[PATCH] NFS: use a constant value for TCP retransmit timeouts
Implement a best practice: don't use exponential backoff when computing
retransmit timeout values on TCP connections, but simply retransmit
at regular intervals.
This also fixes a bug introduced when xprt_reset_majortimeo() was added.
Test-plan:
Enable RPC debugging and watch timeout behavior on a NFS/TCP mount.
Version: Thu, 11 Aug 2005 16:02:19 -0400
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index 6922469d..b6a1ca5 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -358,6 +358,35 @@
return no_root_error;
}
+static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
+{
+ to->to_initval = timeo * HZ / 10;
+ to->to_retries = retrans;
+ if (!to->to_retries)
+ to->to_retries = 2;
+
+ switch (proto) {
+ case IPPROTO_TCP:
+ if (!to->to_initval)
+ to->to_initval = 60 * HZ;
+ if (to->to_initval > RPC_MAX_TCP_TIMEOUT)
+ to->to_initval = RPC_MAX_TCP_TIMEOUT;
+ to->to_increment = to->to_initval;
+ to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
+ to->to_exponential = 0;
+ break;
+ case IPPROTO_UDP:
+ default:
+ if (!to->to_initval)
+ to->to_initval = 11 * HZ / 10;
+ if (to->to_initval > RPC_MAX_UDP_TIMEOUT)
+ to->to_initval = RPC_MAX_UDP_TIMEOUT;
+ to->to_maxval = RPC_MAX_UDP_TIMEOUT;
+ to->to_exponential = 1;
+ break;
+ }
+}
+
/*
* Create an RPC client handle.
*/
@@ -367,22 +396,12 @@
struct rpc_timeout timeparms;
struct rpc_xprt *xprt = NULL;
struct rpc_clnt *clnt = NULL;
- int tcp = (data->flags & NFS_MOUNT_TCP);
+ int proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
- /* Initialize timeout values */
- timeparms.to_initval = data->timeo * HZ / 10;
- timeparms.to_retries = data->retrans;
- timeparms.to_maxval = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
- timeparms.to_exponential = 1;
-
- if (!timeparms.to_initval)
- timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
- if (!timeparms.to_retries)
- timeparms.to_retries = 5;
+ nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
/* create transport and client */
- xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
- &server->addr, &timeparms);
+ xprt = xprt_create_proto(proto, &server->addr, &timeparms);
if (IS_ERR(xprt)) {
dprintk("%s: cannot create RPC transport. Error = %ld\n",
__FUNCTION__, PTR_ERR(xprt));
@@ -1674,7 +1693,7 @@
struct rpc_clnt *clnt = NULL;
struct rpc_timeout timeparms;
rpc_authflavor_t authflavour;
- int proto, err = -EIO;
+ int err = -EIO;
sb->s_blocksize_bits = 0;
sb->s_blocksize = 0;
@@ -1692,30 +1711,8 @@
server->acdirmax = data->acdirmax*HZ;
server->rpc_ops = &nfs_v4_clientops;
- /* Initialize timeout values */
- timeparms.to_initval = data->timeo * HZ / 10;
- timeparms.to_retries = data->retrans;
- timeparms.to_exponential = 1;
- if (!timeparms.to_retries)
- timeparms.to_retries = 5;
-
- proto = data->proto;
- /* Which IP protocol do we use? */
- switch (proto) {
- case IPPROTO_TCP:
- timeparms.to_maxval = RPC_MAX_TCP_TIMEOUT;
- if (!timeparms.to_initval)
- timeparms.to_initval = 600 * HZ / 10;
- break;
- case IPPROTO_UDP:
- timeparms.to_maxval = RPC_MAX_UDP_TIMEOUT;
- if (!timeparms.to_initval)
- timeparms.to_initval = 11 * HZ / 10;
- break;
- default:
- return -EINVAL;
- }
+ nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
clp = nfs4_get_client(&server->addr.sin_addr);
if (!clp) {
@@ -1740,7 +1737,7 @@
down_write(&clp->cl_sem);
if (IS_ERR(clp->cl_rpcclient)) {
- xprt = xprt_create_proto(proto, &server->addr, &timeparms);
+ xprt = xprt_create_proto(data->proto, &server->addr, &timeparms);
if (IS_ERR(xprt)) {
up_write(&clp->cl_sem);
err = PTR_ERR(xprt);