Here's the output when overlayfs try to do a copy up with the original function calls order and with debug enabled: root@io2200:/etc# mv machine-id machine-id1 mv: cannot move 'machine-id' to 'machine-id1': No such file or directory root@io2200:/etc# dmesg [ 2999.275275] tmpfile(upper.work/work, 0100644) = 0 [ 2999.278789] setxattr(work/#232, "trusted.overlay.origin", "(null)", 0x0) = -2 And here's with inverted function calls: root@io2200:/etc# mv securetty securetty1 root@io2200:/etc# dmesg [ 659.515200] tmpfile(upper.work/work, 0100400) = 0 [ 659.515422] link(work/#164, etc/securetty) = 0 [ 659.518577] setxattr(work/#164, "trusted.overlay.origin", "(null)", 0x0) = 0 [ 659.522153] rename(etc/securetty, etc/securetty1, 0x4) Thanks, Vincent Guérard -----Original Message----- From: Miklos Szeredi [mailto:miklos@xxxxxxxxxx] Sent: December 10, 2018 4:27 AM To: Guerard, Vincent Subject: [EXTERNAL] Re: Copy_up function fail with ubifs on upper layer On Mon, Dec 3, 2018 at 8:51 PM Guerard, Vincent <VincentGuerard@xxxxxxxxx> wrote: > > Hello, > > > > I’m using the Linux kernel 4.14 of Texas Instruments, which have this > patch in the ubifs file system code: > https://lore.kernel.org/patchwork/patch/960502/ > > I’m using a ubifs file system as the upper layer of an overlay, so when overlayfs try to do a copy up of a file, it fail with ENOENT. Could you please enable debugging with: echo "file fs/overlayfs/* +p" > <debugfs>/dynamic_debug/control ? Also would you mind adding <linux-unionfs@xxxxxxxxxxxxxxx> to the discussion? > > The workaround I found was to invert the call of copy_up_inode and ovl_install_temp in the function ovl_copy_up_locked. > > > > I’ve run the basic overlayfs tests which I found at : https://github.com/amir73il/overlayfs/wiki/Overlayfs-testing and all tests passed. > > > > So my questions are : > > Is it okay to invert these call in this case? Inverting the calls make the copy-up operation non-atomic. This shouldn't normally be an issue, only when e.g. power is lost in the middle of the operation and after reboot the overlay will be corrupted. Thanks, Miklos
