On 06. 05. 24, 20:55, Justin Stitt wrote:
Using the signed overflow sanitizer with syzkaller produces this UBSAN
report:
[ 31.304043] ------------[ cut here ]------------
[ 31.304048] UBSAN: signed-integer-overflow in ../drivers/tty/vt/vt.c:309:19
[ 31.304055] -2147483648 + -1073741824 cannot be represented in type 'int'
[ 31.304066] CPU: 1 PID: 3894 Comm: syz-executor Not tainted 6.8.0-rc2-00035-gb3ef86b5a957 #1
[ 31.304073] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014
[ 31.304077] Call Trace:
[ 31.304080] <TASK>
[ 31.304083] dump_stack_lvl+0x93/0xd0
[ 31.304177] handle_overflow+0x171/0x1b0
[ 31.304186] scrollfront+0xcb/0xd0
[ 31.304196] tioclinux+0x3cc/0x450
[ 31.304205] tty_ioctl+0x7fc/0xc00
The rest of the stack trace can be trimmed:
[ 31.304212] ? __pfx_tty_ioctl+0x10/0x10
[ 31.304219] __se_sys_ioctl+0xe0/0x140
[ 31.304228] do_syscall_64+0xd7/0x1b0
[ 31.304236] ? arch_exit_to_user_mode_prepare+0x11/0x60
[ 31.304244] entry_SYSCALL_64_after_hwframe+0x6f/0x77
[ 31.304254] RIP: 0033:0x7fc3902ae539
[ 31.304263] Code: 28 00 00 00 75 05 48 83 c4 28 c3 e8 f1 14 00 00 90 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 8
[ 31.304282] RSP: 002b:00007ffc8a457998 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
[ 31.304289] RAX: ffffffffffffffda RBX: 00007fc3903e2f80 RCX: 00007fc3902ae539
[ 31.304293] RDX: 0000000020000040 RSI: 000000000000541c RDI: 0000000000000003
[ 31.304297] RBP: 00007fc39030d496 R08: 0000000000000000 R09: 0000000000000000
[ 31.304300] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
[ 31.304304] R13: 0000000000000800 R14: 00007fc3903e2f80 R15: 00007fc3903e2f80
[ 31.304310] </TASK>
[ 31.304371] ---[ end trace ]---
This is caused by the scrollback_delta overflowing. Historically, the
signed integer overflow sanitizer did not work in the kernel due to its
interaction with `-fwrapv` but this has since been changed [1] in the
newest version of Clang; It being re-enabled in the kernel with Commit
557f8c582a9ba8ab ("ubsan: Reintroduce signed overflow sanitizer").
Note that it would be difficult to reproduce this bug in a non-fuzzing
scenario as it requires inputting tons of scroll inputs via keyboard
before the scheduled console callback has had a chance to update.
Nonetheless, let's saturate scrollback_delta so it stays clamped to
integer bounds without wrapping around.
And what is actually broken, given signed overflow is well defined under
the -fwrapv wings?
[1]: https://github.com/llvm/llvm-project/pull/82432
Closes: https://github.com/KSPP/linux/issues/351
Signed-off-by: Justin Stitt <justinstitt@xxxxxxxxxx>
---
Note: I am using Kees' SIO tree as a base:
https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git/log/?h=dev/v6.8-rc2/signed-overflow-sanitizer
---
drivers/tty/vt/vt.c | 9 ++++++++-
1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/drivers/tty/vt/vt.c b/drivers/tty/vt/vt.c
index 9b5b98dfc8b4..b4768336868e 100644
--- a/drivers/tty/vt/vt.c
+++ b/drivers/tty/vt/vt.c
@@ -308,7 +308,14 @@ static inline void scrolldelta(int lines)
/* FIXME */
/* scrolldelta needs some kind of consistency lock, but the BKL was
and still is not protecting versus the scheduled back end */
- scrollback_delta += lines;
+
+ /* saturate scrollback_delta so that it never wraps around */
+ if (lines > 0 && unlikely(INT_MAX - lines < scrollback_delta))
+ scrollback_delta = INT_MAX;
+ else if (lines < 0 && unlikely(INT_MIN - lines > scrollback_delta))
+ scrollback_delta = INT_MIN;
+ else
+ scrollback_delta += lines;
NACK, this is horrid.
Introduce a helper for this in overflow.h if we have none yet.
thanks,
--
js
suse labs
