Re: [PATCH] nfsd: make a copy of struct iattr before calling notify_change

[Jeff Layton]

On Wed, 2023-05-17 at 19:13 +0000, Chuck Lever III wrote:
> 
> > On May 17, 2023, at 3:05 PM, Jeff Layton <jlayton@xxxxxxxxxx> wrote:
> > 
> > On Wed, 2023-05-17 at 17:47 +0000, Chuck Lever III wrote:
> > > 
> > > > On May 17, 2023, at 12:26 PM, Jeff Layton <jlayton@xxxxxxxxxx> wrote:
> > > > 
> > > > notify_change can modify the iattr structure. In particular it can can
> > > > end up setting ATTR_MODE when ATTR_KILL_SUID is already set, causing a
> > > > BUG() if the same iattr is passed to notify_change more than once.
> > > > 
> > > > Make a copy of the struct iattr before calling notify_change.
> > > > 
> > > > Fixes: 34b91dda7124 NFSD: Make nfsd4_setattr() wait before returning NFS4ERR_DELAY
> > > > Link: https://bugzilla.redhat.com/show_bug.cgi?id=2207969
> > > > Reported-by: Zhi Li <yieli@xxxxxxxxxx>
> > > > Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx>
> > > > ---
> > > > fs/nfsd/vfs.c | 4 +++-
> > > > 1 file changed, 3 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
> > > > index c4ef24c5ffd0..ad0c5cd900b1 100644
> > > > --- a/fs/nfsd/vfs.c
> > > > +++ b/fs/nfsd/vfs.c
> > > > @@ -538,7 +538,9 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp,
> > > > 
> > > > inode_lock(inode);
> > > > for (retries = 1;;) {
> > > > - host_err = __nfsd_setattr(dentry, iap);
> > > > + struct iattr attrs = *iap;
> > > 
> > > This construct always makes me queazy. I'm never sure if an
> > > initializer inside a loop is "only once" or "every time". I
> > > fixed a bug like this once.
> > > 
> > > But if you've tested it and it addresses the BUG, then let's
> > > go with this. I can apply it to nfsd-fixes.
> > > 
> > 
> > 
> > I've done some light testing with this kernel, but this was found by Zhi
> > while testing with the lustre racer test, so it involves some raciness.
> > I've never hit this myself.
> 
> Has Zhi tested this fix?
> 

Not yet. I just cooked it up this morning. I built a test kernel but
testing it will take some time since it depends on load.

> 
> > I'm pretty sure though that this has to be initialized every time. The
> > assignment is inside the loop, after all. I'm ok with moving the
> > assignment to a different line if you like though:
> > 
> > struct iattr attrs;
> > 
> > attrs = *iap;
> > ...
> 
> Yeah I could do that. I find that easier to read when a loop is
> involved; it's unambiguous then what is going on.
> 

Your call. I'm fairly certain that the patch does the right thing as-is,
but if you think it makes it more readable, then OK.
-- 
Jeff Layton <jlayton@xxxxxxxxxx>