Mysterious string truncation in 2.4.25 kernel
Aloha,
I'm running a patched 2.4.25 kernel and I noticed truncated strings in various procfs files. e.g.
# cat /proc/tty/drivers se /dev/cua/%d 5 64-127 serial:callout se /dev/tts/%d 4 64-127 se pty_slave /dev/pts/%d 136 0-255 pty:slave pty_master /dev/ptm 128 0-255 pty:master pty_slave /dev/pty/s%d 3 0-255 pty:slave pty_master /dev/pty/m%d 2 0-255 pty:master /dev/vc/0 /dev/vc/0 4 0 system:vtmaster /dev/ptmx /dev/ptmx 5 2 system /dev/console /dev/console 5 1 system:console /dev/tty /dev/tty 5 0 system:/dev/tty unknown /dev/vc/%d 4 1-63 console
Notice that "serial" is output as "se" in the top two lines. I was able to produce similar results on a vanilla 2.4.25 kernel by simply changing the length of a few strings in fs/proc/proc_tty.c. I believe the problem is caused by the following patch:
[email protected] 2004-01-31 21:16:34-02:00 [email protected]
Quoted 11 lines
[PATCH] Zero last byte of mount option page
Hi Al,
Here's a patch which zeroes the last byte of the mount option data copied from userspace during mount(2).
For filesystems which parse mount options as strings (the majority), lack of a zero terminator could cause the page to be overrun. The source code comments specify that the maximum size of the mount data is PAGE_SIZE-1, so this patch will not affect any valid binary-formatted mount data.
fs/namespace.c +3 -0
@@ -715,6 +715,9 @@if (dev_name && !memchr (dev_name,0 ,PAGE_SIZE ))return -EINVAL ; +if (data_page) + ((char *)data_page)[PAGE_SIZE -1 ] =0 ; +/* Separate the per-mountpoint flags */ if (flags &MS_NOSUID ) mnt_flags |=MNT_NOSUID ;
Could someone please comment on the correctness of the above patch especially regarding procfs?
- glen
Quoted 2 lines · Glen Nakamura
Could someone please comment on the correctness of the above patch especially regarding procfs?
I don't see how the patch could be related to the problem you are seeing.
- James
Signature · 2 lines
James Morris <[email protected]>
Quoted 1 line · James Morris
I don't see how the patch could be related to the problem you are seeing.
Thanks for the response... I took another look and my current theory is that the problem occurs in the following invocation of do_mount:
void __init mount_devfs_fs (void)
{
int err;
if ( !(boot_options & OPTION_MOUNT) ) return;
if (err == 0) printk (KERN_INFO "Mounted devfs on /dev\n");
else PRINTK ("(): unable to mount devfs, err: %d\n", err);
} /* End Function mount_devfs_fs */This call to do_mount is on line 3552 of fs/devfs/base.c and passes a const string as the data_page parameter. Then in do_mount in fs/namespace.c on line 718:
if (data_page) ((char *)data_page)[PAGE_SIZE - 1] = 0;
The above statement zeros a byte that is out of bounds and corrupts another string in the same section of memory. In my build, this happens to truncate the "serial" string to "se".
So is it really safe to simply zero the byte at [PAGE_SIZE - 1]? The comment says "up to PAGE_SIZE-1 bytes", _not_ "exactly PAGE_SIZE-1 bytes". It doesn't mention anything about padding, etc.
Of course, perhaps 0 should passed instead of "" for data_page?
- err = do_mount ("none", "/dev", "devfs", 0, "");
+ err = do_mount ("none", "/dev", "devfs", 0, 0);
Comments?
- glen
Quoted 9 lines · Glen Nakamura
void
__initmount_devfs_fs(void){ int err; if ( !(boot_options & OPTION_MOUNT) ) return; + err = do_mount ("none", "/dev", "devfs", 0, ""); if (err == 0) printk (KERN_INFO "Mounted devfs on /dev\n"); else PRINTK ("(): unable to mount devfs, err: %d\n", err); } /* End Function mount_devfs_fs */
Whoops... I deleted the do_mount line by accident.
- glen
Quoted 6 lines · Glen Nakamura
Of course, perhaps 0 should passed instead of "" for
data_page?(quoted diff) +1
-1- err = do_mount ("none", "/dev", "devfs", 0, ""); + err = do_mount ("none", "/dev", "devfs", 0, 0);Comments?
Yes, the devfs fix above is needed if the data_page patch has been applied.
This is the case in 2.6, but not 2.4.25.
- James
Signature · 2 lines
James Morris <[email protected]>
Quoted 7 lines · James Morris
Quoted 5 lines · Glen Nakamura
Of course, perhaps 0 should passed instead of "" for
data_page?(quoted diff) +1
-1- err = do_mount ("none", "/dev", "devfs", 0, ""); + err = do_mount ("none", "/dev", "devfs", 0, 0);
Quoted 6 lines
Quoted 1 line
Comments?
Yes, the devfs fix above is needed if the
data_pagepatch has been applied.This is the case in 2.6, but not 2.4.25.
Hmm the data_page line is in my copy of 2.4.25...
Signature · 4 lines
Debian GNU/Linux 3.0 is out! ( http://www.debian.org/ )
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txtQuoted 7 lines · James Morris
Quoted 5 lines · Glen Nakamura
Of course, perhaps 0 should passed instead of "" for
data_page?(quoted diff) +1
-1- err = do_mount ("none", "/dev", "devfs", 0, ""); + err = do_mount ("none", "/dev", "devfs", 0, 0);
Quoted 6 lines
Quoted 1 line
Comments?
Yes, the devfs fix above is needed if the
data_pagepatch has been applied.This is the case in 2.6, but not 2.4.25.
My bad :(
Changed last argument of fs/devfs/base.c do_mount() call to NULL, as per 2.6.
James, wonder if your change can't cause other similar problems in 2.4?
Quoted 16 lines · Marcelo Tosatti
Quoted 12 lines · James Morris
Quoted 6 lines · Glen Nakamura
Of course, perhaps 0 should passed instead of "" for
data_page?(quoted diff) +1
-1- err = do_mount ("none", "/dev", "devfs", 0, ""); + err = do_mount ("none", "/dev", "devfs", 0, 0);Comments?
Yes, the devfs fix above is needed if the
data_pagepatch has been applied.
Quoted 8 lines
Quoted 1 line
This is the case in 2.6, but not 2.4.25.
My bad :(
Changed last argument of
fs/devfs/base.cdo_mount()call to NULL, as per 2.6.James, wonder if your change can't cause other similar problems in 2.4?
Looks like sunos_nfs_mount() needs to be fixed to pass a page as the last argument of do_mount().
- James
Signature · 2 lines
James Morris <[email protected]>