Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <dd7c6170def7159b558b79d34520c3d5290e36b9.camel@redhat.com>
Subject: Re: [RFC PATCH 0/3] scsi: qedf: sanitise uaccess
From:   Laurence Oberman <loberman@redhat.com>
To:     Oleksandr Natalenko <oleksandr@redhat.com>,
        linux-kernel@vger.kernel.org
Cc:     linux-scsi@vger.kernel.org, Saurav Kashyap <skashyap@marvell.com>,
        Javed Hasan <jhasan@marvell.com>,
        GR-QLogic-Storage-Upstream@marvell.com,
        "James E.J. Bottomley" <jejb@linux.ibm.com>,
        "Martin K. Petersen" <martin.petersen@oracle.com>,
        Jozef Bacik <jobacik@redhat.com>, Rob Evers <revers@redhat.com>
Date:   Mon, 24 Jul 2023 09:03:32 -0400
In-Reply-To: <20230724120241.40495-1-oleksandr@redhat.com>
References: <20230724120241.40495-1-oleksandr@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.46.4 (3.46.4-1.fc37) 
MIME-Version: 1.0
Precedence: bulk

On Mon, 2023-07-24 at 14:02 +0200, Oleksandr Natalenko wrote:
> qedf driver, debugfs part of it specifically, touches __user pointers
> directly for printing out info to userspace via sprintf(), which may
> cause crash like this:
>=20
> BUG: unable to handle kernel paging request at 00007ffd1d6b43a0
> IP: [<ffffffffaa7a882a>] string.isra.7+0x6a/0xf0
> Oops: 0003 [#1] SMP
> Call Trace:
> =C2=A0[<ffffffffaa7a9f31>] vsnprintf+0x201/0x6a0
> =C2=A0[<ffffffffaa7aa556>] sprintf+0x56/0x80
> =C2=A0[<ffffffffc04227ed>] qedf_dbg_stop_io_on_error_cmd_read+0x6d/0x90
> [qedf]
> =C2=A0[<ffffffffaa65bb2f>] vfs_read+0x9f/0x170
> =C2=A0[<ffffffffaa65cb82>] SyS_pread64+0x92/0xc0
>=20
> Avoid this by preparing the info in a kernel buffer first, either
> allocated on stack for small printouts, or via vmalloc() for big
> ones,
> and then copying it to the userspace properly.
>=20
> I'm not sure how big the vmalloc()'ed buffer should be, and also
> whether
> vmalloc()'ing it directly in the _read() function is a good idea,
> hence
> RFC prefix.
>=20
> The qedf_dbg_stop_io_on_error_cmd_read()-related patch is actually
> tested,
> the rest is compile-tested only.
>=20
> Oleksandr Natalenko (3):
> =C2=A0 scsi: qedf: do not touch __user pointer in
> =C2=A0=C2=A0=C2=A0 qedf_dbg_stop_io_on_error_cmd_read() directly
> =C2=A0 scsi: qedf: do not touch __user pointer in
> qedf_dbg_debug_cmd_read()
> =C2=A0=C2=A0=C2=A0 directly
> =C2=A0 scsi: qedf: do not touch __user pointer in
> qedf_dbg_fp_int_cmd_read()
> =C2=A0=C2=A0=C2=A0 directly
>=20
> =C2=A0drivers/scsi/qedf/qedf_dbg.h=C2=A0=C2=A0=C2=A0=C2=A0 |=C2=A0 2 ++
> =C2=A0drivers/scsi/qedf/qedf_debugfs.c | 35 +++++++++++++++++++----------=
-
> --
> =C2=A02 files changed, 23 insertions(+), 14 deletions(-)
>=20
> --
> 2.41.0
>=20

I will test the series, the one patch was already tested.
This was reproduced in our LAB
Will report back after testing

Regards
Laurence