[Cluster-devel] [PATCH 2/3] dlm: add TCP multihoming/failover support
David Windsor
dwindsor at redhat.com
Tue Jan 8 15:08:54 UTC 2019
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.
Signed-off-by: David Windsor <dwindsor at redhat.com>
---
fs/dlm/config.c | 21 +++++++++++++++++++++
fs/dlm/lowcomms.c | 34 +++++++++++++++++++++++-----------
fs/dlm/lowcomms.h | 1 +
3 files changed, 45 insertions(+), 11 deletions(-)
diff --git a/fs/dlm/config.c b/fs/dlm/config.c
index 1270551d24e3..23d2677e10bd 100644
--- a/fs/dlm/config.c
+++ b/fs/dlm/config.c
@@ -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 {
@@ -662,8 +664,26 @@ 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_, error);
CONFIGFS_ATTR_WO(comm_, addr);
CONFIGFS_ATTR_RO(comm_, addr_list);
@@ -672,6 +692,7 @@ static struct configfs_attribute *comm_attrs[] = {
[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,
};
diff --git a/fs/dlm/lowcomms.c b/fs/dlm/lowcomms.c
index 905cbdbd31bc..d37af1372ed0 100644
--- a/fs/dlm/lowcomms.c
+++ b/fs/dlm/lowcomms.c
@@ -159,6 +159,7 @@ static DEFINE_SPINLOCK(dlm_node_addrs_spin);
static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
static int dlm_local_count;
static int dlm_allow_conn;
+static int dlm_local_idx;
/* Work queues */
static struct workqueue_struct *recv_workqueue;
@@ -330,7 +331,7 @@ static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
if (!sa_out)
return 0;
- if (dlm_local_addr[0]->ss_family == AF_INET) {
+ if (dlm_local_addr[dlm_local_idx]->ss_family == AF_INET) {
struct sockaddr_in *in4 = (struct sockaddr_in *) &sas;
struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
@@ -519,6 +520,8 @@ static void lowcomms_error_report(struct sock *sk)
dlm_config.ci_tcp_port, sk->sk_err,
sk->sk_err_soft);
}
+
+ dlm_lowcomms_next_addr();
out:
read_unlock_bh(&sk->sk_callback_lock);
if (orig_report)
@@ -572,7 +575,7 @@ static void add_sock(struct socket *sock, struct connection *con)
static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
int *addr_len)
{
- saddr->ss_family = dlm_local_addr[0]->ss_family;
+ saddr->ss_family = dlm_local_addr[dlm_local_idx]->ss_family;
if (saddr->ss_family == AF_INET) {
struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
in4_addr->sin_port = cpu_to_be16(port);
@@ -1169,7 +1172,7 @@ static void tcp_connect_to_sock(struct connection *con)
/* Bind to our cluster-known address connecting to avoid
routing problems */
- memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
+ memcpy(&src_addr, dlm_local_addr[dlm_local_idx], sizeof(src_addr));
make_sockaddr(&src_addr, 0, &addr_len);
result = sock->ops->bind(sock, (struct sockaddr *) &src_addr,
addr_len);
@@ -1213,6 +1216,7 @@ static void tcp_connect_to_sock(struct connection *con)
con->retries, result);
mutex_unlock(&con->sock_mutex);
msleep(1000);
+ dlm_lowcomms_next_addr();
lowcomms_connect_sock(con);
return;
}
@@ -1293,6 +1297,7 @@ static void init_local(void)
struct sockaddr_storage sas, *addr;
int i;
+ dlm_local_idx = 0;
dlm_local_count = 0;
for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
if (dlm_our_addr(&sas, i))
@@ -1375,16 +1380,9 @@ static int tcp_listen_for_all(void)
if (!con)
return -ENOMEM;
- /* We don't support multi-homed hosts */
- if (dlm_local_addr[1] != NULL) {
- log_print("TCP protocol can't handle multi-homed hosts, "
- "try SCTP");
- return -EINVAL;
- }
-
log_print("Using TCP for communications");
- sock = tcp_create_listen_sock(con, dlm_local_addr[0]);
+ sock = tcp_create_listen_sock(con, dlm_local_addr[dlm_local_idx]);
if (sock) {
add_sock(sock, con);
result = 0;
@@ -1716,6 +1714,20 @@ static void work_flush(void)
} while (!ok);
}
+void dlm_lowcomms_next_addr(void)
+{
+ if (!dlm_local_count)
+ init_local();
+
+ if (dlm_local_idx < dlm_local_count)
+ dlm_local_idx++;
+ else
+ dlm_local_idx = 0;
+
+ dlm_lowcomms_stop();
+ dlm_lowcomms_start();
+}
+
void dlm_lowcomms_stop(void)
{
/* Set all the flags to prevent any
diff --git a/fs/dlm/lowcomms.h b/fs/dlm/lowcomms.h
index 67462e54fc2f..7c7c5e4b5549 100644
--- a/fs/dlm/lowcomms.h
+++ b/fs/dlm/lowcomms.h
@@ -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__ */
--
2.20.1
More information about the Cluster-devel
mailing list