nvme/070: add a test for Identify CNS 07h NULL pointer dereference
PATCH v2 1/1 nvme/070: add a test for Identify CNS 07h NULL pointer dereference
nvmet_execute_identify_nslist() handles both the Active Namespace ID list (CNS 02h) and the per-command-set variant (CNS 07h). For CNS 07h it filtered the list on req->ns->csi, but this handler never resolves req->ns, so it is always NULL. As soon as an enabled namespace with an NSID above the requested value exists, the target dereferenced a NULL pointer and oopsed.
This test connects a target with a single namespace and issues an Identify with CNS 07h starting from NSID 0, which is exactly the condition that triggered the crash. Without the kernel fix [0] the target oopses; with it the command completes normally.
[0] https://lore.kernel.org/linux-nvme/[email protected]/
---
v1 -> v2:
- Use "nvme list-ns --csi=0" instead of a raw admin-passthru to issue Identify CNS 07h, as suggested by Nilay and Shinichiro; it drives the same target code path via the existing nvme-cli subcommand.
tests/nvme/070 | 52 ++++++++++++++++++++++++++++++++++++++++++++++ tests/nvme/070.out | 2 ++ 2 files changed, 54 insertions(+) create mode 100755 tests/nvme/070 create mode 100644 tests/nvme/070.out
tests/nvme/070 new file +52 -0
@@ -0,0 +1,52 @@ +#!/bin/bash +# SPDX-License-Identifier: GPL-3.0+ +# Copyright (C) 2026 Guixin Liu +# +# Regression test for the NULL pointer dereference in +# nvmet_execute_identify_nslist() when handling Identify CNS 07h (Active +# Namespace ID List for the specified I/O Command Set). The CNS 07h handler +# filtered the list on req->ns->csi, but this handler never resolves req->ns +# so it is always NULL. As soon as an enabled namespace with an NSID above the +# requested value exists, the target dereferenced a NULL pointer and oopsed. + +. tests/nvme/rc + +DESCRIPTION="issue Identify CNS 07h (per-command-set active NS list)" +QUICK=1 + +requires() { + _nvme_requires + _have_loop + _require_nvme_trtype_is_fabrics +} + +set_conditions() { + _set_nvme_trtype "$@" +} + +test() { + echo "Running ${TEST_NAME}" + + _setup_nvmet + + _nvmet_target_setup + + _nvme_connect_subsys + + local nvmedev + nvmedev=$(_find_nvme_dev "${def_subsysnqn}") + + # "nvme list-ns --csi=0" issues Identify CNS 07h (Active Namespace ID + # list for the specified I/O Command Set, here NVM), starting from NSID + # 0 so the enabled namespace (NSID 1) is listed. That is exactly the + # condition that used to dereference the NULL req->ns in the target. + if ! nvme list-ns "/dev/${nvmedev}" --csi=0 >> "${FULL}" 2>&1; then + echo "Error: Identify CNS 07h (list-ns --csi=0) failed" + fi + + _nvme_disconnect_subsys + + _nvmet_target_cleanup + + echo "Test complete" +}
tests/nvme/070.out new file +2 -0
@@ -0,0 +1,2 @@ +Running nvme/070 +Test complete
Signature · 1 line
2.43.7
Quoted 92 lines · Guixin Liu
nvmet_execute_identify_nslist()handles both the Active Namespace ID list (CNS 02h) and the per-command-set variant (CNS 07h). For CNS 07h it filtered the list onreq->ns->csi, but this handler never resolvesreq->ns, so it is always NULL. As soon as an enabled namespace with an NSID above the requested value exists, the target dereferenced a NULL pointer and oopsed.This test connects a target with a single namespace and issues an Identify with CNS 07h starting from NSID 0, which is exactly the condition that triggered the crash. Without the kernel fix [0] the target oopses; with it the command completes normally.
[0] https://lore.kernel.org/linux-nvme/[email protected]/
Suggested-by Christoph Hellwig <[email protected]>Signed-off-by Guixin Liu <[email protected]>---
v1 -> v2:
- Use "nvme list-ns --csi=0" instead of a raw admin-passthru to issue Identify CNS 07h, as suggested by Nilay and Shinichiro; it drives the same target code path via the existing nvme-cli subcommand.tests/nvme/070 | 52 ++++++++++++++++++++++++++++++++++++++++++++++ tests/nvme/070.out | 2 ++ 2 files changed, 54 insertions(+)create mode 100755 tests/nvme/070
create mode 100644 tests/nvme/070.outtests/nvme/070 new file +52
-0@@ -0,0 +1,52 @@ +#!/bin/bash +# SPDX-License-Identifier: GPL-3.0+ +# Copyright (C) 2026 Guixin Liu +# +# Regression test for the NULL pointer dereference in +# nvmet_execute_identify_nslist() when handling Identify CNS 07h (Active +# Namespace ID List for the specified I/O Command Set). The CNS 07h handler +# filtered the list on req->ns->csi, but this handler never resolves req->ns +# so it is always NULL. As soon as an enabled namespace with an NSID above the +# requested value exists, the target dereferenced a NULL pointer and oopsed. + +. tests/nvme/rc + +DESCRIPTION="issue Identify CNS 07h (per-command-set active NS list)" +QUICK=1 + +requires() { + _nvme_requires + _have_loop + _require_nvme_trtype_is_fabrics +} + +set_conditions() { + _set_nvme_trtype "$@" +} + +test() { + echo "Running ${TEST_NAME}" + + _setup_nvmet + + _nvmet_target_setup + + _nvme_connect_subsys + + local nvmedev + nvmedev=$(_find_nvme_dev "${def_subsysnqn}") + + # "nvme list-ns --csi=0" issues Identify CNS 07h (Active Namespace ID + # list for the specified I/O Command Set, here NVM), starting from NSID + # 0 so the enabled namespace (NSID 1) is listed. That is exactly the + # condition that used to dereference the NULL req->ns in the target. + if ! nvme list-ns "/dev/${nvmedev}" --csi=0 >> "${FULL}" 2>&1; then + echo "Error: Identify CNS 07h (list-ns --csi=0) failed" + fi + + _nvme_disconnect_subsys + + _nvmet_target_cleanup + + echo "Test complete" +}tests/nvme/070.out new file +0
-0
tests/nvme/070.out line 2
Quoted 2 lines
tests/nvme/070.out new file +2
-0@@ -0,0 +1,2 @@ +Running nvme/070 +Test complete
Looks good to me.
Quoted 14 lines · Guixin Liu
nvmet_execute_identify_nslist()handles both the Active Namespace ID list (CNS 02h) and the per-command-set variant (CNS 07h). For CNS 07h it filtered the list onreq->ns->csi, but this handler never resolvesreq->ns, so it is always NULL. As soon as an enabled namespace with an NSID above the requested value exists, the target dereferenced a NULL pointer and oopsed.This test connects a target with a single namespace and issues an Identify with CNS 07h starting from NSID 0, which is exactly the condition that triggered the crash. Without the kernel fix [0] the target oopses; with it the command completes normally.
Quoted 8 lines
Suggested-by Christoph Hellwig <[email protected]>Signed-off-by Guixin Liu <[email protected]>---
v1 -> v2:
- Use "nvme list-ns --csi=0" instead of a raw admin-passthru to issue Identify CNS 07h, as suggested by Nilay and Shinichiro; it drives the same target code path via the existing nvme-cli subcommand.
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich
Quoted 14 lines · Guixin Liu
nvmet_execute_identify_nslist()handles both the Active Namespace ID list (CNS 02h) and the per-command-set variant (CNS 07h). For CNS 07h it filtered the list onreq->ns->csi, but this handler never resolvesreq->ns, so it is always NULL. As soon as an enabled namespace with an NSID above the requested value exists, the target dereferenced a NULL pointer and oopsed.This test connects a target with a single namespace and issues an Identify with CNS 07h starting from NSID 0, which is exactly the condition that triggered the crash. Without the kernel fix [0] the target oopses; with it the command completes normally.
Quoted 7 lines
Suggested-by Christoph Hellwig <[email protected]>Signed-off-by Guixin Liu <[email protected]>---
v1 -> v2:
- Use "nvme list-ns --csi=0" instead of a raw admin-passthru to issue Identify CNS 07h, as suggested by Nilay and Shinichiro; it drives the same target code path via the existing nvme-cli subcommand.
Thanks for this v2 patch. It looks good to me. I will wait for the kernel side fix get settled on Linus master branch before I apply this patch.
Quoted 22 lines · Shin'ichiro Kawasaki
Quoted 14 lines · Guixin Liu
nvmet_execute_identify_nslist()handles both the Active Namespace ID list (CNS 02h) and the per-command-set variant (CNS 07h). For CNS 07h it filtered the list onreq->ns->csi, but this handler never resolvesreq->ns, so it is always NULL. As soon as an enabled namespace with an NSID above the requested value exists, the target dereferenced a NULL pointer and oopsed.This test connects a target with a single namespace and issues an Identify with CNS 07h starting from NSID 0, which is exactly the condition that triggered the crash. Without the kernel fix [0] the target oopses; with it the command completes normally.
Quoted 7 lines
Suggested-by Christoph Hellwig <[email protected]>Signed-off-by Guixin Liu <[email protected]>---
v1 -> v2:
- Use "nvme list-ns --csi=0" instead of a raw admin-passthru to issue Identify CNS 07h, as suggested by Nilay and Shinichiro; it drives the same target code path via the existing nvme-cli subcommand.
Quoted 2 lines
Thanks for this v2 patch. It looks good to me. I will wait for the kernel side fix get settled on Linus master branch before I apply this patch.
Hi Shin'ichiro, the kernel side fix patch is already applied to master, could you please apply this patch?
Best Regards,
Guixin Liu
On Sep 16, 2026 / 19:40, Guixin Liu wrote:
[...]
Quoted 2 lines
Hi Shin'ichiro, the kernel side fix patch is already applied to master, could you please apply this patch?
Thanks for the heads-up. I applied the patch.