A recent kernel change has change the behavior of chmod on fat filesystems:
commit 19c561a60ffe52df88dd63de0bff480ca094efe4 Author: Jan Engelhardt jengelh@computergmbh.de Date: Wed Feb 6 01:36:08 2008 -0800
fs/fat/: refine chmod checks
After this change rpm installs of kernels fail on ia64. This is because efi systems use vfat for the boot filesystem. RPM installs fail because cpio gets a error on chmod (before this kernel change it would fail silently).
My suggestion was to have anaconda set the "quiet" option (which makes chmod failures silent) for /boot/efi in fstab however they disagree (see BZ 435319).
So, it seems like the options are: revert the patch (probably not a good idea) make "quiet" the default for fat filesystems, or, convince anaconda to add "quiet" to the fstab. Or, hopefully someone else has a better idea.
thanks,
- Doug
On Thu, 2008-02-28 at 16:03 -0500, Doug Chapman wrote:
A recent kernel change has change the behavior of chmod on fat filesystems:
commit 19c561a60ffe52df88dd63de0bff480ca094efe4 Author: Jan Engelhardt jengelh@computergmbh.de Date: Wed Feb 6 01:36:08 2008 -0800
fs/fat/: refine chmod checksAfter this change rpm installs of kernels fail on ia64. This is because efi systems use vfat for the boot filesystem. RPM installs fail because cpio gets a error on chmod (before this kernel change it would fail silently).
My suggestion was to have anaconda set the "quiet" option (which makes chmod failures silent) for /boot/efi in fstab however they disagree (see BZ 435319).
So, it seems like the options are: revert the patch (probably not a good idea) make "quiet" the default for fat filesystems, or, convince anaconda to add "quiet" to the fstab. Or, hopefully someone else has a better idea.
thanks,
- Doug
Actually I came up with what I think is a cleaner fix for this. Since the default file permission on files on vfat are 755 anyway if the kernel is mode 755 rpm doesn't complain.
Anybody have thoughts on this specfile change? I build this as a scratch build on our ia64 koji server and it installs cleanly.
- Doug
*** kernel.spec.bad 2008-02-28 19:58:55.000000000 -0500 --- kernel.spec 2008-02-28 21:39:57.000000000 -0500 *************** BuildKernel() { *** 1301,1306 **** --- 1301,1310 ---- $CopyKernel $KernelImage \ $RPM_BUILD_ROOT/%{image_install_path}/$InstallName-$KernelVer
+ %ifarch ia64 + chmod 755 $RPM_BUILD_ROOT/%{image_install_path}/$InstallName-$KernelVer + %endif + mkdir -p $RPM_BUILD_ROOT/lib/modules/$KernelVer make -s ARCH=$Arch INSTALL_MOD_PATH=$RPM_BUILD_ROOT modules_install KERNELRELEASE=$KernelVer %ifarch %{vdso_arches}
On Thu, Feb 28, 2008 at 10:00:32PM -0500, Doug Chapman wrote:
Actually I came up with what I think is a cleaner fix for this. Since the default file permission on files on vfat are 755 anyway if the kernel is mode 755 rpm doesn't complain.
Anybody have thoughts on this specfile change? I build this as a scratch build on our ia64 koji server and it installs cleanly.
- Doug
*** kernel.spec.bad 2008-02-28 19:58:55.000000000 -0500 --- kernel.spec 2008-02-28 21:39:57.000000000 -0500 *************** BuildKernel() { *** 1301,1306 **** --- 1301,1310 ---- $CopyKernel $KernelImage \ $RPM_BUILD_ROOT/%{image_install_path}/$InstallName-$KernelVer
- %ifarch ia64
chmod 755 $RPM_BUILD_ROOT/%{image_install_path}/$InstallName-$KernelVer- %endif
mkdir -p $RPM_BUILD_ROOT/lib/modules/$KernelVer make -s ARCH=$Arch INSTALL_MOD_PATH=$RPM_BUILD_ROOT modules_install KERNELRELEASE=$KernelVer
There are systems with EFI32 and EFI64 out there, that aren't ia64, but that will likewise be dropping files into a vfat file system.
On Fri, Feb 29, 2008 at 10:16:00AM -0600, Matt Domsch wrote:
On Thu, Feb 28, 2008 at 10:00:32PM -0500, Doug Chapman wrote:
Actually I came up with what I think is a cleaner fix for this. Since the default file permission on files on vfat are 755 anyway if the kernel is mode 755 rpm doesn't complain.
Anybody have thoughts on this specfile change? I build this as a scratch build on our ia64 koji server and it installs cleanly.
- Doug
*** kernel.spec.bad 2008-02-28 19:58:55.000000000 -0500 --- kernel.spec 2008-02-28 21:39:57.000000000 -0500 *************** BuildKernel() { *** 1301,1306 **** --- 1301,1310 ---- $CopyKernel $KernelImage \ $RPM_BUILD_ROOT/%{image_install_path}/$InstallName-$KernelVer
- %ifarch ia64
chmod 755 $RPM_BUILD_ROOT/%{image_install_path}/$InstallName-$KernelVer- %endif
mkdir -p $RPM_BUILD_ROOT/lib/modules/$KernelVer make -s ARCH=$Arch INSTALL_MOD_PATH=$RPM_BUILD_ROOT modules_install KERNELRELEASE=$KernelVerThere are systems with EFI32 and EFI64 out there, that aren't ia64, but that will likewise be dropping files into a vfat file system.
I don't see any problem in unconditionally doing the chmod. Anyone else?
Dave
On Fri, 2008-02-29 at 11:36 -0500, Dave Jones wrote:
On Fri, Feb 29, 2008 at 10:16:00AM -0600, Matt Domsch wrote:
On Thu, Feb 28, 2008 at 10:00:32PM -0500, Doug Chapman wrote:
Actually I came up with what I think is a cleaner fix for this. Since the default file permission on files on vfat are 755 anyway if the kernel is mode 755 rpm doesn't complain.
Anybody have thoughts on this specfile change? I build this as a scratch build on our ia64 koji server and it installs cleanly.
- Doug
*** kernel.spec.bad 2008-02-28 19:58:55.000000000 -0500 --- kernel.spec 2008-02-28 21:39:57.000000000 -0500 *************** BuildKernel() { *** 1301,1306 **** --- 1301,1310 ---- $CopyKernel $KernelImage \ $RPM_BUILD_ROOT/%{image_install_path}/$InstallName-$KernelVer
- %ifarch ia64
chmod 755 $RPM_BUILD_ROOT/%{image_install_path}/$InstallName-$KernelVer- %endif
mkdir -p $RPM_BUILD_ROOT/lib/modules/$KernelVer make -s ARCH=$Arch INSTALL_MOD_PATH=$RPM_BUILD_ROOT modules_install KERNELRELEASE=$KernelVerThere are systems with EFI32 and EFI64 out there, that aren't ia64, but that will likewise be dropping files into a vfat file system.
I don't see any problem in unconditionally doing the chmod. Anyone else?
I can't actually think of any reason this would break on other platforms. I added the %ifarch ia64 "just in case" because I really don't want to be "that ia64 guy who breaks everyone else's stuff".
- Doug
On Fri, Feb 29, 2008 at 01:23:02PM -0500, Doug Chapman wrote:
chmod 755 $RPM_BUILD_ROOT/%{image_install_path}/$InstallName-$KernelVerThere are systems with EFI32 and EFI64 out there, that aren't ia64, but that will likewise be dropping files into a vfat file system.
I don't see any problem in unconditionally doing the chmod. Anyone else?
I can't actually think of any reason this would break on other platforms. I added the %ifarch ia64 "just in case" because I really don't want to be "that ia64 guy who breaks everyone else's stuff".
I just committed the change to devel. We'll see tomorrow, but I'll be really surprised if anything at all cares that the vmlinuz grew an x bit.
Dave
kernel@lists.fedoraproject.org
-
Dave Jones -
Doug Chapman -
Matt Domsch