Android gpg-agent crashes in libgcrypt when signing, decrypting, importing secret keys

[Hans-Christoph Steiner]

On 01/18/2014 06:31 AM, Jussi Kivilinna wrote:
> On 17.01.2014 20:34, Hans-Christoph Steiner wrote:
>>
>> On GPG for Android, I've updated to the latest libgcrypt in master (or close
>> to it, its commit 4b7db51ad5d1bf98fd08ca3048f258059eca61a4).  Now it seems
>> that any operation that needs a passphrase is crashing somewhere in libgcrypt.
>>  I've tried building with auto-detection of CPU which enables Padlock, Intelt
>> DRNG, and NEON.  I also tried with --disable-padlock-support
>> --disable-drng-support --disable-neon-support, and seemed to get the same thing.
>>
>> I've also tried running gpg-agent with and without --enable-ssh-support, and
>> same result each time.
>>
>> Here's the basic backtrace:
> <..snip..>
>> From the bug report in our tracker, you can download the complete build log, a
>> debug log from the Android app, a log from gpg-agent, and a log from gpgme:
>>
>> https://dev.guardianproject.info/issues/2888
> 
> Have you configured gcc flags correctly for target platform? It seems that
> compiler (and libgcrypt assembly) are configured to allow unaligned memory
> accesses, but target does not support them.
> 
> Disassembly of crash site:
> 
>    0:   e1866469        orr     r6, r6, r9, ror #8
>    4:   e8900f00        ldm     r0, {r8, r9, sl, fp}
>    8:   e0244008        eor     r4, r4, r8
>    c:   e0255009        eor     r5, r5, r9
>   10:   e026600a        eor     r6, r6, sl
>   14:   e027700b        eor     r7, r7, fp
>   18:   eafffded        b       0xfffff7d4
>   1c:   e92d5ff0        push    {r4, r5, r6, r7, r8, r9, sl, fp, ip, lr}
> !!20:   e89200f0        ldm     r2, {r4, r5, r6, r7}
>   24:   e24dd010        sub     sp, sp, #16
>   28:   e59fe864        ldr     lr, [pc, #2148] ; 0x894
>   2c:   e3a0c0ff        mov     ip, #255        ; 0xff
>   30:   e58d1004        str     r1, [sp, #4]
>   34:   e1a0c18c        lsl     ip, ip, #3
>   38:   e353000c        cmp     r3, #12
>   3c:   aa000215        bge     0x898
> 
> Crash happens in rinjdael_arm.S:_gcry_aes_arm_decrypt_block, line 496:
> 	/* aligned load */
> 	ldm	%r2, {RA, RB, RC, RD};
> 
> This just loads four 32-bit words from input buffer (pointer in r2). The pointer
> in r2 is 0x013ebf9f, not aligned to 32-bit word boundary. Above disassembly
> shows that code is compiled with __ARM_FEATURE_UNALIGNED (-munaligned-access)
> and unaligned memory accesses are assumed to be ok. But clearly unaligned
> memory accesses are not allowed as programs crashes with "signal 7 (SIGBUS),
> code 1 (BUS_ADRALN), fault addr 013ebf9f" - Invalid address alignment.
> 
> GCC documentation says [1]:
>  -munaligned-access
>  -mno-unaligned-access
>    Enables (or disables) reading and writing of 16- and 32- bit values from
>    addresses that are not 16- or 32- bit aligned. By default unaligned access
>    is disabled for all pre-ARMv6 and all ARMv6-M architectures, and enabled for
>    all other architectures. If unaligned access is not enabled then words in
>    packed data structures will be accessed a byte at a time.
> 
>    The ARM attribute Tag_CPU_unaligned_access will be set in the generated
>    object file to either true or false, depending upon the setting of this
>    option. If unaligned access is enabled then the preprocessor symbol
>    __ARM_FEATURE_UNALIGNED will also be defined.
> 
> -Jussi
> 
> [1] http://gcc.gnu.org/onlinedocs/gcc/ARM-Options.html

I forget if I mentioned this before: the build flags are set by the default
Android build system.

So I built the whole thing again, manually adding -mno-unaligned-access to the
libgcrypt build, and the tests seem to be failing in the same place.  I tested
head of master on the armv7a emulator, which failed a lot more, and the head
of LIBGCRYPT-1-6-BRANCH on the Nexus 7 ARMv7 tablet, which failed in the same
places. Any pointers for next steps?

FYI, I'm gathering all these log files on our bug tracker:
https://dev.guardianproject.info/issues/2888

Attached are the latest test logs, including the full build log for head of
master running tests on the armv7a emulator.

.hc


-- 
PGP fingerprint: 5E61 C878 0F86 295C E17D  8677 9F0F E587 374B BE81
-------------- next part --------------
A non-text attachment was scrubbed...
Name: gnupg-tests-emulator-armv7a-mno-unaligned-access.txt.bz2
Type: application/x-bzip
Size: 5371 bytes
Desc: not available
URL: </pipermail/attachments/20140118/92bfb505/attachment-0002.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: gnupg-tests-nexus-7-mno-unaligned-access.txt.bz2
Type: application/x-bzip
Size: 11934 bytes
Desc: not available
URL: </pipermail/attachments/20140118/92bfb505/attachment-0003.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 969 bytes
Desc: OpenPGP digital signature
URL: </pipermail/attachments/20140118/92bfb505/attachment-0001.sig>