From c2459dc46269728e4a080ec8d5a316b2bba2e142 Mon Sep 17 00:00:00 2001 From: Chuck Lever Date: Mon, 1 Feb 2010 14:17:14 -0500 Subject: NFS: Proper accounting for NFS VFS calls Nit: The VFSOPEN and VFSFLUSH counters are function call counters. Count every call to these routines. Signed-off-by: Chuck Lever Signed-off-by: Trond Myklebust --- fs/nfs/file.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'fs/nfs/file.c') diff --git a/fs/nfs/file.c b/fs/nfs/file.c index 63f2071d6445..57cf94f129ba 100644 --- a/fs/nfs/file.c +++ b/fs/nfs/file.c @@ -123,11 +123,11 @@ nfs_file_open(struct inode *inode, struct file *filp) filp->f_path.dentry->d_parent->d_name.name, filp->f_path.dentry->d_name.name); + nfs_inc_stats(inode, NFSIOS_VFSOPEN); res = nfs_check_flags(filp->f_flags); if (res) return res; - nfs_inc_stats(inode, NFSIOS_VFSOPEN); res = nfs_open(inode, filp); return res; } @@ -237,9 +237,9 @@ nfs_file_flush(struct file *file, fl_owner_t id) dentry->d_parent->d_name.name, dentry->d_name.name); + nfs_inc_stats(inode, NFSIOS_VFSFLUSH); if ((file->f_mode & FMODE_WRITE) == 0) return 0; - nfs_inc_stats(inode, NFSIOS_VFSFLUSH); /* Flush writes to the server and return any errors */ return nfs_do_fsync(ctx, inode); -- cgit v1.2.3 From 4184dcf2dbde481b34d370e1704f2b91a8c9f0d1 Mon Sep 17 00:00:00 2001 From: Chuck Lever Date: Mon, 1 Feb 2010 14:17:23 -0500 Subject: NFS: Fix byte accounting for generic NFS reads Currently, the NFS I/O counters count the number of bytes requested by applications, rather than the number of bytes actually read by the system calls. The number of bytes requested for reads is actually not that useful, because the value is usually a buffer size for reads. That is, that requested number is usually a maximum, and frequently doesn't reflect the actual number of bytes read. Signed-off-by: Chuck Lever Signed-off-by: Trond Myklebust --- fs/nfs/file.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'fs/nfs/file.c') diff --git a/fs/nfs/file.c b/fs/nfs/file.c index 57cf94f129ba..7f4910c98c7c 100644 --- a/fs/nfs/file.c +++ b/fs/nfs/file.c @@ -262,9 +262,11 @@ nfs_file_read(struct kiocb *iocb, const struct iovec *iov, (unsigned long) count, (unsigned long) pos); result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); - nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count); - if (!result) + if (!result) { result = generic_file_aio_read(iocb, iov, nr_segs, pos); + if (result > 0) + nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); + } return result; } -- cgit v1.2.3 From aa2f1ef10e6ad65c9138ec576f82c08f32e6f32c Mon Sep 17 00:00:00 2001 From: Chuck Lever Date: Mon, 1 Feb 2010 14:17:32 -0500 Subject: NFS: Account for NFS bytes read via the splice API Bytes read via the splice API should be accounted for in the NFS performance statistics. Signed-off-by: Chuck Lever Signed-off-by: Trond Myklebust --- fs/nfs/file.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'fs/nfs/file.c') diff --git a/fs/nfs/file.c b/fs/nfs/file.c index 7f4910c98c7c..abbc20281ea4 100644 --- a/fs/nfs/file.c +++ b/fs/nfs/file.c @@ -284,8 +284,11 @@ nfs_file_splice_read(struct file *filp, loff_t *ppos, (unsigned long) count, (unsigned long long) *ppos); res = nfs_revalidate_mapping(inode, filp->f_mapping); - if (!res) + if (!res) { res = generic_file_splice_read(filp, ppos, pipe, count, flags); + if (res > 0) + nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res); + } return res; } -- cgit v1.2.3 From 7e381172cf6e0282a56374e50667515aed55166a Mon Sep 17 00:00:00 2001 From: Chuck Lever Date: Mon, 1 Feb 2010 14:17:41 -0500 Subject: NFS: Improve NFS iostat byte count accuracy for writes The bytes counted by the performance counters for NFS writes should reflect write and sync errors. If the write(2) system call reports an error, the bytes should not be counted. And, if the write is short, the actual number of bytes that was written should be counted, not the number of bytes that was requested. Signed-off-by: Chuck Lever Signed-off-by: Trond Myklebust --- fs/nfs/file.c | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-) (limited to 'fs/nfs/file.c') diff --git a/fs/nfs/file.c b/fs/nfs/file.c index abbc20281ea4..ae8d02294e46 100644 --- a/fs/nfs/file.c +++ b/fs/nfs/file.c @@ -601,6 +601,7 @@ static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov, { struct dentry * dentry = iocb->ki_filp->f_path.dentry; struct inode * inode = dentry->d_inode; + unsigned long written = 0; ssize_t result; size_t count = iov_length(iov, nr_segs); @@ -627,14 +628,18 @@ static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov, if (!count) goto out; - nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count); result = generic_file_aio_write(iocb, iov, nr_segs, pos); + if (result > 0) + written = result; + /* Return error values for O_DSYNC and IS_SYNC() */ if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) { int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode); if (err < 0) result = err; } + if (result > 0) + nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); out: return result; @@ -649,6 +654,7 @@ static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe, { struct dentry *dentry = filp->f_path.dentry; struct inode *inode = dentry->d_inode; + unsigned long written = 0; ssize_t ret; dprintk("NFS splice_write(%s/%s, %lu@%llu)\n", @@ -659,14 +665,17 @@ static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe, * The combination of splice and an O_APPEND destination is disallowed. */ - nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count); - ret = generic_file_splice_write(pipe, filp, ppos, count, flags); + if (ret > 0) + written = ret; + if (ret >= 0 && nfs_need_sync_write(filp, inode)) { int err = nfs_do_fsync(nfs_file_open_context(filp), inode); if (err < 0) ret = err; } + if (ret > 0) + nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); return ret; } -- cgit v1.2.3 From d812e575822a2b7ab1a7cadae2571505ec6ec2bd Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Fri, 19 Mar 2010 13:55:17 -0400 Subject: NFS: Prevent another deadlock in nfs_release_page() We should not attempt to free the page if __GFP_FS is not set. Otherwise we can deadlock as per http://bugzilla.kernel.org/show_bug.cgi?id=15578 Signed-off-by: Trond Myklebust Cc: stable@kernel.org --- fs/nfs/file.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'fs/nfs/file.c') diff --git a/fs/nfs/file.c b/fs/nfs/file.c index ae8d02294e46..ae0d92736531 100644 --- a/fs/nfs/file.c +++ b/fs/nfs/file.c @@ -491,7 +491,8 @@ static int nfs_release_page(struct page *page, gfp_t gfp) { dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); - if (gfp & __GFP_WAIT) + /* Only do I/O if gfp is a superset of GFP_KERNEL */ + if ((gfp & GFP_KERNEL) == GFP_KERNEL) nfs_wb_page(page->mapping->host, page); /* If PagePrivate() is set, then the page is not freeable */ if (PagePrivate(page)) -- cgit v1.2.3