diff mbox series

fs: don't overwrite the nsec field if I_CTIME_QUERIED is already set

Message ID 20240809-mgtime-v1-1-b2cab4f1558d@kernel.org (mailing list archive)
State New
Headers show
Series fs: don't overwrite the nsec field if I_CTIME_QUERIED is already set | expand

Commit Message

Jeff Layton Aug. 9, 2024, 1:39 p.m. UTC 
When fetching the ctime's nsec value for a stat-like operation, do a
simple fetch first and avoid the atomic_fetch_or if the flag is already
set.

Suggested-by: Mateusz Guzik <mjguzik@gmail.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
I'm running tests on this now, but I don't expect any problems.

This is based on top of Christian's vfs.mgtime branch. It may be best to
squash this into 6feb43ecdd8e ("fs: add infrastructure for multigrain
timestamps").
---
To: Alexander Viro <viro@zeniv.linux.org.uk>
To: Christian Brauner <brauner@kernel.org>
To: Jan Kara <jack@suse.cz>
Cc: linux-fsdevel@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
---
 fs/stat.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)


---
base-commit: 9000eec2bdc08a0b9027427f9d3d9a3545694258
change-id: 20240809-mgtime-cbc2aa6dc0fe

Best regards,

Comments

Christian Brauner Aug. 9, 2024, 2:55 p.m. UTC | #1 
On Fri, Aug 09, 2024 at 09:39:43AM GMT, Jeff Layton wrote:
> When fetching the ctime's nsec value for a stat-like operation, do a
> simple fetch first and avoid the atomic_fetch_or if the flag is already
> set.
> 
> Suggested-by: Mateusz Guzik <mjguzik@gmail.com>
> Signed-off-by: Jeff Layton <jlayton@kernel.org>
> ---
> I'm running tests on this now, but I don't expect any problems.
> 
> This is based on top of Christian's vfs.mgtime branch. It may be best to
> squash this into 6feb43ecdd8e ("fs: add infrastructure for multigrain
> timestamps").

Squashed it. Can you double-check that things look correct?
Jeff Layton Aug. 9, 2024, 3:05 p.m. UTC | #2 
On Fri, 2024-08-09 at 16:55 +0200, Christian Brauner wrote:
> On Fri, Aug 09, 2024 at 09:39:43AM GMT, Jeff Layton wrote:
> > When fetching the ctime's nsec value for a stat-like operation, do
> > a
> > simple fetch first and avoid the atomic_fetch_or if the flag is
> > already
> > set.
> > 
> > Suggested-by: Mateusz Guzik <mjguzik@gmail.com>
> > Signed-off-by: Jeff Layton <jlayton@kernel.org>
> > ---
> > I'm running tests on this now, but I don't expect any problems.
> > 
> > This is based on top of Christian's vfs.mgtime branch. It may be
> > best to
> > squash this into 6feb43ecdd8e ("fs: add infrastructure for
> > multigrain
> > timestamps").
> 
> Squashed it. Can you double-check that things look correct?

One minor issue in fill_mg_cmtime:

-------------8<-----------------
        if (!(stat->ctime.tv_nsec & I_CTIME_QUERIED))
                stat->ctime.tv_nsec = ((u32)atomic_fetch_or(I_CTIME_QUERIED, pcn));
        trace_fill_mg_cmtime(inode, &stat->ctime, &stat->mtime);
        stat->ctime.tv_nsec &= ~I_CTIME_QUERIED;
}
-------------8<-----------------

I'd swap the last two lines of the function. We print the ctime in the
tracepoint as a timestamp, so if the QUERIED bit is present it's going
to look funny. We _know_ that it's flagged after this function, so
leaving it set is not terribly helpful.
Christian Brauner Aug. 9, 2024, 3:17 p.m. UTC | #3 
On Fri, Aug 09, 2024 at 11:05:54AM GMT, Jeff Layton wrote:
> On Fri, 2024-08-09 at 16:55 +0200, Christian Brauner wrote:
> > On Fri, Aug 09, 2024 at 09:39:43AM GMT, Jeff Layton wrote:
> > > When fetching the ctime's nsec value for a stat-like operation, do
> > > a
> > > simple fetch first and avoid the atomic_fetch_or if the flag is
> > > already
> > > set.
> > > 
> > > Suggested-by: Mateusz Guzik <mjguzik@gmail.com>
> > > Signed-off-by: Jeff Layton <jlayton@kernel.org>
> > > ---
> > > I'm running tests on this now, but I don't expect any problems.
> > > 
> > > This is based on top of Christian's vfs.mgtime branch. It may be
> > > best to
> > > squash this into 6feb43ecdd8e ("fs: add infrastructure for
> > > multigrain
> > > timestamps").
> > 
> > Squashed it. Can you double-check that things look correct?
> 
> One minor issue in fill_mg_cmtime:
> 
> -------------8<-----------------
>         if (!(stat->ctime.tv_nsec & I_CTIME_QUERIED))
>                 stat->ctime.tv_nsec = ((u32)atomic_fetch_or(I_CTIME_QUERIED, pcn));
>         trace_fill_mg_cmtime(inode, &stat->ctime, &stat->mtime);
>         stat->ctime.tv_nsec &= ~I_CTIME_QUERIED;
> }
> -------------8<-----------------
> 
> I'd swap the last two lines of the function. We print the ctime in the
> tracepoint as a timestamp, so if the QUERIED bit is present it's going
> to look funny. We _know_ that it's flagged after this function, so
> leaving it set is not terribly helpful.

Bah, fat-fingered on my end. I did intend to place it after. Thanks!
diff mbox series

Patch

diff --git a/fs/stat.c b/fs/stat.c
index f2bf7cca64b2..9eb6d9b2d010 100644
--- a/fs/stat.c
+++ b/fs/stat.c
@@ -35,8 +35,8 @@ 
  * @inode: inode from which to grab the c/mtime
  *
  * Given @inode, grab the ctime and mtime out if it and store the result
- * in @stat. When fetching the value, flag it as queried so the next write
- * will ensure a distinct timestamp.
+ * in @stat. When fetching the value, flag it as QUERIED (if not already)
+ * so the next write will record a distinct timestamp.
  */
 void fill_mg_cmtime(struct kstat *stat, u32 request_mask, struct inode *inode)
 {
@@ -50,7 +50,10 @@  void fill_mg_cmtime(struct kstat *stat, u32 request_mask, struct inode *inode)
 
 	stat->mtime = inode_get_mtime(inode);
 	stat->ctime.tv_sec = inode->i_ctime_sec;
-	stat->ctime.tv_nsec = ((u32)atomic_fetch_or(I_CTIME_QUERIED, pcn)) & ~I_CTIME_QUERIED;
+	stat->ctime.tv_nsec = (u32)atomic_read(pcn);
+	if (!(stat->ctime.tv_nsec & I_CTIME_QUERIED))
+		stat->ctime.tv_nsec = ((u32)atomic_fetch_or(I_CTIME_QUERIED, pcn));
+	stat->ctime.tv_nsec &= ~I_CTIME_QUERIED;
 	trace_fill_mg_cmtime(inode, &stat->ctime, &stat->mtime);
 }
 EXPORT_SYMBOL(fill_mg_cmtime);