The branch, 1.0.69 has been updated
via 9a00035c9c815736d0a5909dfc4de7f4c3ac41d4 (commit)
via 8ae93da7f6341f1cddb15786df67644f29648b9b (commit)
via b06d4c6ee7da97aaa810aa4c95952b6505cf720f (commit)
from d26d522fd0a1df7f06be056e7458ed5349a00658 (commit)
http://gitweb.samba.org/?p=sahlberg/ctdb.git;a=shortlog;h=1.0.69
- Log -----------------------------------------------------------------
commit 9a00035c9c815736d0a5909dfc4de7f4c3ac41d4
Author: root <r...@rcn1.vsofs1.com>
Date: Tue May 5 16:33:21 2009 +1000
change the talloc hierarchy for the main transaction_start context and the
individual transaction_all handles
commit 8ae93da7f6341f1cddb15786df67644f29648b9b
Author: Andrew Tridgell <tri...@samba.org>
Date: Tue May 5 12:17:21 2009 +1000
fixed an error with newer autotools versions
commit b06d4c6ee7da97aaa810aa4c95952b6505cf720f
Author: Andrew Tridgell <tri...@samba.org>
Date: Tue May 5 12:17:07 2009 +1000
fixed a problem with clients disconnecting during a traverse
When a client (such as smbstatus) is killed, it may have outstanding
traverse children on remote nodes. We need to catch the client
disconnect in ctdbd and send a control to all nodes telling them to
kill those outstanding traverse children.
-----------------------------------------------------------------------
Summary of changes:
include/ctdb_private.h | 6 ++-
lib/replace/autoconf-2.60.m4 | 2 +
server/ctdb_control.c | 6 ++-
server/ctdb_traverse.c | 135 ++++++++++++++++++++++++++++++++---------
4 files changed, 117 insertions(+), 32 deletions(-)
Changeset truncated at 500 lines:
diff --git a/include/ctdb_private.h b/include/ctdb_private.h
index 5121cd9..e14a272 100644
--- a/include/ctdb_private.h
+++ b/include/ctdb_private.h
@@ -439,6 +439,7 @@ struct ctdb_db_context {
struct ctdb_registered_call *calls; /* list of registered calls */
uint32_t seqnum;
struct timed_event *te;
+ struct ctdb_traverse_local_handle *traverse;
};
@@ -564,6 +565,7 @@ enum ctdb_controls {CTDB_CONTROL_PROCESS_EXISTS =
0,
CTDB_CONTROL_TAKEOVER_IP = 89,
CTDB_CONTROL_GET_PUBLIC_IPS = 90,
CTDB_CONTROL_GET_NODEMAP = 91,
+ CTDB_CONTROL_TRAVERSE_KILL = 92,
};
/*
@@ -1131,9 +1133,11 @@ struct ctdb_client_call_state {
int32_t ctdb_control_traverse_start(struct ctdb_context *ctdb, TDB_DATA
indata,
- TDB_DATA *outdata, uint32_t srcnode);
+ TDB_DATA *outdata, uint32_t srcnode,
uint32_t client_id);
int32_t ctdb_control_traverse_all(struct ctdb_context *ctdb, TDB_DATA data,
TDB_DATA *outdata);
int32_t ctdb_control_traverse_data(struct ctdb_context *ctdb, TDB_DATA data,
TDB_DATA *outdata);
+int32_t ctdb_control_traverse_kill(struct ctdb_context *ctdb, TDB_DATA indata,
+ TDB_DATA *outdata, uint32_t srcnode);
int ctdb_dispatch_message(struct ctdb_context *ctdb, uint64_t srvid, TDB_DATA
data);
diff --git a/lib/replace/autoconf-2.60.m4 b/lib/replace/autoconf-2.60.m4
index acdcd38..2d5dbc1 100644
--- a/lib/replace/autoconf-2.60.m4
+++ b/lib/replace/autoconf-2.60.m4
@@ -179,6 +179,7 @@ AC_DEFUN([AC_PROG_CC_C99],
# ------------------------
# Enable extensions on systems that normally disable them,
# typically due to standards-conformance issues.
+m4_ifndef([AC_USE_SYSTEM_EXTENSIONS],[
AC_DEFUN([AC_USE_SYSTEM_EXTENSIONS],
[
AC_BEFORE([$0], [AC_COMPILE_IFELSE])
@@ -208,3 +209,4 @@ AC_DEFUN([AC_USE_SYSTEM_EXTENSIONS],
AC_DEFINE([__EXTENSIONS__])
AC_DEFINE([_POSIX_PTHREAD_SEMANTICS])
])
+])
diff --git a/server/ctdb_control.c b/server/ctdb_control.c
index 15f5000..73d5d46 100644
--- a/server/ctdb_control.c
+++ b/server/ctdb_control.c
@@ -217,7 +217,7 @@ static int32_t ctdb_control_dispatch(struct ctdb_context
*ctdb,
case CTDB_CONTROL_TRAVERSE_START:
CHECK_CONTROL_DATA_SIZE(sizeof(struct ctdb_traverse_start));
- return ctdb_control_traverse_start(ctdb, indata, outdata,
srcnode);
+ return ctdb_control_traverse_start(ctdb, indata, outdata,
srcnode, client_id);
case CTDB_CONTROL_TRAVERSE_ALL:
return ctdb_control_traverse_all(ctdb, indata, outdata);
@@ -225,6 +225,10 @@ static int32_t ctdb_control_dispatch(struct ctdb_context
*ctdb,
case CTDB_CONTROL_TRAVERSE_DATA:
return ctdb_control_traverse_data(ctdb, indata, outdata);
+ case CTDB_CONTROL_TRAVERSE_KILL:
+ CHECK_CONTROL_DATA_SIZE(sizeof(struct ctdb_traverse_start));
+ return ctdb_control_traverse_kill(ctdb, indata, outdata,
srcnode);
+
case CTDB_CONTROL_REGISTER_SRVID:
return daemon_register_message_handler(ctdb, client_id, srvid);
diff --git a/server/ctdb_traverse.c b/server/ctdb_traverse.c
index 35ae318..dbb4b4f 100644
--- a/server/ctdb_traverse.c
+++ b/server/ctdb_traverse.c
@@ -24,6 +24,7 @@
#include "db_wrap.h"
#include "lib/tdb/include/tdb.h"
#include "../include/ctdb_private.h"
+#include "lib/util/dlinklist.h"
typedef void (*ctdb_traverse_fn_t)(void *private_data, TDB_DATA key, TDB_DATA
data);
@@ -32,9 +33,12 @@ typedef void (*ctdb_traverse_fn_t)(void *private_data,
TDB_DATA key, TDB_DATA da
terminate the traverse
*/
struct ctdb_traverse_local_handle {
+ struct ctdb_traverse_local_handle *next, *prev;
struct ctdb_db_context *ctdb_db;
int fd[2];
pid_t child;
+ uint64_t srvid;
+ uint32_t client_reqid;
void *private_data;
ctdb_traverse_fn_t callback;
struct timeval start_time;
@@ -73,6 +77,7 @@ static void ctdb_traverse_local_handler(uint8_t *rawdata,
size_t length, void *p
*/
static int traverse_local_destructor(struct ctdb_traverse_local_handle *h)
{
+ DLIST_REMOVE(h->ctdb_db->traverse, h);
kill(h->child, SIGKILL);
return 0;
}
@@ -87,7 +92,6 @@ static int ctdb_traverse_local_fn(struct tdb_context *tdb,
TDB_DATA key, TDB_DAT
struct ctdb_rec_data *d;
struct ctdb_ltdb_header *hdr;
-
hdr = (struct ctdb_ltdb_header *)data.dptr;
if (h->ctdb_db->persistent == 0) {
@@ -114,6 +118,14 @@ static int ctdb_traverse_local_fn(struct tdb_context *tdb,
TDB_DATA key, TDB_DAT
return 0;
}
+struct traverse_all_state {
+ struct ctdb_context *ctdb;
+ struct ctdb_traverse_local_handle *h;
+ uint32_t reqid;
+ uint32_t srcnode;
+ uint32_t client_reqid;
+ uint64_t srvid;
+};
/*
setup a non-blocking traverse of a local ltdb. The callback function
@@ -124,12 +136,12 @@ static int ctdb_traverse_local_fn(struct tdb_context
*tdb, TDB_DATA key, TDB_DAT
*/
static struct ctdb_traverse_local_handle *ctdb_traverse_local(struct
ctdb_db_context *ctdb_db,
ctdb_traverse_fn_t callback,
- void
*private_data)
+ struct
traverse_all_state *all_state)
{
struct ctdb_traverse_local_handle *h;
int ret;
- h = talloc_zero(ctdb_db, struct ctdb_traverse_local_handle);
+ h = talloc_zero(all_state, struct ctdb_traverse_local_handle);
if (h == NULL) {
return NULL;
}
@@ -151,8 +163,10 @@ static struct ctdb_traverse_local_handle
*ctdb_traverse_local(struct ctdb_db_con
}
h->callback = callback;
- h->private_data = private_data;
+ h->private_data = all_state;
h->ctdb_db = ctdb_db;
+ h->client_reqid = all_state->client_reqid;
+ h->srvid = all_state->srvid;
if (h->child == 0) {
/* start the traverse in the child */
@@ -164,6 +178,8 @@ static struct ctdb_traverse_local_handle
*ctdb_traverse_local(struct ctdb_db_con
close(h->fd[1]);
talloc_set_destructor(h, traverse_local_destructor);
+ DLIST_ADD(ctdb_db->traverse, h);
+
/*
setup a packet queue between the child and the parent. This
copes with all the async and packet boundary issues
@@ -202,6 +218,8 @@ struct ctdb_traverse_all {
uint32_t db_id;
uint32_t reqid;
uint32_t pnn;
+ uint32_t client_reqid;
+ uint64_t srvid;
};
/* called when a traverse times out */
@@ -210,12 +228,23 @@ static void ctdb_traverse_all_timeout(struct
event_context *ev, struct timed_eve
{
struct ctdb_traverse_all_handle *state = talloc_get_type(private_data,
struct ctdb_traverse_all_handle);
+ DEBUG(DEBUG_ERR,(__location__ " Traverse all timeout on database:%s\n",
state->ctdb_db->db_name));
state->ctdb->statistics.timeouts.traverse++;
state->callback(state->private_data, tdb_null, tdb_null);
- talloc_free(state);
}
+
+struct traverse_start_state {
+ struct ctdb_context *ctdb;
+ struct ctdb_traverse_all_handle *h;
+ uint32_t srcnode;
+ uint32_t reqid;
+ uint32_t db_id;
+ uint64_t srvid;
+};
+
+
/*
setup a cluster-wide non-blocking traverse of a ctdb. The
callback function will be called on every record in the local
@@ -226,7 +255,7 @@ static void ctdb_traverse_all_timeout(struct event_context
*ev, struct timed_eve
*/
static struct ctdb_traverse_all_handle *ctdb_daemon_traverse_all(struct
ctdb_db_context *ctdb_db,
ctdb_traverse_fn_t callback,
- void
*private_data)
+ struct
traverse_start_state *start_state)
{
struct ctdb_traverse_all_handle *state;
struct ctdb_context *ctdb = ctdb_db->ctdb;
@@ -235,7 +264,7 @@ static struct ctdb_traverse_all_handle
*ctdb_daemon_traverse_all(struct ctdb_db_
struct ctdb_traverse_all r;
uint32_t destination;
- state = talloc(ctdb_db, struct ctdb_traverse_all_handle);
+ state = talloc(start_state, struct ctdb_traverse_all_handle);
if (state == NULL) {
return NULL;
}
@@ -244,7 +273,7 @@ static struct ctdb_traverse_all_handle
*ctdb_daemon_traverse_all(struct ctdb_db_
state->ctdb_db = ctdb_db;
state->reqid = ctdb_reqid_new(ctdb_db->ctdb, state);
state->callback = callback;
- state->private_data = private_data;
+ state->private_data = start_state;
state->null_count = 0;
talloc_set_destructor(state, ctdb_traverse_all_destructor);
@@ -252,6 +281,8 @@ static struct ctdb_traverse_all_handle
*ctdb_daemon_traverse_all(struct ctdb_db_
r.db_id = ctdb_db->db_id;
r.reqid = state->reqid;
r.pnn = ctdb->pnn;
+ r.client_reqid = start_state->reqid;
+ r.srvid = start_state->srvid;
data.dptr = (uint8_t *)&r;
data.dsize = sizeof(r);
@@ -300,13 +331,6 @@ static struct ctdb_traverse_all_handle
*ctdb_daemon_traverse_all(struct ctdb_db_
return state;
}
-struct traverse_all_state {
- struct ctdb_context *ctdb;
- struct ctdb_traverse_local_handle *h;
- uint32_t reqid;
- uint32_t srcnode;
-};
-
/*
called for each record during a traverse all
*/
@@ -351,7 +375,7 @@ int32_t ctdb_control_traverse_all(struct ctdb_context
*ctdb, TDB_DATA data, TDB_
struct ctdb_db_context *ctdb_db;
if (data.dsize != sizeof(struct ctdb_traverse_all)) {
- DEBUG(DEBUG_ERR,("Invalid size in
ctdb_control_traverse_all\n"));
+ DEBUG(DEBUG_ERR,(__location__ " Invalid size in
ctdb_control_traverse_all\n"));
return -1;
}
@@ -368,6 +392,8 @@ int32_t ctdb_control_traverse_all(struct ctdb_context
*ctdb, TDB_DATA data, TDB_
state->reqid = c->reqid;
state->srcnode = c->pnn;
state->ctdb = ctdb;
+ state->client_reqid = c->client_reqid;
+ state->srvid = c->srvid;
state->h = ctdb_traverse_local(ctdb_db, traverse_all_callback, state);
if (state->h == NULL) {
@@ -423,21 +449,59 @@ int32_t ctdb_control_traverse_data(struct ctdb_context
*ctdb, TDB_DATA data, TDB
private_data = state->private_data;
callback(private_data, key, data);
- if (key.dsize == 0 && data.dsize == 0) {
- /* we've received all of the null replies, so all
- nodes are finished */
- talloc_free(state);
- }
return 0;
}
-struct traverse_start_state {
- struct ctdb_context *ctdb;
- struct ctdb_traverse_all_handle *h;
- uint32_t srcnode;
- uint32_t reqid;
- uint64_t srvid;
-};
+/*
+ kill a in-progress traverse, used when a client disconnects
+ */
+int32_t ctdb_control_traverse_kill(struct ctdb_context *ctdb, TDB_DATA data,
+ TDB_DATA *outdata, uint32_t srcnode)
+{
+ struct ctdb_traverse_start *d = (struct ctdb_traverse_start *)data.dptr;
+ struct ctdb_db_context *ctdb_db;
+ struct ctdb_traverse_local_handle *t;
+
+ ctdb_db = find_ctdb_db(ctdb, d->db_id);
+ if (ctdb_db == NULL) {
+ return -1;
+ }
+
+ for (t=ctdb_db->traverse; t; t=t->next) {
+ if (t->client_reqid == d->reqid &&
+ t->srvid == d->srvid) {
+ talloc_free(t);
+ break;
+ }
+ }
+
+ return 0;
+}
+
+
+/*
+ this is called when a client disconnects during a traverse
+ we need to notify all the nodes taking part in the search that they
+ should kill their traverse children
+ */
+static int ctdb_traverse_start_destructor(struct traverse_start_state *state)
+{
+ struct ctdb_traverse_start r;
+ TDB_DATA data;
+
+ DEBUG(DEBUG_ERR,(__location__ " Traverse cancelled by client disconnect
for database:0x%08x\n", state->db_id));
+ r.db_id = state->db_id;
+ r.reqid = state->reqid;
+ r.srvid = state->srvid;
+
+ data.dptr = (uint8_t *)&r;
+ data.dsize = sizeof(r);
+
+ ctdb_daemon_send_control(state->ctdb, CTDB_BROADCAST_CONNECTED, 0,
+ CTDB_CONTROL_TRAVERSE_KILL,
+ 0, CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL);
+ return 0;
+}
/*
callback which sends records as messages to the client
@@ -461,19 +525,27 @@ static void traverse_start_callback(void *p, TDB_DATA
key, TDB_DATA data)
ctdb_dispatch_message(state->ctdb, state->srvid, cdata);
if (key.dsize == 0 && data.dsize == 0) {
/* end of traverse */
+ talloc_set_destructor(state, NULL);
talloc_free(state);
}
}
+
/*
start a traverse_all - called as a control from a client
*/
int32_t ctdb_control_traverse_start(struct ctdb_context *ctdb, TDB_DATA data,
- TDB_DATA *outdata, uint32_t srcnode)
+ TDB_DATA *outdata, uint32_t srcnode,
uint32_t client_id)
{
struct ctdb_traverse_start *d = (struct ctdb_traverse_start *)data.dptr;
struct traverse_start_state *state;
struct ctdb_db_context *ctdb_db;
+ struct ctdb_client *client = ctdb_reqid_find(ctdb, client_id, struct
ctdb_client);
+
+ if (client == NULL) {
+ DEBUG(DEBUG_ERR,(__location__ " No client found\n"));
+ return -1;
+ }
if (data.dsize != sizeof(*d)) {
DEBUG(DEBUG_ERR,("Bad record size in
ctdb_control_traverse_start\n"));
@@ -485,7 +557,7 @@ int32_t ctdb_control_traverse_start(struct ctdb_context
*ctdb, TDB_DATA data,
return -1;
}
- state = talloc(ctdb_db, struct traverse_start_state);
+ state = talloc(client, struct traverse_start_state);
if (state == NULL) {
return -1;
}
@@ -493,6 +565,7 @@ int32_t ctdb_control_traverse_start(struct ctdb_context
*ctdb, TDB_DATA data,
state->srcnode = srcnode;
state->reqid = d->reqid;
state->srvid = d->srvid;
+ state->db_id = d->db_id;
state->ctdb = ctdb;
state->h = ctdb_daemon_traverse_all(ctdb_db, traverse_start_callback,
state);
@@ -501,5 +574,7 @@ int32_t ctdb_control_traverse_start(struct ctdb_context
*ctdb, TDB_DATA data,
return -1;
}
+ talloc_set_destructor(state, ctdb_traverse_start_destructor);
+
return 0;
}
--
CTDB repository
