On Sat, Nov 09, 2024 at 08:40:30PM +0800, Kuan-Wei Chiu wrote:
> Hi Jinjie,
>
> On Thu, Oct 24, 2024 at 05:43:03PM +0800, Jinjie Ruan wrote:
> > In kunit_debugfs_create_suite(), if alloc_string_stream() fails in the
> > kunit_suite_for_each_test_case() loop, the "suite->log = stream"
> > has assigned before, and the error path only free the suite->log's stream
> > memory but not set it to NULL in string_stream_destroy(), so the later
> > string_stream_clear() of suite->log in kunit_init_suite() will cause
> > below UAF bug.
> >
> > Set stream pointer to NULL after free in string_stream_destroy()
> > to fix it.
> >
> > Unable to handle kernel paging request at virtual address 006440150000030d
> > Mem abort info:
> > ESR = 0x0000000096000004
> > EC = 0x25: DABT (current EL), IL = 32 bits
> > SET = 0, FnV = 0
> > EA = 0, S1PTW = 0
> > FSC = 0x04: level 0 translation fault
> > Data abort info:
> > ISV = 0, ISS = 0x00000004, ISS2 = 0x00000000
> > CM = 0, WnR = 0, TnD = 0, TagAccess = 0
> > GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
> > [006440150000030d] address between user and kernel address ranges
> > Internal error: Oops: 0000000096000004 [#1] PREEMPT SMP
> > Dumping ftrace buffer:
> > (ftrace buffer empty)
> > Modules linked in: iio_test_gts industrialio_gts_helper cfg80211 rfkill ipv6 [last unloaded: iio_test_gts]
> > CPU: 5 UID: 0 PID: 6253 Comm: modprobe Tainted: G B W N 6.12.0-rc4+ #458
> > Tainted: [B]=BAD_PAGE, [W]=WARN, [N]=TEST
> > Hardware name: linux,dummy-virt (DT)
> > pstate: 40000005 (nZcv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> > pc : string_stream_clear+0x54/0x1ac
> > lr : string_stream_clear+0x1a8/0x1ac
> > sp : ffffffc080b47410
> > x29: ffffffc080b47410 x28: 006440550000030d x27: ffffff80c96b5e98
> > x26: ffffff80c96b5e80 x25: ffffffe461b3f6c0 x24: 0000000000000003
> > x23: ffffff80c96b5e88 x22: 1ffffff019cdf4fc x21: dfffffc000000000
> > x20: ffffff80ce6fa7e0 x19: 032202a80000186d x18: 0000000000001840
> > x17: 0000000000000000 x16: 0000000000000000 x15: ffffffe45c355cb4
> > x14: ffffffe45c35589c x13: ffffffe45c03da78 x12: ffffffb810168e75
> > x11: 1ffffff810168e74 x10: ffffffb810168e74 x9 : dfffffc000000000
> > x8 : 0000000000000004 x7 : 0000000000000003 x6 : 0000000000000001
> > x5 : ffffffc080b473a0 x4 : 0000000000000000 x3 : 0000000000000000
> > x2 : 0000000000000001 x1 : ffffffe462fbf620 x0 : dfffffc000000000
> > Call trace:
> > string_stream_clear+0x54/0x1ac
> > __kunit_test_suites_init+0x108/0x1d8
> > kunit_exec_run_tests+0xb8/0x100
> > kunit_module_notify+0x400/0x55c
> > notifier_call_chain+0xfc/0x3b4
> > blocking_notifier_call_chain+0x68/0x9c
> > do_init_module+0x24c/0x5c8
> > load_module+0x4acc/0x4e90
> > init_module_from_file+0xd4/0x128
> > idempotent_init_module+0x2d4/0x57c
> > __arm64_sys_finit_module+0xac/0x100
> > invoke_syscall+0x6c/0x258
> > el0_svc_common.constprop.0+0x160/0x22c
> > do_el0_svc+0x44/0x5c
> > el0_svc+0x48/0xb8
> > el0t_64_sync_handler+0x13c/0x158
> > el0t_64_sync+0x190/0x194
> > Code: f9400753 d2dff800 f2fbffe0 d343fe7c (38e06b80)
> > ---[ end trace 0000000000000000 ]---
> > Kernel panic - not syncing: Oops: Fatal exception
> >
> > Cc: stable@xxxxxxxxxxxxxxx
> > Fixes: a3fdf784780c ("kunit: string-stream: Decouple string_stream from kunit")
> > Signed-off-by: Jinjie Ruan <ruanjinjie@xxxxxxxxxx>
> > ---
> > lib/kunit/string-stream.c | 1 +
> > 1 file changed, 1 insertion(+)
> >
> > diff --git a/lib/kunit/string-stream.c b/lib/kunit/string-stream.c
> > index 54f4fdcbfac8..00ad518b730b 100644
> > --- a/lib/kunit/string-stream.c
> > +++ b/lib/kunit/string-stream.c
> > @@ -178,6 +178,7 @@ void string_stream_destroy(struct string_stream *stream)
> >
> > string_stream_clear(stream);
> > kfree(stream);
> > + stream = NULL;
>
> Thanks for proposing the fix. However, I don't believe it fully
> resolves the UAF issue. Changing the stream pointer value within
> string_stream_destroy() doesn't update the caller's stream pointer, so
> the issue remains.
>
> I think the correct approach would be to set the log pointer to NULL in
> the error path of kunit_debugfs_create_suite() to address the issue.
> Let me know if I've missed something or overlooked an obvious detail.
>
BTW, since alloc_string_stream() returns -ENOMEM on error rather than
NULL, it's best for the caller to check it with IS_ERR() instead of
IS_ERR_OR_NULL(). It's not a big issue, but I can send a cleanup patch
later to address this.
Regards,
Kuan-Wei
