On Tue, Sep 24, 2024 at 7:25 AM Krishna Vivek Vitta <kvitta@xxxxxxxxxxxxx> wrote: > > Hi Amir > > Thanks for the reply. > > We have another image with kernel version: 6.6.36.3. git clone operation fails there as well. Do we need to still try with 5.15.154 kernel version ? No need. > > Currently, we are marking the mount points with mask(FAN_CLOSE_WRITE) to handle only close_write events. Do we need to add any other flag in mask and check ? No need. > > Following is the mount entry in /proc/mounts file: > C:\134 /mnt/c 9p rw,noatime,aname=drvfs;path=C:\;uid=0;gid=0;symlinkroot=/mnt/,cache=5,access=client,msize=65536,trans=fd,rfd=4,wfd=4 0 0 I don't know this symlinkroot feature. It looks like a WSL2 feature (?) and my guess is that the failure might be related. Not sure how fanotify mount mark affects this, maybe because the close_write events open the file for reporting the event, but maybe you should try to ask your question also the WSL2 kernel maintainers. I have tried to reproduce your test case on the 9p mount on my test box: v_tmp on /vtmp type 9p (rw,relatime,access=client,msize=262144,trans=virtio) with fanotify examples program: https://manpages.debian.org/unstable/manpages/fanotify.7.en.html#Example_program:_fanotify_example.c and could not reproduce the issue with plain: echo 123 > x && mv x y && cat y > > Attached is the strace for failed git clone operation(line: 419, 420). All the failures are ENOENT, which is why I suspect maybe related to the symlinkroot thing. > Even I wrote a small program to invoke rename, followed with open. > The open fails immediately and succeeds after 3-4 iterations. > This exercise was performed on p9 file system marked with fanotify. Please share your reproducer program. The difference could be in the details. Can you test in a 9p mount without those WSL options? Can you test on upstream or LTS kernel? Can you test with the fanotify example? > > Am not reporting this as regression. We havent checked this behavior before. > Ok. patience, we will try to get to the bottom of this. Thanks, Amir.
