* [PATCH] ksmbd: read request header for compounded CANCEL
@ 2026-11-08 22:30 Noah Weinberger
2026-11-08 23:46 ` Ian Bridges
0 siblings, 1 reply; 3+ messages in thread
From: Noah Weinberger @ 2026-11-08 22:30 UTC (permalink / raw)
To: linkinjeon; +Cc: linux-cifs, linux-kernel, security
Hello,
For a compounded CANCEL, smb2_cancel() matches the cancel against the
*response* header:
if (work->next_smb2_rcv_hdr_off)
hdr = ksmbd_resp_buf_next(work);
The chained response header is initialized with Flags =
SERVER_TO_REDIR|RELATED, so SMB2_FLAGS_ASYNC_COMMAND is never set there:
an async-form CANCEL therefore always takes the MessageId branch and
matches nothing, i.e. a compounded async CANCEL is a silent no-op and
the deferred work keeps running until it completes, times out or the
connection closes. MS-SMB2 forbids compounding CANCEL, so only
non-conforming clients are affected, and the read itself is always
in-bounds (the response header is fully initialized).
Read the request header for the compounded case.
Affected version/commit: mainline fa4ad798f6f681031775266044ba1b0cf3c670bd (v7.4-rc1, 2026-11-07), fs/smb/server/smb2pdu.c:10515-10524 (smb2_cancel).
Pre-existing in v7.3.
This was found during an assisted (AI-supported) static source review of
mainline fa4ad798f6f681031775266044ba1b0cf3c670bd (v7.4-rc1, 2026-11-07); every code path cited below was re-verified by the
reporter against that tree. The patch below has NOT been compiled and has
NOT been runtime tested, and no reproducer was produced - the triggering
conditions are described above instead. Please treat this as a report for
verification rather than a battle-tested fix.
Reported-by: Noah Weinberger <noahweinberger2003@proton.me>
---
diff --git a/fs/smb/server/smb2pdu.c b/fs/smb/server/smb2pdu.c
index b084cca..36c5a40 100644
--- a/fs/smb/server/smb2pdu.c
+++ b/fs/smb/server/smb2pdu.c
@@ -10521,7 +10521,7 @@ int smb2_cancel(struct ksmbd_work *work)
struct list_head *command_list;
if (work->next_smb2_rcv_hdr_off)
- hdr = ksmbd_resp_buf_next(work);
+ hdr = ksmbd_req_buf_next(work);
ksmbd_debug(SMB, "smb2 cancel called on mid %llu, async flags 0x%x\n",
le64_to_cpu(hdr->MessageId),
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] ksmbd: read request header for compounded CANCEL
2026-11-08 22:30 [PATCH] ksmbd: read request header for compounded CANCEL Noah Weinberger
@ 2026-11-08 23:46 ` Ian Bridges
2026-11-08 23:54 ` Noah Weinberger
0 siblings, 1 reply; 3+ messages in thread
From: Ian Bridges @ 2026-11-08 23:46 UTC (permalink / raw)
To: Noah Weinberger; +Cc: linkinjeon, linux-cifs, linux-kernel, Ian Bridges
On Sun, 08 Nov 2026 22:30:59 +0000, Noah Weinberger wrote:
> The chained response header is initialized with Flags =
> SERVER_TO_REDIR|RELATED, so SMB2_FLAGS_ASYNC_COMMAND is never set there:
> an async-form CANCEL therefore always takes the MessageId branch and
> matches nothing, i.e. a compounded async CANCEL is a silent no-op
The bug is real, but narrower than this says. init_chained_smb2_rsp()
copies MessageId and the Id union from the request header into the
response header, so the async flag is the only field the two headers
disagree on, and smb2_cancel() reads the response header, which never
has it. A compounded async CANCEL therefore takes the MessageId branch.
That branch still finds the pending request when the CANCEL reuses the
request's MessageId.
I tested this on ksmbd-for-next with a blocking LOCK that went async and
an async CANCEL sent as the second command of a compound [ECHO, CANCEL].
With the LOCK's MessageId, the LOCK was cancelled on the current code
too. With MessageId 0, it was not: the LOCK stayed pending until the
holder unlocked. With your diff applied, both forms cancelled it, and so
did a standalone CANCEL with either MessageId.
> MS-SMB2 forbids compounding CANCEL, so only
> non-conforming clients are affected
I could not find this in MS-SMB2. Which section did you have in mind?
Thanks,
Ian
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] ksmbd: read request header for compounded CANCEL
2026-11-08 23:46 ` Ian Bridges
@ 2026-11-08 23:54 ` Noah Weinberger
0 siblings, 0 replies; 3+ messages in thread
From: Noah Weinberger @ 2026-11-08 23:54 UTC (permalink / raw)
To: Ian Bridges; +Cc: linkinjeon, linux-cifs, linux-kernel
Hi Ian,
Thanks for testing this and catching the overstatement. The failure is narrower than I described: using the response header loses SMB2_FLAGS_ASYNC_COMMAND and selects MessageId lookup, which can still succeed when the CANCEL reuses the pending request's MessageId. Your MessageId 0 case demonstrates the missed cancellation.
I don't have a supporting MS-SMB2 citation for the claim that CANCEL cannot be compounded. Please disregard that statement and the resulting “only non-conforming clients” qualification. Section 3.3.5.16 describes AsyncId lookup for async CANCEL, and Appendix A's note for 3.2.4.24 documents Windows clients setting MessageId to 0 in that case.
Thanks for confirming that reading the request header fixes both forms.
Noah
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-11-08 23:54 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-08 22:30 [PATCH] ksmbd: read request header for compounded CANCEL Noah Weinberger
2026-11-08 23:46 ` Ian Bridges
2026-11-08 23:54 ` Noah Weinberger
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®