On Sun, Oct 01, 2000 at 02:27:27AM +0200, xfs@xxxxxxxxxx wrote:
> hi
>
> i just checked it with a vanila 2.4.0-test5. it works.
>
> open("/", O_RDONLY|O_NONBLOCK|O_DIRECTORY) = 4
> fstat64(4, {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
> fcntl64(4, F_SETFD, FD_CLOEXEC) = -1 ENOSYS (Function not
> implemented)
> fcntl(4, F_SETFD, FD_CLOEXEC) = 0
>
> seems that the xfs patch does something wrong.
> or the new glibc is not fault tolerant enough. i made the ls test with my
> redhat 6.2 installation on a xfs root partition. fcntl64 is not used.
I think I know what the problem is. The old XFS beta5 kernel has attr_*
system calls in the call slots that are fstat64() et.al. on the
RH7.0/later 2.4 kernels. The RH7 glibc seems to expect LFS calls there
(rightfully, because later 2.4 after test5 put them there). Calling
the attr_* calls as different system calls does no good.
Does it work when you replace the
.long SYMBOL_NAME(sys_attr_get)
.long SYMBOL_NAME(sys_attr_set)
.long SYMBOL_NAME(sys_attr_remove)
.long SYMBOL_NAME(sys_attr_list)
.long SYMBOL_NAME(sys_attr_getf)
.long SYMBOL_NAME(sys_attr_setf)
.long SYMBOL_NAME(sys_attr_removef)
.long SYMBOL_NAME(sys_attr_listf)
lines at the end of arch/i386/kernel/entry.S with
.long SYMBOL_NAME(sys_ni_syscall)
for each attr line ?
-Andi
|