On Mon, Mar 15, 2010 at 01:46:10PM +0100, Jan-Simon Möller wrote:
> We're still investigating the topic. So far, we tracked one failure down to
> ldconfig.real which is a static arm binary executed after the chroot got the
> basic packages installed. If we setup a chroot with an older qemu-arm and
> exchange it afterwards with the new qemu and rerun just the build step, it
> works. Thus it seems to be an issue with static arm binaries atm.
>
> Strace of a call of ldconfig.real with the qemu-arm failing:
>
> http://pastie.org/870189
>
> Sort version:
>
> r...@frodo:/# qemu-arm -strace /sbin/ldconfig.real
> 16359 uname(0x403fef78) = 0
> 16359 brk(NULL) = 0x000a9000
> 16359 brk(0x000a9d08) = 0x000a9d08
> 16359 open("/dev/urandom",O_RDONLY) = 3
> 16359 read(3,0x403ff27d,3) = 3
> 16359 close(3) = 0
> [...]
> 16359 stat64("/usr/lib/libgettextlib.so",0x403fdf28) = 0
> 16359 stat64("/usr/lib/libgettextpo.so.0",0x403fdec0) = 0
> 16359 stat64("/usr/lib/libgettextpo.so.0.4.0",0x403fdf28) = 0
> 16359 stat64("/usr/lib/libpython2.6.so.1.0",0x403fdec0) = 0
> 16359 stat64("/usr/lib/libpython2.6.so.1.0",0x403fdf28) = 0
> 16359 open("/etc/ld.so.cache~",O_WRONLY|O_CREAT|O_NOFOLLOW|O_TRUNC,0600) = 3
> 16359 write(3,0xb03d0,1288) = 1288
> 16359 write(3,0x403ff0a0,0) = -1 errno=14 (Bad address)
A zero sized write. According to manpage ok.
In qemu we do a lock_user to to get the string to write. Richards change changes
the access checks the get called by lock_user:
page_check_range:
- if (start + len < start)
- /* we've wrapped around */
...
+ if (start + len - 1 < start) {
+ /* We've wrapped around. */
This now blows up with len = 0;
> 16359 write(2,0x403fca08,21)/sbin/ldconfig.real: = 21
> 16359 write(2,0x403fc9e8,28)Writing of cache data failed = 28
> 16359 write(2,0x403fc5b8,13): Bad address = 13
> 16359 write(2,0x403fc9c0,1)
> = 1
> 16359 exit_group(1)
>
> Best,
> Jan-Simon
>
