From: Martin Schwidefsky <[email protected]>
The calculation of the value nr in do_xip_mapping_read is incorrect. If
the copy required more than one iteration in the do while loop the
copies variable will be non-zero. The maximum length that may be passed
to the call to copy_to_user(buf+copied, xip_mem+offset, nr) is len-copied
but the check only compares against (nr > len).
This bug is the cause for the heap corruption Carsten has been chasing
for so long:
*** glibc detected *** /bin/bash: free(): invalid next size (normal): 0x00000000800e39f0 ***
======= Backtrace: =========
/lib64/libc.so.6[0x200000b9b44]
/lib64/libc.so.6(cfree+0x8e)[0x200000bdade]
/bin/bash(free_buffered_stream+0x32)[0x80050e4e]
/bin/bash(close_buffered_stream+0x1c)[0x80050ea4]
/bin/bash(unset_bash_input+0x2a)[0x8001c366]
/bin/bash(make_child+0x1d4)[0x8004115c]
/bin/bash[0x8002fc3c]
/bin/bash(execute_command_internal+0x656)[0x8003048e]
/bin/bash(execute_command+0x5e)[0x80031e1e]
/bin/bash(execute_command_internal+0x79a)[0x800305d2]
/bin/bash(execute_command+0x5e)[0x80031e1e]
/bin/bash(reader_loop+0x270)[0x8001efe0]
/bin/bash(main+0x1328)[0x8001e960]
/lib64/libc.so.6(__libc_start_main+0x100)[0x200000592a8]
/bin/bash(clearerr+0x5e)[0x8001c092]
With this bug fix the commit 0e4a9b59282914fe057ab17027f55123964bc2e2
"ext2/xip: refuse to change xip flag during remount with busy inodes"
can be removed again.
Cc: Carsten Otte <[email protected]>
Cc: Nick Piggin <[email protected]>
Cc: Jared Hulbert <[email protected]>
Cc: Andrew Morton <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
---
mm/filemap_xip.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff -urpN linux-2.6/mm/filemap_xip.c linux-2.6-patched/mm/filemap_xip.c
--- linux-2.6/mm/filemap_xip.c 2009-03-24 00:12:14.000000000 +0100
+++ linux-2.6-patched/mm/filemap_xip.c 2009-03-31 15:25:53.000000000 +0200
@@ -89,8 +89,8 @@ do_xip_mapping_read(struct address_space
}
}
nr = nr - offset;
- if (nr > len)
- nr = len;
+ if (nr > len - copied)
+ nr = len - copied;
error = mapping->a_ops->get_xip_mem(mapping, index, 0,
&xip_mem, &xip_pfn);
On Tue, 31 Mar 2009 15:32:23 +0200
Martin Schwidefsky <[email protected]> wrote:
> From: Martin Schwidefsky <[email protected]>
>
> The calculation of the value nr in do_xip_mapping_read is incorrect. If
> the copy required more than one iteration in the do while loop the
> copies variable will be non-zero. The maximum length that may be passed
> to the call to copy_to_user(buf+copied, xip_mem+offset, nr) is len-copied
> but the check only compares against (nr > len).
>
> This bug is the cause for the heap corruption Carsten has been chasing
> for so long:
>
> *** glibc detected *** /bin/bash: free(): invalid next size (normal): 0x00000000800e39f0 ***
> ======= Backtrace: =========
> /lib64/libc.so.6[0x200000b9b44]
> /lib64/libc.so.6(cfree+0x8e)[0x200000bdade]
> /bin/bash(free_buffered_stream+0x32)[0x80050e4e]
> /bin/bash(close_buffered_stream+0x1c)[0x80050ea4]
> /bin/bash(unset_bash_input+0x2a)[0x8001c366]
> /bin/bash(make_child+0x1d4)[0x8004115c]
> /bin/bash[0x8002fc3c]
> /bin/bash(execute_command_internal+0x656)[0x8003048e]
> /bin/bash(execute_command+0x5e)[0x80031e1e]
> /bin/bash(execute_command_internal+0x79a)[0x800305d2]
> /bin/bash(execute_command+0x5e)[0x80031e1e]
> /bin/bash(reader_loop+0x270)[0x8001efe0]
> /bin/bash(main+0x1328)[0x8001e960]
> /lib64/libc.so.6(__libc_start_main+0x100)[0x200000592a8]
> /bin/bash(clearerr+0x5e)[0x8001c092]
Please get into the habit of adding Cc: <[email protected]> to the
changelogs?
I believe I personally am pretty good at picking up stable things, but
other patch-mergers are quite unreliable. We all need as much help as
we can get on this, because things are falling through cracks.
> With this bug fix the commit 0e4a9b59282914fe057ab17027f55123964bc2e2
> "ext2/xip: refuse to change xip flag during remount with busy inodes"
> can be removed again.
OK, please send a standalone patch to do this at an appropriate time.
I guess that this second patch won't be needed in -stable.
On Wed, 1 Apr 2009 14:17:00 -0700
Andrew Morton <[email protected]> wrote:
> On Tue, 31 Mar 2009 15:32:23 +0200
> Martin Schwidefsky <[email protected]> wrote:
>
> Please get into the habit of adding Cc: <[email protected]> to the
> changelogs?
Ok, I've made me a sticky note and attached it to my monitor. Perhaps
that will help me to remember ..
> I believe I personally am pretty good at picking up stable things, but
> other patch-mergers are quite unreliable. We all need as much help as
> we can get on this, because things are falling through cracks.
Ok, makes sense. Better safe than sorry.
> > With this bug fix the commit 0e4a9b59282914fe057ab17027f55123964bc2e2
> > "ext2/xip: refuse to change xip flag during remount with busy inodes"
> > can be removed again.
>
> OK, please send a standalone patch to do this at an appropriate time.
> I guess that this second patch won't be needed in -stable.
The revert should be done either with the bug fix or after the bug fix
hit mainline. Guess I just create the patch and send it.
--
blue skies,
Martin.
"Reality continues to ruin my life." - Calvin.
> This bug is the cause for the heap corruption Carsten has been chasing
> for so long:
Finally saw this issue on ARM. Some programs in Android on ARM were
crashing with AXFS. This fixed it.
Nice catch!
I tested this patch on ext2's ramdisk with xip.
It fixed following bug.
*** stack smashing detected ***: <unknown> terminated
======= Backtrace: =========
/lib/tls/i686/cmov/libc.so.6(__fortify_fail+0x48)[0xb7fe5138]
/lib/tls/i686/cmov/libc.so.6(__fortify_fail+0x0)[0xb7fe50f0]
[0x80485f8]
[0x0]
======= Memory map: ========
08048000-08049000 r-xp 00000000 01:00 12 /mnt/xip_test
08049000-0804a000 rw-p 00000000 01:00 12 /mnt/xip_test
0804a000-0806b000 rw-p 0804a000 00:00 0 [heap]
b7ef7000-b7ef8000 rw-p b7ef7000 00:00 0
b7ef8000-b8041000 r-xp 00000000 08:03 4219915 /lib/tls/i686/cmov/libc-2.7.so
b8041000-b8042000 r--p 00149000 08:03 4219915 /lib/tls/i686/cmov/libc-2.7.so
b8042000-b8044000 rw-p 0014a000 08:03 4219915 /lib/tls/i686/cmov/libc-2.7.so
b8044000-b8047000 rw-p b8044000 00:00 0
b8047000-b8051000 r-xp 00000000 08:03 4202508 /lib/libgcc_s.so.1
b8051000-b8052000 rw-p 0000a000 08:03 4202508 /lib/libgcc_s.so.1
b8052000-b8055000 rw-p b8052000 00:00 0
b8055000-b8056000 r-xp b8055000 00:00 0 [vdso]
b8056000-b8070000 r-xp 00000000 08:03 4202565 /lib/ld-2.7.so
b8070000-b8072000 rw-p 00019000 08:03 4202565 /lib/ld-2.7.so
bf95c000-bf971000 rw-p bffeb000 00:00 0 [stack]
Aborted
Reviewed-by: Minchan kim <[email protected]>
Tested-by: Minchan kim <[email protected]>
On Tue, Mar 31, 2009 at 10:32 PM, Martin Schwidefsky
<[email protected]> wrote:
> From: Martin Schwidefsky <[email protected]>
>
> The calculation of the value nr in do_xip_mapping_read is incorrect. If
> the copy required more than one iteration in the do while loop the
> copies variable will be non-zero. The maximum length that may be passed
> to the call to copy_to_user(buf+copied, xip_mem+offset, nr) is len-copied
> but the check only compares against (nr > len).
>
> This bug is the cause for the heap corruption Carsten has been chasing
> for so long:
>
> *** glibc detected *** /bin/bash: free(): invalid next size (normal): 0x00000000800e39f0 ***
> ======= Backtrace: =========
> /lib64/libc.so.6[0x200000b9b44]
> /lib64/libc.so.6(cfree+0x8e)[0x200000bdade]
> /bin/bash(free_buffered_stream+0x32)[0x80050e4e]
> /bin/bash(close_buffered_stream+0x1c)[0x80050ea4]
> /bin/bash(unset_bash_input+0x2a)[0x8001c366]
> /bin/bash(make_child+0x1d4)[0x8004115c]
> /bin/bash[0x8002fc3c]
> /bin/bash(execute_command_internal+0x656)[0x8003048e]
> /bin/bash(execute_command+0x5e)[0x80031e1e]
> /bin/bash(execute_command_internal+0x79a)[0x800305d2]
> /bin/bash(execute_command+0x5e)[0x80031e1e]
> /bin/bash(reader_loop+0x270)[0x8001efe0]
> /bin/bash(main+0x1328)[0x8001e960]
> /lib64/libc.so.6(__libc_start_main+0x100)[0x200000592a8]
> /bin/bash(clearerr+0x5e)[0x8001c092]
>
> With this bug fix the commit 0e4a9b59282914fe057ab17027f55123964bc2e2
> "ext2/xip: refuse to change xip flag during remount with busy inodes"
> can be removed again.
>
> Cc: Carsten Otte <[email protected]>
> Cc: Nick Piggin <[email protected]>
> Cc: Jared Hulbert <[email protected]>
> Cc: Andrew Morton <[email protected]>
> Signed-off-by: Martin Schwidefsky <[email protected]>
> ---
>
> mm/filemap_xip.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff -urpN linux-2.6/mm/filemap_xip.c linux-2.6-patched/mm/filemap_xip.c
> --- linux-2.6/mm/filemap_xip.c 2009-03-24 00:12:14.000000000 +0100
> +++ linux-2.6-patched/mm/filemap_xip.c 2009-03-31 15:25:53.000000000 +0200
> @@ -89,8 +89,8 @@ do_xip_mapping_read(struct address_space
> }
> }
> nr = nr - offset;
> - if (nr > len)
> - nr = len;
> + if (nr > len - copied)
> + nr = len - copied;
>
> error = mapping->a_ops->get_xip_mem(mapping, index, 0,
> &xip_mem, &xip_pfn);
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to [email protected]. For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
>
--
Kinds regards,
Minchan Kim