cifs: Fix encryption of cleared, but unset rq_iter data buffers
authorDavid Howells <dhowells@redhat.com>
Mon, 6 Nov 2023 14:40:11 +0000 (14:40 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 28 Nov 2023 17:20:09 +0000 (17:20 +0000)
commit0c00e422bf3dea146cd02e7f4bea951a66f4646e
treed3ab478ea548d8d666d3f238d8e82d69829f5436
parent739bf98ce9deebbd12000cba81d2b48ee1d009de
cifs: Fix encryption of cleared, but unset rq_iter data buffers

commit 37de5a80e932f828c34abeaae63170d73930dca3 upstream.

Each smb_rqst struct contains two things: an array of kvecs (rq_iov) that
contains the protocol data for an RPC op and an iterator (rq_iter) that
contains the data payload of an RPC op.  When an smb_rqst is allocated
rq_iter is it always cleared, but we don't set it up unless we're going to
use it.

The functions that determines the size of the ciphertext buffer that will
be needed to encrypt a request, cifs_get_num_sgs(), assumes that rq_iter is
always initialised - and employs user_backed_iter() to check that the
iterator isn't user-backed.  This used to incidentally work, because
->user_backed was set to false because the iterator has never been
initialised, but with commit f1b4cb650b9a0eeba206d8f069fcdc532bfbcd74[1]
which changes user_backed_iter() to determine this based on the iterator
type insted, a warning is now emitted:

        WARNING: CPU: 7 PID: 4584 at fs/smb/client/cifsglob.h:2165 smb2_get_aead_req+0x3fc/0x420 [cifs]
        ...
        RIP: 0010:smb2_get_aead_req+0x3fc/0x420 [cifs]
        ...
         crypt_message+0x33e/0x550 [cifs]
         smb3_init_transform_rq+0x27d/0x3f0 [cifs]
         smb_send_rqst+0xc7/0x160 [cifs]
         compound_send_recv+0x3ca/0x9f0 [cifs]
         cifs_send_recv+0x25/0x30 [cifs]
         SMB2_tcon+0x38a/0x820 [cifs]
         cifs_get_smb_ses+0x69c/0xee0 [cifs]
         cifs_mount_get_session+0x76/0x1d0 [cifs]
         dfs_mount_share+0x74/0x9d0 [cifs]
         cifs_mount+0x6e/0x2e0 [cifs]
         cifs_smb3_do_mount+0x143/0x300 [cifs]
         smb3_get_tree+0x15e/0x290 [cifs]
         vfs_get_tree+0x2d/0xe0
         do_new_mount+0x124/0x340
         __se_sys_mount+0x143/0x1a0

The problem is that rq_iter was never set, so the type is 0 (ie. ITER_UBUF)
which causes user_backed_iter() to return true.  The code doesn't
malfunction because it checks the size of the iterator - which is 0.

Fix cifs_get_num_sgs() to ignore rq_iter if its count is 0, thereby
bypassing the warnings.

It might be better to explicitly initialise rq_iter to a zero-length
ITER_BVEC, say, as it can always be reinitialised later.

Fixes: d08089f649a0 ("cifs: Change the I/O paths to use an iterator rather than a page list")
Reported-by: Damian Tometzki <damian@riscv-rocks.de>
Closes: https://lore.kernel.org/r/ZUfQo47uo0p2ZsYg@fedora.fritz.box/
Tested-by: Damian Tometzki <damian@riscv-rocks.de>
Cc: stable@vger.kernel.org
cc: Eric Biggers <ebiggers@kernel.org>
cc: linux-cifs@vger.kernel.org
cc: linux-fsdevel@vger.kernel.org
Link: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f1b4cb650b9a0eeba206d8f069fcdc532bfbcd74
Reviewed-by: Paulo Alcantara (SUSE) <pc@manguebit.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/smb/client/cifsglob.h