[PATCH 5.15.y] net/smc: check v2_ext_offset/eid_cnt/ism_gid_cnt when receiving proposal msg

[PATCH 5.15.y] net/smc: check v2_ext_offset/eid_cnt/ism_gid_cnt when receiving proposal msg

[Junjie Cao]

From: Guangguan Wang <guangguan.wang@linux.alibaba.com>

commit 7863c9f3d24ba49dbead7e03dfbe40deb5888fdf upstream.

When receiving proposal msg in server, the fields v2_ext_offset/
eid_cnt/ism_gid_cnt in proposal msg are from the remote client
and can not be fully trusted. Especially the field v2_ext_offset,
once exceed the max value, there has the chance to access wrong
address, and crash may happen.

This patch checks the fields v2_ext_offset/eid_cnt/ism_gid_cnt
before using them.

Fixes: 8c3dca341aea ("net/smc: build and send V2 CLC proposal")
Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Reviewed-by: Wen Gu <guwen@linux.alibaba.com>
Reviewed-by: D. Wythe <alibuda@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
[ jcao: drop the af_smc.c hunk, smc_find_rdma_v2_device_serv() came
  with SMC-Rv2 in e49300a6bf62 (v5.16); take SMC_CLC_MAX_UEID from
  fa0866625543 (v5.16); SMC_MAX_ISM_DEVS is the pre-b40584d14570 name
  of SMCD_CLC_MAX_V2_GID_ENTRIES ]
Signed-off-by: Junjie Cao <junjie.cao@intel.com>
---
5.15.y backport of the CVE-2024-49568 fix, requested in the 6.1.y
thread:
https://lore.kernel.org/r/20260918045353.686837-1-junjie.cao@intel.com

Differences from the 6.1.y version: no af_smc.c hunk; the callers of
smc_get_clc_v2_ext() in af_smc.c already handle NULL. The
SMC_CLC_MAX_UEID define comes along. 5.15 has no UEID support and
uses the received eid_cnt only in the length check of
smc_clc_msg_prop_valid(); 8 is the limit newer kernels apply to the
same field.

net/smc builds on v5.15.221, x86_64, CONFIG_SMC=m, W=1 clean. Not
runtime-tested.

 net/smc/smc_clc.c | 8 +++++++-
 net/smc/smc_clc.h | 9 ++++++++-
 2 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/net/smc/smc_clc.c b/net/smc/smc_clc.c
index fb214faf5f984..0c2c0b2edfdff 100644
--- a/net/smc/smc_clc.c
+++ b/net/smc/smc_clc.c
@@ -47,7 +47,6 @@ static bool smc_clc_msg_prop_valid(struct smc_clc_msg_proposal *pclc)
 	struct smc_clc_msg_hdr *hdr = &pclc->hdr;
 	struct smc_clc_v2_extension *v2_ext;
 
-	v2_ext = smc_get_clc_v2_ext(pclc);
 	pclc_prfx = smc_clc_proposal_get_prefix(pclc);
 	if (!pclc_prfx ||
 	    pclc_prfx->ipv6_prefixes_cnt > SMC_CLC_MAX_V6_PREFIX)
@@ -64,6 +63,13 @@ static bool smc_clc_msg_prop_valid(struct smc_clc_msg_proposal *pclc)
 			sizeof(struct smc_clc_msg_trail))
 			return false;
 	} else {
+		v2_ext = smc_get_clc_v2_ext(pclc);
+		if ((hdr->typev2 != SMC_TYPE_N &&
+		     (!v2_ext || v2_ext->hdr.eid_cnt > SMC_CLC_MAX_UEID)) ||
+		    (smcd_indicated(hdr->typev2) &&
+		     v2_ext->hdr.ism_gid_cnt > SMC_MAX_ISM_DEVS))
+			return false;
+
 		if (ntohs(hdr->length) !=
 			sizeof(*pclc) +
 			sizeof(struct smc_clc_msg_smcd) +
diff --git a/net/smc/smc_clc.h b/net/smc/smc_clc.h
index 78a94b9122b64..bda7c0384b845 100644
--- a/net/smc/smc_clc.h
+++ b/net/smc/smc_clc.h
@@ -158,6 +158,7 @@ struct smc_clc_msg_proposal {	/* clc proposal message sent by Linux */
 } __aligned(4);
 
 #define SMC_CLC_MAX_V6_PREFIX		8
+#define SMC_CLC_MAX_UEID		8
 
 struct smc_clc_msg_proposal_area {
 	struct smc_clc_msg_proposal		pclc_base;
@@ -297,8 +298,14 @@ static inline struct smc_clc_v2_extension *
 smc_get_clc_v2_ext(struct smc_clc_msg_proposal *prop)
 {
 	struct smc_clc_msg_smcd *prop_smcd = smc_get_clc_msg_smcd(prop);
+	u16 max_offset;
 
-	if (!prop_smcd || !ntohs(prop_smcd->v2_ext_offset))
+	max_offset = offsetof(struct smc_clc_msg_proposal_area, pclc_v2_ext) -
+		     offsetof(struct smc_clc_msg_proposal_area, pclc_smcd) -
+		     offsetofend(struct smc_clc_msg_smcd, v2_ext_offset);
+
+	if (!prop_smcd || !ntohs(prop_smcd->v2_ext_offset) ||
+	    ntohs(prop_smcd->v2_ext_offset) > max_offset)
 		return NULL;
 
 	return (struct smc_clc_v2_extension *)
-- 
2.43.0

Re: [PATCH 5.15.y] net/smc: check v2_ext_offset/eid_cnt/ism_gid_cnt when receiving proposal msg

[Sasha Levin]

> 5.15.y backport of the CVE-2024-49568 fix, requested in the 6.1.y
> thread:

Queued for 5.15, thanks.

-- 
Thanks,
Sasha