[PATCH] NFS: Judge the file access cache's timestamp in rcu path

[PATCH] NFS: Judge the file access cache's timestamp in rcu path

[Chengen Du]

If the user's login time is newer than the cache's timestamp,
we expect the cache may be stale and need to clear.
The stale cache will remain in the list's tail if no other
users operate on that inode.
Once the user accesses the inode, the stale cache will be
returned in rcu path.

Signed-off-by: Chengen Du <chengen.du@canonical.com>
---
 fs/nfs/dir.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c
index ea1ceffa1d3a..d490d64a9ebd 100644
--- a/fs/nfs/dir.c
+++ b/fs/nfs/dir.c
@@ -3023,6 +3023,7 @@ static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cre
 	 * but do it without locking.
 	 */
 	struct nfs_inode *nfsi = NFS_I(inode);
+	u64 login_time = nfs_access_login_time(current, cred);
 	struct nfs_access_entry *cache;
 	int err = -ECHILD;
 	struct list_head *lh;
@@ -3037,6 +3038,8 @@ static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cre
 		cache = NULL;
 	if (cache == NULL)
 		goto out;
+	if ((s64)(login_time - cache->timestamp) > 0)
+		goto out;
 	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
 		goto out;
 	*mask = cache->mask;
-- 
2.37.2

Re: [PATCH] NFS: Judge the file access cache's timestamp in rcu path

[Chengen Du]

Hi,

The patch follows the commit 0eb43812c027 but handles the stale cache
in rcu path.
The stale cache will remain in the list's tail if no other users
operate on that inode.
May I ask if this patch has any other concerns or drawbacks that need
to be considered?

Best regards,
Chengen Du

On Fri, Dec 30, 2022 at 11:05 AM Chengen Du <chengen.du@canonical.com> wrote:
>
> If the user's login time is newer than the cache's timestamp,
> we expect the cache may be stale and need to clear.
> The stale cache will remain in the list's tail if no other
> users operate on that inode.
> Once the user accesses the inode, the stale cache will be
> returned in rcu path.
>
> Signed-off-by: Chengen Du <chengen.du@canonical.com>
> ---
>  fs/nfs/dir.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c
> index ea1ceffa1d3a..d490d64a9ebd 100644
> --- a/fs/nfs/dir.c
> +++ b/fs/nfs/dir.c
> @@ -3023,6 +3023,7 @@ static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cre
>          * but do it without locking.
>          */
>         struct nfs_inode *nfsi = NFS_I(inode);
> +       u64 login_time = nfs_access_login_time(current, cred);
>         struct nfs_access_entry *cache;
>         int err = -ECHILD;
>         struct list_head *lh;
> @@ -3037,6 +3038,8 @@ static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cre
>                 cache = NULL;
>         if (cache == NULL)
>                 goto out;
> +       if ((s64)(login_time - cache->timestamp) > 0)
> +               goto out;
>         if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
>                 goto out;
>         *mask = cache->mask;
> --
> 2.37.2
>

Re: [PATCH] NFS: Judge the file access cache's timestamp in rcu path

[Trond Myklebust]

On Thu, 2023-01-12 at 09:26 +0800, Chengen Du wrote:
> Hi,
> 
> The patch follows the commit 0eb43812c027 but handles the stale cache
> in rcu path.
> The stale cache will remain in the list's tail if no other users
> operate on that inode.
> May I ask if this patch has any other concerns or drawbacks that need
> to be considered?
> 
> Best regards,
> Chengen Du

No complaints. The patch has already been merged into 6.2-rc3.

Thanks!
  Trond

> 
> On Fri, Dec 30, 2022 at 11:05 AM Chengen Du
> <chengen.du@canonical.com> wrote:
> > 
> > If the user's login time is newer than the cache's timestamp,
> > we expect the cache may be stale and need to clear.
> > The stale cache will remain in the list's tail if no other
> > users operate on that inode.
> > Once the user accesses the inode, the stale cache will be
> > returned in rcu path.
> > 
> > Signed-off-by: Chengen Du <chengen.du@canonical.com>
> > ---
> >  fs/nfs/dir.c | 3 +++
> >  1 file changed, 3 insertions(+)
> > 
> > diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c
> > index ea1ceffa1d3a..d490d64a9ebd 100644
> > --- a/fs/nfs/dir.c
> > +++ b/fs/nfs/dir.c
> > @@ -3023,6 +3023,7 @@ static int nfs_access_get_cached_rcu(struct
> > inode *inode, const struct cred *cre
> >          * but do it without locking.
> >          */
> >         struct nfs_inode *nfsi = NFS_I(inode);
> > +       u64 login_time = nfs_access_login_time(current, cred);
> >         struct nfs_access_entry *cache;
> >         int err = -ECHILD;
> >         struct list_head *lh;
> > @@ -3037,6 +3038,8 @@ static int nfs_access_get_cached_rcu(struct
> > inode *inode, const struct cred *cre
> >                 cache = NULL;
> >         if (cache == NULL)
> >                 goto out;
> > +       if ((s64)(login_time - cache->timestamp) > 0)
> > +               goto out;
> >         if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
> >                 goto out;
> >         *mask = cache->mask;
> > --
> > 2.37.2
> > 

-- 
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@hammerspace.com