Skip to contentSkip to conversation

dlm: check if workqueues are NULL before flushing/destroying

PATCH New David Windsor 2 participants 4 messages Apr 2, 2019 – Apr 5, 2019 Distributed lock manager (DLM)

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/3 dlm: check if workqueues are NULL before flushing/destroying

If the DLM lowcomms stack is shut down before any DLM traffic can be generated, flush_workqueue() and destroy_workqueue() can be called on empty send and/or recv workqueues.

Insert guard conditionals to only call flush_workqueue() and destroy_workqueue() on workqueues that are not NULL.

---
 fs/dlm/lowcomms.c | 18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)
fs/dlm/lowcomms.c +12 -6
index a5e4a221435c..a93ebffe84b3 100644
@@ -1630,8 +1630,10 @@ static void clean_writequeues(void)
 
 static void work_stop(void)
 {
-	destroy_workqueue(recv_workqueue);
-	destroy_workqueue(send_workqueue);
+	if (recv_workqueue)
+		destroy_workqueue(recv_workqueue);
+	if (send_workqueue)
+		destroy_workqueue(send_workqueue);
 }
 
 static int work_start(void)
@@ -1691,13 +1693,17 @@ static void work_flush(void)
 	struct hlist_node *n;
 	struct connection *con;
 
-	flush_workqueue(recv_workqueue);
-	flush_workqueue(send_workqueue);
+	if (recv_workqueue)
+		flush_workqueue(recv_workqueue);
+	if (send_workqueue)
+		flush_workqueue(send_workqueue);
 	do {
 		ok = 1;
 		foreach_conn(stop_conn);
-		flush_workqueue(recv_workqueue);
-		flush_workqueue(send_workqueue);
+		if (recv_workqueue)
+			flush_workqueue(recv_workqueue);
+		if (send_workqueue)
+			flush_workqueue(send_workqueue);
 		for (i = 0; i < CONN_HASH_SIZE && ok; i++) {
 			hlist_for_each_entry_safe(con, n,
 						  &connection_hash[i], list) {
Signature · 1 line
2.20.1
#1 lore raw

PATCH v2 2/3 dlm: add TCP multihoming/failover support

Add the ability to specify multiple source addresses for DLM nodes so that multihomed configurations can use multiple addresses and still be recognized by the receiving node.

While each node is capable of being configured for multiple IPs, DLM requires each node have only one active address at a time.

This patch introduces a round-robin heuristic for selecting the next active interface, but other heuristics could easily be added later.

To support failover, a new configfs node is added by this patch:
/sys/kernel/config/dlm/cluster/comms/<comm>/error
This node is write-only, and is provided so that userspace may signal the kernel when it detects a communications error. The kernel will switch to the next local network interface after 1 is written to the new configfs node.

---
 fs/dlm/config.c   | 21 +++++++++++++++++
 fs/dlm/lowcomms.c | 60 +++++++++++++++++++++++++++++++++++++----------
 fs/dlm/lowcomms.h |  1 +
 3 files changed, 69 insertions(+), 13 deletions(-)
fs/dlm/config.c +21 -0
index 1270551d24e3..96db7b1346f9 100644
@@ -31,6 +31,7 @@
  * /config/dlm/<cluster>/comms/<comm>/local
  * /config/dlm/<cluster>/comms/<comm>/addr      (write only)
  * /config/dlm/<cluster>/comms/<comm>/addr_list (read only)
+ * /config/dlm/<cluster>/comms/<comm>/error	(write only)
  * The <cluster> level is useless, but I haven't figured out how to avoid it.
  */
 
@@ -198,6 +199,7 @@ enum {
 	COMM_ATTR_LOCAL,
 	COMM_ATTR_ADDR,
 	COMM_ATTR_ADDR_LIST,
+	COMM_ATTR_ERROR,
 };
 
 enum {
@@ -662,16 +664,35 @@ static ssize_t comm_addr_list_show(struct config_item *item, char *buf)
 	return 4096 - allowance;
 }
 
+static ssize_t comm_error_store(struct config_item *item, const char *buf,
+				size_t len)
+{
+	int ret, i;
+
+	ret = kstrtoint(buf, 0, &i);
+	if (ret < 0)
+		return ret;
+
+	if (i == 0)
+		return 0;
+
+	dlm_lowcomms_next_addr();
+
+	return len;
+}
+
 CONFIGFS_ATTR(comm_, nodeid);
 CONFIGFS_ATTR(comm_, local);
 CONFIGFS_ATTR_WO(comm_, addr);
 CONFIGFS_ATTR_RO(comm_, addr_list);
+CONFIGFS_ATTR_WO(comm_, error);
 
 static struct configfs_attribute *comm_attrs[] = {
 	[COMM_ATTR_NODEID] = &comm_attr_nodeid,
 	[COMM_ATTR_LOCAL] = &comm_attr_local,
 	[COMM_ATTR_ADDR] = &comm_attr_addr,
 	[COMM_ATTR_ADDR_LIST] = &comm_attr_addr_list,
+	[COMM_ATTR_ERROR] = &comm_attr_error,
 	NULL,
 };
 
fs/dlm/lowcomms.c +47 -13

140 lines not on this page · raw message

fs/dlm/lowcomms.h +1 -0
index 67462e54fc2f..7c7c5e4b5549 100644
@@ -22,6 +22,7 @@ void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc)
 void dlm_lowcomms_commit_buffer(void *mh);
 int dlm_lowcomms_connect_node(int nodeid);
 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len);
+void dlm_lowcomms_next_addr(void);
 
 #endif				/* __LOWCOMMS_DOT_H__ */
 
Signature · 1 line
2.20.1
#2 lore raw

PATCH v2 3/3 dlm: allow binding to all network interfaces

In reply to David Windsor

Currently, in the kernel, DLM only is able to bind its listen socket to a single network interface. To support more robust network configurations, DLM should be able to bind to all network interfaces.

This patch adds a configfs node to enable/disable binding to all network interfaces. When 1 is written to this configfs node, the DLM listen socket will bind to all network interfaces. When 0 is written to the node, DLM will bind only to its current local network interface.

---
 fs/dlm/config.c   | 21 +++++++++++++++++++++
 fs/dlm/config.h   |  3 ++-
 fs/dlm/lowcomms.c | 19 ++++++++++++++++++-
 3 files changed, 41 insertions(+), 2 deletions(-)
fs/dlm/config.c +21 -0
index 96db7b1346f9..16b83d61b060 100644
@@ -29,6 +29,7 @@
  * /config/dlm/<cluster>/spaces/<space>/nodes/<node>/weight
  * /config/dlm/<cluster>/comms/<comm>/nodeid
  * /config/dlm/<cluster>/comms/<comm>/local
+ * /config/dlm/<cluster>/comms/<comm>/bind_all
  * /config/dlm/<cluster>/comms/<comm>/addr      (write only)
  * /config/dlm/<cluster>/comms/<comm>/addr_list (read only)
  * /config/dlm/<cluster>/comms/<comm>/error	(write only)
@@ -39,6 +40,7 @@ static struct config_group *space_list;
 static struct config_group *comm_list;
 static struct dlm_comm *local_comm;
 static uint32_t dlm_comm_count;
+static int bind_all;
 
 struct dlm_clusters;
 struct dlm_cluster;
@@ -200,6 +202,7 @@ enum {
 	COMM_ATTR_ADDR,
 	COMM_ATTR_ADDR_LIST,
 	COMM_ATTR_ERROR,
+	COMM_ATTR_BIND_ALL,
 };
 
 enum {
@@ -681,11 +684,23 @@ static ssize_t comm_error_store(struct config_item *item, const char *buf,
 	return len;
 }
 
+static ssize_t comm_bind_all_show(struct config_item *item, char *buf)
+{
+	return sprintf(buf, "%d\n", bind_all);
+}
+
+static ssize_t comm_bind_all_store(struct config_item *item, const char *buf,
+				   size_t len)
+{
+	return kstrtoint(buf, 0, &bind_all);
+}
+
 CONFIGFS_ATTR(comm_, nodeid);
 CONFIGFS_ATTR(comm_, local);
 CONFIGFS_ATTR_WO(comm_, addr);
 CONFIGFS_ATTR_RO(comm_, addr_list);
 CONFIGFS_ATTR_WO(comm_, error);
+CONFIGFS_ATTR(comm_, bind_all);
 
 static struct configfs_attribute *comm_attrs[] = {
 	[COMM_ATTR_NODEID] = &comm_attr_nodeid,
@@ -693,6 +708,7 @@ static struct configfs_attribute *comm_attrs[] = {
 	[COMM_ATTR_ADDR] = &comm_attr_addr,
 	[COMM_ATTR_ADDR_LIST] = &comm_attr_addr_list,
 	[COMM_ATTR_ERROR] = &comm_attr_error,
+	[COMM_ATTR_BIND_ALL] = &comm_attr_bind_all,
 	NULL,
 };
 
@@ -868,6 +884,11 @@ int dlm_our_addr(struct sockaddr_storage *addr, int num)
 	return 0;
 }
 
+int dlm_bind_all(void)
+{
+	return bind_all;
+}
+
 /* Config file defaults */
 #define DEFAULT_TCP_PORT       21064
 #define DEFAULT_BUFFER_SIZE     4096
fs/dlm/config.h +2 -1
index 6041eec886ab..e3fd8ce45874 100644
@@ -21,7 +21,7 @@ struct dlm_config_node {
 	uint32_t comm_seq;
 };
 
-#define DLM_MAX_ADDR_COUNT 3
+#define DLM_MAX_ADDR_COUNT 9
 
 struct dlm_config_info {
 	int ci_tcp_port;
@@ -49,6 +49,7 @@ int dlm_config_nodes(char *lsname, struct dlm_config_node **nodes_out,
 int dlm_comm_seq(int nodeid, uint32_t *seq);
 int dlm_our_nodeid(void);
 int dlm_our_addr(struct sockaddr_storage *addr, int num);
+int dlm_bind_all(void);
 
 #endif				/* __CONFIG_DOT_H__ */
 
fs/dlm/lowcomms.c +18 -1
index be0e134d4fc4..60ae7c53a8a1 100644
@@ -1394,6 +1394,9 @@ static int sctp_listen_for_all(void)
 static int tcp_listen_for_all(void)
 {
 	struct socket *sock = NULL;
+	struct sockaddr_in *sin4;
+	struct sockaddr_in6 *sin6;
+	struct sockaddr_storage sas, laddr;
 	struct connection *con = nodeid2con(0, GFP_NOFS);
 	int result = -EINVAL;
 
@@ -1402,7 +1405,21 @@ static int tcp_listen_for_all(void)
 
 	log_print("Using TCP for communications");
 
-	sock = tcp_create_listen_sock(con, dlm_local_addr[dlm_local_idx]);
+	memcpy(&sas, dlm_local_addr[dlm_local_idx], sizeof(sas));
+	memcpy(&laddr, dlm_local_addr[dlm_local_idx], sizeof(laddr));
+	if (dlm_bind_all()) {
+		if (sas.ss_family == AF_INET) {
+			sin4 = (struct sockaddr_in *) &sas;
+			sin4->sin_addr.s_addr = htonl(INADDR_ANY);
+			memcpy(&laddr, sin4, sizeof(laddr));
+		} else {
+			sin6 = (struct sockaddr_in6 *) &sas;
+			sin6->sin6_addr = in6addr_any;
+			memcpy(&laddr, sin6, sizeof(laddr));
+		}
+	}
+
+	sock = tcp_create_listen_sock(con, &laddr);
 	if (sock) {
 		add_sock(sock, con);
 		result = 0;
Signature · 1 line
2.20.1
maintainer
#3 lore raw

PATCH v2 1/3 dlm: check if workqueues are NULL before flushing/destroying

In reply to David Windsor

I tried these with one address and had no problem, so I've pushed them to the linux-dlm next branch.
Dave