mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH] dlm: fix RCOM_LOOKUP length underflow
@ 2026-09-11  6:53 Aohan Mei
  0 siblings, 0 replies; 2+ messages in thread
From: Aohan Mei @ 2026-09-11  6:53 UTC (permalink / raw)
  To: gfs2
  Cc: aahringo, teigland, linux-kernel, Aohan Mei, TencentOS Corvus AI, stable

From: Aohan Mei <henrymei@tencent.com>

receive_rcom_lookup() derives the resource name length from the
peer-supplied 16-bit h_length header field:

    int len = le16_to_cpu(rc_in->rc_header.h_length) -
              sizeof(struct dlm_rcom);

The 3.1 receive path only bounds h_length to the range
[sizeof(struct dlm_header), DLM_MAX_SOCKET_BUFSIZE] in
dlm_validate_incoming_buffer(), and the RCOM case in
dlm_midcomms_receive_buffer_3_1() performs no further length check
despite its "length already checked" comment.  A peer can therefore
send an RCOM_LOOKUP message whose h_length is smaller than
sizeof(struct dlm_rcom) (48).  The subtraction is evaluated in size_t
arithmetic and wraps, and the result is assigned to an int as a
negative value.

Both consumers of that length only guard the upper bound
(len > DLM_RESNAME_MAXLEN), which a negative len passes as a signed
comparison:

- with rc_id == 0xffffffff, len reaches dlm_dump_rsb_name() ->
  dlm_search_rsb_tree(), where memcpy() converts it to a huge size_t
  and overflows the on-stack 64-byte key buffer;
- otherwise len reaches dlm_master_lookup() -> _dlm_master_lookup(),
  where jhash() converts it to a huge u32 and reads out of bounds.

Drop the message when the computed name length falls outside
[0, DLM_RESNAME_MAXLEN], which is exactly the range valid lookups
use.

Fixes: e7fd41792fc0 ("[DLM] The core of the DLM for GFS2/CLVM")
Reported-by: TencentOS Corvus AI <corvus@tencent.com>
Cc: stable@vger.kernel.org
Assisted-by: CodeBuddy:Kimi-K3
Signed-off-by: Aohan Mei <henrymei@tencent.com>
---
 fs/dlm/rcom.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/fs/dlm/rcom.c b/fs/dlm/rcom.c
index be1a71a6303a..14d5ab658f60 100644
--- a/fs/dlm/rcom.c
+++ b/fs/dlm/rcom.c
@@ -385,6 +385,9 @@ static void receive_rcom_lookup(struct dlm_ls *ls,
 	int len = le16_to_cpu(rc_in->rc_header.h_length) -
 		sizeof(struct dlm_rcom);
 
+	if (len < 0 || len > DLM_RESNAME_MAXLEN)
+		return;
+
 	/* Old code would send this special id to trigger a debug dump. */
 	if (rc_in->rc_id == cpu_to_le64(0xFFFFFFFF)) {
 		log_error(ls, "receive_rcom_lookup dump from %d", nodeid);
-- 
2.43.7


^ permalink raw reply	[flat|nested] 2+ messages in thread
* [PATCH] dlm: fix RCOM_LOOKUP length underflow
@ 2026-08-21  2:28 Aohan Mei
  0 siblings, 0 replies; 2+ messages in thread
From: Aohan Mei @ 2026-08-21  2:28 UTC (permalink / raw)
  To: gfs2
  Cc: aahringo, teigland, linux-kernel, Aohan Mei, TencentOS Corvus AI, stable

From: Aohan Mei <henrymei@tencent.com>

receive_rcom_lookup() derives the resource name length from the
peer-supplied 16-bit h_length header field:

    int len = le16_to_cpu(rc_in->rc_header.h_length) -
              sizeof(struct dlm_rcom);

The 3.1 receive path only bounds h_length to the range
[sizeof(struct dlm_header), DLM_MAX_SOCKET_BUFSIZE] in
dlm_validate_incoming_buffer(), and the RCOM case in
dlm_midcomms_receive_buffer_3_1() performs no further length check
despite its "length already checked" comment.  A peer can therefore
send an RCOM_LOOKUP message whose h_length is smaller than
sizeof(struct dlm_rcom) (48).  The subtraction is evaluated in size_t
arithmetic and wraps, and the result is assigned to an int as a
negative value.

Both consumers of that length only guard the upper bound
(len > DLM_RESNAME_MAXLEN), which a negative len passes as a signed
comparison:

- with rc_id == 0xffffffff, len reaches dlm_dump_rsb_name() ->
  dlm_search_rsb_tree(), where memcpy() converts it to a huge size_t
  and overflows the on-stack 64-byte key buffer;
- otherwise len reaches dlm_master_lookup() -> _dlm_master_lookup(),
  where jhash() converts it to a huge u32 and reads out of bounds.

Drop the message when the computed name length falls outside
[0, DLM_RESNAME_MAXLEN], which is exactly the range valid lookups
use.

Fixes: e7fd41792fc0 ("[DLM] The core of the DLM for GFS2/CLVM")
Reported-by: TencentOS Corvus AI <corvus@tencent.com>
Cc: stable@vger.kernel.org
Assisted-by: CodeBuddy:Kimi-K3
Signed-off-by: Aohan Mei <henrymei@tencent.com>
---
 fs/dlm/rcom.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/fs/dlm/rcom.c b/fs/dlm/rcom.c
index be1a71a6303a..14d5ab658f60 100644
--- a/fs/dlm/rcom.c
+++ b/fs/dlm/rcom.c
@@ -385,6 +385,9 @@ static void receive_rcom_lookup(struct dlm_ls *ls,
 	int len = le16_to_cpu(rc_in->rc_header.h_length) -
 		sizeof(struct dlm_rcom);
 
+	if (len < 0 || len > DLM_RESNAME_MAXLEN)
+		return;
+
 	/* Old code would send this special id to trigger a debug dump. */
 	if (rc_in->rc_id == cpu_to_le64(0xFFFFFFFF)) {
 		log_error(ls, "receive_rcom_lookup dump from %d", nodeid);
-- 
2.43.7


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-09-11  6:54 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-09-11  6:53 [PATCH] dlm: fix RCOM_LOOKUP length underflow Aohan Mei
  -- strict thread matches above, loose matches on Subject: below --
2026-08-21  2:28 Aohan Mei

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®