#include <stdlib.h>
#include <sys/socket.h>
#include <sys/un.h>
+#include <sys/ioctl.h>
#include "libctdb_private.h"
#include "io_elem.h"
#include "local_tdb.h"
struct ctdb_db *ctdb_db;
TDB_DATA key;
- /* This will always be true by the time user sees this. */
- bool held;
+ /* This will always be set by the time user sees this. */
+ unsigned long held_magic;
struct ctdb_ltdb_header *hdr;
/* For convenience, we stash original callback here. */
ctdb_rrl_callback_t callback;
};
+struct ctdb_db {
+ struct ctdb_connection *ctdb;
+ bool persistent;
+ uint32_t tdb_flags;
+ uint32_t id;
+ struct tdb_context *tdb;
+
+ ctdb_callback_t callback;
+ void *private_data;
+};
+
static void remove_lock(struct ctdb_connection *ctdb, struct ctdb_lock *lock)
{
DLIST_REMOVE(ctdb->locks, lock);
DLIST_ADD(ctdb->locks, lock);
}
+static void cleanup_locks(struct ctdb_connection *ctdb, struct ctdb_db *db)
+{
+ struct ctdb_lock *i, *next;
+
+ for (i = ctdb->locks; i; i = next) {
+ /* Grab next pointer, as release_lock will free i */
+ next = i->next;
+ if (i->ctdb_db == db) {
+ ctdb_release_lock(db, i);
+ }
+ }
+}
+
/* FIXME: Could be in shared util code with rest of ctdb */
static void close_noerr(int fd)
{
ctdb->outq = NULL;
ctdb->doneq = NULL;
ctdb->in = NULL;
+ ctdb->inqueue = NULL;
ctdb->message_handlers = NULL;
ctdb->next_id = 0;
ctdb->broken = false;
return NULL;
}
+void ctdb_disconnect(struct ctdb_connection *ctdb)
+{
+ struct ctdb_request *i;
+
+ DEBUG(ctdb, LOG_DEBUG, "ctdb_disconnect");
+
+ while ((i = ctdb->outq) != NULL) {
+ DLIST_REMOVE(ctdb->outq, i);
+ ctdb_request_free(ctdb, i);
+ }
+
+ while ((i = ctdb->doneq) != NULL) {
+ DLIST_REMOVE(ctdb->doneq, i);
+ ctdb_request_free(ctdb, i);
+ }
+
+ if (ctdb->in)
+ free_io_elem(ctdb->in);
+
+ remove_message_handlers(ctdb);
+
+ close(ctdb->fd);
+ /* Just in case they try to reuse */
+ ctdb->fd = -1;
+ free(ctdb);
+}
+
int ctdb_get_fd(struct ctdb_connection *ctdb)
{
return ctdb->fd;
void ctdb_request_free(struct ctdb_connection *ctdb, struct ctdb_request *req)
{
+ if (req->next || req->prev) {
+ DEBUG(ctdb, LOG_ALERT,
+ "ctdb_request_free: request not complete! ctdb_cancel? %p (id %u)",
+ req, req->hdr.hdr ? req->hdr.hdr->reqid : 0);
+ ctdb_cancel(ctdb, req);
+ return;
+ }
if (req->extra_destructor) {
req->extra_destructor(ctdb, req);
}
/* Library user error if this isn't a reply to a call. */
if (req->hdr.hdr->operation != CTDB_REQ_CALL) {
errno = EINVAL;
- DEBUG(ctdb, LOG_ERR,
+ DEBUG(ctdb, LOG_ALERT,
"This was not a ctdbd call request: operation %u",
req->hdr.hdr->operation);
return NULL;
if (req->hdr.call->callid != callid) {
errno = EINVAL;
- DEBUG(ctdb, LOG_ERR,
+ DEBUG(ctdb, LOG_ALERT,
"This was not a ctdbd %u call request: %u",
callid, req->hdr.call->callid);
return NULL;
/* Library user error if this isn't a reply to a call. */
if (len < sizeof(*inhdr)) {
errno = EINVAL;
- DEBUG(ctdb, LOG_CRIT,
+ DEBUG(ctdb, LOG_ALERT,
"Short ctdbd control reply: %zu bytes", len);
return NULL;
}
if (req->hdr.hdr->operation != CTDB_REQ_CONTROL) {
errno = EINVAL;
- DEBUG(ctdb, LOG_ERR,
+ DEBUG(ctdb, LOG_ALERT,
"This was not a ctdbd control request: operation %u",
req->hdr.hdr->operation);
return NULL;
/* ... or if it was a different control from what we expected. */
if (req->hdr.control->opcode != control) {
errno = EINVAL;
- DEBUG(ctdb, LOG_ERR,
+ DEBUG(ctdb, LOG_ALERT,
"This was not an opcode %u ctdbd control request: %u",
control, req->hdr.control->opcode);
return NULL;
}
if (holding_lock(ctdb)) {
- DEBUG(ctdb, LOG_WARNING, "Do not block while holding lock!");
+ DEBUG(ctdb, LOG_ALERT, "Do not block while holding lock!");
}
if (revents & POLLOUT) {
while (revents & POLLIN) {
int ret;
+ int num_ready = 0;
+
+ if (ioctl(ctdb->fd, FIONREAD, &num_ready) != 0) {
+ DEBUG(ctdb, LOG_ERR,
+ "ctdb_service: ioctl(FIONREAD) %d", errno);
+ ctdb->broken = true;
+ return false;
+ }
+ if (num_ready == 0) {
+ /* the descriptor has been closed or we have all our data */
+ break;
+ }
+
if (!ctdb->in) {
ctdb->in = new_io_elem(sizeof(struct ctdb_req_header));
return false;
} else if (ret < 0) {
/* No progress, stop loop. */
- revents = 0;
+ break;
} else if (io_elem_finished(ctdb->in)) {
- handle_incoming(ctdb, ctdb->in);
+ io_elem_queue(ctdb, ctdb->in);
ctdb->in = NULL;
}
}
+
+ while (ctdb->inqueue != NULL) {
+ struct io_elem *io = ctdb->inqueue;
+
+ io_elem_dequeue(ctdb, io);
+ handle_incoming(ctdb, io);
+ }
+
return true;
}
void ctdb_cancel(struct ctdb_connection *ctdb, struct ctdb_request *req)
{
+ if (!req->next && !req->prev) {
+ DEBUG(ctdb, LOG_ALERT,
+ "ctdb_cancel: request completed! ctdb_request_free? %p (id %u)",
+ req, req->hdr.hdr ? req->hdr.hdr->reqid : 0);
+ ctdb_request_free(ctdb, req);
+ return;
+ }
+
DEBUG(ctdb, LOG_DEBUG, "ctdb_cancel: %p (id %u)",
req, req->hdr.hdr ? req->hdr.hdr->reqid : 0);
req->callback = ctdb_cancel_callback;
}
-struct ctdb_db {
- struct ctdb_connection *ctdb;
- bool persistent;
- uint32_t tdb_flags;
- uint32_t id;
- struct tdb_context *tdb;
-
- /* The lock we are holding, if any (we can only have one!) */
- struct ctdb_lock *lock;
-
- ctdb_callback_t callback;
- void *private_data;
-};
+void ctdb_detachdb(struct ctdb_connection *ctdb, struct ctdb_db *db)
+{
+ cleanup_locks(ctdb, db);
+ tdb_close(db->tdb);
+ free(db);
+}
static void attachdb_getdbpath_done(struct ctdb_connection *ctdb,
struct ctdb_request *req,
struct ctdb_reply_control *reply;
struct ctdb_db *db = req->priv_data;
uint32_t tdb_flags = db->tdb_flags;
+ struct tdb_logging_context log;
/* Never sent the dbpath request? We've failed. */
if (!dbpath_req) {
tdb_flags = db->persistent ? TDB_DEFAULT : TDB_NOSYNC;
tdb_flags |= TDB_DISALLOW_NESTING;
- /* FIXME: Setup logging to go through our logging. */
- db->tdb = tdb_open((char *)reply->data, 0, tdb_flags, O_RDWR, 0);
+ log.log_fn = ctdb_tdb_log_bridge;
+ log.log_private = ctdb;
+ db->tdb = tdb_open_ex((char *)reply->data, 0, tdb_flags, O_RDWR, 0,
+ &log, NULL);
if (db->tdb == NULL) {
DEBUG(db->ctdb, LOG_ERR,
"ctdb_attachdb_recv: failed to tdb_open %s",
struct ctdb_request *
ctdb_attachdb_send(struct ctdb_connection *ctdb,
- const char *name, int persistent, uint32_t tdb_flags,
+ const char *name, bool persistent, uint32_t tdb_flags,
ctdb_callback_t callback, void *private_data)
{
struct ctdb_request *req;
req = new_ctdb_control_request(ctdb, opcode, CTDB_CURRENT_NODE, name,
strlen(name) + 1, attachdb_done, db);
if (!req) {
- DEBUG(db->ctdb, LOG_ERR,
+ DEBUG(ctdb, LOG_ERR,
"ctdb_attachdb_send: failed allocating DB_ATTACH");
free(db);
return NULL;
return req;
}
-void ctdb_release_lock(struct ctdb_lock *lock)
+static unsigned long lock_magic(struct ctdb_lock *lock)
+{
+ /* A non-zero magic specific to this structure. */
+ return ((unsigned long)lock->key.dptr
+ ^ (((unsigned long)lock->key.dptr) << 16)
+ ^ 0xBADC0FFEEBADC0DEULL)
+ | 1;
+}
+
+/* This is only called on locks before they're held. */
+static void free_lock(struct ctdb_lock *lock)
{
- if (lock->held) {
+ if (lock->held_magic) {
+ DEBUG(lock->ctdb_db->ctdb, LOG_ALERT,
+ "free_lock invalid lock %p", lock);
+ }
+ free(lock->hdr);
+ free(lock);
+}
+
+
+void ctdb_release_lock(struct ctdb_db *ctdb_db, struct ctdb_lock *lock)
+{
+ if (lock->held_magic != lock_magic(lock)) {
+ DEBUG(lock->ctdb_db->ctdb, LOG_ALERT,
+ "ctdb_release_lock invalid lock %p", lock);
+ } else if (lock->ctdb_db != ctdb_db) {
+ errno = EBADF;
+ DEBUG(ctdb_db->ctdb, LOG_ALERT,
+ "ctdb_release_lock: wrong ctdb_db.");
+ } else {
tdb_chainunlock(lock->ctdb_db->tdb, lock->key);
DEBUG(lock->ctdb_db->ctdb, LOG_DEBUG,
"ctdb_release_lock %p", lock);
- lock->held = false;
remove_lock(lock->ctdb_db->ctdb, lock);
}
- free(lock->hdr);
- free(lock);
+ lock->held_magic = 0;
+ free_lock(lock);
}
+
/* We keep the lock if local node is the dmaster. */
static bool try_readrecordlock(struct ctdb_lock *lock, TDB_DATA *data)
{
if (hdr && hdr->dmaster == lock->ctdb_db->ctdb->pnn) {
DEBUG(lock->ctdb_db->ctdb, LOG_DEBUG,
"ctdb_readrecordlock_async: got local lock");
- lock->held = true;
+ lock->held_magic = lock_magic(lock);
lock->hdr = hdr;
add_lock(lock->ctdb_db->ctdb, lock);
return true;
static void destroy_lock(struct ctdb_connection *ctdb,
struct ctdb_request *req)
{
- ctdb_release_lock(req->extra);
+ free_lock(req->extra);
}
static void readrecordlock_retry(struct ctdb_connection *ctdb,
/* Now it's their responsibility to free lock & request! */
req->extra_destructor = NULL;
lock->callback(lock->ctdb_db, lock, data, private);
+ ctdb_request_free(ctdb, req);
return;
}
TDB_DATA data;
if (holding_lock(ctdb_db->ctdb)) {
- DEBUG(ctdb_db->ctdb, LOG_ERR,
+ DEBUG(ctdb_db->ctdb, LOG_ALERT,
"ctdb_readrecordlock_async: already holding lock");
return false;
}
lock->key.dsize = key.dsize;
lock->ctdb_db = ctdb_db;
lock->hdr = NULL;
- lock->held = false;
+ lock->held_magic = 0;
/* Fast path. */
if (try_readrecordlock(lock, &data)) {
if (!req) {
DEBUG(ctdb_db->ctdb, LOG_ERR,
"ctdb_readrecordlock_async: allocation failed");
- ctdb_release_lock(lock);
+ free_lock(lock);
return NULL;
}
req->extra = lock;
return true;
}
-int ctdb_writerecord(struct ctdb_lock *lock, TDB_DATA data)
+bool ctdb_writerecord(struct ctdb_db *ctdb_db,
+ struct ctdb_lock *lock, TDB_DATA data)
{
- if (lock->ctdb_db->persistent) {
+ if (lock->ctdb_db != ctdb_db) {
+ errno = EBADF;
+ DEBUG(ctdb_db->ctdb, LOG_ALERT,
+ "ctdb_writerecord: Can not write, wrong ctdb_db.");
+ return false;
+ }
+
+ if (lock->held_magic != lock_magic(lock)) {
+ errno = EBADF;
+ DEBUG(ctdb_db->ctdb, LOG_ALERT,
+ "ctdb_writerecord: Can not write. Lock has been released.");
+ return false;
+ }
+
+ if (ctdb_db->persistent) {
errno = EINVAL;
- DEBUG(lock->ctdb_db->ctdb, LOG_ERR,
+ DEBUG(ctdb_db->ctdb, LOG_ALERT,
"ctdb_writerecord: cannot write to persistent db");
- return -1;
+ return false;
}
- return ctdb_local_store(lock->ctdb_db->tdb, lock->key, lock->hdr,
- data);
+ switch (ctdb_local_store(ctdb_db->tdb, lock->key, lock->hdr, data)) {
+ case 0:
+ DEBUG(ctdb_db->ctdb, LOG_DEBUG,
+ "ctdb_writerecord: optimized away noop write.");
+ /* fall thru */
+ case 1:
+ return true;
+
+ default:
+ switch (errno) {
+ case ENOMEM:
+ DEBUG(ctdb_db->ctdb, LOG_CRIT,
+ "ctdb_writerecord: out of memory.");
+ break;
+ case EINVAL:
+ DEBUG(ctdb_db->ctdb, LOG_ALERT,
+ "ctdb_writerecord: record changed under lock?");
+ break;
+ default: /* TDB already logged. */
+ break;
+ }
+ return false;
+ }
}