glibc 2.14-4 eats my data (Re: F15 ext3, eCryptfs + samba = data corruption (Re: F15 "Error mounting eCryptfs: [-5] Input/output error" on different disks))

[Michal Hlavinka]

>>>> W dniu 8 lipca 2011 18:21 użytkownik Michał Piotrowski
>>>>
>>>> <mkkp4x4 at gmail.com>  napisał:
>>>>> Hi,
>>>>>
>>>>> 2011/7/8 Andreas Schwab<schwab at redhat.com>:
>>>>>> Use valgrind.
>>>>>
>>>>> I attach valgrind output.
>>>>>
>>>>> ==1312== 1 errors in context 1 of 116:
>>>>> ==1312== Source and destination overlap in memcpy(0xaef1590, 0xaef1593,
>>>>> 76) ==1312==    at 0x4C283B6: memcpy@@GLIBC_2.14
>>>>> (mc_replace_strmem.c:653) ==1312==    by 0x401835: ??? (in
>>>>> /sbin/mount.ecryptfs)
>>>>> ==1312==    by 0x5E3039C: (below main) (in /lib64/libc-2.14.so)
>>>>
>>>> I installed ecryptfs-utils-debuginfo package and now it's more readable
>>>>
>>>> ==1815== 1 errors in context 1 of 116:
>>>> ==1815== Source and destination overlap in memcpy(0xaef1590, 0xaef1593, 76)
>>>> ==1815==    at 0x4C283B6: memcpy@@GLIBC_2.14 (mc_replace_strmem.c:653)
>>>> ==1815==    by 0x401835: main (string3.h:52)
>>>>
>>>>> Could this be related to
>>>>>   - Fix static linking with checking x86/x86-64 memcpy (BZ#12653)
>>>>> or is it an eCryptfs problem?
 >> W dniu 8 lipca 2011 20:08 użytkownik Michal Hlavinka
 >> <mhlavink at redhat.com>  napisał:
 >>> Hi,
 >>>
 >>> please check if this package changes anything for you:
 >>>
 >>> http://kojipkgs.fedoraproject.org/scratch/mhlavink/task_3187528/
 >>
 >> unfortunately there is no difference
 >
 > I'm attaching valgrind output. I checked your patch and it removes
 > correctly all uses of memcpy so it seems that memcpy only covered the
 > root of the problem.

ok, complain about memcpy in ecryptfs-utils is gone. I've checked all of 
them and I think I've fixed those which needed it. I was not able to 
reproduce original problem nor valgrind complaint, so please test if 
following package produces memcpy complain in valgrind output or not:

http://kojipkgs.fedoraproject.org/scratch/mhlavink/task_3190860/

Anyway, I don't think there is any problem in ecryptfs-utils, because 
it's just mount helper. It's not running when files are being 
encrypted/decrypted and you said it works fine when you use ecryptfs 
directly (without samba). We've fixed memcpy bug in ecryptfs which is 
definitely a good think, but problem is elsewhere.

If you want, you can test following build of samba which has all 
occurrences of memcpy replaced by memmove. I don't dare to guess if it 
changes anything, but give it a try if you want:

http://kojipkgs.fedoraproject.org/scratch/mhlavink/task_3190918/

Could you describe your environment in more details so we can try to 
reproduce it? For example what ecryptfs options you use (including fstab 
line if you have any), samba configuration etc...

Michal