crypt issue with 2.6.30/gcc 4.4.0?
Hello,
I am having trouble decrypting my dm-crypted disks with 2.6.30. 2.6.29.4 works fine with the same encrypted disks.
I've been upgrading to Fedora 11 in the past few days so now GCC 4.4.0 is the compiler, that may (also) be a factor.
On 2.6.92.4 I see the encryption password prompt being buried by a lot of USB-messages from connected gear. I can 'blindly' type my password and asterisks appear for each letter. At the end of the password I press Enter; a complaint about the partition table for md1 is displayed and booting continues.
With 2.6.30 I see the Password prompt being repeated, after all the USB messages, as soon as I press the first letter of the password. Each time the password is entered it is not accepted and a new Password: prompt is printed up to a total of three tries.
For all of these boots into 2.6.29.4 or 2.6.30 the userland etc is the same (Fedora 11), it is just that 2.6.29.4 was built while running Fedora 10 and 2.6.30 is now built with the tools in 2.6.30. I did boot into 2.6.30 (built while running Fedora 10) successfully once, right after the upgrade to Fedora 11. Later I made small changes to the kernel config and rebuilt the kernel and this made the encryption password fail. A make clean and rebuild does not fix the issue
How can I find out what's the cause for the password issue?
Kind regards,
Udo
Quoted 25 lines · Udo van den Heuvel
Hello,
I am having trouble decrypting my dm-crypted disks with 2.6.30. 2.6.29.4 works fine with the same encrypted disks.
I've been upgrading to Fedora 11 in the past few days so now GCC 4.4.0 is the compiler, that may (also) be a factor.On 2.6.92.4 I see the encryption password prompt being buried by a lot of USB-messages from connected gear. I can 'blindly' type my password and asterisks appear for each letter. At the end of the password I press Enter; a complaint about the partition table for md1 is displayed and booting continues.
With 2.6.30 I see the Password prompt being repeated, after all the USB messages, as soon as I press the first letter of the password. Each time the password is entered it is not accepted and a new Password: prompt is printed up to a total of three tries.For all of these boots into 2.6.29.4 or 2.6.30 the userland etc is the same (Fedora 11), it is just that 2.6.29.4 was built while running Fedora 10 and 2.6.30 is now built with the tools in 2.6.30. I did boot into 2.6.30 (built while running Fedora 10) successfully once, right after the upgrade to Fedora 11. Later I made small changes to the kernel config and rebuilt the kernel and this made the encryption password fail. A
make cleanand rebuild does not fix the issue
Quoted 1 line
How can I find out what's the cause for the password issue?
Instead of changing two variables at once (kernel version AND gcc version) you keep one constant and vary the other.
IOW, rebuild 2.6.29.4 with gcc-4.4 and test. Then rebuild 2.6.30 with gcc-4.3 or older and test.
Quoted 8 lines · Mikael Pettersson
Quoted 1 line · Udo van den Heuvel
How can I find out what's the cause for the password issue?
Instead of changing two variables at once (kernel version AND gcc version) you keep one constant and vary the other.
IOW, rebuild 2.6.29.4 with gcc-4.4 and test. Then rebuild 2.6.30 with gcc-4.3 or older and test.
I can do the 2.6.29.4 build with gcc-4.4 but the gcc-4.3 case is harder to do: I need to convince Fedora to install a 2nd compiler-version (and rest of toolchain).
I'll post back with the 2.6.29.4 results using gcc-4.4.
Udo
Quoted 7 lines · Udo van den Heuvel
Quoted 6 lines · Mikael Pettersson
Quoted 1 line · Udo van den Heuvel
How can I find out what's the cause for the password issue?
Instead of changing two variables at once (kernel version AND gcc version) you keep one constant and vary the other.
Quoted 6 lines
Quoted 2 lines
IOW, rebuild 2.6.29.4 with gcc-4.4 and test. Then rebuild 2.6.30 with gcc-4.3 or older and test.
I can do the 2.6.29.4 build with gcc-4.4 but the gcc-4.3 case is harder to do: I need to convince Fedora to install a 2nd compiler-version (and rest of toolchain).
Not true. Just as you can choose from multiple self-built kernels you can build and choose dynamically from multiple self-built GCCs.
E.g., for this particular case, download gcc-4.3.3.tar.bz2, tar jxf it, mkdir objdir, cd objdir, /path/to/gcc-4.3.3/configure --prefix=/tmp/gcc43 --enable-languages=c, make, make install, cd /path/to/linux-2.6.30, PATH=/tmp/gcc43/bin:$PATH make, and so on.
Quoted 1 line · Mikael Pettersson
IOW, rebuild 2.6.29.4 with gcc-4.4 and test.
This testcase gives the same results as 2.6.30 with gcc-4.4, i.e.:
password is not accepted.
So it is not the kernel I guess. Perhaps gcc or something that goes in the ramdisk?
Udo
Quoted 5 lines · Udo van den Heuvel
I am having trouble decrypting my dm-crypted disks with 2.6.30. 2.6.29.4 works fine with the same encrypted disks.
I've been upgrading to Fedora 11 in the past few days so now GCC 4.4.0 is the compiler, that may (also) be a factor.
Quoted 9 lines
On 2.6.92.4 I see the encryption password prompt being buried by a lot of USB-messages from connected gear. I can 'blindly' type my password and asterisks appear for each letter. At the end of the password I press Enter; a complaint about the partition table for md1 is displayed and booting continues.
With 2.6.30 I see the Password prompt being repeated, after all the USB messages, as soon as I press the first letter of the password. Each time the password is entered it is not accepted and a new Password: prompt is printed up to a total of three tries.
Hi,
password prompt in Fedora is handled completely in userspace in initrd by plymouth, then it send password to cryptsetup.
You did provide neither kernel log messages nor description of changes to your kernel config... Try run it in text mode with full kernel log, also try enable some debug mode in plymouth and check exact error message.
Anyway, the problem is probably in some misconfiguration in you initrd (like missing kernel module - some new module maybe needed now for your config, bad order of module loading, device name changes of underlying device, maybe some timing problem initialization of USB keyboard etc.) and not in kernel itself. (Are you using mkinitrd from F11? Also see fedora bug 470740)
Milan
--
[email protected]
Quoted 7 lines · Milan Broz
Quoted 4 lines
With 2.6.30 I see the Password prompt being repeated, after all the USB messages, as soon as I press the first letter of the password. Each time the password is entered it is not accepted and a new Password: prompt is printed up to a total of three tries.
password prompt in Fedora is handled completely in userspace in initrd by plymouth, then it send password to cryptsetup.
Yes. I saw that in mkinitrd source...
Quoted 1 line
You did provide neither kernel log messages
The kernel is just booting up, messages scroll by fast, no strange stuff appears as far as I can see.
Quoted 2 lines
nor description of changes to
your kernel config...
I got the 2.6.29.4 .config, copied it into 2.6.30 tree and added ext4 as fs and nfs4 server support. Nothing much in the area of where the problem is?
Quoted 1 line
Try run it in text mode with full kernel log,
How to retrieve the log?
Quoted 1 line
also try enable some debug mode in plymouth and check exact error message.
Will have look into possibilties there.
Quoted 1 line
(Are you using mkinitrd from F11? Also see fedora bug 470740)
Indeed. This looks very much like it, although I use a normal PS/2 keyboard and I'll have to count the asterisks.
Thanks for this tip!!
Udo