[Devel] [PATCH] nfsd: return file system superblock time granulaty on FSINFO request
Stanislav Kinsburskiy
skinsbursky at virtuozzo.com
Fri Jun 30 20:16:42 MSK 2017
Current NFS client implementation can zap inode caches or just re-validate the
inode after successful lock request using it as a synchronization point.
Inode caches zapping or re-validation depends on server's file system time
granularity status: if time granularity is less than second, then
re-validation
is assumed to be enough. But Linux NFS server always return "1 second" value
for any file system. This looks like an old artefact.
So, let's so the following: return "sb->s_time_gran" to NFS client.
But only for EXT4 file system for a while.
This will remove bottleneck when many processes are concurrently trying to
lock large file like SQL database.
Ported from rhel6.
https://jira.sw.ru/browse/PSBM-66572
Signed-off-by: Stanislav Kinsburskiy <skinsbursky at virtuozzo.com>
---
fs/nfsd/nfs3proc.c | 3 +++
fs/nfsd/nfs3xdr.c | 9 +++++++--
fs/nfsd/xdr3.h | 1 +
3 files changed, 11 insertions(+), 2 deletions(-)
diff --git a/fs/nfsd/nfs3proc.c b/fs/nfsd/nfs3proc.c
index 88afc45..3199757 100644
--- a/fs/nfsd/nfs3proc.c
+++ b/fs/nfsd/nfs3proc.c
@@ -559,6 +559,9 @@ nfsd3_proc_fsinfo(struct svc_rqst * rqstp, struct nfsd_fhandle *argp,
resp->f_properties = NFS3_FSF_BILLYBOY;
}
resp->f_maxfilesize = sb->s_maxbytes;
+ resp->f_time_gran = 0;
+ if (!strcmp(sb->s_type->name, "ext4"))
+ resp->f_time_gran = sb->s_time_gran;
}
fh_put(&argp->fh);
diff --git a/fs/nfsd/nfs3xdr.c b/fs/nfsd/nfs3xdr.c
index 73d21f0..ff41760 100644
--- a/fs/nfsd/nfs3xdr.c
+++ b/fs/nfsd/nfs3xdr.c
@@ -1057,8 +1057,13 @@ nfs3svc_encode_fsinfores(struct svc_rqst *rqstp, __be32 *p,
*p++ = htonl(resp->f_wtmult);
*p++ = htonl(resp->f_dtpref);
p = xdr_encode_hyper(p, resp->f_maxfilesize);
- *p++ = xdr_one;
- *p++ = xdr_zero;
+ if (resp->f_time_gran) {
+ *p++ = xdr_zero;
+ *p++ = htonl(resp->f_time_gran);
+ } else {
+ *p++ = xdr_one;
+ *p++ = xdr_zero;
+ }
*p++ = htonl(resp->f_properties);
}
diff --git a/fs/nfsd/xdr3.h b/fs/nfsd/xdr3.h
index 335e04a..2744607 100644
--- a/fs/nfsd/xdr3.h
+++ b/fs/nfsd/xdr3.h
@@ -206,6 +206,7 @@ struct nfsd3_fsinfores {
__u32 f_dtpref;
__u64 f_maxfilesize;
__u32 f_properties;
+ __u32 f_time_gran;
};
struct nfsd3_pathconfres {
More information about the Devel
mailing list