+David Gow - just FYI since you are the only KUnit person who touched VFAT. PTAL at other emails in thread too, please.

On Fri, Sep 10, 2021 at 4:57 AM Arthur Marsh arthur.marsh@internode.on.net wrote:

Hi, I have been sharing an old VFAT formatted hard disk on one pc to another using Samba and sometime after kernel 5.14.0 it stopped working (apparently no longer being shared as the mount.smbfs command on the client failed with error -13 yet mount.smbfs still worked for ext3 filesytems shared from the same machine which had the VFAT filesystem). The only error I saw on the machine with the VFAT formatted hard disk was the output of the mount command had truncated the name of the mount to only include the first 4 characters of the base name of the mount point. e.g. when VFAT filesystem was mounted on /mnt/victoria, the output of the mount command showed the filesytem mounted on /mnt/vict

The kernel build used was i386 with gcc 11.2.0-4 using

make - j2 menuconfig bindeb-pkg

.config available on request.

The git-bisect was: victoria:/usr/src/linux# git bisect loggit bisect start '--' 'fs/fat'# good: [7d2a07b769330c34b4deabeed939325c77a7ec2f] Linux 5.14git bisect good 7d2a07b769330c34b4deabeed939325c77a7ec2f# bad: [a3fa7a101dcff93791d1b1bdb3affcad1410c8c1] Merge branches 'akpm' and 'akpm-hotfixes' (patches from Andrew)git bisect bad a3fa7a101dcff93791d1b1bdb3affcad1410c8c1# good: [edb0872f44ec9976ea6d052cb4b93cd2d23ac2ba] block: move the bdi from the request_queue to the gendiskgit bisect good edb0872f44ec9976ea6d052cb4b93cd2d23ac2ba# good: [b0d4adaf3b3c4402d9c3b6186e02aa1e4f7985cd] fat: Add KUnit tests for checksums and timestampsgit bisect good b0d4adaf3b3c4402d9c3b6186e02aa1e4f7985cd# bad: [c815f04ba94940fbc303a6ea9669e7da87f8e77d] Merge tag 'linux-kselftest-kunit-5.15-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftestgit bisect bad c815f04ba94940fbc303a6ea9669e7da87f8e77d# first bad commit: [c815f04ba94940fbc303a6ea9669e7da87f8e77d] Merge tag 'linux-kselftest-kunit-5.15-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest

amarsh04@victoria:~$ mount|grep vic/dev/sdb6 on /vict type vfat (rw,relatime,uid=65534,gid=65534,fmask=0000,dmask=0000,allow_utime=0022,codepage=437,iocharset=utf8,shortname=mixed,errors=remount-ro)

I don't see how our patch could have caused this either. Nevertheless, I CC'ed David who has worked on VFAT who may be able to provide more info.

Happy to run any further tests but kernel builds are slow on this machine (Pentium Dl.

Arthur.

On Fri, Sep 10, 2021 at 8:02 PM Arthur Marsh arthur.marsh@internode.on.net wrote:

Hi, I have been sharing an old VFAT formatted hard disk on one pc to another using Samba and sometime after kernel 5.14.0 it stopped working (apparently no longer being shared as the mount.smbfs command on the client failed with error -13 yet mount.smbfs still worked for ext3 filesytems shared from the same machine which had the VFAT filesystem). The only error I saw on the machine with the VFAT formatted hard disk was the output of the mount command had truncated the name of the mount to only include the first 4 characters of the base name of the mount point. e.g. when VFAT filesystem was mounted on /mnt/victoria, the output of the mount command showed the filesytem mounted on /mnt/vict

I can't reproduce this on my machine (which is openSUSE Tumbleweed with their "vanilla" 5.14 kernel package on x86_64, mounting a FAT16 filesystem).

# mount /dev/sda1 /mnt/victoria # mount | grep vic /dev/sda1 on /mnt/victoria type vfat (rw,relatime,fmask=0022,dmask=0022,codepage=437,iocharset=iso8859-1,shortname=mixed,errors=remount-ro) # uname -a Linux patpat 5.14.0-1-vanilla #1 SMP Mon Aug 30 07:01:36 UTC 2021 (dc06e24) x86_64 x86_64 x86_64 GNU/Linux

I can try it again on an older i386 machine, but I doubt that'd change things: this doesn't smell architecture-specific to me.

This seems a lot more like it's something to do with /proc/mounts or similar, rather than a FAT specific issue (and, unless something really strange has happened with the CONFIG_FAT_DEFAULT_CODEPAGE config option, which I doubt), this change shouldn't affect anything at all when KUnit isn't enabled and used. I suspect it just shows up in the bisect because it's basically the only change in fs/fat for a while.

The bisect against the whole kernel tree seems likely to be of more use.

-- David