Hello Miklos, Thank you so much for the patch! I finally managed to build RHEL9 with your patch and I confirm that truncation of the symlinks is not happening anymore. However, I will need to do some further testing as I see that sometimes the updating of symlinks is not propagated correctly. This should not be a kernel problem. I believe this is more a CVMFS problem, its revision-based update process, that there is no way to pause user interaction with the fuse mount, and the asynchronous removal from inodes/dentries. Thanks again Laura ________________________________________ From: Al Viro <viro@xxxxxxxxxxxxxxxx> on behalf of Al Viro <viro@xxxxxxxxxxxxxxxxxx> Sent: Sunday, February 23, 2025 20:41 To: Miklos Szeredi <miklos@xxxxxxxxxx> Cc: Miklos Szeredi <mszeredi@xxxxxxxxxx>; linux-fsdevel@xxxxxxxxxxxxxxx <linux-fsdevel@xxxxxxxxxxxxxxx>; Christian Brauner <brauner@xxxxxxxxxx>; Bernd Schubert <bernd.schubert@xxxxxxxxxxx>; Laura Promberger <laura.promberger@xxxxxxx>; Sam Lewis <samclewis@xxxxxxxxxx> Subject: Re: [PATCH] fuse: don't truncate cached, mutated symlink On Sun, Feb 23, 2025 at 08:12:21PM +0100, Miklos Szeredi wrote: > On Sun, 23 Feb 2025 at 01:28, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote: > > > > On Thu, Feb 20, 2025 at 11:02:58AM +0100, Miklos Szeredi wrote: > > > > > The solution is to just remove this truncation. This can cause a > > > regression in a filesystem that relies on supplying a symlink larger than > > > the file size, but this is unlikely. If that happens we'd need to make > > > this behavior conditional. > > > > Note, BTW, that page *contents* must not change at all, so truncation is > > only safe if we have ->i_size guaranteed to be stable. > > > > Core pathwalk really counts upon the string remaining immutable, and that > > does include the string returned by ->get_link(). > > Page contents will not change after initial readlink, but page could > get invalidated and a fresh page filled with new value for the same > object. *nod* My point is that truncation of something that might be traversed by another pathwalk is a hard bug - we only get away with doing that in page_get_link() because it's idempotent. If ->i_size can change, we are not allowed to do that.
