On Thu, May 8, 2025 at 10:38 PM Amir Goldstein <amir73il@xxxxxxxxx> wrote: > > On Thu, May 8, 2025 at 9:38 PM Zorro Lang <zlang@xxxxxxxxxx> wrote: > > > > On Wed, Apr 09, 2025 at 01:52:20PM +0200, Amir Goldstein wrote: > > > This is a variant of generic/477 with connectable file handles. > > > This test uses load and store of file handles from a temp file to test > > > decoding connectable file handles after cycle mount and after renames. > > > Decoding connectable file handles after being moved to a new parent > > > is expected to fail. > > > > > > Signed-off-by: Amir Goldstein <amir73il@xxxxxxxxx> > > > --- > > > > Hi Amir, > > > > This test case fails on some filesystems, e.g. nfs [1] and tmpfs [2]. > > Is this as your expected? > > No. I will look into this failure. So what happens is that on filesystems that really get unmounted in mount cycle, trying to open a connectable file handle after file was moved to another parent is expected to fail and this is what the test expects, but tmpfs does not really get unmounted on mount cycle so opening the file handle does work. Similarly for nfs, the file remains in cache on the local server so opening by handle works. I have removed that test case in v2 because it is unpredictable and added verification that opened files have a connected path. Thanks, Amir.
