem_data(req->io, NULL);
195 req->priv_data = cbdata;
197 req->extra_destructor = NULL;
201 void ctdb_request_free(struct ctdb_connection *ctdb, struct ctdb_request *req)
203 if (req->extra_destructor) {
204 req->extra_destructor(ctdb, req);
207 free_io_elem(req->reply);
209 free_io_elem(req->io);
213 /* Sanity-checking wrapper for reply. */
214 static struct ctdb_reply_call *unpack_reply_call(struct ctdb_connection *ctdb,
215 struct ctdb_request *req,
219 struct ctdb_reply_call *inhdr = io_elem_data(req->reply, &len);
221 /* Library user error if this isn't a reply to a call. */
222 if (req->hdr.hdr->operation != CTDB_REQ_CALL) {
225 "This was not a ctdbd call request: operation %u",
226 req->hdr.hdr->operation);
230 if (req->hdr.call->callid != callid) {
233 "This was not a ctdbd %u call request: %u",
234 callid, req->hdr.call->callid);
238 /* ctdbd or our error if this isn't a reply call. */
239 if (len < sizeof(*inhdr) || inhdr->hdr.operation != CTDB_REPLY_CALL) {
241 DEBUG(ctdb, LOG_CRIT,
242 "Invalid ctdbd call reply: len %zu, operation %u",
243 len, inhdr->hdr.operation);
250 /* Sanity-checking wrapper for reply. */
251 struct ctdb_reply_control *unpack_reply_control(struct ctdb_connection *ctdb,
252 struct ctdb_request *req,
253 enum ctdb_controls control)
256 struct ctdb_reply_control *inhdr = io_elem_data(req->reply, &len);
258 /* Library user error if this isn't a reply to a call. */
259 if (len < sizeof(*inhdr)) {
261 DEBUG(ctdb, LOG_CRIT,
262 "Short ctdbd control reply: %zu bytes", len);
265 if (req->hdr.hdr->operation != CTDB_REQ_CONTROL) {
268 "This was not a ctdbd control request: operation %u",
269 req->hdr.hdr->operation);
273 /* ... or if it was a different control from what we expected. */
274 if (req->hdr.control->opcode != control) {
277 "This was not an opcode %u ctdbd control request: %u",
278 control, req->hdr.control->opcode);
282 /* ctdbd or our error if this isn't a reply call. */
283 if (inhdr->hdr.operation != CTDB_REPLY_CONTROL) {
285 DEBUG(ctdb, LOG_CRIT,
286 "Invalid ctdbd control reply: operation %u",
287 inhdr->hdr.operation);
294 static void handle_incoming(struct ctdb_connection *ctdb, struct io_elem *in)
296 struct ctdb_req_header *hdr;
298 struct ctdb_request *i;
300 hdr = io_elem_data(in, &len);
301 /* FIXME: use len to check packet! */
303 if (hdr->operation == CTDB_REQ_MESSAGE) {
304 deliver_message(ctdb, hdr);
308 for (i = ctdb->doneq; i; i = i->next) {
309 if (i->hdr.hdr->reqid == hdr->reqid) {
310 DLIST_REMOVE(ctdb->doneq, i);
312 i->callback(ctdb, i, i->priv_data);
316 DEBUG(ctdb, LOG_WARNING,
317 "Unexpected ctdbd request reply: operation %u reqid %u",
318 hdr->operation, hdr->reqid);
322 /* Remove "harmless" errors. */
323 static ssize_t real_error(ssize_t ret)
325 if (ret < 0 && (errno == EINTR || errno == EWOULDBLOCK))
330 bool ctdb_service(struct ctdb_connection *ctdb, int revents)
336 if (holding_lock(ctdb)) {
337 DEBUG(ctdb, LOG_WARNING, "Do not block while holding lock!");
340 if (revents & POLLOUT) {
342 if (real_error(write_io_elem(ctdb->fd,
343 ctdb->outq->io)) < 0) {
345 "ctdb_service: error writing to ctdbd");
349 if (io_elem_finished(ctdb->outq->io)) {
350 struct ctdb_request *done = ctdb->outq;
351 DLIST_REMOVE(ctdb->outq, done);
352 /* We add at the head: any dead ones
354 DLIST_ADD(ctdb->doneq, done);
359 while (revents & POLLIN) {
363 ctdb->in = new_io_elem(sizeof(struct ctdb_req_header));
366 "ctdb_service: allocating readbuf");
372 ret = read_io_elem(ctdb->fd, ctdb->in);
373 if (real_error(ret) < 0 || ret == 0) {
374 /* They closed fd? */
378 "ctdb_service: error reading from ctdbd");
381 } else if (ret < 0) {
382 /* No progress, stop loop. */
384 } else if (io_elem_finished(ctdb->in)) {
385 handle_incoming(ctdb, ctdb->in);
393 /* This is inefficient. We could pull in idtree.c. */
394 static bool reqid_used(const struct ctdb_connection *ctdb, uint32_t reqid)
396 struct ctdb_request *i;
398 for (i = ctdb->outq; i; i = i->next) {
399 if (i->hdr.hdr->reqid == reqid) {
403 for (i = ctdb->doneq; i; i = i->next) {
404 if (i->hdr.hdr->reqid == reqid) {
411 uint32_t new_reqid(struct ctdb_connection *ctdb)
413 while (reqid_used(ctdb, ctdb->next_id)) {
416 return ctdb->next_id++;
419 struct ctdb_request *new_ctdb_control_request(struct ctdb_connection *ctdb,
422 const void *extra_data,
424 ctdb_callback_t callback,
427 struct ctdb_request *req;
428 struct ctdb_req_control *pkt;
430 req = new_ctdb_request(offsetof(struct ctdb_req_control, data) + extra, callback, cbdata);
434 io_elem_init_req_header(req->io,
435 CTDB_REQ_CONTROL, destnode, new_reqid(ctdb));
437 pkt = req->hdr.control;
439 pkt->opcode = opcode;
443 pkt->datalen = extra;
444 memcpy(pkt->data, extra_data, extra);
445 DLIST_ADD(ctdb->outq, req);
449 void ctdb_cancel_callback(struct ctdb_connection *ctdb,
450 struct ctdb_request *req,
453 ctdb_request_free(ctdb, req);
456 void ctdb_cancel(struct ctdb_connection *ctdb, struct ctdb_request *req)
458 DEBUG(ctdb, LOG_DEBUG, "ctdb_cancel: %p (id %u)",
459 req, req->hdr.hdr ? req->hdr.hdr->reqid : 0);
461 /* FIXME: If it's not sent, we could just free it right now. */
462 req->callback = ctdb_cancel_callback;
466 struct ctdb_connection *ctdb;
470 struct tdb_context *tdb;
472 /* The lock we are holding, if any (we can only have one!) */
473 struct ctdb_lock *lock;
475 ctdb_callback_t callback;
479 static void attachdb_getdbpath_done(struct ctdb_connection *ctdb,
480 struct ctdb_request *req,
483 struct ctdb_db *db = _db;
485 /* Do callback on original request. */
486 db->callback(ctdb, req->extra, db->private_data);
489 struct ctdb_db *ctdb_attachdb_recv(struct ctdb_connection *ctdb,
490 struct ctdb_request *req)
492 struct ctdb_request *dbpath_req = req->extra;
493 struct ctdb_reply_control *reply;
494 struct ctdb_db *db = req->priv_data;
495 uint32_t tdb_flags = db->tdb_flags;
497 /* Never sent the dbpath request? We've failed. */
499 /* FIXME: Save errno? */
504 reply = unpack_reply_control(ctdb, dbpath_req, CTDB_CONTROL_GETDBPATH);
508 if (reply->status != 0) {
509 DEBUG(db->ctdb, LOG_ERR,
510 "ctdb_attachdb_recv: reply status %i", reply->status);
514 tdb_flags = db->persistent ? TDB_DEFAULT : TDB_NOSYNC;
515 tdb_flags |= TDB_DISALLOW_NESTING;
517 /* FIXME: Setup logging to go through our logging. */
518 db->tdb = tdb_open((char *)reply->data, 0, tdb_flags, O_RDWR, 0);
519 if (db->tdb == NULL) {
520 DEBUG(db->ctdb, LOG_ERR,
521 "ctdb_attachdb_recv: failed to tdb_open %s",
522 (char *)reply->data);
526 /* Finally, separate the db from the request (see destroy_req_db). */
527 req->priv_data = NULL;
528 DEBUG(db->ctdb, LOG_DEBUG,
529 "ctdb_attachdb_recv: db %p, tdb %s", db, (char *)reply->data);
533 static void attachdb_done(struct ctdb_connection *ctdb,
534 struct ctdb_request *req,
537 struct ctdb_db *db = _db;
538 struct ctdb_request *req2;
539 struct ctdb_reply_control *reply;
540 enum ctdb_controls control = CTDB_CONTROL_DB_ATTACH;
542 if (db->persistent) {
543 control = CTDB_CONTROL_DB_ATTACH_PERSISTENT;
546 reply = unpack_reply_control(ctdb, req, control);
547 if (!reply || reply->status != 0) {
550 "ctdb_attachdb_send(async): DB_ATTACH status %i",
553 /* We failed. Hand request to user and have them discover it
554 * via ctdb_attachdb_recv. */
555 db->callback(ctdb, req, db->private_data);
558 db->id = *(uint32_t *)reply->data;
560 /* Now we do another call, to get the dbpath. */
561 req2 = new_ctdb_control_request(db->ctdb, CTDB_CONTROL_GETDBPATH,
563 &db->id, sizeof(db->id),
564 attachdb_getdbpath_done, db);
566 DEBUG(db->ctdb, LOG_ERR,
567 "ctdb_attachdb_send(async): failed to allocate");
568 db->callback(ctdb, req, db->private_data);
573 DEBUG(db->ctdb, LOG_DEBUG,
574 "ctdb_attachdb_send(async): created getdbpath request");
577 static void destroy_req_db(struct ctdb_connection *ctdb,
578 struct ctdb_request *req)
580 /* Incomplete db is in priv_data. */
581 free(req->priv_data);
582 /* second request is chained off this one. */
584 ctdb_request_free(ctdb, req->extra);
588 struct ctdb_request *
589 ctdb_attachdb_send(struct ctdb_connection *ctdb,
590 const char *name, int persistent, uint32_t tdb_flags,
591 ctdb_callback_t callback, void *private_data)
593 struct ctdb_request *req;
597 /* FIXME: Search if db already open. */
598 db = malloc(sizeof(*db));
604 opcode = CTDB_CONTROL_DB_ATTACH_PERSISTENT;
606 opcode = CTDB_CONTROL_DB_ATTACH;
609 req = new_ctdb_control_request(ctdb, opcode, CTDB_CURRENT_NODE, name,
610 strlen(name) + 1, attachdb_done, db);
612 DEBUG(db->ctdb, LOG_ERR,
613 "ctdb_attachdb_send: failed allocating DB_ATTACH");
619 db->tdb_flags = tdb_flags;
620 db->persistent = persistent;
621 db->callback = callback;
622 db->private_data = private_data;
624 req->extra_destructor = destroy_req_db;
625 /* This is set non-NULL when we succeed, see ctdb_attachdb_recv */
628 /* Flags get overloaded into srvid. */
629 req->hdr.control->srvid = tdb_flags;
630 DEBUG(db->ctdb, LOG_DEBUG,
631 "ctdb_attachdb_send: DB_ATTACH request %p", req);
635 void ctdb_release_lock(struct ctdb_lock *lock)
638 tdb_chainunlock(lock->ctdb_db->tdb, lock->key);
639 DEBUG(lock->ctdb_db->ctdb, LOG_DEBUG,
640 "ctdb_release_lock %p", lock);
642 remove_lock(lock->ctdb_db->ctdb, lock);
648 /* We keep the lock if local node is the dmaster. */
649 static bool try_readrecordlock(struct ctdb_lock *lock, TDB_DATA *data)
651 struct ctdb_ltdb_header *hdr;
653 if (tdb_chainlock(lock->ctdb_db->tdb, lock->key) != 0) {
654 DEBUG(lock->ctdb_db->ctdb, LOG_WARNING,
655 "ctdb_readrecordlock_async: failed to chainlock");
659 hdr = ctdb_local_fetch(lock->ctdb_db->tdb, lock->key, data);
660 if (hdr && hdr->dmaster == lock->ctdb_db->ctdb->pnn) {
661 DEBUG(lock->ctdb_db->ctdb, LOG_DEBUG,
662 "ctdb_readrecordlock_async: got local lock");
665 add_lock(lock->ctdb_db->ctdb, lock);
669 tdb_chainunlock(lock->ctdb_db->tdb, lock->key);
674 /* If they shutdown before we hand them the lock, we free it here. */
675 static void destroy_lock(struct ctdb_connection *ctdb,
676 struct ctdb_request *req)
678 ctdb_release_lock(req->extra);
681 static void readrecordlock_retry(struct ctdb_connection *ctdb,
682 struct ctdb_request *req, void *private)
684 struct ctdb_lock *lock = req->extra;
685 struct ctdb_reply_call *reply;
688 /* OK, we've received reply to noop migration */
689 reply = unpack_reply_call(ctdb, req, CTDB_NULL_FUNC);
690 if (!reply || reply->status != 0) {
693 "ctdb_readrecordlock_async(async):"
694 " NULL_FUNC returned %i", reply->status);
696 lock->callback(lock->ctdb_db, NULL, tdb_null, private);
697 ctdb_request_free(ctdb, req); /* Also frees lock. */
701 /* Can we get lock now? */
702 if (try_readrecordlock(lock, &data)) {
703 /* Now it's their responsibility to free lock & request! */
704 req->extra_destructor = NULL;
705 lock->callback(lock->ctdb_db, lock, data, private);
709 /* Retransmit the same request again (we lost race). */
710 io_elem_reset(req->io);
711 DLIST_ADD(ctdb->outq, req);
715 ctdb_readrecordlock_async(struct ctdb_db *ctdb_db, TDB_DATA key,
716 ctdb_rrl_callback_t callback, void *cbdata)
718 struct ctdb_request *req;
719 struct ctdb_lock *lock;
722 if (holding_lock(ctdb_db->ctdb)) {
723 DEBUG(ctdb_db->ctdb, LOG_ERR,
724 "ctdb_readrecordlock_async: already holding lock");
729 lock = malloc(sizeof(*lock) + key.dsize);
731 DEBUG(ctdb_db->ctdb, LOG_ERR,
732 "ctdb_readrecordlock_async: lock allocation failed");
735 lock->key.dptr = (void *)(lock + 1);
736 memcpy(lock->key.dptr, key.dptr, key.dsize);
737 lock->key.dsize = key.dsize;
738 lock->ctdb_db = ctdb_db;
743 if (try_readrecordlock(lock, &data)) {
744 callback(ctdb_db, lock, data, cbdata);
748 /* Slow path: create request. */
749 req = new_ctdb_request(offsetof(struct ctdb_req_call, data)
750 + key.dsize, readrecordlock_retry, cbdata);
752 DEBUG(ctdb_db->ctdb, LOG_ERR,
753 "ctdb_readrecordlock_async: allocation failed");
754 ctdb_release_lock(lock);
758 req->extra_destructor = destroy_lock;
759 /* We store the original callback in the lock, and use our own. */
760 lock->callback = callback;
762 io_elem_init_req_header(req->io, CTDB_REQ_CALL, CTDB_CURRENT_NODE,
763 new_reqid(ctdb_db->ctdb));
765 req->hdr.call->flags = CTDB_IMMEDIATE_MIGRATION;
766 req->hdr.call->db_id = ctdb_db->id;
767 req->hdr.call->callid = CTDB_NULL_FUNC;
768 req->hdr.call->hopcount = 0;
769 req->hdr.call->keylen = key.dsize;
770 req->hdr.call->calldatalen = 0;
771 memcpy(req->hdr.call->data, key.dptr, key.dsize);
772 DLIST_ADD(ctdb_db->ctdb->outq, req);
776 int ctdb_writerecord(struct ctdb_lock *lock, TDB_DATA data)
778 if (lock->ctdb_db->persistent) {
780 DEBUG(lock->ctdb_db->ctdb, LOG_ERR,
781 "ctdb_writerecord: cannot write to persistent db");
785 return ctdb_local_store(lock->ctdb_db->tdb, lock->key, lock->hdr,