On Thu, 26 Jan 2012, Gleb Smirnoff wrote:
Log:
Although aio_nbytes is size_t, later is is signed to
casted types: to ssize_t in filesystem code and to
int in buf code, thus supplying a negative argument
leads to kernel panic later.
And supplying a large positive argument leads to undefined
behaviour later in the buf case.
To fix that check user
supplied argument in the beginning of syscall.
Now, supplying a large positive argument gives implementation-
defined behaviour immediately. Sometimes the implementation-
defined behaviour is to have no effect immediately and allow
a panic later.
Modified: head/sys/kern/vfs_aio.c
==============================================================================
--- head/sys/kern/vfs_aio.c Thu Jan 26 11:15:12 2012 (r230582)
+++ head/sys/kern/vfs_aio.c Thu Jan 26 11:59:48 2012 (r230583)
@@ -1552,6 +1552,12 @@ aio_aqueue(struct thread *td, struct aio
return (error);
}
+ /* XXX: aio_nbytes is later casted to signed types. */
+ if ((int)aiocbe->uaiocb.aio_nbytes < 0) {
This should avoid implementation-defined behaviour by checking if
(uncast)aiocbe->uaiocb.aio_nbytes > INT_MAX.
It isn't clear what the effects of the implementation-defined behaviour
are even when you know what it is. It certainly results in the
check passing values that exceed INT_MAX. For example, with 32-bit
int and 64-bit size_t, and everything 2's complement, and no perverse
implementation-defined behaviour like "the result of converting to
int when the argment exceeds INT_MAX is always 42", then
(int)0x10000000 is 0.
+ uma_zfree(aiocb_zone, aiocbe);
+ return (EINVAL);
+ }
+
if (aiocbe->uaiocb.aio_sigevent.sigev_notify != SIGEV_KEVENT &&
aiocbe->uaiocb.aio_sigevent.sigev_notify != SIGEV_SIGNAL &&
aiocbe->uaiocb.aio_sigevent.sigev_notify != SIGEV_THREAD_ID &&
Bruce
_______________________________________________
svn-src-head@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/svn-src-head
To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"
