[PATCH] RDMA/rxe: bound the ODP page index against the umem in rxe_check_pagefault()

[PATCH] RDMA/rxe: bound the ODP page index against the umem in rxe_check_pagefault()

[Youngsung Ahn]

An ODP MR's page array map.pfn_list[] is sized from the umem's
start, but rxe_check_pagefault() indexes it with an iova taken
relative to that start, while the access was range-checked against
the MR's iova (ibmr.iova). reg_mr only requires start and hca_va to
share a page offset, so a user can register with hca_va =
start + N*PAGE, and the index skew (hca_va - start) >> PAGE_SHIFT
then lands out of bounds of pfn_list[], which is read with no bound
check. An unprivileged local user reaches this with a single RDMA
operation on a self-connected rxe QP: observed as a KASAN
slab-out-of-bounds read and, when the out-of-bounds qword looks like
a writable HMM pfn, a write into an attacker-influenced page.

Reject an index that leaves the umem and force the fault path, which
fails the out-of-range access via the umem range check in
ib_umem_odp_map_dma_and_lock(). The downstream pfn_list[] users
(__rxe_odp_mr_copy() and the atomic helpers) run only after a
successful map, so this check protects them as well.

Fixes: 2fae67ab63db ("RDMA/rxe: Add support for Send/Recv/Write/Read with ODP")
Signed-off-by: Youngsung Ahn <ays511.kr@gmail.com>
---
Notes (not part of the commit):
Reproduced on a KASAN x86-64 build of 7.3-rc4 as uid 1000; the data-only write
primitive was used to reach uid 0 on a fully hardened build.
drivers/infiniband/sw/rxe/ is unchanged between 7.3-rc4 (93f51579e7df) and
mainline 551c722f4080 (2026-09-29) and rdma for-next, so both are affected.
Compile-tested (KASAN+RDMA_RXE), not runtime-tested. Found through AI-assisted
review; per Documentation/process/security-bugs.rst this is public and a
reproducer can be shared on request. A maintainer may prefer to also index
relative to ibmr.iova (as the non-ODP rxe_mr_iova_to_index() does) so
legitimately skewed MRs keep working; this patch takes the minimal, safe route.

 drivers/infiniband/sw/rxe/rxe_odp.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/drivers/infiniband/sw/rxe/rxe_odp.c b/drivers/infiniband/sw/rxe/rxe_odp.c
index ab21b620e94c..07a30c60fc62 100644
--- a/drivers/infiniband/sw/rxe/rxe_odp.c
+++ b/drivers/infiniband/sw/rxe/rxe_odp.c
@@ -136,6 +136,11 @@ static inline bool rxe_check_pagefault(struct ib_umem_odp *umem_odp, u64 iova,
 	while (addr < iova + length) {
 		idx = (addr - ib_umem_start(umem_odp)) >> umem_odp->page_shift;

+		if (idx >= ib_umem_odp_num_pages(umem_odp)) {
+			need_fault = true;
+			break;
+		}
+
 		if ((umem_odp->map.pfn_list[idx] & access) != access) {
 			need_fault = true;
 			break;
--
2.43.0

Re: [PATCH] RDMA/rxe: bound the ODP page index against the umem in rxe_check_pagefault()

[sashiko-bot]

> An ODP MR's page array map.pfn_list[] is sized from the umem's
> start, but rxe_check_pagefault() indexes it with an iova taken
> relative to that start, while the access was range-checked against
> the MR's iova (ibmr.iova). reg_mr only requires start and hca_va to
> share a page offset, so a user can register with hca_va =
> start + N*PAGE, and the index skew (hca_va - start) >> PAGE_SHIFT
> then lands out of bounds of pfn_list[], which is read with no bound
> check. An unprivileged local user reaches this with a single RDMA
> operation on a self-connected rxe QP: observed as a KASAN
> slab-out-of-bounds read and, when the out-of-bounds qword looks like
> a writable HMM pfn, a write into an attacker-influenced page.
> 
> Reject an index that leaves the umem and force the fault path, which
> fails the out-of-range access via the umem range check in
> ib_umem_odp_map_dma_and_lock(). The downstream pfn_list[] users
> (__rxe_odp_mr_copy() and the atomic helpers) run only after a
> successful map, so this check protects them as well.
> 
> Fixes: 2fae67ab63db ("RDMA/rxe: Add support for Send/Recv/Write/Read with ODP")
> Signed-off-by: Youngsung Ahn <ays511.kr@gmail.com>

Sashiko has reviewed this patch and found no issues. It looks great!

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260930175145.2491352-1-ays511.kr@gmail.com?part=1

Re: [PATCH] RDMA/rxe: bound the ODP page index against the umem in rxe_check_pagefault()

[Willy Tarreau]

Hello,

On Thu, Oct 01, 2026 at 02:51:45AM +0900, Youngsung Ahn wrote:
> An ODP MR's page array map.pfn_list[] is sized from the umem's
> start, but rxe_check_pagefault() indexes it with an iova taken
> relative to that start, while the access was range-checked against
> the MR's iova (ibmr.iova). reg_mr only requires start and hca_va to
> share a page offset, so a user can register with hca_va =
> start + N*PAGE, and the index skew (hca_va - start) >> PAGE_SHIFT
> then lands out of bounds of pfn_list[], which is read with no bound
> check. An unprivileged local user reaches this with a single RDMA
> operation on a self-connected rxe QP: observed as a KASAN
> slab-out-of-bounds read and, when the out-of-bounds qword looks like
> a writable HMM pfn, a write into an attacker-influenced page.
> 
> Reject an index that leaves the umem and force the fault path, which
> fails the out-of-range access via the umem range check in
> ib_umem_odp_map_dma_and_lock(). The downstream pfn_list[] users
> (__rxe_odp_mr_copy() and the atomic helpers) run only after a
> successful map, so this check protects them as well.
> 
> Fixes: 2fae67ab63db ("RDMA/rxe: Add support for Send/Recv/Write/Read with ODP")
> Signed-off-by: Youngsung Ahn <ays511.kr@gmail.com>

Are you sure you're not missing an Assisted-by: tag here ?

I didn't notice that you sent this to a public list, that's fine but in
this case there's no need to Cc security@k.o since the essential role of
the team is to help route the message to its destination :-)

thanks,
Willy

[PATCH v2] RDMA/rxe: bound the ODP page index against the umem in rxe_check_pagefault()

[Youngsung Ahn]

An ODP MR's page array map.pfn_list[] is sized from the umem's
start, but rxe_check_pagefault() indexes it with an iova taken
relative to that start, while the access was range-checked against
the MR's iova (ibmr.iova). reg_mr only requires start and hca_va to
share a page offset, so a user can register with hca_va =
start + N*PAGE, and the index skew (hca_va - start) >> PAGE_SHIFT
then lands out of bounds of pfn_list[], which is read with no bound
check. An unprivileged local user reaches this with a single RDMA
operation on a self-connected rxe QP: observed as a KASAN
slab-out-of-bounds read and, when the out-of-bounds qword looks like
a writable HMM pfn, a write into an attacker-influenced page.

Reject an index that leaves the umem and force the fault path, which
fails the out-of-range access via the umem range check in
ib_umem_odp_map_dma_and_lock(). The downstream pfn_list[] users
(__rxe_odp_mr_copy() and the atomic helpers) run only after a
successful map, so this check protects them as well.

Fixes: 2fae67ab63db ("RDMA/rxe: Add support for Send/Recv/Write/Read with ODP")
Cc: stable@vger.kernel.org
Signed-off-by: Youngsung Ahn <ays511.kr@gmail.com>
Assisted-by: LLM
---
Notes (not part of the commit):
Reproduced on a KASAN x86-64 build of 7.3-rc4 as uid 1000; the data-only write
primitive was used to reach uid 0 on a fully hardened build.
drivers/infiniband/sw/rxe/ is unchanged between 7.3-rc4 (93f51579e7df) and
mainline 551c722f4080 (2026-09-29) and rdma for-next, so both are affected.
Compile-tested (KASAN+RDMA_RXE), not runtime-tested. Found through AI-assisted
review; per Documentation/process/security-bugs.rst this is public and a
reproducer can be shared on request. A maintainer may prefer to also index
relative to ibmr.iova (as the non-ODP rxe_mr_iova_to_index() does) so
legitimately skewed MRs keep working; this patch takes the minimal, safe route.

v2: no code change. Added Assisted-by: LLM and dropped the security@kernel.org
Cc, per review of v1.

 drivers/infiniband/sw/rxe/rxe_odp.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/drivers/infiniband/sw/rxe/rxe_odp.c b/drivers/infiniband/sw/rxe/rxe_odp.c
index ab21b620e94c..07a30c60fc62 100644
--- a/drivers/infiniband/sw/rxe/rxe_odp.c
+++ b/drivers/infiniband/sw/rxe/rxe_odp.c
@@ -136,6 +136,11 @@ static inline bool rxe_check_pagefault(struct ib_umem_odp *umem_odp, u64 iova,
 	while (addr < iova + length) {
 		idx = (addr - ib_umem_start(umem_odp)) >> umem_odp->page_shift;

+		if (idx >= ib_umem_odp_num_pages(umem_odp)) {
+			need_fault = true;
+			break;
+		}
+
 		if ((umem_odp->map.pfn_list[idx] & access) != access) {
 			need_fault = true;
 			break;
--
2.43.0

Re: [PATCH v2] RDMA/rxe: bound the ODP page index against the umem in rxe_check_pagefault()

[sashiko-bot]

> An ODP MR's page array map.pfn_list[] is sized from the umem's
> start, but rxe_check_pagefault() indexes it with an iova taken
> relative to that start, while the access was range-checked against
> the MR's iova (ibmr.iova). reg_mr only requires start and hca_va to
> share a page offset, so a user can register with hca_va =
> start + N*PAGE, and the index skew (hca_va - start) >> PAGE_SHIFT
> then lands out of bounds of pfn_list[], which is read with no bound
> check. An unprivileged local user reaches this with a single RDMA
> operation on a self-connected rxe QP: observed as a KASAN
> slab-out-of-bounds read and, when the out-of-bounds qword looks like
> a writable HMM pfn, a write into an attacker-influenced page.
> 
> Reject an index that leaves the umem and force the fault path, which
> fails the out-of-range access via the umem range check in
> ib_umem_odp_map_dma_and_lock(). The downstream pfn_list[] users
> (__rxe_odp_mr_copy() and the atomic helpers) run only after a
> successful map, so this check protects them as well.
> 
> Fixes: 2fae67ab63db ("RDMA/rxe: Add support for Send/Recv/Write/Read with ODP")
> Cc: stable@vger.kernel.org
> Signed-off-by: Youngsung Ahn <ays511.kr@gmail.com>

Sashiko has reviewed this patch and found no issues. It looks great!

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261001152325.3258902-1-ays511.kr@gmail.com?part=1