On Thu, Dec 17, 2020 at 9:46 PM Michael Labriola
<michael.d.labriola@xxxxxxxxx> wrote:
>
> On Thu, Dec 17, 2020 at 1:07 PM Amir Goldstein <amir73il@xxxxxxxxx> wrote:
> >
> > On Thu, Dec 17, 2020 at 6:22 PM Michael Labriola
> *snip*
> > > On Thu, Dec 17, 2020 at 7:00 AM Amir Goldstein <amir73il@xxxxxxxxx> wrote:
> > > Thanks, Amir. I didn't have CONFIG_DYNAMIC_DEBUG enabled, so
> >
> > I honestly don't expect to find much in the existing overlay debug prints
> > but you never know..
> > I suspect you will have to add debug prints to find the problem.
>
> Ok, here goes. I had to setup a new virtual machine that doesn't use
> overlayfs for its root filesystem because turning on dynamic debug
> gave way too much output for a nice controlled test. It's exhibiting
> the same behavior as my previous tests (5.8 good, 5.9 bad). The is
> with a freshly compiled 5.9.15 w/ CONFIG_OVERLAY_FS_XINO_AUTO turned
> off and CONFIG_DYNAMIC_DEBUG turned on. Here's what we get:
>
> echo "file fs/overlayfs/* +p" > /sys/kernel/debug/dynamic_debug/control
> mount borky2.sqsh t
> mount -t tmpfs tmp tt
> mkdir -p tt/upper/{upper,work}
> mount -t overlay -o \
> lowerdir=t,upperdir=tt/upper/upper,workdir=tt/upper/work blarg ttt
> [ 164.505193] overlayfs: mkdir(work/work, 040000) = 0
> [ 164.505204] overlayfs: tmpfile(work/work, 0100000) = 0
> [ 164.505209] overlayfs: create(work/#3, 0100000) = 0
> [ 164.505210] overlayfs: rename(work/#3, work/#4, 0x4)
> [ 164.505216] overlayfs: unlink(work/#3) = 0
> [ 164.505217] overlayfs: unlink(work/#4) = 0
> [ 164.505221] overlayfs: setxattr(work/work,
> "trusted.overlay.opaque", "0", 1, 0x0) = 0
>
> touch ttt/FOO
> touch: cannot touch 'ttt/FOO': No data available
> [ 191.919498] overlayfs: setxattr(upper/upper,
> "trusted.overlay.impure", "y", 1, 0x0) = 0
> [ 191.919523] overlayfs: tmpfile(work/work, 0100644) = 0
> [ 191.919788] overlayfs: tmpfile(work/work, 0100644) = 0
>
> That give you any hints? I'll start reading through the overlayfs
> code. I've never actually looked at it, so I'll be planting printk
> calls at random. ;-)
We have seen that open("FOO", O_WRONLY) fails
We know that FOO is lower at that time so that brings us to
ovl_open
ovl_maybe_copy_up
ovl_copy_up_flags
ovl_copy_up_one
ovl_do_copy_up
ovl_set_impure
[ 191.919498] overlayfs: setxattr(upper/upper,
"trusted.overlay.impure", "y", 1, 0x0) = 0
ovl_copy_up_tmpfile
ovl_do_tmpfile
[ 191.919523] overlayfs: tmpfile(work/work, 0100644) = 0
ovl_copy_up_inode
This must be were we fail and likely in:
ovl_copy_xattr
vfs_getxattr
which can return -ENODATA, but it is not expected because the
xattrs returned by vfs_listxattr should exist...
So first guess would be to add a debug print for xattr 'name'
and return value of vfs_getxattr().
Good luck!
Thanks,
Amir.
