On Wed, Dec 16, 2015 at 07:38:33PM +0100, Dmitry Vyukov wrote:
> > Yup, looks correct. Will you send a patch?
>
> I've drafted the verification:
>
> @@ -1269,6 +1269,8 @@ static long read_events(struct kioctx *ctx, long
> min_nr, long nr,
>
> if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
> return -EFAULT;
> + if (!timespec_valid_strict(&strict))
> + return -EINVAL;
>
> until = timespec_to_ktime(ts);
> }
>
> But now I am thinking whether it is the right solution.
> First, user does not know about KTIME_MAX, so it is not unreasonable
> to pass timespec{INT64_MAX, INT64_MAX} as timeout expecting that it
> will block for a long time. And it actually probably mostly works now,
> because after the overflow you still get something large with high
> probability. If we do the fix, then users will need to pass seconds <
> KTIME_MAX, while they don't know KTIME_MAX value.
> Second, there seems to be more serious issue in ktime_set() which
> checks seconds for KTIME_MAX, but on the next line addition still
> overflows int64.
> Thoughts?
I finally had some time to look over this after the holidays, and I
don't think using timespec_valid_strict() is the right approach here,
as userspace will have no idea what KTIME_MAX is. Instead, I think the
right approach is to -EINVAL for negative values (which should avoid
the overflow), and to allow too large values to be silently truncated
by timespec_to_ktime(). The truncation doesn't matter all that much
given that it's in the hundreds of years ballpark. I'll push the patch
below if there are no objections.
-ben
--
"Thought is the essence of where you are now."
commit 4304367826d0df42086ef24428c6c277acd822a9
Author: Benjamin LaHaise <
bc...@kvack.org>
Date: Wed Jan 6 12:46:12 2016 -0500
aio: handle integer overflow in io_getevents() timespec usage
Dmitry Vyukov reported an integer overflow in io_getevents() when
running a fuzzer. Upon investigation, the triggers appears to be that a
negative value for the tv_sec or tv_nsec was passed in which is not
handled by timespec_to_ktime(). This patch fixes that by making
io_getevents() return -EINVAL when negative timeouts are passed in.
Signed-off-by: Benjamin LaHaise <
bc...@kvack.org>
diff --git a/fs/aio.c b/fs/aio.c
index 155f842..f325ed4 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -1269,6 +1269,8 @@ static long read_events(struct kioctx *ctx, long min_nr, long nr,
if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
return -EFAULT;
+ if ((ts.tv_sec < 0) || (ts.tv_nsec < 0))