Linux CIFS filesystem development
 help / color / mirror / Atom feed
From: Michael Bommarito <michael.bommarito@gmail.com>
To: Namjae Jeon <linkinjeon@kernel.org>, Steve French <smfrench@gmail.com>
Cc: Sergey Senozhatsky <senozhatsky@chromium.org>,
	Tom Talpey <tom@talpey.com>, Hyunchul Lee <hyc.lee@gmail.com>,
	Ronnie Sahlberg <lsahlber@redhat.com>,
	Marios Makassikis <mmakassikis@freebox.fr>,
	linux-cifs@vger.kernel.org, linux-kernel@vger.kernel.org,
	stable@vger.kernel.org
Subject: [PATCH] ksmbd: reject unsigned non-exempt requests when session requires signing
Date: Tue, 19 May 2026 23:19:23 -0400	[thread overview]
Message-ID: <20260520031923.3679744-1-michael.bommarito@gmail.com> (raw)

This is the ksmbd SMB2 sibling of Samba CVE-2023-3347 ("SMB2
packet signing not enforced"): when the server-side policy requires
SMB2 signing, unsigned non-exempt requests must be rejected instead
of processed. CVE-2016-2114 is the older Samba SMB1 analogue for
mandatory server signing not being enforced.

When a ksmbd target is configured with "server signing = mandatory"
and a client completes a signed SESSION_SETUP, an attacker on the
network path of that connection can read the cleartext SessionId
from the SMB2 header and inject unsigned TREE_CONNECT, CREATE, and
WRITE requests on the victim's TCP flow. ksmbd processes the
unsigned PDUs and commits attacker-supplied content to the share
under the victim session's authority, without the attacker holding
the victim's credentials.

In fs/smb/server/server.c, __process_request() verifies signatures
only when conn->ops->is_sign_req() returns true. The predicate
(smb2_is_sign_req() at fs/smb/server/smb2pdu.c:8936-8947) reads
only the client-set SMB2_FLAGS_SIGNED and never work->sess->sign --
the per-session "signing required" state set during SESSION_SETUP
at smb2pdu.c:1546 and :1645. MS-SMB2 3.3.5.2.4 requires the server
to reject any unsigned non-exempt request when
Session.SigningRequired is TRUE.

Reject directly in __process_request() when work->sess->sign is
set and the inbound non-exempt request lacks SMB2_FLAGS_SIGNED,
bypassing check_sign_req() to avoid emitting
pr_err("bad smb2 signature") at KERN_ERR for each unsigned PDU.
The set of commands exempt from signing (NEGOTIATE, SESSION_SETUP,
OPLOCK_BREAK) is unchanged.

Fixes: e2f34481b24d ("cifsd: add server-side procedures for SMB3")
Cc: stable@vger.kernel.org
Signed-off-by: Michael Bommarito <michael.bommarito@gmail.com>
Assisted-by: Claude:claude-opus-4-7
---

Reproduction
============

Tree: linux-mainline 2f448dd9ef4e, x86_64 QEMU+KVM,
CONFIG_SMB_SERVER=y, CONFIG_KASAN=y. Target ksmbd.conf has
"server signing = mandatory", "map to guest = Never",
"valid users = alice"; SMB user alice was added via
ksmbd.adduser.

Conditions: ksmbd configured with "server signing = mandatory"
or a client requesting SMB2_NEGOTIATE_SIGNING_REQUIRED, so
work->sess->sign is set to TRUE after SESSION_SETUP completes.
Encryption MUST NOT be negotiated on the session (smb2_sess_setup
clears sess->sign to FALSE at smb2pdu.c:1558/:1657 when
smb3_encryption_negotiated() is TRUE). The PoCs force SMB 2.1 to
keep the session in the signing-required state.

Harness: a wire-level Python client establishes a normal NTLMv2
SESSION_SETUP on a fresh TCP connection, then on the same
connection issues TREE_CONNECT, CREATE, and WRITE with
SMB2_FLAGS_SIGNED cleared and the Signature field zeroed. A
companion transparent TCP MITM proxy demonstrates the same
primitive without holding credentials: it observes SessionId from
the cleartext SESSION_SETUP response and injects an unsigned
TREE_CONNECT, CREATE, and WRITE into the victim's TCP flow using
that SessionId; the victim's own SMB session continues normally
and observes the attacker's file appear on the share.

Stock kernel: unsigned TREE_CONNECT returns STATUS_SUCCESS with a
tree_id; CREATE returns STATUS_SUCCESS with a file id; WRITE
returns STATUS_SUCCESS and the attacker payload lands on the
share path. The response header carries SMB2_FLAGS_SIGNED with a
non-zero MAC, confirming work->sess->sign was TRUE on this
session.

Patched kernel: same harness, the dispatcher returns
STATUS_ACCESS_DENIED on the first unsigned TREE_CONNECT and
nothing reaches the share. dmesg shows zero
"bad smb2 signature" entries; the rejection bypasses
check_sign_req() so the existing pr_err path is not exercised.
Regression: a legitimate smbclient session with
--client-protection=sign continues to read and write to the
share with no observable change.

Mitigations: until patched, deployments can leave
"server signing = mandatory" disabled in ksmbd.conf (the bypass
is meaningful only when the session sets work->sess->sign), or
configure clients to negotiate SMB3 encryption (sess->enc=TRUE
clears sess->sign and the session uses TRANSFORM_HEADER
encryption for the data path).

Selftests: grep over tools/testing/selftests/ on the patched tree
returns 0 references to ksmbd or smb2_is_sign_req; there is no
in-tree selftest binary that exercises fs/smb/server/server.c's
dispatch gate. The trigger Python client and the TCP MITM proxy
are available off-list on request.
---
 fs/smb/server/server.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/fs/smb/server/server.c b/fs/smb/server/server.c
index 58ef02c423fce..5755f907f29f4 100644
--- a/fs/smb/server/server.c
+++ b/fs/smb/server/server.c
@@ -143,6 +143,15 @@ static int __process_request(struct ksmbd_work *work, struct ksmbd_conn *conn,
 			conn->ops->set_rsp_status(work, STATUS_ACCESS_DENIED);
 			return SERVER_HANDLER_ABORT;
 		}
+	} else if (work->sess && work->sess->sign &&
+		   command != SMB2_NEGOTIATE_HE &&
+		   command != SMB2_SESSION_SETUP_HE &&
+		   command != SMB2_OPLOCK_BREAK_HE) {
+		/* MS-SMB2 3.3.5.2.4: Session.SigningRequired==TRUE,
+		 * reject unsigned non-exempt request.
+		 */
+		conn->ops->set_rsp_status(work, STATUS_ACCESS_DENIED);
+		return SERVER_HANDLER_ABORT;
 	}
 
 	ret = cmds->proc(work);
-- 
2.53.0

                 reply	other threads:[~2026-05-20  3:19 UTC|newest]

Thread overview: [no followups] expand[flat|nested]  mbox.gz  Atom feed

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20260520031923.3679744-1-michael.bommarito@gmail.com \
    --to=michael.bommarito@gmail.com \
    --cc=hyc.lee@gmail.com \
    --cc=linkinjeon@kernel.org \
    --cc=linux-cifs@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=lsahlber@redhat.com \
    --cc=mmakassikis@freebox.fr \
    --cc=senozhatsky@chromium.org \
    --cc=smfrench@gmail.com \
    --cc=stable@vger.kernel.org \
    --cc=tom@talpey.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox