Re: [RFC PATCH] aio: add check for timeout to aviod invalid value

[Jeff Moyer]

Tan Xiaojun <tanxiaojun@xxxxxxxxxx> writes:

> (When I was testing with syzkaller, I found a lot of ubsan problems. Here 
> is one of them. I am not sure if it needs to be fixed and how it will be 
> fixed. So I sent this patch to ask your opinion.)
>
> Syzkaller reported a UBSAN bug below, which was mainly caused by a large
> negative number passed to the timeout of the io_getevents system call.
>
> ================================================================================
> UBSAN: Undefined behaviour in ./include/linux/ktime.h:42:14
> signed integer overflow:
> -8427032702788048137 * 1000000000 cannot be represented in type 'long long int'
> CPU: 3 PID: 11668 Comm: syz-executor0 Not tainted 4.19.18-514.55.6.9.x86_64+ #1
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014
> Call Trace:
>  __dump_stack lib/dump_stack.c:77 [inline]
>  dump_stack+0xca/0x13e lib/dump_stack.c:113
>  ubsan_epilogue+0xe/0x81 lib/ubsan.c:159
>  handle_overflow+0x193/0x1e2 lib/ubsan.c:190
>  ktime_set include/linux/ktime.h:42 [inline]
>  timespec64_to_ktime include/linux/ktime.h:78 [inline]
>  do_io_getevents+0x307/0x390 fs/aio.c:2043
>  __do_sys_io_getevents fs/aio.c:2080 [inline]
>  __se_sys_io_getevents fs/aio.c:2068 [inline]
>  __x64_sys_io_getevents+0x163/0x250 fs/aio.c:2068
>  do_syscall_64+0xc8/0x580 arch/x86/entry/common.c:290
>  entry_SYSCALL_64_after_hwframe+0x49/0xbe
> RIP: 0033:0x462589
> Code: f7 d8 64 89 02 b8 ff ff ff ff c3 66 0f 1f 44 00 00 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 bc ff ff ff f7 d8 64 89 01 48
> RSP: 002b:00007fde9b04ec58 EFLAGS: 00000246 ORIG_RAX: 00000000000000d0
> RAX: ffffffffffffffda RBX: 000000000072bf00 RCX: 0000000000462589
> RDX: 0000000000000006 RSI: 0000000000000000 RDI: 0000000000000000
> RBP: 0000000000000005 R08: 0000000020000100 R09: 0000000000000000
> R10: 0000000020000040 R11: 0000000000000246 R12: 00007fde9b04f6bc
> R13: 00000000004bd1f0 R14: 00000000006f6b60 R15: 00000000ffffffff
> ================================================================================
> bond0 (unregistering): Released all slaves
>
> The timeout described in "man io_getevents" does not say whether it
> can be negative or not, but as a waiting time, a negative number has
> no meaning. So I add check to avoid this case.

It's embarrassing that this bug is still present.  See, for example,
this discussion, started in 2015:
  https://lore.kernel.org/lkml/CACT4Y+bBxVYLQ6LtOKrKtnLthqLHcw-BMp3aqP3mjdAvr9FULQ@xxxxxxxxxxxxxx/

I could swear it was brought up again since then, but I can't find
records of that.

> Signed-off-by: Tan Xiaojun <tanxiaojun@xxxxxxxxxx>
> ---
>  fs/aio.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/fs/aio.c b/fs/aio.c
> index aaaaf4d..28e0fa6 100644
> --- a/fs/aio.c
> +++ b/fs/aio.c
> @@ -2078,10 +2078,15 @@ static long do_io_getevents(aio_context_t ctx_id,
>  		struct io_event __user *events,
>  		struct timespec64 *ts)
>  {
> -	ktime_t until = ts ? timespec64_to_ktime(*ts) : KTIME_MAX;
> +	ktime_t until;
>  	struct kioctx *ioctx = lookup_ioctx(ctx_id);
>  	long ret = -EINVAL;
>  
> +	if (ts && !timespec64_valid(ts))
> +		return -EINVAL;
> +
> +	until = ts ? timespec64_to_ktime(*ts) : KTIME_MAX;
> +
>  	if (likely(ioctx)) {
>  		if (likely(min_nr <= nr && min_nr >= 0))
>  			ret = read_events(ioctx, min_nr, nr, events, until);

Looks good to me.  Thanks for fixing this.

Reviewed-by: Jeff Moyer <jmoyer@xxxxxxxxxx>