Re: scsi: BUG in scsi_init_io

From: Al Viro
Date: Sun Feb 19 2017 - 02:16:36 EST

On Tue, Jan 31, 2017 at 07:41:51AM -0800, James Bottomley wrote:

> > Please-please-please, let's not use WARN for something that is not a
> > kernel bug and is user-triggerable.
> It is a kernel bug and it should not be user triggerable, so it should
> have a warn_on or bug_on. It means something called a data setup
> function with no data. There's actually a root cause that patches like
> this won't fix, can we find it?

The root cause is unfixable without access to TARDIS and dose of
antipsychotics sufficient to prevent /dev/sg API creation.

What happens is that write to /dev/sg is given a request with non-zero
->iovec_count combined with zero ->dxfer_len. Or with ->dxferp pointing
to an array full of empty iovecs.

AFAICS, the minimal fix would be something like this:

YAMissingSanityCheck in /dev/sg

write permission to /dev/sg shouldn't be equivalent to the ability to trigger
BUG_ON() while holding spinlocks...

Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Al Viro <viro@xxxxxxxxxxxxxxxxxx>

diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c
index dbe5b4b95df0..121de0aaa6ad 100644
--- a/drivers/scsi/sg.c
+++ b/drivers/scsi/sg.c
@@ -1753,6 +1753,10 @@ sg_start_req(Sg_request *srp, unsigned char *cmd)
return res;

iov_iter_truncate(&i, hp->dxfer_len);
+ if (!iov_iter_count(&i)) {
+ kfree(iov);
+ return -EINVAL;
+ }

res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);