Bug 144688

[root@sizzler temp]# gunzip -c initrd-2.6.9-1.906_EL.img | ./cpio -id
5657 blocks
[root@sizzler temp]# find -type f -exec ls -l {} \;
-rwxr-xr-x  1 root root 1251 Jan 12 08:49 ./init
-rw-r--r--  1 root root 1202580 Jan 12 08:35 ./initrd-2.6.9-1.906_EL.img
-rw-r--r--  1 root root 302992 Jan 12 08:49 ./lib/lpfc.ko
-rw-r--r--  1 root root 268240 Jan 12 08:49 ./lib/scsi_mod.ko
-rw-r--r--  1 root root 264456 Jan 12 08:49 ./lib/ext3.ko
-rw-r--r--  1 root root 148112 Jan 12 08:49 ./lib/jbd.ko
-rw-r--r--  1 root root 160056 Jan 12 08:49 ./lib/sym53c8xx.ko
-rw-r--r--  1 root root 41600 Jan 12 08:49 ./lib/sd_mod.ko
-rw-r--r--  1 root root 41680 Jan 12 08:49 ./lib/scsi_transport_spi.ko
-rw-r--r--  1 root root 24984 Jan 12 08:49 ./lib/xor.ko
-rw-r--r--  1 root root 50912 Jan 12 08:49 ./lib/raid5.ko
-rw-r--r--  1 root root 28512 Jan 12 08:49 ./lib/scsi_transport_fc.ko
-rw-r--r--  1 root root 1129 Jan 12 08:49 ./etc/udev/udev.conf
-rwxr-xr-x  1 root root 426016 Jan 12 08:49 ./bin/insmod
-rwxr-xr-x  1 root root 483380 Jan 12 08:49 ./bin/nash
-rwxr-xr-x  1 root root 648120 Jan 12 08:49 ./bin/udev
-rwxr-xr-x  1 root root 139013 Jan 12 08:43 ./cpio
[root@sizzler temp]# uname -a
Linux sizzler.build.redhat.com 2.6.9-1.906_EL #1 SMP Sun Dec 12
22:54:40 EST 2004 ppc ppc ppc GNU/Linux
                           
Version-Release number of cpio
cpio-2.5-10

Same result with ppc64

I played around with gdb a little, and I think I found the reason for
the misbehaviour I am seeing, although it totally puzzles me:

(gdb) run -id < initrd
Starting program: /bin/cpio -id < initrd

Breakpoint 5, copy_files_tape_to_disk (in_des=0, out_des=8,
num_bytes=warning: Unhandled dwarf expresion opcode DW_OP_piece
639964) at util.c:459
459           disk_buffered_write (in_buff, out_des, size);
(gdb) print size
warning: Unhandled dwarf expresion opcode DW_OP_piece
warning: Unhandled dwarf expresion opcode DW_OP_piece
$5 = 164
(gdb) c
Continuing.

Breakpoint 6, disk_buffered_write (in_buf=0x9c338 <Address 0x9c338 out
of bounds>, out_des=8, num_bytes=0) at util.c:276
276     {
(gdb) where
#0  disk_buffered_write (in_buf=0x9c338 <Address 0x9c338 out of
bounds>, out_des=8, num_bytes=0) at util.c:276
#1  0x10009920 in copy_files_tape_to_disk (in_des=0, out_des=8,
num_bytes=warning: Unhandled dwarf expresion opcode DW_OP_piece
639800) at util.c:459
#2  0x100022d4 in copyin_regular_file (file_hdr=0x7ffff650,
in_file_des=0) at copyin.c:961
#3  0x10003208 in process_copy_in () at copyin.c:768
#4  0x10006dd8 in main (argc=2, argv=0x7ffffa74) at main.c:533
#5  0x0fe9b840 in generic_start_main () from /lib/tls/libc.so.6
#6  0x0fe9b988 in __libc_start_main () from /lib/tls/libc.so.6
#7  0x00000000 in ?? ()



As you can see, disk_buffered_write() is called from
copyin_regular_file() to write out "size" bytes to disk (breakpoint 5)
"size" in the context of copyin_regular_file is 164. But in the actual
function call (breakpoint 6), "num_bytes" suddenly is 0, so nothing is
being written to disk.

I am quite out of my waters here. Is this a compiler bug?

[root@sizzler /]# gcc --version
gcc (GCC) 3.4.3 20050104 (Red Hat 3.4.3-13)

OK, I think I've got it.

copy_files_tape_to_disk [util.c] calls disk_buffered_write(in_buff,
out_des, size) [util.c:459]. In this context, size is an off_t, which
resolves to a long long int.

disk_buffered_write (in_buf, out_des, num_bytes) [util.c:272] defines
num_bytes as a long. Depending on whether the compiler chooses to pass
arguments via the stack or via registers, the endianess of the machine
and the phase of the moon, this breaks. I think it works on ix86,
because parameters are passed on the stack, and the memory layout is
right, and it works on your machines because they have 64-bit
registers (unless I am seroiusly mistaken). My machine is a G3,
without large registers.

I've built a small patch, replacing long by off_t in
disk_buffered_write for at the appropriate places, and suddenly
unpacking works.

I think there are more places where this happens, the code is a mess.

I'll attach the patch, it also adds some printf()s I added for debugging.

Created attachment 109712 [details]
patch to disk_buffered_write

Thx. for your patch. 

I integrate your patch and cpio-2.5-lfs.patch into new
cpio-2.6-lfs.patch, also I fixed other places where this happens. Hope
it will work fine.

Created attachment 109776 [details]
new lfs patch

An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.

http://rhn.redhat.com/errata/RHSA-2005-080.html