Kernel Security Transparency

• 1: Compiled-in CVEs — what each score means
• 2: Not Affected — disabled features

1 - Compiled-in CVEs — what each score means

Overview: Each row is a kernel CVE for a path in the Root Lock kernel. The word you file is on Kernel Security Transparency: Not affected when the option is unset, Affected when the code is in the kernel you boot, or Fixed on 6.18.9-hs when that kernel already has the upstream fix. An Affected row stays on the patch date.

A Score on Root Lock of 0.0 is printed when the program that reaches the bug has no allowlist entry, so that program does not run, and when the hardware the bug needs is not present. Approving the program makes the row Affected. Not affected is the other word, for an option that is unset. Application CVEs, such as a bug in a web server or database you approved, are outside this catalog. No exploit was run to produce these rows. Groups whose option is unset are on Not Affected — Disabled Features.

Module loads after boot. Where a write-up says a module cannot be loaded, the control is heartsuite-kernel-latch.service. Early in every Root Lock kernel boot, Setup Mode included, before sshd starts, it loads the netfilter modules the firewall needs and sets kernel.modules_disabled=1. Until the next reboot, a later modprobe stays refused (init_module, finit_module, and delete_module). The latch script does not unload modules already in memory. OpenRC has no latch service and gets that setting from HS_lockdown.sh when Lockdown is applied. A 0.0 on an =m option does not take that row off the patch date. Not Affected means the option is unset.

CVE-2026-31431

Status: Not Affected
Component: algif_aead — the in-kernel AEAD interface exposed by the AF_ALG socket family (CONFIG_CRYPTO_USER_API_AEAD)
Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) — CNA (kernel.org); NVD assessment pending
Upstream fix: Linux 6.12.85 (LTS), 6.18.22 (LTS), 6.19.12 (LTS)

This CVE describes a privilege escalation through the AF_ALG socket interface. An attacker who can open an AF_ALG socket reaches algif_aead_copy_sgl(), exploits a copy-on-write failure in the scatter-gather list handling, and gains root.

CONFIG_CRYPTO_USER_API_AEAD is not compiled into the Root Lock kernel. The AF_ALG socket family is not available. An attempt to open an AF_ALG socket returns EAFNOSUPPORT — there is no algif_aead code present in the running kernel and therefore no reachable code path. The Root Lock kernel predates the upstream fix versions listed above, but the fix is not required: the fix removes a vulnerability in code that was never compiled in.

CVE-2026-43284

Status: Not exploitable
Component: XFRM framework and IPv6 ESP (CONFIG_XFRM, CONFIG_INET6_ESP)
Base Score: 8.8 HIGH — NVD full vector assessment pending
Score on Root Lock: 0.0 — esp_output is unreachable; no XFRM security association can be established on a default Root Lock deployment
Upstream fix: merged; backported to active stable series by 2026-05-09 (5.19 branch is EOL; no backport — not required for HS)

This CVE describes a write-what-where condition in the esp_output page-write path. The vulnerable code is at net/ipv6/esp6.c:524: tail = page_address(page) + pfrag->offset followed by esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto). If pfrag->offset is corrupted or attacker-influenced, the trailer write reaches an arbitrary kernel page address. The identical pattern exists in net/ipv4/esp4.c:489 (CONFIG_INET_ESP, not compiled), but the absence of IPv4 ESP is irrelevant — esp6.c carries the same code. The bug is one half of the “Dirty Frag” exploit chain; chaining it with CVE-2026-43500 produces a deterministic privilege escalation.

CONFIG_INET6_ESP=y is compiled in and esp6.c:524 is present in the running kernel. The esp_output function is called only when the kernel encrypts an outgoing packet that matches a configured XFRM security association. With no security association configured, esp_output is never reached — by any user, at any privilege level. Configuring a XFRM security association requires XFRM management tooling: ip xfrm (iproute2), setkey, strongSwan, libreswan, or an equivalent IKE daemon. None of these are in the Root Lock default allowlist. Under Lockdown, the allowlist is chattr +i immutable and FS_IOC_SETFLAGS returns EPERM for all callers — root cannot add management tools and therefore cannot establish a security association. The esp_output page-write path is unreachable for the lifetime of the boot.

The Dirty Frag chain has no second link on this system regardless: CONFIG_AF_RXRPC is not compiled (see CVE-2026-43500).

The trigger cannot be reached on any default Root Lock deployment.

If your deployment adds XFRM management tooling (ip xfrm, setkey, strongSwan, libreswan, or an equivalent IKE daemon) to the Root Lock allowlist, a security association can be established and esp_output becomes reachable. In that configuration this CVE applies at its base score of 8.8 HIGH. Treat it as Affected and apply the standard backstop logic.

CVE-2026-23395

Status: Not exploitable while no HCI device is up
Component: Bluetooth L2CAP enhanced credit connect (CONFIG_BT, CONFIG_BT_LE_L2CAP_ECRED=y)
Base Score: 8.8 HIGH (AV:A/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H) — Linux CNA
Score on Root Lock: 0.0 while no LE HCI connection is in BT_CONNECTED
Upstream fix: 6.18.20

l2cap_ecred_conn_req (net/bluetooth/l2cap_core.c) runs only after l2cap_recv_frame has an hci_conn in BT_CONNECTED and a listening PSM. The overflow write is in l2cap_ecred_rsp_defer. Opening an AF_BLUETOOTH socket does not enter either function. The CNA trigger is a nearby radio.

CONFIG_BT_HCIVHCI=m is a second way to get an HCI device. vhci_open in drivers/bluetooth/hci_vhci.c has no capable() check, and the misc device sets no mode in this tree. This note does not confirm who can open /dev/vhci. If that device is opened and an LE connection reaches BT_CONNECTED, the base score applies.

CVE-2026-31500

Status: Not exploitable on a server with no Intel USB Bluetooth controller
Component: btintel_hw_error (CONFIG_BT_INTEL)
Base Score: Linux CNA published no CVSS base score
Score on Root Lock: 0.0 — the callback is not installed without that controller
Upstream fix: 6.18.21

btintel_hw_error (drivers/bluetooth/btintel.c) is assigned only from btintel_configure_setup, and the only caller of that setup in this tree is btusb under BTUSB_INTEL_COMBINED. hci_hardware_error_evt is what queues it. A socket does not call it. vhci does not install this callback.

CVE-2026-23234

Status: Not exploitable while no f2fs filesystem is mounted
Component: f2fs_write_end_io (CONFIG_F2FS_FS)
Base Score: Linux CNA published no CVSS base score
Score on Root Lock: 0.0 — no f2fs superblock on the ext4 or xfs install root
Upstream fix: 6.18.13

f2fs_write_end_io (fs/f2fs/data.c) is the end I/O of a write bio on an f2fs superblock. The use-after-free is that completion against kfree(sbi) in kill_f2fs_super. The function does not run for an ext4 or xfs root. The Nitro install note records that root as ext4. The installer module list names ext4 and xfs, not f2fs.

A new mount needs CAP_SYS_ADMIN in the init user namespace. f2fs_fs_type.fs_flags is FS_REQUIRES_DEV | FS_ALLOW_IDMAP, with no FS_USERNS_MOUNT. A user namespace does not pass that check. This note does not treat the program allowlist as the gate.

CVE-2026-23235

Status: Not exploitable while no f2fs filesystem is mounted
Component: f2fs per-mount sysfs (CONFIG_F2FS_FS)
Base Score: Linux CNA published no CVSS base score
Score on Root Lock: 0.0 — the attributes are created in f2fs_fill_super
Upstream fix: 6.18.13

f2fs_sbi_show and __sbi_store (fs/f2fs/sysfs.c) are the show and store for carve_out and atgc_age_threshold. Those attributes are registered from f2fs_fill_super. Module load does not create them. The same mount rule as CVE-2026-23234 applies. Once a filesystem is mounted, the attributes are mode 0644, so a non-owner can read them.

CVE-2026-31532

Status: Stays on the patch date. Not scored 0.0
Component: CAN raw receive (CONFIG_CAN_RAW)
Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) — Linux CNA
Score on Root Lock: not scored 0.0
Upstream fix: 6.18.24

raw_rcv (net/can/raw.c) runs when a frame arrives on an ARPHRD_CAN netdevice. socket(AF_CAN, SOCK_RAW, CAN_RAW) succeeds with no interface and does not call raw_rcv. can.ko autoloads from that socket. can-raw.ko autoloads from can_create. Neither load delivers a frame.

Creating vcan is an rtnetlink newlink. Non-GET rtnetlink messages require CAP_NET_ADMIN. The 6.18.9 pin has CONFIG_USER_NS=y and CONFIG_CAN_VCAN=m. The CNA says a user namespace can supply that capability. This note does not call the trigger absent.

CVE-2026-43500

Status: Not Affected
Component: rxrpc — RxRPC transport protocol (CONFIG_AF_RXRPC)
Base Score: 7.8 HIGH — NVD full vector assessment pending
Upstream fix: merged; backported to active stable series by 2026-05-09 (5.19 branch is EOL; no backport — not required for HS)

This CVE describes a local privilege escalation through an out-of-bounds write in the rxrpc transport protocol implementation. It is the second half of the “Dirty Frag” exploit chain (paired with CVE-2026-43284); chaining both produces a deterministic privilege escalation to root.

CONFIG_AF_RXRPC is not compiled into the Root Lock kernel. The rxrpc address family is not available; an attempt to open an AF_RXRPC socket returns EAFNOSUPPORT. The vulnerable code in net/rxrpc/ is entirely absent from the running kernel. The Root Lock kernel predates the upstream fix, but the fix is not required: there is no reachable code path for this bug on any Root Lock deployment. The Dirty Frag chain has no second link on this system.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-46242

Status: Not Affected on 5.19.6; Not exploitable on 6.18.9-hs
Component: epoll — event polling subsystem (CONFIG_EPOLL)
Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) — CNA (kernel.org)
Score on Root Lock: 0.0 — 5.19.6 is outside the affected range; on 6.18.9-hs the linked-epoll close race is not constructible from the allowlist
Affected range: 6.4 through 6.18.32; also narrow LTS windows 5.15.209–5.15.x and 6.1.175–6.1.177. 5.19.6 is not in range. Production 6.18.9-hs remains in range until base ≥ 6.18.33
Upstream fix: a6dc643c6931 (mainline); stable 6.18.33+

This CVE describes a use-after-free in ep_remove(). When one epoll instance monitors another, a concurrent close() can clear file->f_ep and then keep using that struct file while __fput() frees the watched struct eventpoll. The construction turns the race into an attacker-controlled free against the wrong slab cache and privilege escalation to root.

CONFIG_EPOLL=y is compiled in on both fielded kernels. That is not enough for 5.19.6: the mainline introduction is 6.4 (58c9b016e128). 5.19.6 predates that change, so the 5.19.6 ep_remove path is not the vulnerable one.

On 6.18.9-hs the vulnerable interleaving is present until a 6.18.33+ base. Reaching it requires allocating two specifically linked epoll instances and driving the close ordering from a dedicated program. No such program appears in the Root Lock allowlist. An attacker who has already gained root cannot add one: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot. The kernel therefore refuses to run a dropped exploit binary.

The trigger cannot be reached on any default Root Lock deployment.

If your 6.18.9-hs deployment adds a program that performs the linked-epoll close pattern to the allowlist, treat this CVE as Affected at 7.8 HIGH and apply the standard backstop.

CVE-2024-47685

Status: Score on Root Lock 0.0 — trigger not present in default configuration Component: nf_reject_ipv6 — IPv6 netfilter TCP RST generation (CONFIG_NF_REJECT_IPV6, CONFIG_IP6_NF_TARGET_REJECT)
Base Score: 9.1 CRITICAL (AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:H)
Score on Root Lock: 0.0 — trigger not present; HeartSuite installs no ip6tables REJECT rules
Upstream fix: Linux 4.19.323, 5.4.285, 5.10.227, 5.15.168, 6.1.113, 6.6.54, 6.10.13, 6.11.2 (5.19 branch is EOL; no backport — not required for HS)

This CVE describes an information disclosure in the IPv6 netfilter TCP reset path. When the kernel sends a TCP RST packet in response to a connection rejected by an ip6tables rule, nf_reject_ip6_tcphdr_put() allocates a TCP header via skb_put() without zeroing the buffer. The function then writes every field in the header explicitly except the four reserved bits (th->res1) in byte 12. Those bits retain whatever value was in the allocated kernel memory region. The RST packet is sent with that uninitialized content on the wire.

CONFIG_NF_REJECT_IPV6=y and CONFIG_IP6_NF_TARGET_REJECT=y are compiled in. The code path exists in this kernel. The vulnerable function has five callers across the kernel source. In this configuration only ip6t_REJECT.c is built in. The other four callers (nft_reject_ipv6, nft_reject_inet, nft_reject_bridge, nft_reject_netdev) are gated by CONFIG_NF_TABLES=m. The boot latch preloads nft_reject_ipv6 and nft_reject_inet before it sets kernel.modules_disabled=1. It does not preload nft_reject_bridge or nft_reject_netdev. Reaching the vulnerable code therefore requires an active ip6tables rule using REJECT --reject-with tcp-reset on IPv6 traffic. The Root Lock install scripts and service unit contain no ip6tables rules of any kind. If you manually add such a rule, this path becomes exposed.

Lockdown does not patch the vulnerability mechanism — the kernel still places uninitialized bits into the packet header if the path is reached. However, the program allowlist and Lockdown together make the triggering condition unreachable in practice.

To trigger this CVE, you must first add an ip6tables rule with REJECT --reject-with tcp-reset. That requires running ip6tables with root privilege. In Lockdown, HeartSuite’s program allowlist is enforced at the kernel level for every user including root: a program without a valid allowlist entry cannot execute regardless of the caller’s privilege level. Network management utilities such as ip6tables have no allowlist entry on a production HeartSuite deployment, so root cannot run them and the rule cannot be added.

Lockdown closes the remaining path. Even if an attacker gained root and attempted to add ip6tables to the allowlist first, Lockdown blocks every mechanism for doing so: FS_IOC_SETFLAGS (the ioctl used by chattr) returns EPERM for all callers during lockdown, so immutable flags cannot be cleared from the allowlist database files; mount(), fsmount(), and move_mount() all return EPERM, blocking any bind-mount or remount workaround; and the HeartSuite reactivation path is disabled, preventing the service from being reconfigured to accept new entries.

The result is a two-layer guarantee: the program allowlist prevents the trigger from being established, and Lockdown ensures the allowlist cannot be modified to enable the tools that would establish it. A 9.1 CRITICAL CVE that requires setting up an ip6tables REJECT rule becomes unreachable by any user, including root, once Lockdown is in force.

CVE-2022-41674, CVE-2022-42719, CVE-2022-42720

Status: Not exploitable Component: mac80211 — 802.11 wireless stack (CONFIG_MAC80211)
Base Scores: CVE-2022-42719: 8.8 HIGH (AV:A); CVE-2022-41674: 8.1 HIGH (AV:A); CVE-2022-42720: 7.8 HIGH (AV:A)
Score on Root Lock: 0.0 — no WiFi hardware present; attack vector (frame injection via wireless NIC) has no path to execution
Affected range: Linux 5.19.x before 5.19.16
Upstream fix: Linux 5.4.218–219, 5.10.148–149, 5.15.74, 5.19.16, 6.0.2

These three CVEs cover memory corruption in the mac80211 multi-BSSID scanning path, exploitable by an attacker who can inject 802.11 management frames:

• CVE-2022-41674 (CVSS 8.1) — buffer overflow in ieee80211_bss_info_update() in net/mac80211/scan.c triggered by a crafted beacon or probe response with a malformed multi-BSSID element
• CVE-2022-42719 (CVSS 8.8) — use-after-free when parsing a multi-BSSID element, exploitable to crash the kernel or gain privilege
• CVE-2022-42720 (CVSS 7.8) — refcounting bugs in multi-BSS handling reachable through the same scanning path

CONFIG_MAC80211=y is compiled in and 5.19.6 is within the affected version range for all three. The entry point is ieee80211_scan_rx() in net/mac80211/rx.c, which has a single caller: the hardware NIC interrupt RX path. A physical WiFi NIC must be present, registered, and receiving frames for any of these paths to execute. CONFIG_MAC80211_HWSIM (software WiFi simulator) is not set. On server deployments without a WiFi interface the code paths are unreachable.

If exploited on a deployment with WiFi hardware, all three CVEs lead to kernel memory corruption that can escalate to root. At that point Lockdown constrains everything the attacker can do with that root access.

HeartSuite makes the allowlist database files immutable before Lockdown is engaged. Once Lockdown is active, FS_IOC_SETFLAGS returns EPERM for all callers (kernel/ioctl.c), so root cannot use chattr to clear those immutable flags and rewrite the allowlist. mount(), fsmount(), and move_mount() all return EPERM (kernel/namespace.c), blocking any bind-mount or remount attempt to shadow or replace the allowlist files. HeartSuite reactivation is disabled during Lockdown, so the service cannot be reconfigured to accept new entries through any path.

Lockdown’s allowlist adds a further constraint on program execution: every execution is checked at the kernel level, applying equally to root. An attacker who has gained root cannot execute a backdoor program they drop onto the filesystem — it has no allowlist entry, and the kernel refuses to run it regardless of file ownership or permission bits.

CVE-2023-0266

Status: Not exploitable Component: ALSA PCM — in-kernel sound subsystem (CONFIG_SND)
Base Score: 7.9 HIGH (AV:L/AC:H/PR:L/UI:N/S:C/C:H/I:H/A:H)
Score on Root Lock: 0.0 — no audio hardware present; no /dev/snd devices; ioctl path unreachable
Affected range: Linux 5.16 through 6.1.5
Upstream fix: Linux 4.14.303, 4.19.270, 5.4.229, 5.10.163, 5.15.88, 6.1.6 (5.19 branch is EOL; no backport — not required for HS)

This CVE describes a use-after-free in the ALSA PCM control interface. SNDRV_CTL_IOCTL_ELEM_READ and SNDRV_CTL_IOCTL_ELEM_WRITE (32-bit compat variants) are missing locks that allow a local user to trigger a use-after-free and gain elevated privilege.

CONFIG_SND=y is compiled in and 5.19.6 falls within the affected range. Reaching the vulnerable code requires an ALSA-accessible sound device. Server deployments without audio hardware have no /dev/snd devices and no reachable path to this ioctl.

If exploited on a deployment with audio hardware, the CVE achieves local privilege escalation to root. At that point Lockdown constrains everything the attacker can do with that root access.

The allowlist database files are made immutable before Lockdown is engaged. Once Lockdown is active, FS_IOC_SETFLAGS returns EPERM for all callers (kernel/ioctl.c), so root cannot use chattr to clear those immutable flags and rewrite the allowlist. mount(), fsmount(), and move_mount() all return EPERM (kernel/namespace.c), blocking any bind-mount or remount attempt to shadow or replace the allowlist files. HeartSuite reactivation is disabled during Lockdown, so the service cannot be reconfigured to accept new entries through any path.

Lockdown’s allowlist adds a further constraint on program execution: every execution is checked at the kernel level, applying equally to root. An attacker who has gained root cannot execute a backdoor program they drop onto the filesystem — it has no allowlist entry, and the kernel refuses to run it regardless of file ownership or permission bits.

CVE-2022-4139

Status: Not exploitable Component: i915 GPU driver (CONFIG_DRM_I915)
Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H)
Score on Root Lock: 0.0 — no i915 GPU present; GPU context entry point unreachable
Affected range: Linux 5.16 through 6.0.10
Upstream fix: Linux 5.4.226, 5.10.157, 5.15.81, 6.0.11 (5.19 branch is EOL; no backport — not required for HS)

This CVE describes an incorrect TLB flush in the Intel i915 GPU driver. When GPU memory mappings are changed, a missing or incorrect TLB invalidation can leave stale translation entries active, allowing writes to land in the wrong physical pages. This can corrupt kernel memory and is exploitable by a local user with access to a GPU context to gain elevated privilege.

CONFIG_DRM_I915=y is compiled in and 5.19.6 falls within the affected range. Reaching the vulnerable path requires an Intel i915 GPU to be present and accessible. Deployments without i915 hardware have no reachable path to this driver.

The vulnerable path never opens. The bug exists in the source — not on this system.

CVE-2023-2236, CVE-2022-3910

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: io_uring — asynchronous I/O subsystem (CONFIG_IO_URING)
Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H)
Score on Root Lock: 7.1–7.3 HIGH — Lockdown reduces MI: High→Low (no allowlist modification, no persistence, no backdoors); C and A remain High; score stays within the HIGH band
Affected ranges: CVE-2023-2236: 5.19 through 6.0.10; CVE-2022-3910: 5.18 through 5.19.10
Upstream fix: CVE-2023-2236: 6.0.11; CVE-2022-3910: 5.19.11 (5.19 branch is EOL for CVE-2023-2236; CVE-2022-3910 fix was in-branch but 5.19.6 predates it)

What this means for an attacker:

Both CVEs describe use-after-free conditions in io_uring’s fixed file management, exploitable by a local user to gain root:

• CVE-2023-2236 — double fput() in the io_install_fixed_file() path. When an async open operation installs a fixed file and encounters an error, io_install_fixed_file() calls fput(file) at its error label; the caller then calls fput(file) a second time. The file’s reference count reaches zero while the object is still referenced, producing a use-after-free.
• CVE-2022-3910 — improper reference count update in io_uring’s fixed file handling that leads to a use-after-free and local privilege escalation.

Why the score is not 0.0:

CONFIG_IO_URING=y is compiled in on 5.19.6 and on the 6.18.9-hs kernel that ships. The upstream fix is already in 6.18.9-hs. On 5.19.6 the io_uring_setup syscall has no capability gate — any local user can create an io_uring ring and reach both vulnerable paths. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

These constraints are why the Score on Root Lock reflects a reduced MI (High→Low): root cannot modify the allowlist, cannot install persistent backdoors, and cannot survive a reboot. Confidentiality and Availability impacts remain High, reflecting that an attacker with a live root session can still read data and disrupt services within the bounds of already-permitted processes.

A more sophisticated exploit could use the kernel use-after-free to directly corrupt kernel data structures before surfacing in userspace. In that scenario Lockdown’s API-level restrictions are not the binding constraint — the corruption happens below the layer where those checks operate. This is why the Score on Root Lock does not reach 0.0: the io_uring path is reachable by any local user, and pre-userspace kernel corruption is outside the scope of what Lockdown addresses.

CVE-2024-0775

Status: Not exploitable Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 6.7 MEDIUM (AV:L/AC:L/PR:H/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — mount(MS_REMOUNT) blocked by Lockdown; ext4 remount entry point unreachable Affected range: kernels through 6.7.2, 6.6.15, 6.1.79, 5.15.148, 5.10.211, 5.4.270, 4.19.308 (5.19 branch is EOL; no backport) Upstream fix: Linux 6.7.3, 6.6.16, 6.1.80, 5.15.149, 5.10.212, 5.4.271, 4.19.309

This CVE describes a use-after-free in the __ext4_remount() error path in fs/ext4/super.c. When a remount operation fails and rolls back to saved options, the function restores quota file name pointers via rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]) and then frees the displaced current pointer via kfree(to_free[i]). If the success path has already freed those names at the earlier kfree(old_opts.s_qf_names[i]) call, the error path operates on already-freed memory. The CVE requires CAP_SYS_ADMIN (implicit in PR:H) because mount(MS_REMOUNT) is a privileged operation.

CONFIG_EXT4_FS=y is compiled in and 5.19.6 falls within the affected range. ext4 is the primary filesystem on a Debian 11 server. __ext4_remount() is reached exclusively via mount(MS_REMOUNT) — a privileged operation that Lockdown blocks unconditionally. do_mount() returns EPERM whenever HS_locked_down() is true (kernel/namespace.c:4218), so root cannot call mount() at all; the CVE’s entry point is blocked at the syscall level before any ext4 code is reached. In Lockdown, the allowlist additionally prevents execution of any exploit program that would invoke the remount path.

CVE-2023-52530

Status: Not exploitable Component: mac80211 wireless stack (CONFIG_MAC80211) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no WiFi NIC present; WoWLAN path unreachable Affected range: kernels through 6.7.3, 6.6.18, 6.1.81, 5.15.150, 5.10.214, 5.4.273, 4.19.311 (5.19 branch is EOL; no backport) Upstream fix: Linux 6.7.4, 6.6.19, 6.1.82, 5.15.151, 5.10.215, 5.4.274, 4.19.312

This CVE describes a use-after-free in the mac80211 WoWLAN (Wake on Wireless LAN) GTK rekey path. When ieee80211_gtk_rekey_add() installs a new group temporal key, it calls ieee80211_key_link(). If the new key is identical to the one already installed — the KRACK protection path — ieee80211_key_link() frees the new key via ieee80211_key_free_unused(key) and returns 0 to signal that the reinstall was silently accepted. ieee80211_gtk_rekey_add() treats the 0 return as success, skips the error branch, and returns &key->conf — a pointer into the object that was just freed. The caller receives a dangling pointer to freed ieee80211_key memory.

CONFIG_MAC80211=y is compiled in. The entry point ieee80211_gtk_rekey_add() guards itself with WARN_ON(!local->wowlan): it requires WoWLAN to be active, which in turn requires a WiFi NIC with WoWLAN firmware support, a wireless interface, and an active station association. No WiFi network interface card is present on a server deployment. Without WiFi hardware, mac80211 creates no wireless interfaces and neither the rekey path nor any other mac80211 code path is reachable.

If exploited on a deployment with WiFi hardware and WoWLAN active, the CVE leads to kernel memory corruption that can escalate to root. At that point Lockdown constrains everything the attacker can do with that root access.

The allowlist database files are made immutable before Lockdown is engaged. FS_IOC_SETFLAGS returns EPERM for all callers (kernel/ioctl.c), so root cannot use chattr to clear those immutable flags and rewrite the allowlist. mount(), fsmount(), and move_mount() all return EPERM (kernel/namespace.c), blocking any bind-mount or remount attempt to shadow or replace the allowlist files. HeartSuite reactivation is disabled during Lockdown, so the service cannot be reconfigured to accept new entries through any path.

Lockdown’s allowlist adds a further constraint on program execution: every execution is checked at the kernel level, applying equally to root. An attacker who has gained root cannot execute a backdoor program they drop onto the filesystem — it has no allowlist entry, and the kernel refuses to run it regardless of file ownership or permission bits.

CVE-2023-52612

Status: Not exploitable Component: kernel crypto framework — scomp interface (CONFIG_CRYPTO) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_INET_IPCOMP not compiled; no compression algorithm registered; scomp_acomp_comp_decomp() unreachable Affected range: kernels prior to stable fixes in the 6.7.x, 6.6.x, 6.1.x, 5.15.x, 5.10.x, and 5.4.x series (5.19 branch is EOL; no backport) Upstream fix: merged in Linux 6.8-rc; backported across active stable series

This CVE describes a buffer overflow in the kernel software compression (scomp) interface in crypto/scompress.c. The scomp_acomp_comp_decomp() function uses a per-CPU scratch buffer of SCOMP_SCRATCH_SIZE bytes as working space. If the caller provides a req->dst scatter list smaller than SCOMP_SCRATCH_SIZE, the function still caps req->dlen to SCOMP_SCRATCH_SIZE and then copies the full output — up to that size — into req->dst via scatterwalk_map_and_copy(). No check verifies that req->dst can hold req->dlen bytes before the copy. A caller who controls req->dst and triggers a compression or decompression that fills the scratch buffer can write beyond the end of the destination scatter list.

The scomp interface is the software-side of the kernel’s acomp (asynchronous compression) API. It is not a general-purpose path used by dm-crypt, TLS, or cipher operations — it exists exclusively to service IPsec compression transforms (IPCOMP, RFC 3173). scomp_acomp_comp_decomp() is only reached when a compression algorithm is registered with the scomp backend and a caller submits a request to it. On Root Lock there are no such callers and no such registrations:

• # CONFIG_INET_IPCOMP is not set — the IPv4/IPv6 IPsec compression module is not compiled; no IPCOMP transform can be configured
• # CONFIG_CRYPTO_DEFLATE is not set — DEFLATE not compiled; not registered with scomp
• # CONFIG_CRYPTO_LZ4 is not set — LZ4 not compiled; not registered with scomp
• # CONFIG_CRYPTO_ZSTD is not set — ZSTD not compiled; not registered with scomp

With no compression algorithm registered, the scomp backend has no handler to dispatch to. CONFIG_CRYPTO=y means the crypto framework is present, but framework presence is not trigger reachability. The trigger cannot be reached on any Root Lock deployment.

CVE-2024-26654

Status: Not exploitable Component: ALSA AICA Dreamcast sound driver (CONFIG_SND_AICA) Base Score: 7.0 HIGH (AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — driver not compiled in; no code path exists Upstream fix: merged in Linux 6.8; backported across active stable series (5.19 branch is EOL; no backport — not required for HS)

This CVE describes a use-after-free caused by a circular scheduling race between dreamcastcard->timer and spu_dma_work in the AICA Yamaha sound chip driver (sound/sh/aica.c). The timer callback aica_period_elapsed() schedules spu_dma_work via schedule_work(); the work handler then re-arms the timer via mod_timer(). spu_begin_dma() independently schedules the work and arms the timer in the same call. These two execution paths can race against each other and against card teardown, producing a use-after-free on the snd_card_aica object while the timer or work item is still pending.

CONFIG_SND_AICA is not set in the Root Lock kernel. sound/sh/aica.c is gated by obj-$(CONFIG_SND_AICA) in sound/sh/Makefile and is not compiled. There is no AICA driver code present in the running kernel — not merely absent hardware, but absent code. An attempt to reach this path has no code to execute. The Root Lock kernel predates the upstream fix, but the fix is not required: it patches code that was never compiled in.

CVE-2024-26704

Status: Not exploitable Component: ext4 filesystem — online defragmentation (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — EXT4_IOC_MOVE_EXT ioctl only reached by defragmentation tools; none in Root Lock allowlist Affected range: kernels prior to stable fixes in the 6.8.x, 6.7.x, 6.6.x, 6.1.x, 5.15.x, 5.10.x, and 5.4.x series (5.19 branch is EOL; no backport) Upstream fix: merged in Linux 6.8; backported across active stable series

This CVE describes a use-after-free in ext4_move_extents() in fs/ext4/move_extent.c, reachable via the EXT4_IOC_MOVE_EXT ioctl. The function moves file extents between an original inode and a donor inode. If the first move operation fails, o_start has not advanced past orig_blk, so *moved_len is set to zero. Preallocation blocks set up for orig_inode and donor_inode are discarded only when *moved_len is non-zero — the guard at move_extent.c:692. With *moved_len == 0, those preallocations are never discarded, leaving stale preallocation state that produces a use-after-free when the preallocations are later released. The EXT4_IOC_MOVE_EXT ioctl requires only write access to the file — no CAP_SYS_ADMIN, consistent with the PR:L CVSS score.

CONFIG_EXT4_FS=y is compiled in and 5.19.6 falls within the affected range. The EXT4_IOC_MOVE_EXT ioctl is the sole entry point to the vulnerable ext4_move_extents() path; it is invoked by extent-defragmentation tools (e4defrag) and not by normal filesystem read or write operations. No defragmentation tool appears in the Root Lock allowlist, and the kernel blocks any process without an allowlist entry from executing. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

If your deployment adds e4defrag or any other extent-defragmentation tool to the Root Lock allowlist, the EXT4_IOC_MOVE_EXT ioctl becomes reachable and this CVE applies at its base score of 7.8 HIGH. Treat it as Affected and apply the standard backstop logic.

CVE-2024-26842

Status: Not exploitable Component: UFS host controller driver (CONFIG_SCSI_UFSHCD) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — driver not compiled in; no code path exists Upstream fix: merged in Linux 6.8; backported across active stable series (5.19 branch is EOL; no backport — not required for HS)

This CVE describes an out-of-bounds memory access in the UFS host controller driver’s MCQ (Multi-Circular Queue) mode. When task_tag >= 32 and sizeof(unsigned int) == 4, the expression 1U << task_tag is undefined behaviour in C — shifting a 32-bit value by 32 or more positions. In practice this produces incorrect bitmask values in the per-queue task tracking, allowing the computed mask to index outside the valid task range and corrupt adjacent memory.

CONFIG_SCSI_UFSHCD is not set in the Root Lock kernel. The UFS host controller driver is not compiled, and no UFS source files are present under drivers/scsi/ufs/ in the kernel tree. The prior claim that “ufshcd is compiled in but never bound to hardware” was incorrect — the driver does not exist in the running kernel image at all. The Root Lock kernel predates the upstream fix, but the fix is not required: it patches code that was never compiled in.

CVE-2022-48662

Status: Not exploitable Component: Intel i915 DRM driver — i915_perf (CONFIG_DRM_I915) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no Intel display GPU present Affected range: Linux 5.19.x before 5.19.16; 5.15.x before 5.15.74; earlier stable series also affected Upstream fix: Linux 5.19.16, 5.15.74, 5.10.148, 5.4.218, 4.19.263 (fix landed within the 5.19 branch before it reached EOL; 5.19.6 predates it)

This CVE describes a use-after-free in the i915 performance monitoring subsystem (i915_perf.c). During OA register reconfiguration, i915_perf iterates i915->gem.contexts.list under i915->gem.contexts.lock. For each entry it acquires a reference via kref_get_unless_zero() and then drops the spin lock to call gen8_configure_context(). After the call it re-acquires the lock and calls list_safe_reset_next(ctx, cn, link) to advance the iteration cursor — dereferencing ctx->link. The assumption is that holding a reference prevents the context from being unlinked. It does not: a concurrent thread can remove ctx from the list while its refcount is non-zero. When list_safe_reset_next dereferences ctx->link after the lock is re-acquired, it reads from freed or repurposed list-head memory.

CONFIG_DRM_I915=y is compiled in and 5.19.6 falls within the affected range. No Intel integrated or discrete display GPU is present on a server deployment. Without GPU hardware, DRM device nodes are not created and the i915_perf entry point is unreachable. This follows the established pattern for i915 CVEs — see CVE-2022-4139.

If exploited on a deployment with i915 hardware, the CVE leads to kernel memory corruption that can escalate to root. At that point Lockdown constrains everything the attacker can do with that root access.

The allowlist database files are made immutable before Lockdown is engaged. FS_IOC_SETFLAGS returns EPERM for all callers (kernel/ioctl.c), so root cannot use chattr to clear those immutable flags and rewrite the allowlist. mount(), fsmount(), and move_mount() all return EPERM (kernel/namespace.c), blocking any bind-mount or remount attempt to shadow or replace the allowlist files. HeartSuite reactivation is disabled during Lockdown, so the service cannot be reconfigured to accept new entries through any path.

Lockdown’s allowlist adds a further constraint on program execution: every execution is checked at the kernel level, applying equally to root. An attacker who has gained root cannot execute a backdoor program they drop onto the filesystem — it has no allowlist entry, and the kernel refuses to run it regardless of file ownership or permission bits.

CVE-2024-26934

Status: Not exploitable Component: USB core (CONFIG_USB) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no USB interface device on headless HS server; deadlock race unreachable Affected range: 4.11–6.8 Upstream fix: 6.8.2 series

Among the attribute file callback routines in drivers/usb/core/sysfs.c, interface_authorized_store() is the only one that acquires a device lock on an ancestor device. It delegates immediately to usb_deauthorize_interface() (drivers/usb/core/message.c), which takes device_lock(dev->parent) first (line 1792) and then device_lock(dev) (line 1795). This lock ordering diverges from other USB subsystem paths, creating an ABBA deadlock when a concurrent bind or configuration operation holds the interface device lock and waits to acquire the parent lock while usb_deauthorize_interface() holds the parent lock and waits for the child. The deadlock stalls the USB subsystem and can produce a kernel hang. The HS 5.19.6 kernel carries the unpatched interface_authorized_store() at drivers/usb/core/sysfs.c:1172 and the unchanged usb_deauthorize_interface() at drivers/usb/core/message.c:1792.

CONFIG_USB=y is compiled in and 5.19.6 falls within the affected range. Triggering the ABBA deadlock race requires writing to the /sys/.../authorized sysfs attribute of an enumerated USB interface device while a concurrent USB operation is in progress. Root Lock runs on headless server hardware with no external USB devices connected; no USB interface device is enumerated, so the sysfs path does not exist and the race condition is unreachable. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-26939

Status: Not exploitable Component: Intel i915 DRM driver (CONFIG_DRM_I915) Base Score: 7.0 HIGH (AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no Intel display GPU present Affected range: pre-6.8 Upstream fix: 6.8 series

Object debugging tools were sporadically reporting illegal attempts to free a still-active i915 VMA object when parking a GT believed to be idle: [161.359441] ODEBUG: free active object type: i915_active. When the GPU’s Graphics Tile (GT) transitions to the parked (powered-down) state, i915_vma_parked() (drivers/gpu/drm/i915/i915_vma.c:1729) iterates the gt->closed_vma list of VMAs marked for deferred destruction. For each candidate it calls i915_gem_object_trylock() (line 1758) and, on success, calls i915_vma_destroy() (line 1760) immediately — without checking whether the VMA’s embedded i915_active tracker has reached zero. If outstanding GPU command-buffer work still holds a live reference through that tracker, the object is freed while completion callbacks continue to dereference it, producing a use-after-free with attacker-controlled timing on the GPU side.

CONFIG_DRM_I915=y is compiled in. No Intel integrated or discrete display GPU is present on this server deployment. Without display hardware, DRM device nodes are not created and the GT power-management paths that call i915_vma_parked() are never reached. The environmental score reflects this: the vulnerable code path is structurally unreachable on the deployed hardware configuration.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2022-48689

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: TCP receive zerocopy (CONFIG_INET) Base Score: 7.0 HIGH (AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 6.5 MEDIUM — Lockdown reduces MI: High→Low; AC:H reduces exploitability (Exp=1.05 vs 1.83 for AC:L) Affected range: 4.14–pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

What this means for an attacker:

A syzbot report identified a misuse of pfmemalloc page status in TCP zerocopy receive paths. In tcp_zerocopy_receive() (net/ipv4/tcp.c:2086), socket buffer fragment pages are collected into a batch (line 2178: page = skb_frag_page(frags)) and mapped directly into userspace via vm_insert_pages(). No page_is_pfmemalloc() check is performed before adding a fragment page to the batch. Pages allocated from pfmemalloc reserves (used to break memory-pressure deadlocks in the network receive path) carry special lifecycle accounting; mapping them into userspace circumvents that accounting. A local attacker who can induce a pfmemalloc allocation into the TCP receive path can map a reserve page into their own address space, potentially corrupting page refcount state in ways that lead to privilege escalation.

Why the score is not 0.0:

CONFIG_INET=y is compiled in and 5.19.6 falls within the affected range. The TCP zerocopy receive path (TCP_ZEROCOPY_RECEIVE ioctl on a connected socket) is reachable by any local user with network access. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2022-48701

Status: Not exploitable Component: USB audio driver (CONFIG_SND_USB_AUDIO) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — driver not compiled in Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

There may be a bad USB audio device with a USB ID of (0x04fa, 0x4201) and fewer than 4 interfaces; an out-of-bounds read bug occurs when the USB audio stream parser iterates altsettings. The Dallas DS4201 workaround at sound/usb/stream.c:1108 unconditionally caps num = 4 regardless of how many altsettings the device actually reports. If a malicious or malformed device presents that USB ID with fewer than 4 altsettings, the loop at line 1111 accesses iface->altsetting[i] beyond the bounds of the array, leaking kernel memory.

CONFIG_SND_USB_AUDIO is not set in the HS 5.19.6 configuration. The USB audio driver — including the vulnerable sound/usb/stream.c altsetting parser — is not compiled into the kernel image. A USB device with this ID cannot be claimed by any USB audio driver, and the vulnerable code path does not exist on this system.

CVE-2022-48702

Status: Not exploitable Component: EMU10K1 audio driver (CONFIG_SND_EMU10K1) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — driver not compiled in Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

The voice allocator sometimes begins allocating from near the end of the array and then wraps around; however snd_emu10k1_pcm_channel_alloc() accesses the voices array without the wrapping modulo that the allocator itself uses. The round-robin allocator in sound/pci/emu10k1/voice.c:42 uses i %= NUM_G to keep indices in bounds, but sound/pci/emu10k1/emupcm.c:127 assigns multichannel voices as &emu->voices[epcm->voices[0]->number + i] with no % NUM_G guard. When the allocator places the first voice near the end of the 64-entry array and more than one voice is requested, the addition exceeds array bounds, producing an out-of-bounds read and write that can corrupt adjacent kernel memory.

CONFIG_SND_EMU10K1 is not set in the HS 5.19.6 configuration. The EMU10K1 driver — including the vulnerable sound/pci/emu10k1/emupcm.c channel allocator — is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2022-48695

Status: Not exploitable Component: mpt3sas SCSI driver (CONFIG_SCSI_MPT3SAS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — driver not compiled in Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

A use-after-free occurs during controller reset in the mpt3sas firmware event cleanup path. In drivers/scsi/mpt3sas/mpt3sas_scsih.c, the reset handler iterates queued firmware events and calls cancel_work_sync() on each. When cancel_work_sync() returns non-zero (the work was never executed), the handler calls fw_event_work_put() at line 3752 to release the work’s reference — then unconditionally calls fw_event_work_put() again at line 3754. This double decrement underflows the kref reference count, freeing the fw_event_work object while other paths may still hold pointers to it.

CONFIG_SCSI_MPT3SAS is not set in the HS 5.19.6 configuration. The mpt3sas driver — including the vulnerable firmware event cleanup path — is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2024-35789

Status: Not exploitable Component: mac80211 wireless stack (CONFIG_MAC80211) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no WiFi NIC present Affected range: pre-fix Upstream fix: 6.9 series

When moving a station out of a VLAN and deleting the VLAN afterwards, the fast_rx entry still holds a pointer to the VLAN’s netdev, which can cause use-after-free. In net/mac80211/cfg.c, the station change path at line 1949 calls __ieee80211_check_fast_rx_iface(vlansdata), which builds a new fast_rx structure with dev = vlansdata->dev (the target VLAN’s netdev). The original VLAN’s fast_rx is cleared at line 1955 via ieee80211_clear_fast_rx(sta), but that function uses RCU: the old fast_rx object — containing dev = original_vlan->dev — is not freed until after a grace period. If the original VLAN interface is deleted before that grace period expires, any CPU still reading the old fast_rx entry will dereference a freed netdev. The HS 5.19.6 kernel carries the unpatched station change path at net/mac80211/cfg.c:1939–1970.

CONFIG_MAC80211=y is compiled in. No WiFi network interface card is present on a server deployment. Without WiFi hardware, mac80211 creates no wireless interfaces and the relevant code paths are never reached.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2024-35886

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: IPv6 networking stack (CONFIG_IPV6) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: pre-fix Upstream fix: 6.9 series

What this means for an attacker:

syzkaller reported infinite recursive calls of fib6_dump_done() during netlink socket destruction. From the log, syzkaller sent an AF_UNSPEC RTM_GETROUTE message, and then closed the netlink socket. The IPv6 FIB dump handler at net/ipv6/ip6_fib.c:652 hooks the callback destructor by setting cb->done = fib6_dump_done (saving the original callback in cb->args[3]). When the netlink socket closes, netlink core invokes the destructor, calling fib6_dump_done() at line 570. This function calls cb->done(cb) — but cb->done is now fib6_dump_done itself, creating infinite recursion that exhausts the kernel stack. The HS 5.19.6 kernel carries the unpatched FIB dump callback at net/ipv6/ip6_fib.c:645–684.

Why the score is not 0.0:

CONFIG_IPV6=y is compiled in and 5.19.6 falls within the affected range. Triggering the infinite recursion requires sending an AF_UNSPEC RTM_GETROUTE netlink message and then closing the socket — reachable by any local user with a netlink socket. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2023-52835

Status: Not exploitable Component: perf events subsystem (CONFIG_PERF_EVENTS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — perf_event_paranoid=3 restricts perf_event_open(); no profiling tool in Root Lock allowlist Affected range: pre-fix Upstream fix: 6.8 series

When perf-record with a large AUX area, e.g. 4GB, it fails with: #perf record -C 0 -m ,4G -e arm_spe_0// -- sleep 1 failed to mmap with 12 (Cannot allocate memory). The perf AUX area mmap handler in kernel/events/core.c:6269–6345 calculates memory accounting limits and calls rb_alloc_aux() to allocate the backing pages. For very large AUX areas (gigabytes), the accounting arithmetic at line 6285 (user_locked += user_extra) can underflow or produce incorrect values when user_extra is extremely large (e.g., 1M pages for 4GB). The mmap() still succeeds despite the accounting failure, allowing unprivileged users to bypass RLIMIT_MEMLOCK restrictions and exhaust kernel memory. The HS 5.19.6 kernel carries the unpatched AUX area accounting at kernel/events/core.c:6269–6345.

CONFIG_PERF_EVENTS=y is compiled in and 5.19.6 falls within the affected range. On a Root Lock system, perf_event_paranoid=3 restricts perf_event_open() to processes with CAP_SYS_ADMIN; no profiling or performance analysis tool appears in the Root Lock allowlist. The exploitation path — loading and executing a non-allowlisted program — is blocked at the kernel execution gate before any perf subsystem interaction is possible. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2023-52868

Status: Not exploitable Component: thermal management (CONFIG_THERMAL) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — thermal sysfs not accessible in Root Lock allowlist; Lockdown blocks the trigger Affected range: pre-fix Upstream fix: 6.9 series

The dev->id value comes from ida_alloc(), so it is a number between zero and INT_MAX. In drivers/thermal/thermal_core.c, this ID is formatted into fixed-size THERMAL_NAME_LENGTH (20-byte) buffers using sprintf(). At line 681, sprintf(dev->attr_name, "cdev%d_trip_point", dev->id) produces a string of the form "cdev<N>_trip_point". For large IDs, the full string exceeds 20 bytes: "cdev2147483647_trip_point" is 25 characters plus a null terminator (26 bytes total), overflowing attr_name by 6 bytes. The same overflow applies at line 690 for sprintf(dev->weight_attr_name, "cdev%d_weight", dev->id), which produces up to 22 bytes into a 20-byte buffer. Both overflows corrupt adjacent kernel heap memory and can be leveraged for privilege escalation.

CONFIG_THERMAL=y is compiled in and 5.19.6 falls within the affected range. Thermal management is present on all x86 servers for CPU temperature control. Triggering the overflow requires registering a thermal cooling device with a sufficiently large ID — this path requires access to the thermal sysfs interface, which is not included in the Root Lock allowlist. On a Root Lock system in Lockdown, the kernel blocks any process without an allowlist entry from executing, so a standalone exploit tool cannot reach the thermal registration interface. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-36916

Status: Not exploitable Component: block I/O cost controller (CONFIG_BLK_CGROUP_IOCOST) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — iocost cgroup paths not in Root Lock allowlist; Lockdown blocks the trigger Affected range: pre-fix Upstream fix: 6.9 series

UBSAN catches undefined behavior in blk-iocost, where sometimes iocg->delay is shifted right by a number that is too large, resulting in undefined behavior on some architectures. Two sites in block/blk-iocost.c are affected: line 1338 computes iocg->delay >> div64_u64(tdelta, USEC_PER_SEC), where the divisor is elapsed time in seconds — if the delay has been active for 64 or more seconds, the shift amount reaches or exceeds 64, which is undefined behavior for a 64-bit type under the C standard. Line 2112 performs iocg->delay >> nr_cycles, where nr_cycles can similarly exceed 63. On x86 the shift wraps, but on other architectures the result is indeterminate. Incorrect delay values can bypass I/O throttling controls or cause the cgroup I/O cost model to make scheduling decisions based on garbage data.

CONFIG_BLK_CGROUP_IOCOST=y is compiled in and 5.19.6 falls within the affected range. The blk-iocost controller is active whenever cgroups are in use with I/O cost weighting enabled. Configuring iocost requires writing to cgroup control files under /sys/fs/cgroup/ — no cgroup management tool that exposes iocost configuration appears in the Root Lock allowlist. On a Root Lock system in Lockdown, the kernel blocks any process without an allowlist entry from executing, so a standalone exploit tool cannot reach the iocost configuration path. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-38560

Status: Not exploitable Component: Brocade bfa SCSI driver (CONFIG_SCSI_BFA) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — driver not compiled in Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

Currently, we allocate a nbytes-sized kernel buffer and copy nbytes from userspace to that buffer. In drivers/scsi/bfa/bfad_bsg.c, the BSG passthrough handler at line 3373 allocates kzalloc(bsg_data->payload_len, GFP_KERNEL) where payload_len comes directly from the user-supplied BSG request structure, with no upper-bound validation. Line 3379 then calls copy_from_user(..., bsg_data->payload_len) using the same unchecked value. An attacker with access to the BSG device node can supply an oversized payload_len to exhaust kernel memory or, with a carefully chosen value, produce a heap overflow.

CONFIG_SCSI_BFA is not set in the HS 5.19.6 configuration. The Brocade bfa Fibre Channel HBA driver — including the vulnerable bfad_bsg.c BSG handler — is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2024-38588

Status: Not exploitable Component: kprobes (CONFIG_KPROBES) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — kprobe registration not in Root Lock allowlist; Lockdown blocks the exploitation trigger Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In kernel/trace/ftrace.c, ftrace_location() at line 1577 calls lookup_rec(ip, ip) at line 1583 to obtain a dyn_ftrace *rec pointer without holding ftrace_lock. On a concurrent path, module unloading frees the pages that back ftrace records for module functions. If a module is removed between the lookup_rec() return and the return rec->ip dereference at line 1594, the pointer references freed memory. The race is reached through the kprobe registration path: check_kprobe_address_safe() → check_ftrace_location() → ftrace_location() — all called without the lock that serialises ftrace record lifetime.

CONFIG_KPROBES=y is compiled in. Triggering the bug requires CAP_SYS_ADMIN to register a kprobe — the attack path runs through check_kprobe_address_safe() → check_ftrace_location() → ftrace_location(). No Root Lock deployment permits any service to register kprobes. Without an allowlist entry covering the kprobes interface, the kernel refuses access. An attacker who has already gained root cannot add one: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2024-40901

Status: Not exploitable Component: LSI/Avago mpt3sas SCSI driver (CONFIG_SCSI_MPT3SAS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — driver not compiled in Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In drivers/scsi/mpt3sas/mpt3sas_scsih.c, the pd_handles bitmap is allocated as (ioc->facts.MaxDevHandle / 8) bytes (rounded up) via kzalloc() at mpt3sas_base.c:8312. The test_bit() function accesses bitmaps in unsigned long-sized units (8 bytes on 64-bit kernels). When the allocation is smaller than sizeof(unsigned long) — for example a single byte when MaxDevHandle is 8 — calls such as test_bit(sas_device->handle, ioc->pd_handles) at line 1942 and test_bit(handle, ioc->pd_handles) at line 4106 read 7 bytes beyond the heap allocation, producing a slab out-of-bounds read.

CONFIG_SCSI_MPT3SAS is not set in the HS 5.19.6 configuration. The LSI/Avago mpt3sas SAS/SATA/NVMe HBA driver — including the vulnerable mpt3sas_scsih.c bitmap access paths — is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2024-40978

Status: Not exploitable Component: QLogic qedi iSCSI driver (CONFIG_SCSI_QEDI) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — driver not compiled in Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In drivers/scsi/qedi/qedi_debugfs.c, qedi_dbg_do_not_recover_cmd_read() at line 128 calls sprintf(buffer, "do_not_recover=%d\n", qedi_do_not_recover) where buffer is the char __user * argument passed directly from the debugfs file read handler. sprintf() writes to a kernel virtual address rather than staging data in a kernel buffer first; on a system with SMAP (Supervisor Mode Access Prevention) enabled, the kernel write to a userspace pointer faults immediately and panics the kernel. The correct fix is to stage into a kernel buffer and use simple_read_from_buffer() to copy to userspace.

CONFIG_SCSI_QEDI is not set in the HS 5.19.6 configuration. The QLogic qedi iSCSI HBA driver — including the vulnerable qedi_dbg_do_not_recover_cmd_read() debugfs handler — is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2024-41092

Status: Not exploitable Component: Intel i915 DRM driver (CONFIG_DRM_I915) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no Intel display GPU present Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In the i915 GT reset path, intel_gt_handle_error() at intel_reset.c:1309 calls synchronize_srcu_expedited() at line 1285 on gt->reset.backoff_srcu to drain concurrent SRCU readers before the GPU reset proceeds. The GuC engine failure worker (reset_fail_worker_func at intel_guc_submission.c:4485) queues via queue_work() at line 4545 and calls intel_gt_handle_error() asynchronously. A race between this deferred reset path and the hangcheck heartbeat — as reproduced by igt@i915_selftest@live@hangcheck on ADL-P (GuC submission) — can reach reset_prepare_engine() at intel_reset.c:743 and the WW-mutex backoff context via i915_gem_ww_ctx_backoff() (i915_gem_ww.c:42) after the owning structure has already been freed, producing a use-after-free.

CONFIG_DRM_I915=y is compiled in and HS 5.19.6 falls within the affected range. No Intel integrated or discrete display GPU is present on a standard Debian 11 server deployment. Without display hardware the DRM device nodes are not created, the GPU submission paths are not initialised, and the GuC engine failure worker that triggers this race is never scheduled. The vulnerable code path cannot be reached.

On a HeartSuite system with this hardware installed, Lockdown’s constraints would still apply after any escalation: FS_IOC_SETFLAGS returns EPERM (kernel/ioctl.c:561–569), every mount path returns EPERM (kernel/namespace.c:4218, 4300, 4453), and allowlist modification is blocked at hs_sandbox_caching.c:1942. Lockdown independently prevents any unauthorised program — including a backdoor dropped post-exploit — from executing regardless of file ownership or capability bits.

CVE-2024-42136

Status: Not exploitable Component: CD-ROM subsystem (CONFIG_CDROM) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CD-ROM drive absent on server Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In drivers/cdrom/cdrom.c, cdrom_read_cd() at line 2080 computes cgc->buflen = blocksize * nblocks and cdrom_read_block() at line 2104 computes cgc->buflen = blksize * nblocks. Both operands are int parameters, so the multiplication is evaluated as a signed 32-bit expression before being stored in the unsigned int buflen field of struct packet_command. When syzkaller passes a large nblocks value — for example, greater than 912,000 with the common CD_FRAMESIZE_RAW = 2352 block size — the intermediate product exceeds INT_MAX, signed integer overflow occurs, and an incorrect (smaller) buffer length is stored in cgc->buflen.

CONFIG_CDROM=y is compiled in and HS 5.19.6 falls within the affected range. No optical drive is present on a standard Debian 11 server deployment. Without this hardware the CD-ROM device nodes are not created and the ioctl paths that call cdrom_read_cd() and cdrom_read_block() are never reached. The vulnerable code path cannot be triggered.

On a HeartSuite system with an optical drive installed, Lockdown’s constraints would still apply after any escalation: FS_IOC_SETFLAGS returns EPERM (kernel/ioctl.c:561–569), every mount path returns EPERM (kernel/namespace.c:4218, 4300, 4453), and allowlist modification is blocked at hs_sandbox_caching.c:1942. Lockdown independently prevents any unauthorised program — including a backdoor dropped post-exploit — from executing regardless of file ownership or capability bits.

CVE-2024-44985

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: IPv6 networking stack (CONFIG_IPV6) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

What this means for an attacker:

In net/ipv6/ip6_output.c, ip6_finish_output2() saves idev = ip6_dst_idev(dst) at line 63. At line 72, skb_expand_head(skb, hh_len) makes room for the link-layer header; when allocation fails, skb_expand_head() frees the original skb and returns NULL. The idev pointer saved before the call now references memory associated with the freed skb. At line 74, IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS) dereferences the stale idev — a use-after-free.

Why the score is not 0.0:

CONFIG_IPV6=y is compiled in and HS 5.19.6 falls within the affected range. Any local process that sends IPv6 network traffic can trigger the vulnerable allocation failure paths; no capability gate is required. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-44986

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: IPv6 networking stack (CONFIG_IPV6) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

What this means for an attacker:

In net/ipv6/ip6_output.c, ip6_xmit() saves idev = ip6_dst_idev(dst) at line 256. At line 271, skb_expand_head(skb, head_room) expands the buffer to accommodate the IPv6 header and IP options; when allocation fails, the original skb is freed and NULL is returned. The idev pointer is now stale. At line 273, IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS) dereferences freed memory — a use-after-free.

Why the score is not 0.0:

CONFIG_IPV6=y is compiled in and HS 5.19.6 falls within the affected range. Any local process that sends IPv6 network traffic can trigger the vulnerable allocation failure paths; no capability gate is required. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-44987

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: IPv6 networking stack (CONFIG_IPV6) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

What this means for an attacker:

In net/ipv6/ip6_output.c, ip6_send_skb() at line 1943 saves rt = (struct rt6_info *)skb_dst(skb) without holding rcu_read_lock(). At line 1946, ip6_local_out() transmits the packet and may consume the skb, releasing the associated route. If ip6_local_out() returns a non-zero error code, lines 1951–1952 dereference rt->rt6i_idev — but rt is an RCU-protected pointer and may be freed before the dereference. Holding rcu_read_lock() for the duration of the rt dereference is required.

Why the score is not 0.0:

CONFIG_IPV6=y is compiled in and HS 5.19.6 falls within the affected range. Any local process that sends IPv6 network traffic can trigger the vulnerable allocation failure paths; no capability gate is required. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-46673

Status: Not exploitable Component: Adaptec aacraid SCSI driver (CONFIG_SCSI_AACRAID) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — driver not compiled in Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

aac_probe_one() at drivers/scsi/aacraid/linit.c:1577 calls the hardware-specific init function pointer from aac_driver_ident, which eventually calls aac_init_adapter() at comminit.c:510. On failure, aac_init_adapter() frees dev->queues internally at line 644 (on aac_comm_init() failure) or line 651 (on aac_fib_setup() failure) before returning NULL. The aac_probe_one() error path at linit.c:1798 then calls kfree(aac->queues) a second time on the same pointer — a double-free.

CONFIG_SCSI_AACRAID is not set in the HS 5.19.6 configuration. The Adaptec aacraid RAID controller driver — including the vulnerable aac_probe_one() and aac_init_adapter() paths — is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2024-46746

Status: Not exploitable Component: AMD Sensor Fusion Hub HID driver (CONFIG_AMD_SFH_HID) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — driver not compiled in Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In drivers/hid/amd-sfh-hid/amd_sfh_client.c, the error cleanup path calls devm_kfree(dev, cl_data->report_descr[i]) at lines 259 and 276 to free the HID report descriptor before hid_destroy_device() at line 178. The amdtp_hid_parse() callback at amd_sfh_hid.c:32 accesses cli_data->report_descr[hid_data->index] during device initialisation or tear-down. If the descriptor is freed before hid_destroy_device() has completed its disconnect sequence — and the callback fires in that window — it dereferences freed memory.

CONFIG_AMD_SFH_HID is not set in the HS 5.19.6 configuration. The AMD Sensor Fusion Hub HID driver — including the vulnerable amd_sfh_client.c cleanup path and amdtp_hid_parse() callback — is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2024-46747

Status: Not exploitable Component: Cougar HID driver (CONFIG_HID_COUGAR) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — driver not compiled in Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

cougar_report_fixup() at drivers/hid/hid-cougar.c:109 reads rdesc[2], rdesc[3], rdesc[115], and rdesc[116], and conditionally writes to rdesc[115]–rdesc[116] at lines 113–114, without first checking that *rsize >= 117. If the Cougar 500k Gaming Keyboard presents a report descriptor shorter than 117 bytes, the fixed-offset accesses go beyond the descriptor buffer, producing an out-of-bounds memory read/write.

CONFIG_HID_COUGAR is not set in the HS 5.19.6 configuration. The Cougar HID driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2024-46798

Status: Not exploitable Component: ALSA rawmidi subsystem (CONFIG_SND_RAWMIDI) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SND_RAWMIDI not compiled Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In sound/core/rawmidi.c, snd_rawmidi_drain_output() at line 224 saves runtime = substream->runtime at line 228, then calls wait_event_interruptible_timeout(runtime->sleep, ...) at line 232, waiting up to 10 seconds for the output buffer to drain. If close_substream() runs concurrently and calls snd_rawmidi_runtime_free(substream) at line 528 — freeing substream->runtime — while the drain wait is still sleeping, the runtime pointer saved at line 228 becomes dangling. When the wait exits, accesses to runtime->avail and runtime->buffer_size at line 237 use freed memory.

CONFIG_SND_RAWMIDI is not compiled in the HS 5.19.6 configuration — no enabled driver selects it. The vulnerable rawmidi.c code path does not exist on this system.

CVE-2024-46849

Status: Not exploitable Component: Amlogic Meson ASoC driver (CONFIG_SND_MESON_CARD_UTILS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — driver not compiled in Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In sound/soc/meson/axg-card.c, axg_card_add_loopback() at line 107 saves pad = &card->dai_link[*index] — a pointer into the current dai_link array. At line 113, meson_card_reallocate_links() calls krealloc() on card->dai_link, potentially moving the array to a new address and freeing the original buffer. At lines 119 and 133, pad->name and pad->cpus->of_node are accessed through the now-dangling pad pointer. The fix moves the pad assignment to after the reallocation, where card->dai_link has been updated.

CONFIG_SND_MESON_CARD_UTILS is not compiled in the HS 5.19.6 configuration — the Amlogic Meson ASoC platform requires ARCH_MESON which is not set on x86. The vulnerable code path does not exist on this system.

CVE-2024-47682

Status: Not exploitable Component: SCSI subsystem (CONFIG_SCSI) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — non-conformant VPD firmware absent; standard SAS/SATA drives conform to SCSI spec Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

If a device returns VPD page 0xb1 with a length of exactly 8 bytes (as QEMU v2.x does), sd_read_block_characteristics() proceeds past the guard at drivers/scsi/sd.c:2921 (vpd->len < 8), then reads vpd->data[8] at line 2927. With len == 8 the valid indices are 0–7; index 8 is one byte past the end of the buffer.

CONFIG_SCSI=y is compiled in and HS 5.19.6 falls within the affected range. The OOB read occurs during device enumeration when a SCSI disk returns VPD page 0xb1 with a length of exactly 8 bytes — behaviour documented in QEMU v2.x, not present on production SAS/SATA/NVMe drives. Standard enterprise storage conforms to the SCSI VPD specification and returns page 0xb1 with the correct length. On a Root Lock server deployment, no non-conformant storage device is present; the OOB read path in sd_read_block_characteristics() is never reached. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-47701

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

What this means for an attacker:

When ext4 searches an inlined directory, ext4_find_inline_entry() at fs/ext4/inline.c:1709 calls ext4_get_inline_xattr_pos() to locate the extended-attribute portion of the inline data. At inline.c:1077, that function returns IFIRST(header) + le16_to_cpu(entry->e_value_offs) without validating that the offset stays within the inode body buffer. A crafted block device can supply an e_value_offs that pushes the resulting pointer out of bounds; that pointer is then passed directly to ext4_search_dir() at line 1712, causing an OOB memory access.

Why the score is not 0.0:

CONFIG_EXT4_FS=y is compiled in and HS 5.19.6 falls within the affected range. ext4 is the primary filesystem on a Debian 11 server; inlined directory processing runs for any small directory during normal operation. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-49852

Status: Not Affected — CONFIG_SCSI_EFCT not compiled Component: Emulex EFC FC driver (CONFIG_SCSI_EFCT) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_EFCT not compiled in Root Lock kernel

The kref_put() function will call nport->release if the refcount drops to zero. The nport->release release function is_efc_nport_free() which frees “nport”

CONFIG_SCSI_EFCT is not set in the HS 5.19.6 configuration. The Emulex EFC Fibre Channel target driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2024-49882

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

What this means for an attacker:

In ext4_ext_try_to_merge_up() at fs/ext4/extents.c:1871, brelse(path[1].p_bh) releases the depth-1 extent block buffer but leaves path[1].p_bh non-NULL. When the caller subsequently runs cleanup via ext4_ext_drop_refs(), it iterates the path and calls brelse() on every non-NULL p_bh, releasing the same buffer head a second time — a use-after-free.

Why the score is not 0.0:

CONFIG_EXT4_FS=y is compiled in and HS 5.19.6 falls within the affected range. ext4 is the primary filesystem on a Debian 11 server; extent tree merge-up runs during any truncate or extent modification on a two-level extent tree. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-49883

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

What this means for an attacker:

In ext4_ext_insert_extent() at fs/ext4/extents.c:2094, the call to ext4_ext_create_new_leaf() may internally call ext4_ext_grow_indepth(), which reallocates the path array via kcalloc(). After the call returns, the caller continues using the original path pointer — now stale — causing a use-after-free.

Why the score is not 0.0:

CONFIG_EXT4_FS=y is compiled in and HS 5.19.6 falls within the affected range. ext4 is the primary filesystem on a Debian 11 server; extent insertion runs during any file write that extends or modifies the extent tree. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-49884

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

What this means for an attacker:

In ext4_split_extent_at() at fs/ext4/extents.c:3178, the function saves the path pointer as path = *ppath. At line 3248 it calls ext4_ext_insert_extent(handle, inode, ppath, ...), which may reallocate *ppath, freeing the memory that path still points to. Subsequent uses of path at lines 3281, 3282, 3301, and 3304 — in both the success and error-recovery branches — dereference the now-freed pointer, constituting a use-after-free.

Why the score is not 0.0:

CONFIG_EXT4_FS=y is compiled in and HS 5.19.6 falls within the affected range. ext4 is the primary filesystem on a Debian 11 server; extent splitting is triggered during any write that bisects an existing extent. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-49889

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

What this means for an attacker:

ext4_find_extent() at fs/ext4/extents.c:874 takes an optional **orig_path argument allowing callers to reuse an existing path allocation. On two code paths it frees the old allocation: when the tree depth has grown beyond the cached maximum (lines 898–901, kfree(path); *orig_path = NULL) and on any I/O or corruption error (lines 953–957, same sequence). Callers that save a local path = *ppath copy before invoking a sub-function that internally calls ext4_find_extent() — such as ext4_split_convert_extents() — retain a pointer to the freed memory. Subsequent use of that stale pointer constitutes a use-after-free.

Why the score is not 0.0:

CONFIG_EXT4_FS=y is compiled in and HS 5.19.6 falls within the affected range. ext4 is the primary filesystem on a Debian 11 server; any extent-modifying write that triggers a tree depth change or encounters a read error while holding a saved path pointer is a triggering condition. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-49960

Status: Not exploitable Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — mount() blocked by Lockdown; do_mount() returns EPERM Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In ext4_fill_super() (fs/ext4/super.c), timer_setup(&sbi->s_err_report, ...) runs at line 4995 and INIT_WORK(&sbi->s_error_work, flush_stashed_error_work) at line 4997. During the failed_mount3: error-unwind at line 5454, flush_work(&sbi->s_error_work) is called at line 5456 immediately before del_timer_sync(&sbi->s_err_report) at line 5457. The work callback flush_stashed_error_work can call mod_timer on s_err_report, arming the timer during the same unwind that is about to cancel it. When the code path passes through failed_mount_wq: (line 5439), flush_work runs a second time at line 5448 before falling through to failed_mount3:, doubling the exposure. Syzbot detected this as an ODEBUG (Object Debug) object-state inconsistency.

CONFIG_EXT4_FS=y is compiled in and HS 5.19.6 falls within the affected range. The vulnerable path runs during a failed mount — for example when ext4_es_register_shrinker() or journal loading fails partway through ext4_fill_super(). On a Root Lock system, sys_hs_lockdown_hs() blocks all mount paths at kernel/namespace.c:4218, 4300, 4453; do_mount() returns EPERM before any filesystem setup begins. No approved process in the Root Lock allowlist carries a mount allowlist entry, and unapproved programs are refused execution by the kernel’s SPF gate regardless of file ownership or privilege. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-49983

Status: Not exploitable Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — mount() blocked by Lockdown; do_mount() returns EPERM Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In ext4_ext_replay_update_ex() at fs/ext4/extents.c:5860, line 5879 assigns ppath = path, making both local variables alias the same allocation. Line 5881 then calls ext4_force_split_extent_at(NULL, inode, &ppath, start, 1), passing the address of ppath. Inside, ext4_split_extent_at() calls ext4_ext_insert_extent() which may invoke ext4_ext_grow_indepth() and reallocate *ppath via kcalloc(). When that happens, the outer ppath is updated to the new allocation and the original memory is freed — but path still holds the original (now stale) pointer. The kfree(path) call at line 5885 then frees already-freed memory, constituting a double-free/use-after-free. The bug is exercised during fast-commit journal replay.

CONFIG_EXT4_FS=y is compiled in and HS 5.19.6 falls within the affected range. The vulnerable path runs during fast-commit journal replay, triggered on mount after an unclean shutdown of a filesystem with the fast-commit feature enabled. On a Root Lock system, sys_hs_lockdown_hs() blocks all mount paths at kernel/namespace.c:4218, 4300, 4453; do_mount() returns EPERM before any filesystem setup begins. No approved process in the Root Lock allowlist carries a mount allowlist entry, and unapproved programs are refused execution by the kernel’s SPF gate regardless of file ownership or privilege. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-50007

Status: Not Affected Component: ASIHPI soundcard driver (CONFIG_SND_ASIHPI) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SND_ASIHPI not compiled

The ASIHPI driver writes firmware-controlled index values into a static array without bounds-checking the index. CONFIG_SND_ASIHPI is not set in the HS 5.19.6 kernel configuration; the driver and this code path are absent from the compiled kernel image.

CVE-2022-48951

Status: Not Affected Component: ALSA SoC layer (CONFIG_SND_SOC) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SND_SOC not compiled

snd_soc_put_volsw_sx() applies bounds checks only to the first channel, allowing out-of-bounds writes to the second. CONFIG_SND_SOC is not set in the HS 5.19.6 kernel configuration; the ALSA SoC layer and this function are absent from the compiled kernel image.

CVE-2022-48956

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: IPv6 networking stack (CONFIG_IPV6) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

ip6_fragment() at net/ipv6/ip6_output.c:831 handles IPv6 packet fragmentation. The function and its callees access RCU-protected routing and neighbor table entries; a prior commit added an assumption that all callers hold the RCU read lock at entry. For the IPv4-style fast path via ip6_finish_output2() this holds — rcu_read_lock_bh() is acquired at line 119. However the UDP egress path (ip6_send_skb() at line 1940 → ip6_local_out() → ip6_output() → ip6_finish_output() → ip6_fragment()) does not guarantee the lock is held before entry into the fragmentation code. Under concurrent route or neighbor table modification this produces a use-after-free. Syzbot confirmed the race.

CONFIG_IPV6=y is compiled in and HS 5.19.6 falls within the affected range. IPv6 is active on any Debian 11 server that has IPv6 addresses configured; the UDP-over-IPv6 fragmentation path is reachable by any process with a UDP socket. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2022-49022

Status: Not exploitable Component: mac80211 wireless stack (CONFIG_MAC80211) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no WiFi NIC present Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In ieee80211_get_rate_duration() at net/mac80211/airtime.c:455, airtime_mcs_groups[group].duration[idx] is accessed where group is computed from bandwidth, stream count, and encoding mode via the VHT_GROUP_IDX/HT_GROUP_IDX/HE_GROUP_IDX macros. The stream-count bounds check at line 451 guards one dimension, but an invalid combination of bandwidth and stream count can produce a group index that exceeds the airtime_mcs_groups array bounds, triggering a UBSAN array-index-out-of-bounds read.

CONFIG_MAC80211=y is compiled in. No WiFi NIC is present on a Debian 11 server deployment; mac80211 creates no wireless interfaces without hardware and this code path is never reached. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or preserve access across a reboot.

CVE-2022-49023

Status: Not exploitable Component: cfg80211 wireless framework (CONFIG_CFG80211) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no WiFi NIC present Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In net/wireless/scan.c:338, when merging per-STA profile elements in the multi-BSSID path, the code calls memcmp(tmp_old + 2, tmp + 2, 5) to compare the OUI (3 bytes) + type (1 byte) + subtype (1 byte) of a vendor element, without first checking that either IE has at least 5 bytes of data. A vendor element with fewer than 5 data bytes causes an out-of-bounds read beyond the element buffer.

CONFIG_CFG80211=y is compiled in. No WiFi NIC is present on a Debian 11 server deployment; cfg80211 creates no wireless interfaces without hardware and this code path is never reached. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or preserve access across a reboot.

CVE-2024-50278

Status: Not Affected Component: dm-cache (CONFIG_DM_CACHE) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — CONFIG_DM_CACHE not compiled

If the cache device is expanded between initial load and first-time resume, the bitsets (dirty_bitset, discard_bitset) allocated in dm-cache-target.c are sized to the pre-expansion block count. On resume, cache-block indices derived from the new device size exceed the allocated bitset bounds, causing an out-of-bounds access. CONFIG_DM_CACHE is not set in the HS 5.19.6 kernel configuration; the dm-cache target and this code path are absent from the compiled kernel image.

CVE-2024-50279

Status: Not Affected Component: dm-cache (CONFIG_DM_CACHE) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — CONFIG_DM_CACHE not compiled

When shrinking the fast (cache) device, dm-cache iterates the dirty_bitset to identify cache blocks that must be flushed before being dropped. An index error in the bitset iteration produces a bit index that exceeds the allocated bitset bounds, causing an out-of-bounds access. CONFIG_DM_CACHE is not set in the HS 5.19.6 kernel configuration; the dm-cache target and this code path are absent from the compiled kernel image.

CVE-2024-53147

Status: Not exploitable Component: FAT/exFAT filesystem (CONFIG_FAT_FS) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:N/I:H/A:H) Score on Root Lock: 0.0 — Lockdown blocks mount(); no adversary-controlled FAT filesystem reachable Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In fs/exfat/dir.c, when iterating directory entries, the cluster-walk loop at line 105 calls exfat_get_next_cluster(sb, &(clu.dir)) to follow the FAT chain. If a directory’s size is at least one cluster (so clu_offset > 0) and ei->start_clu was set to EXFAT_EOF_CLUSTER (0xFFFFFFFF) due to filesystem corruption, clu.dir starts at 0xFFFFFFFF. The call at line 106 then attempts a FAT table lookup at index 0xFFFFFFFF, which is far outside the FAT table’s num_clusters entries, causing an out-of-bounds read.

CONFIG_FAT_FS=y is compiled in and HS 5.19.6 falls within the affected range. exFAT is compiled in and is used for the EFI system partition; the vulnerable path is triggered by traversing a corrupted exFAT directory. The adversary must be able to present a crafted exFAT image — mounting an external device or network share requires mount(), which Lockdown blocks unconditionally. The EFI system partition is already mounted at boot time and its contents are controlled by the administrator; an external attacker cannot inject a malformed exFAT directory into the in-use ESP. On a Root Lock system in Lockdown, the kernel additionally blocks any process without an allowlist entry from executing, closing the exploitation path before it can reach the vulnerable directory traversal code.

CVE-2024-53150

Status: Not Affected Component: USB audio driver (CONFIG_SND_USB_AUDIO) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — CONFIG_SND_USB_AUDIO not compiled

The USB-audio driver does not validate bLength of each descriptor when traversing clock descriptors, allowing a malformed USB device to cause an out-of-bounds read. CONFIG_SND_USB_AUDIO is not set in the HS 5.19.6 kernel configuration; the USB audio driver and this descriptor-traversal path are absent from the compiled kernel image.

CVE-2024-53170

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: SCSI subsystem (CONFIG_SCSI) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

In blk_mq_exit_hctx() at block/blk-mq.c:3440, the call to blk_mq_clear_flush_rq_mapping() (line 3441) is guarded by if (blk_queue_init_done(q)). During SCSI device probe, the queue is not yet fully initialized, so this condition is false and blk_mq_clear_flush_rq_mapping() is skipped. The function is responsible for atomically clearing the flush_rq pointer from every slot in tags->rqs[]. When skipped, flush_rq is subsequently freed but its pointer remains live in the rqs[] array. Any later iteration over tags->rqs[] — such as during a tag-set teardown or request lookup — dereferences the stale pointer, constituting a use-after-free.

CONFIG_SCSI=y is compiled in and HS 5.19.6 falls within the affected range. The SCSI subsystem underpins block storage on Debian 11 via libata; the vulnerable path is triggered during SCSI probe teardown when initialization does not complete successfully. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-53173

Status: Not exploitable Component: NFS v4 client (CONFIG_NFS_V4) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — mount() blocked by Lockdown; no NFS v4 share reachable on Root Lock deployments Affected range: pre-fix Upstream fix: mainline; not backported to 5.19.x (5.19 EOL)

nfs_release_seqid() at fs/nfs/nfs4state.c:1088 removes a seqid from the sequence wait-list and wakes the next waiter (rpc_wake_up_queued_task() at line 1102). When two threads open the same file concurrently and both abort before receiving a reply, two separate code paths each call nfs_release_seqid() on the same nfs_seqid: the prepare callback at fs/nfs/nfs4proc.c:2462 (when nfs4_setup_sequence() returns non-zero) and the done/release callback at line 2061. The second call finds seqid->list already empty and returns without action, but by this point nfs_free_seqid() may have freed the seqid object. The task woken by the first release can dereference seqid->sequence through the nfs_seqid pointer it holds — now pointing to freed memory — constituting a use-after-free.

CONFIG_NFS_V4=y is compiled in and HS 5.19.6 falls within the affected range. The vulnerable seqid use-after-free path is only reachable when an NFS v4 share is mounted. On a Root Lock system, Lockdown blocks mount() unconditionally — do_mount(), fsmount(), and move_mount() all return EPERM (kernel/namespace.c:4218, 4300, 4453). No NFS v4 filesystem can be mounted by any process, so the vulnerable code path is never reached. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-53214

Status: Not Affected Component: VFIO subsystem (CONFIG_VFIO) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_VFIO not compiled

In drivers/vfio/pci/vfio_pci_config.c, the VFIO PCI extended-capability enumeration loop at line 1638 hides capabilities with unknown length by rewriting the next pointer in the previous entry’s header. When a capability should be hidden but occupies the first position in the extended-capability chain, the pointer fixup path has incorrect behaviour, allowing a misconfigured or malicious guest to reach memory it should not. CONFIG_VFIO is not set in the HS 5.19.6 kernel configuration; the VFIO subsystem and this PCI config-space virtualisation path are absent from the compiled kernel image.

CVE-2024-53227

Status: Not Affected Component: Brocade bfa FC driver (CONFIG_SCSI_BFA_FC) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_BFA_FC not compiled

In the Brocade bfa Fibre Channel adapter driver (drivers/scsi/bfa/), a use-after-free occurs during driver load: an internal object containing an embedded spinlock is freed while lockdep still holds a reference to that lock, producing a KASAN slab-use-after-free splat inside __lock_acquire. CONFIG_SCSI_BFA_FC is not set in the HS 5.19.6 kernel configuration; the Brocade bfa driver is absent from the compiled kernel image.

CVE-2024-53239

Status: Not Affected Component: 6fire USB audio driver (CONFIG_SND_USB_6FIRE) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SND_USB_6FIRE not compiled

In the TerraTec AUREON 6fire USB audio driver (sound/usb/6fire/chip.c), usb6fire_chip_disconnect() calls usb6fire_chip_abort() at line 183 — which schedules a deferred snd_card_free_when_closed() and nulls chip->card — immediately followed by usb6fire_chip_destroy() at line 184, which frees the underlying sub-resources. When userspace still holds the card open, the deferred free races against the destroy path, producing a use-after-free. CONFIG_SND_USB_6FIRE is not set in the HS 5.19.6 kernel configuration; the driver is absent from the compiled kernel image.

CVE-2024-56609

Status: Not Affected Component: Realtek rtw88 WiFi driver (CONFIG_RTW88) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_RTW88 not compiled

In the Realtek rtw88 802.11ac/ax wireless driver (drivers/net/wireless/realtek/rtw88/tx.c), rtw_tx_report_purge_timer() at line 160 calls skb_queue_purge() at line 172 to discard queued TX-report SKBs when the firmware fails to acknowledge them. Because ieee80211_tx_status() is never called for the discarded SKBs, mac80211 retains a reference to the associated station structure after it has been freed, producing a use-after-free during driver teardown. CONFIG_RTW88 is not set in the HS 5.19.6 kernel configuration; the rtw88 driver family is absent from the compiled kernel image.

CVE-2024-56631

Status: Not exploitable Component: SCSI generic driver (CONFIG_CHR_DEV_SG) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — /dev/sg* access not in Root Lock allowlist; Lockdown blocks the exploitation trigger Affected range: Linux ≤ 6.12-rc7 Upstream fix: commit 4a9804207b58 (“scsi: sg: Fix UAF in sg_release()”)

In the SCSI generic device driver (drivers/scsi/sg.c), sg_release() at line 382 acquires sdp->open_rel_lock at line 391, then calls kref_put(&sfp->f_ref, sg_remove_sfp) at line 393. If that kref_put drops the last reference, sg_remove_sfp is invoked, which can free the Sg_device structure that sdp points to — including its embedded mutex. The subsequent mutex_unlock(&sdp->open_rel_lock) at line 404 then operates on freed memory, producing a KASAN slab-use-after-free in lock_release.

CONFIG_CHR_DEV_SG=y is compiled in. Reaching sg_release() in the race window requires an active open of a /dev/sg* device node — SCSI generic pass-through that requires CAP_SYS_RAWIO. No Root Lock deployment includes raw SCSI access in the Lockdown allowlist. Without an allowlist entry, the kernel refuses any process attempting to open /dev/sg*. An attacker who has already gained root cannot add one: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2024-56663

Status: Not exploitable Component: cfg80211 wireless stack (CONFIG_CFG80211) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — no WiFi NIC present

In net/wireless/nl80211.c, the netlink policy for NL80211_ATTR_MLO_LINK_ID at line 797 uses NLA_POLICY_RANGE(NLA_U8, 0, IEEE80211_MLD_MAX_NUM_LINKS) — where IEEE80211_MLD_MAX_NUM_LINKS = 15 (include/linux/ieee80211.h:4349). Since the range check is inclusive, link ID 15 passes validation. Structures such as cfg80211_bss size their links[] array with 15 entries (valid indices 0–14); an attacker-supplied link ID of 15 indexes one element past the end of the array, producing an out-of-bounds access. CONFIG_CFG80211=y is compiled in. No WiFi network interface card is present on a server deployment; without WiFi hardware, no wireless interfaces are created and the MLO link ID path is never reachable.

CVE-2024-57899

Status: Not Affected Component: mac80211 wireless stack (CONFIG_MAC80211) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — 32-bit-specific vulnerability; Root Lock kernel is x86_64

In the mac80211 wireless stack, a type-size mismatch between unsigned long (4 bytes on 32-bit) and u64 (8 bytes) causes incorrect arithmetic or storage on 32-bit architectures. On x86_64, sizeof(unsigned long) == sizeof(u64) == 8; the size mismatch condition cannot arise. CONFIG_X86_64=y in the HS 5.19.6 configuration; additionally, no WiFi hardware is present on a server deployment.

CVE-2025-21863

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: io_uring (CONFIG_IO_URING) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: Linux ≤ 6.13-rc6 Upstream fix: commit 838154be1ea7 (“io_uring: sanitise sqe->opcode against speculations”)

What this means for an attacker:

In io_uring/io_uring.c, io_init_req() reads sqe->opcode from userspace and checks it against IORING_OP_LAST at line 8385. Without a Spectre v1 barrier, the CPU’s speculative execution engine can index into io_op_defs[] at line 8389 before the bounds-check branch resolves, enabling a microarchitectural side-channel read of kernel memory at speculative offsets. The upstream fix inserts array_index_nospec(opcode, IORING_OP_LAST) before the array access.

Why the score is not 0.0:

CONFIG_IO_URING=y is compiled in on 5.19.6 and on the 6.18.9-hs kernel that ships. The upstream fix is already in 6.18.9-hs. On 5.19.6, reaching the vulnerable io_uring path requires a process to submit crafted SQEs via io_uring_enter(); this is a normal operation for any application using io_uring. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2023-52930

Status: Not exploitable Component: Intel i915 DRM driver (CONFIG_DRM_I915) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no Intel display GPU present

In drivers/gpu/drm/i915/gem/i915_gem_tiling.c, i915_gem_object_set_tiling() releases the gem object lock at line 308, then performs an unguarded check-and-free of obj->bit_17 at lines 314–322. Two threads concurrently calling I915_GEM_SET_TILING to set tiling to I915_TILING_NONE can both enter the else branch at line 319 and both call bitmap_free(obj->bit_17) at line 320, producing a double-free. Conversely, two threads setting a swizzled tiling mode can both pass the !obj->bit_17 check at line 315 and both call bitmap_zalloc, leaking the first allocation. CONFIG_DRM_I915=y is compiled in. No Intel integrated or discrete display GPU is present on this server deployment; DRM device nodes are not created and the GEM ioctl path is unreachable.

CVE-2023-52988

Status: Not exploitable Component: Intel HDA audio driver (CONFIG_SND_HDA_INTEL) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no audio hardware present

In sound/pci/hda/patch_via.c, via_auto_init_analog_input() calls snd_hda_get_connections() at line 820 and stores the return value in nums. The function can return a negative error code. The subsequent loop at line 822 (for (i = 0; i < nums; i++)) is a no-op for negative nums, but the conn[nums++] write at line 832 then indexes the conn[] array at a negative offset, producing an out-of-bounds write. CONFIG_SND_HDA_INTEL=y is compiled in. No audio hardware is present on a headless server deployment; HDA codec probing never runs and the vulnerable path is never reached.

CVE-2025-21993

Status: Not Affected — CONFIG_ISCSI_IBFT not set Component: iSCSI iBFT driver (CONFIG_ISCSI_IBFT) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — CONFIG_ISCSI_IBFT not compiled in Root Lock kernel

In the iSCSI Boot Firmware Table (iBFT) kernel driver, the subnet-mask field read from /sys/firmware/ibft/ethernetX/subnet-mask during an IPv6 iSCSI boot contains a memory safety issue. CONFIG_ISCSI_IBFT is not set in the HS 5.19.6 kernel configuration; the iBFT sysfs interface is absent from the compiled kernel image.

CVE-2025-22083

Status: Not Affected Component: vhost-SCSI driver (CONFIG_VHOST_SCSI) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_VHOST_SCSI not compiled

In drivers/vhost/scsi.c, vhost_scsi_set_endpoint() at line 1531 does not guard against being called multiple times without an intervening vhost_scsi_clear_endpoint(). Duplicate invocations corrupt the vs_tpg pointer array and reference counts, triggering use-after-free and null-pointer conditions. CONFIG_VHOST_SCSI is not set in the HS 5.19.6 kernel configuration; the vhost-SCSI virtualisation driver is absent from the compiled kernel image.

CVE-2025-22121

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 7.1 HIGH — base I:N; Lockdown limits post-exploitation persistence Affected range: Linux ≤ 6.13-rc3 Upstream fix: commit 34f96e89f84c (“ext4: fix UAF in ext4_xattr_inode_dec_ref_all()”)

What this means for an attacker:

In fs/ext4/xattr.c, ext4_xattr_inode_dec_ref_all() at line 1127 iterates over xattr entries, calling ext4_xattr_inode_iget() at line 1148 to obtain each ea_inode. If ext4_expand_inode_array() at line 1154 fails, iput(ea_inode) at line 1158 frees the inode. When the journal restart function (ext4_xattr_restart_fn) subsequently runs, it can re-encounter the same entry and dereference the freed inode at line 1182 (ext4_xattr_inode_dec_ref), producing a use-after-free. The published vector is C:H/I:N/A:H — disclosure and crash, not direct privilege escalation.

Why the score is not 0.0:

CONFIG_EXT4_FS=y is compiled in and 5.19.6 falls within the affected range. Reaching the xattr teardown path requires a process to manipulate extended attributes on an ext4 filesystem — a standard operation available to any user with file access. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

The attacker cannot turn this UAF into anything that runs new code. Even if a follow-on memory-corruption bug is chained in to escalate to root, Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2025-37785

Status: Not exploitable Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — mount() blocked by Lockdown; crafted ext4 image cannot be mounted Affected range: Linux ≤ 6.14-rc4 Upstream fix: commit 4f45d4452e6b (“ext4: fix OOB read when mounting corrupted fs”)

In fs/ext4/dir.c, when a corrupted ext4 directory block contains a '.' entry whose rec_len equals the filesystem block size, the iteration offset at line 246 jumps to exactly block_size after the first entry. During directory removal, a subsequent traversal computes a de pointer one block past the buffer boundary, producing an out-of-bounds read.

CONFIG_EXT4_FS=y is compiled in and 5.19.6 falls within the affected range. Triggering the out-of-bounds read requires mounting an ext4 filesystem image containing a corrupted directory block. sys_hs_lockdown_hs() sets HS_lockdown_state = 7, blocking all mount paths at kernel/namespace.c:4218, 4300, 4453 with EPERM; do_mount() returns EPERM before any ext4 directory parsing code is reached. In Lockdown, no approved program in the Root Lock allowlist carries a mount entry — the kernel SPF gate enforces this independently of Lockdown. The trigger cannot be reached on any Root Lock deployment.

CVE-2025-40364

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: io_uring (CONFIG_IO_URING) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: Linux ≤ 6.14-rc5 Upstream fix: commit 0f2122045b94 (“io_uring: don’t import buffers for async preparation”)

What this means for an attacker:

In io_uring/io_uring.c, io_req_prep_async() at line 7829 prepares an asynchronous copy of a request’s state. For requests using provided buffers (IOSQE_BUFFER_SELECT), the function can select and consume a buffer slot during the async preparation phase. If the ring state is then discarded before the I/O completes — for example, when the async path is abandoned and the request is retried — the buffer slot is consumed but the reference is lost, allowing the slot to be selected again by a subsequent request and producing a use-after-free of the shared buffer metadata.

Why the score is not 0.0:

CONFIG_IO_URING=y is compiled in on 5.19.6 and on the 6.18.9-hs kernel that ships. The upstream fix is already in 6.18.9-hs. On 5.19.6, reaching the provided-buffer UAF path requires a process to submit io_uring SQEs with IOSQE_BUFFER_SELECT in a pattern where the async preparation phase selects a buffer slot before the request is discarded. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2025-37738

Status: Not exploitable Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — mount() blocked by Lockdown; crafted xattr image cannot be mounted Affected range: Linux ≤ 6.13-rc3 Upstream fix: commit b631e432b12d (“ext4: fix xattr inode dec ref boundary”)

In fs/ext4/xattr.c, ext4_xattr_inode_dec_ref_all() at line 1143 iterates xattr entries with for (entry = first; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)). The loop has no upper-boundary parameter: it relies solely on the IS_LAST_ENTRY() zero-terminator sentinel. A corrupted xattr block without a valid terminating entry causes the loop to walk past the end of the allocated buffer, reading and dereferencing arbitrary memory.

CONFIG_EXT4_FS=y is compiled in and 5.19.6 falls within the affected range. Triggering the unbounded xattr loop requires mounting a filesystem with a corrupted xattr block that lacks the valid zero-terminator sentinel. sys_hs_lockdown_hs() sets HS_lockdown_state = 7, blocking all mount paths at kernel/namespace.c:4218, 4300, 4453 with EPERM; do_mount() returns EPERM before any ext4 xattr parsing code is reached. In Lockdown, no approved program in the Root Lock allowlist carries a mount entry — the kernel SPF gate enforces this independently of Lockdown. The trigger cannot be reached on any Root Lock deployment.

CVE-2022-49789

Status: Not Affected Component: IBM Z Fibre Channel driver (CONFIG_ZFCP) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_ZFCP not compiled

In drivers/s390/scsi/zfcp_fsf.c, zfcp_fsf_req_send() stores the FSF request ID in a variable of the wrong integer type, causing the ID to be truncated on architectures where the required width exceeds that type. CONFIG_ZFCP is not present in the HS 5.19.6 kernel configuration; the IBM Z Fibre Channel driver is s390-architecture-specific and is absent from the x86_64 compiled kernel image.

CVE-2022-49842

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no audio hardware present

In the ALSA sound subsystem, a use-after-free occurs in device_del() during driver module removal. When an ALSA driver is unloaded, a device object is freed while still referenced by a concurrent access path, producing a KASAN use-after-free report at device_del+0xb5b by the rmmod task.

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2022-49865

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: IPv6 networking stack (CONFIG_IPV6) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 7.1 HIGH — base I:N; Lockdown limits post-exploitation persistence Affected range: Linux 5.4–5.19.6 Upstream fix: kernel.org stable queue (net/ipv6/addrlabel.c)

What this means for an attacker:

In net/ipv6/addrlabel.c, ip6addrlbl_putmsg() (line 438) constructs a struct ifaddrlblmsg for a netlink reply. The function writes ifal_family, ifal_prefixlen, ifal_flags, and ifal_seq but never zeroes the __ifal_reserved padding byte. That uninitialised byte is subsequently copied to userspace via nlmsg_unicast(), leaking one byte of kernel stack memory per IPv6 address-label query.

Why the score is not 0.0:

CONFIG_IPV6=y is compiled in and 5.19.6 falls within the affected range. Any process with access to a NETLINK_ROUTE socket can trigger the infoleak — no elevated privilege is required. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

The attacker cannot turn this leak into anything that runs new code. Even if a follow-on memory-corruption bug is chained in to escalate to root, Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2023-53037

Status: Not Affected — CONFIG_SCSI_MPI3MR not set Component: Broadcom mpi3mr SAS driver (CONFIG_SCSI_MPI3MR) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_MPI3MR not compiled in Root Lock kernel

When the SAS Transport Layer support is enabled and a device exposed to the OS by the driver fails INQUIRY commands, the mpi3mr driver frees the memory allocated for an internal device handle but continues to reference that handle in subsequent SCSI transport operations, causing a use-after-free.

CONFIG_SCSI_MPI3MR is not set in the HS 5.19.6 configuration. The Broadcom mpi3mr SAS 3.0 HBA driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2023-53039

Status: Not Affected Component: Intel ISH HID driver (CONFIG_INTEL_ISH_HID) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_INTEL_ISH_HID not compiled

When a reset notify IPC message is received by the Intel Integrated Sensor Hub Transfer Protocol (ISHTP) subsystem, the ISR schedules a work item and passes the device struct via the global ishtp_dev pointer. A race between the reset notify path and device teardown can leave ishtp_dev pointing to a freed object, triggering a use-after-free.

CONFIG_INTEL_ISH_HID is not set in the HS 5.19.6 configuration. The Intel ISH HID driver (drivers/hid/intel-ish-hid/) is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2023-53065

Status: Not exploitable Component: perf events subsystem (CONFIG_PERF_EVENTS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — perf_event_paranoid=3 restricts perf_event_open(); no profiling tool in Root Lock allowlist

In kernel/events/core.c, a stack-out-of-bounds issue discovered by syzkaller occurs in the perf events sample output path. A crafted perf_event_open() call with specific sample type flags causes the kernel to write beyond the bounds of a stack-allocated buffer during event sampling, overwriting adjacent stack memory.

CONFIG_PERF_EVENTS=y is compiled in. The Root Lock kernel sets /proc/sys/kernel/perf_event_paranoid=3, which restricts perf_event_open() to processes with CAP_PERFMON. No profiling tool (perf, sysdig, or equivalent) is included in the HS Lockdown allowlist — the kernel refuses to execute it. The crafted perf_event_open() call required to trigger the stack overflow is unreachable in a standard HS deployment.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2025-37861

Status: Not Affected — CONFIG_SCSI_MPI3MR not set Component: Broadcom mpi3mr SAS driver (CONFIG_SCSI_MPI3MR) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_MPI3MR not compiled in Root Lock kernel

When the task management thread processes reply queues while the reset thread simultaneously resets them, the task management thread accesses an invalid queue ID (0xFFFF) — a sentinel value indicating a torn-down queue — resulting in an out-of-bounds access during the concurrent reset operation.

CONFIG_SCSI_MPI3MR is not set in the HS 5.19.6 configuration. The Broadcom mpi3mr SAS 3.0 HBA driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2025-37979

Status: Not Affected — CONFIG_SND_SOC_SC7280 not compiled Component: Qualcomm sc7280 ASoC driver (CONFIG_SND_SOC_SC7280) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SND_SOC_SC7280 not compiled in Root Lock kernel

Commit 5f78e1fb7a3e (“ASoC: qcom: Add driver support for audioreach solution”) introduced switch-case values in the Qualcomm sc7280 machine driver that index into fixed-size arrays without bounds checking, causing out-of-bounds access when unexpected codec or CPU DAI link types are encountered during probe.

CONFIG_SND_SOC_SC7280 is not set in the HS 5.19.6 configuration. This driver targets the Qualcomm sc7280 SoC, an ARM-based mobile/embedded platform. It is not selected on x86_64 server builds. The vulnerable code path does not exist on this system.

CVE-2022-49934

Status: Not exploitable Component: mac80211 wireless stack (CONFIG_MAC80211) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no WiFi NIC present

In net/mac80211/scan.c, ieee80211_scan_rx() accesses scan_req->flags after a null check. A use-after-free occurs when scan completion triggers __ieee80211_scan_completed(), which frees the scan request while a concurrent ieee80211_scan_rx() call still dereferences it.

CONFIG_MAC80211=y is compiled in. No WiFi network interface card is present on a server deployment. Without WiFi hardware, mac80211 creates no wireless interfaces and the relevant code paths are never reached.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2025-38103

Status: Not exploitable Component: HID subsystem (CONFIG_HID) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — no USB HID input devices on headless server

Update struct hid_descriptor to better reflect the mandatory and optional parts of the HID Descriptor as per USB HID 1.11 specification.

CONFIG_HID=y is compiled in. No USB human interface devices (keyboard, mouse, or other HID peripherals) are connected to a headless production server. HID device paths are never instantiated, making this code path unreachable.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2025-38206

Status: Not Affected — CONFIG_EXFAT_FS not compiled Component: exFAT filesystem (CONFIG_EXFAT_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_EXFAT_FS not compiled in Root Lock kernel

In fs/exfat/nls.c, exfat_load_upcase_table() frees sbi->vol_utbl via exfat_free_upcase_table() on a checksum-mismatch error (line 706) without NULLing the pointer. If the subsequent exfat_load_default_upcase_table() call fails to allocate a replacement buffer, sbi->vol_utbl retains the stale freed pointer. A later cleanup path calling exfat_free_upcase_table() again frees the same allocation, causing a double free. The trigger is mounting a specially crafted exFAT volume.

CONFIG_EXFAT_FS is not set in the HS 5.19.6 configuration. The exFAT filesystem driver — including fs/exfat/nls.c — is not compiled into the kernel image. Note that CONFIG_FAT_FS=y (VFAT/FAT32) is compiled for EFI system partition support, but that is a separate driver with no shared code. The vulnerable code path does not exist on this system.

CVE-2025-38239

Status: Not Affected — CONFIG_MEGARAID_SAS not set Component: LSI MegaRAID SAS driver (CONFIG_MEGARAID_SAS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_MEGARAID_SAS not compiled in Root Lock kernel

On systems with DRAM interleave enabled, the MegaRAID SAS driver miscalculates the MSI-X poll queue allocation, requesting poll queues beyond the number of available vectors. This results in an out-of-bounds access during driver initialization when the hardware exposes a specific MSI-X configuration.

CONFIG_MEGARAID_SAS is not set in the HS 5.19.6 configuration. The LSI/Broadcom MegaRAID SAS controller driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2025-38249

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — no audio hardware present

In snd_usb_get_audioformat_uac3(), the length value returned from snd_usb_ctl_msg() is used directly for memory allocation without validation.

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2025-38389

Status: Not exploitable Component: Intel i915 DRM driver (CONFIG_DRM_I915) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no Intel display GPU present

On ring submission GPU platforms, unbinding the i915 driver during testing sporadically triggers a kernel warning. A GPU context or ring buffer entry is accessed after being freed during the driver teardown path, detected by the kernel’s warning infrastructure during CI unbind tests.

CONFIG_DRM_I915=y is compiled in. No Intel integrated or discrete display GPU is present on this server deployment. Without display hardware, DRM device nodes may not be created and the GPU context entry points are unreachable. This follows the established pattern for i915 CVEs — see CVE-2022-4139.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2025-38494

Status: Not exploitable Component: HID subsystem (CONFIG_HID) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no USB HID input devices on headless server

hid_hw_raw_request() is actually useful to ensure the provided buffer and length are valid.

CONFIG_HID=y is compiled in. No USB human interface devices (keyboard, mouse, or other HID peripherals) are connected to a headless production server. HID device paths are never instantiated, making this code path unreachable.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2025-38550

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: IPv6 networking stack (CONFIG_IPV6) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low

Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: net/ipv6/mcast.c

What this means for an attacker:

In net/ipv6/mcast.c, mld_clear_delrec() releases the pmc->idev reference before calling ip6_mc_clear_src(), but ip6_mc_clear_src() accesses pmc->idev internally. The reference drop must be deferred until after ip6_mc_clear_src() returns; releasing it early causes a use-after-free when ip6_mc_clear_src() subsequently dereferences the freed pointer.

Why the score is not 0.0:

CONFIG_IPV6=y is compiled in and the IPv6 stack is active on configured interfaces. IPv6 multicast listener discovery (MLD) is reachable via network interfaces that join multicast groups — a common configuration on servers. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root. An attacker cannot execute a new exploit program — it has no allowlist entry and the kernel refuses to run it.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2025-38556

Status: Not exploitable Component: HID subsystem (CONFIG_HID) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — no USB HID input devices on headless server

Testing by the syzbot fuzzer showed that the HID core gets a shift-out-of-bounds exception when it tries to convert a 32-bit quantity to a 0-bit quantity.

CONFIG_HID=y is compiled in. No USB human interface devices (keyboard, mouse, or other HID peripherals) are connected to a headless production server. HID device paths are never instantiated, making this code path unreachable.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2025-38563

Status: Not exploitable Component: perf events subsystem (CONFIG_PERF_EVENTS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — perf_event_paranoid=3 restricts perf_event_open(); no profiling tool in Root Lock allowlist

The perf mmap code is careful about mmap()‘ing the user page with the ringbuffer and additionally the auxiliary buffer, when the event supports it.

CONFIG_PERF_EVENTS=y is compiled in and 5.19.6 falls within the affected range. On a Root Lock system, perf_event_paranoid=3 restricts perf_event_open() to processes with CAP_SYS_ADMIN; no profiling or performance analysis tool appears in the Root Lock allowlist. The exploitation path — loading and executing a non-allowlisted program — is blocked at the kernel execution gate before any perf subsystem interaction is possible. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2025-38565

Status: Not exploitable Component: perf events subsystem (CONFIG_PERF_EVENTS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — perf_event_paranoid=3 restricts perf_event_open(); no profiling tool in Root Lock allowlist

When perf_mmap() fails to allocate a buffer, it still invokes the event_mapped() callback of the related event.

CONFIG_PERF_EVENTS=y is compiled in and 5.19.6 falls within the affected range. On a Root Lock system, perf_event_paranoid=3 restricts perf_event_open() to processes with CAP_SYS_ADMIN; no profiling or performance analysis tool appears in the Root Lock allowlist. The exploitation path — loading and executing a non-allowlisted program — is blocked at the kernel execution gate before any perf subsystem interaction is possible. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2025-38572

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: IPv6 networking stack (CONFIG_IPV6) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low

Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: net/ipv6/

What this means for an attacker:

syzbot demonstrated that a crafted IPv6 packet with excessively long chained extension headers causes skb->transport_header to overflow. The field is a __u16; when the cumulative extension header length wraps past 65535, the kernel misidentifies the transport layer offset when parsing subsequent headers, potentially accessing incorrect memory.

Why the score is not 0.0:

CONFIG_IPV6=y is compiled in and the IPv6 stack processes all inbound IPv6 packets, including those with extension headers. This path is reachable from the network without requiring a local process. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root. An attacker cannot execute a new exploit program to escalate further — it has no allowlist entry and the kernel refuses to run it.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2025-38699

Status: Not Affected — CONFIG_SCSI_BFA_FC not compiled Component: Brocade bfa FC driver (CONFIG_SCSI_BFA_FC) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_BFA_FC not compiled in Root Lock kernel

When the bfad_im_probe() function fails during initialization, the memory pointed to by bfad->im is freed without setting bfad->im to NULL.

CONFIG_SCSI_BFA_FC is not set in the HS 5.19.6 configuration. The Brocade bfa Fibre Channel HBA driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2025-38729

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no audio hardware present

UAC3 power domain descriptors need to be verified with its variable bLength for avoiding the unexpected OOB accesses by malicious firmware, too.

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2025-39702

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: IPv6 networking stack (CONFIG_IPV6) Base Score: 7.0 HIGH (AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 6.5 MEDIUM — Lockdown reduces MI: High→Low; AC:H reduces exploitability (Exp=1.05 vs 1.83 for AC:L)

Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: net/ipv6/

What this means for an attacker:

In net/ipv6/, a Message Authentication Code comparison used a variable-time function rather than a constant-time one (such as crypto_memneq()). An attacker who can observe response timing can iteratively determine whether partial MAC bytes are correct, eventually recovering a valid MAC and bypassing authentication in IPv6 protocol handling.

Why the score is not 0.0:

CONFIG_IPV6=y is compiled in and 5.19.6 falls within the affected range. Exploiting a timing side-channel requires high network precision and repeated measurements (AC:H), which significantly reduces practical exploitability. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root. An attacker cannot execute a new exploit program to follow up on a bypassed MAC check — it has no allowlist entry and the kernel refuses to run it.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2025-39757

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — no audio hardware present

UAC3 class segment descriptors need to be verified whether their sizes match with the declared lengths and whether they fit with the allocated buffer sizes, too.

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2025-39760

Status: Not exploitable Component: USB core (CONFIG_USB) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — no USB device on headless HS server; descriptor parsing path unreachable

usb_parse_ss_endpoint_companion() checks descriptor type before length, enabling a potentially odd read outside of the buffer size.

CONFIG_USB=y is compiled in and 5.19.6 falls within the affected range. The usb_parse_ss_endpoint_companion() descriptor parsing path is triggered during USB device enumeration when a device is connected. Root Lock runs on headless server hardware with no external USB devices; no USB device enumeration occurs, so the vulnerable descriptor parsing code path is never reached. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2025-39788

Status: Not exploitable Component: SCSI subsystem (CONFIG_SCSI) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — UFS flash storage absent on x86 server

On Google gs101, the number of UTP transfer request slots (nutrs) is 32, and in this case the driver ends up programming the UTRL_NEXUS_TYPE incorrectly as 0.

CONFIG_SCSI=y is compiled in. UFS (Universal Flash Storage) is used in mobile and embedded platforms. This bug is in the Samsung Exynos UFS variant (ufs-exynos). A Debian 11 x86 server has no UFS hardware; the driver is never active.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2022-50306

Status: Not exploitable Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — mount() blocked by Lockdown; do_mount() returns EPERM

Affected range: Linux 5.10+; 5.19.6 falls within range
Upstream fix: fs/ext4/fast_commit.c

In fs/ext4/fast_commit.c, the fast commit replay scan loop reads the tag-length header (struct ext4_fc_tl, 4 bytes) before verifying that at least 4 bytes remain in the replay buffer. Mounting a filesystem whose fast commit area has been truncated or crafted to place fewer than 4 bytes at the tail causes an out-of-bounds read when parsing the next tag.

CONFIG_EXT4_FS=y is compiled in and 5.19.6 falls within the affected range. The vulnerable path runs during the fast commit replay scan triggered on mount of a filesystem whose fast commit area has a malformed tag-length header. On a Root Lock system, sys_hs_lockdown_hs() blocks all mount paths at kernel/namespace.c:4218, 4300, 4453; do_mount() returns EPERM before any filesystem setup begins. No approved process in the Root Lock allowlist carries a mount allowlist entry, and unapproved programs are refused execution by the kernel’s SPF gate regardless of file ownership or privilege. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2023-53257

Status: Not exploitable Component: mac80211 wireless stack (CONFIG_MAC80211) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no WiFi NIC present

Before checking the action code, check that it even exists in the frame.

CONFIG_MAC80211=y is compiled in. No WiFi network interface card is present on a server deployment. Without WiFi hardware, mac80211 creates no wireless interfaces and the relevant code paths are never reached.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2023-53282

Status: Not Affected — CONFIG_SCSI_LPFC not compiled Component: Emulex lpfc FC driver (CONFIG_SCSI_LPFC) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_LPFC not compiled in Root Lock kernel

In drivers/scsi/lpfc/, lpfc_wr_object() performs a use-after-free read during the sysfs firmware write process. KFENCE detects that a firmware object buffer is read after being freed during the firmware update write sequence.

CONFIG_SCSI_LPFC is not set in the HS 5.19.6 configuration. The Emulex lpfc Fibre Channel HBA driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2023-53285

Status: Not exploitable Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — direct block device write requires CAP_SYS_RAWIO; no raw-device write tool in Root Lock allowlist

Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: fs/ext4/inode.c

ext4 validates i_extra_isize when an inode is first loaded into memory (fs/ext4/inode.c:4794), confirming that the extra space falls within the inode’s allocated size. If an attacker writes directly to the block device while the filesystem is mounted, the raw on-disk inode can be modified so that i_extra_isize exceeds the previously verified bound. Subsequent access to in-inode extended attributes computes the xattr magic pointer as EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize without re-validating the updated value, allowing a read or write beyond the end of the inode body.

CONFIG_EXT4_FS=y is compiled in and 5.19.6 falls within the affected range. Exploiting this bug requires writing directly to the block device while the filesystem is mounted — an operation that requires root or CAP_SYS_RAWIO and a tool that issues raw writes to the block device (e.g., dd, badblocks, or a custom exploit program). On a Root Lock system, no approved process in the Root Lock allowlist writes raw block device data; the SPF allowlist blocks execution of any unapproved program at the kernel gate before the block device can be reached. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2023-53320

Status: Not Affected — CONFIG_SCSI_MPI3MR not set Component: Broadcom mpi3mr SAS driver (CONFIG_SCSI_MPI3MR) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_MPI3MR not compiled in Root Lock kernel

In the mpi3mr driver, mpi3mr_get_all_tgt_info() has multiple issues in its device map handling: the function miscalculates the valid entry length in alltgt_info by incorrectly sizing the struct mpi3mr_device_map_info header, leading to out-of-bounds reads when iterating target device entries.

CONFIG_SCSI_MPI3MR is not set in the HS 5.19.6 configuration. The Broadcom mpi3mr SAS 3.0 HBA driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2023-53321

Status: Not exploitable Component: mac80211 wireless stack (CONFIG_MAC80211) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — no WiFi NIC present

In net/mac80211/, control frames such as ACK frames that legally omit Address 2 and Address 3 are forwarded through wmediumd or similar userspace interfaces. The mac80211 frame parser does not enforce the full 3-address format before forwarding, potentially causing out-of-bounds reads in userspace frame consumers that assume the standard frame layout.

CONFIG_MAC80211=y is compiled in. No WiFi network interface card is present on a server deployment. Without WiFi hardware, mac80211 creates no wireless interfaces and the relevant code paths are never reached.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2023-53322

Status: Not Affected — CONFIG_SCSI_QLA_FC not compiled Component: QLogic qla2xxx FC driver (CONFIG_SCSI_QLA_FC) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_QLA_FC not compiled in Root Lock kernel

System crash due to use after free. Current code allows terminate_rport_io to exit before making sure all IOs has returned.

CONFIG_SCSI_QLA_FC is not set in the HS 5.19.6 configuration. The QLogic qla2xxx Fibre Channel HBA driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2022-50378

Status: Not exploitable Component: DRM subsystem (CONFIG_DRM) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — Amlogic Meson ARM SoC GPU absent

In drivers/gpu/drm/meson/, unloading the Amlogic Meson display driver triggers a KASAN use-after-free. During driver teardown, a resource allocated during probe is accessed after the teardown path has freed it, producing a KASAN warning at module unload time.

CONFIG_DRM=y is compiled in. drm/meson is the display driver for Amlogic Meson SoC platforms (ARM-based embedded boards such as ODROID, Khadas, etc.). This driver and its hardware are not present on an x86 Debian 11 server.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2023-53376

Status: Not Affected — CONFIG_SCSI_MPI3MR not set Component: Broadcom mpi3mr SAS driver (CONFIG_SCSI_MPI3MR) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_MPI3MR not compiled in Root Lock kernel

To allocate bitmaps, the mpi3mr driver calculates sizes of bitmaps using byte as unit.

CONFIG_SCSI_MPI3MR is not set in the HS 5.19.6 configuration. The Broadcom mpi3mr SAS 3.0 HBA driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2023-53392

Status: Not exploitable Component: HID subsystem (CONFIG_HID) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — no USB HID input devices on headless server

In the Intel ISHTP HID driver, during a warm reset device->fw_client is set to NULL. If a bus driver is registered after this NULL assignment but before ISHTP completes re-enumeration of firmware clients, the driver dereferences the NULL fw_client pointer, triggering a kernel panic.

CONFIG_HID=y is compiled in. No USB human interface devices (keyboard, mouse, or other HID peripherals) are connected to a headless production server. HID device paths are never instantiated, making this code path unreachable.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2025-39841

Status: Not Affected — CONFIG_SCSI_LPFC not compiled Component: Emulex lpfc FC driver (CONFIG_SCSI_LPFC) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_LPFC not compiled in Root Lock kernel

Fix a use-after-free window by correcting the buffer release sequence in the deferred receive path.

CONFIG_SCSI_LPFC is not set in the HS 5.19.6 configuration. The Emulex lpfc Fibre Channel HBA driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2025-39864

Status: Not exploitable Component: cfg80211 wireless framework (CONFIG_CFG80211) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no WiFi NIC present

In net/wireless/scan.c, cfg80211_update_known_bss() frees the last beacon frame of a BSS entry under conditions related to hidden SSID tracking (commit 776b3580178f). A race condition allows this beacon frame to be freed while still referenced by another code path, causing a use-after-free.

CONFIG_CFG80211=y is compiled in. No WiFi network interface card is present on a server deployment. cfg80211 manages wireless interfaces; without hardware, no interface is created and the affected code paths are unreachable.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2025-39866

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: VFS writeback subsystem Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low

Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: fs/fs-writeback.c

What this means for an attacker:

In fs/fs-writeback.c, __mark_inode_dirty() acquires a reference to a bdi_writeback structure. A concurrent bdi_writeback_switch() can free the structure before the reference is dropped, resulting in a use-after-free when the writeback pointer is subsequently accessed.

Why the score is not 0.0:

fs/fs-writeback.c is always compiled in on a system with block device support. The writeback subsystem is active for all block I/O on any mounted filesystem. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root. An attacker cannot execute a new exploit program — it has no allowlist entry and the kernel refuses to run it.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2022-50422

Status: Not Affected — CONFIG_SCSI_SAS_LIBSAS not set Component: SAS libsas library (CONFIG_SCSI_SAS_LIBSAS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_SAS_LIBSAS not compiled in Root Lock kernel

When executing SMP task failed, the smp_execute_task_sg() calls del_timer() to delete “slow_task->timer”.

CONFIG_SCSI_SAS_LIBSAS is not set in the HS 5.19.6 configuration. The SAS libsas library — used by SAS host bus adapter drivers — is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2022-50432

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: kernfs subsystem (CONFIG_KERNFS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low

Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: fs/kernfs/dir.c

What this means for an attacker:

Syzkaller triggered concurrent calls to kernfs_remove_by_name_ns() for the same kernfs node, resulting in a KASAN-detected use-after-free in fs/kernfs/dir.c. The race occurs because kernfs_remove_by_name_ns() does not prevent concurrent removals of the same node from two threads.

Why the score is not 0.0:

CONFIG_KERNFS=y is compiled in and 5.19.6 falls within the affected range. kernfs underpins sysfs and is active on every running Linux system. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root. An attacker cannot execute a new exploit program to trigger this path — it has no allowlist entry and the kernel refuses to run it.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2023-53473

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low

Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: fs/ext4/hash.c

What this means for an attacker:

In fs/ext4/hash.c, __ext4fs_dirhash() returns -1 in two cases: when a directory uses the DX_HASH_SIPHASH algorithm but the inode lacks an encryption key (line 271: “Siphash requires key”), and on an unknown hash version (line 280). Callers of ext4fs_dirhash() did not consistently check for this error and proceeded with a stale or zero hinfo->hash, potentially corrupting hash-tree directory lookups or writes.

Why the score is not 0.0:

CONFIG_EXT4_FS=y is compiled in and 5.19.6 falls within the affected range. ext4 is the primary filesystem on a Debian 11 server and directory lookups occur during normal operation. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root. An attacker cannot execute a new exploit program to trigger this path — it has no allowlist entry and the kernel refuses to run it.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2023-53510

Status: Not exploitable Component: SCSI subsystem (CONFIG_SCSI) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — UFS flash storage absent on x86 server

ufshcd_queuecommand() may be called two times in a row for a SCSI command before it is completed.

CONFIG_SCSI=y is compiled in. UFS (Universal Flash Storage) is mobile/embedded storage. The ufshcd core driver is compiled in but never instantiated on an x86 server; no UFS host controller hardware is present.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2023-53521

Status: Not Affected — CONFIG_ENCLOSURE_SERVICES not set Component: SCSI Enclosure Services (CONFIG_ENCLOSURE_SERVICES) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — CONFIG_ENCLOSURE_SERVICES not compiled in Root Lock kernel

In drivers/scsi/ses.c, ses_intf_remove() performs an out-of-bounds slab read when removing a SCSI Enclosure Services device. At ses_intf_remove+0x23f, a buffer access reads beyond its allocated boundary, as reported by KASAN during module removal by the rmmod task.

CONFIG_ENCLOSURE_SERVICES is not set in the HS 5.19.6 configuration. The SCSI Enclosure Services driver (ses) — and its dependence on SAS HBA infrastructure — is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2022-50488

Status: Not Affected Component: BFQ I/O scheduler (CONFIG_IOSCHED_BFQ) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_IOSCHED_BFQ not compiled in Root Lock kernel

In block/bfq-iosched.c, a use-after-free occurs in bfq_select_queue() involving bfqq->bic. A BFQ I/O queue object is freed while a reference to its bic (BFQ I/O context) is still live, leading to a use-after-free when bfq_select_queue() subsequently accesses the freed bfqq pointer.

CONFIG_IOSCHED_BFQ is not set in the HS 5.19.6 configuration. The BFQ (Budget Fair Queueing) block I/O scheduler is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2022-50496

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: device mapper (CONFIG_BLK_DEV_DM) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low

Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: drivers/md/dm-cache-target.c

What this means for an attacker:

In drivers/md/dm-cache-target.c, cache_resume() (line 2971) calls allow_background_work(), which schedules work on cache->wq. If cache_dtr() runs concurrently, destroy() (line 1881) frees cache->wq at line 1891 while those work items are still active, resulting in a use-after-free.

Why the score is not 0.0:

CONFIG_BLK_DEV_DM=y is compiled in and device mapper is used for LVM on a standard Debian 11 installation. Triggering this race requires concurrent resume and destroy operations on a device mapper target. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root. An attacker cannot execute a new exploit program to set up this race — it has no allowlist entry and the kernel refuses to run it.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2022-50546

Status: Not in this kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — the introducing commit is not in 5.19.6, and 6.18.9-hs contains the fix

Affected range: Introducing commit absent from 5.19.6. Fixed in 6.1.4, which 6.18.9-hs contains. Upstream fix: fs/ext4/inode.c

What this means for an attacker:

In ext4_evict_inode() (fs/ext4/inode.c:180), the function checks EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL to determine whether the inode being evicted is an extended attribute inode. Under certain error paths during inode allocation, the ext4-specific i_flags field in ext4_inode_info is not fully initialized before the inode reaches eviction, causing the flag test to read from uninitialized memory. KMSAN reported the uninitialized-value access at this check.

Why the score is 0.0:

The introducing commit is not in 5.19.6. The fix is in 6.1.4, and 6.18.9-hs contains that fix. CONFIG_EXT4_FS=y does not place this bug in either kernel.

What this means for you as an HS user:

5.19.6 predates this bug, and 6.18.9-hs contains the fix. The backstop on How to read the backstop sections applies to rows that are in the kernel.

CVE-2023-53640

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no audio hardware present

In the ALSA sound subsystem, regcache_flat_read() performs a slab-out-of-bounds read. syzkaller reproduced a KASAN report showing an out-of-bounds read in the flat register cache read path, triggered through the ALSA register map interface.

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2023-53675

Status: Not Affected — CONFIG_ENCLOSURE_SERVICES not set Component: SCSI Enclosure Services (CONFIG_ENCLOSURE_SERVICES) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — CONFIG_ENCLOSURE_SERVICES not compiled in Root Lock kernel

Sanitize possible desc_ptr out-of-bounds accesses in ses_enclosure_data_process().

CONFIG_ENCLOSURE_SERVICES is not set in the HS 5.19.6 configuration. The SCSI Enclosure Services driver (ses) is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2023-53676

Status: Not Affected — CONFIG_ISCSI_TARGET not compiled Component: Linux iSCSI target (CONFIG_ISCSI_TARGET) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_ISCSI_TARGET not compiled in Root Lock kernel

In drivers/target/iscsi/, lio_target_nacl_info_show() uses sprintf() in a loop to print details for every iSCSI connection in a session without checking that the output buffer has sufficient remaining space, leading to a buffer overflow when a session contains many connections.

CONFIG_ISCSI_TARGET is not set in the HS 5.19.6 configuration. The Linux iSCSI target (drivers/target/iscsi/) subsystem is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2025-71075

Status: Not Affected — CONFIG_SCSI_AIC94XX not set Component: Adaptec aic94xx SAS driver (CONFIG_SCSI_AIC94XX) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_AIC94XX not compiled in Root Lock kernel

The asd_pci_remove() function fails to synchronize with pending tasklets before freeing the asd_ha structure, leading to a potential use-after-free vulnerability.

CONFIG_SCSI_AIC94XX is not set in the HS 5.19.6 configuration. The Adaptec aic94xx SAS HBA driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2026-23076

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — no audio hardware present

In the ALSA ctxfi audio driver’s mixer handling code, the conf field is used as a loop index and referenced in the index callbacks amixer_index() and sum_index(). Without a bounds check on conf, these callbacks can access mixer entries outside the allocated range, leading to an out-of-bounds read.

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2026-23078

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no audio hardware present

The scarlett2_usb_get_config() function has a logic error in the endianness conversion code that can cause buffer overflows when count > 1.

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2026-23089

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no audio hardware present

When snd_usb_create_mixer() fails, snd_usb_mixer_free() frees mixer->id_elems but the controls already added to the card still reference the freed memory.

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2026-23191

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no audio hardware present

The PCM trigger callback of aloop driver tries to check the PCM state and stop the stream of the tied substream in the corresponding cable.

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2026-23193

Status: Not Affected — CONFIG_ISCSI_TARGET not compiled Component: Linux iSCSI target (CONFIG_ISCSI_TARGET) Base Score: 8.8 HIGH (AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_ISCSI_TARGET not compiled in Root Lock kernel

In iscsit_dec_session_usage_count(), the function calls complete() while holding the sess->session_usage_lock.

CONFIG_ISCSI_TARGET is not set in the HS 5.19.6 configuration. The Linux iSCSI target (drivers/target/iscsi/) subsystem is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2026-23208

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no audio hardware present

In this case, the user constructed the parameters with maxpacksize 40 for rate 22050 / pps 1000, and packsize[0] 22 packsize[1] 23.

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2026-23216

Status: Not Affected — CONFIG_ISCSI_TARGET not compiled Component: Linux iSCSI target (CONFIG_ISCSI_TARGET) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_ISCSI_TARGET not compiled in Root Lock kernel

In iscsit_dec_conn_usage_count(), the function calls complete() while holding the conn->conn_usage_lock.

CONFIG_ISCSI_TARGET is not set in the HS 5.19.6 configuration. The Linux iSCSI target (drivers/target/iscsi/) subsystem is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2025-71238

Status: Not Affected — CONFIG_SCSI_QLA_FC not compiled Component: QLogic qla2xxx FC driver (CONFIG_SCSI_QLA_FC) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_SCSI_QLA_FC not compiled in Root Lock kernel

In drivers/scsi/qla2xxx/, the QLogic Fibre Channel HBA driver writes to an invalid kernel address during a specific error recovery path, triggering a page fault with a supervisor write access error. The invalid address indicates a use-after-free or uninitialized pointer dereference within the driver’s interrupt or completion handling.

CONFIG_SCSI_QLA_FC is not set in the HS 5.19.6 configuration. The QLogic qla2xxx Fibre Channel HBA driver is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2026-23318

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — no audio hardware present

The entry of the validators table for UAC3 AC header descriptor is defined with the wrong protocol version UAC_VERSION_2, while it should have been UAC_VERSION_3.

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2026-31581

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no audio hardware present

In usb6fire_chip_abort(), the chip struct is allocated as the card’s private data (via snd_card_new with sizeof(struct sfire_chip)).

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2023-3268

Status: Not exploitable Component: relay filesystem (CONFIG_RELAY) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — debugfs relay access not in Root Lock allowlist; Lockdown blocks the exploitation trigger

An out of bounds (OOB) memory access flaw was found in the Linux kernel in relay_file_read_start_pos in kernel/relay.c in the relayfs.

CONFIG_RELAY=y is compiled in. Triggering the bug requires CAP_SYS_ADMIN and read access to relay channel files under debugfs — paths used exclusively by kernel tracing tools (SystemTap, etc.) that have no place in a production server allowlist. Without an allowlist entry covering debugfs relay access, the kernel refuses it. An attacker who has already gained root cannot add one: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2023-3567

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: virtual terminal (VT) (CONFIG_VT) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 7.1 HIGH — base I:N; Lockdown limits post-exploitation persistence Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: drivers/tty/vt/vc_screen.c

What this means for an attacker:

In drivers/tty/vt/vc_screen.c, vcs_read() accesses virtual console screen data through a vc_screen reference without holding appropriate locks for the full duration of the read. A concurrent write or deallocation of the virtual console can free the underlying vc_screen structure while vcs_read() is still referencing it, causing a use-after-free. The published vector is C:H/I:N/A:H — disclosure and crash, not direct privilege escalation.

Why the score is not 0.0:

CONFIG_VT=y is compiled in and 5.19.6 falls within the affected range. Reading /dev/vcs* virtual console screen devices requires membership in the tty group on Debian. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root. Executing a non-allowlisted program requires an allowlist entry; an attacker cannot reach this code path without one.

What this means for you as an HS user:

The attacker cannot turn this UAF into anything that runs new code. Even if a follow-on memory-corruption bug is chained in to escalate to root, Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2023-6531

Status: Not in this kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: Unix domain sockets (CONFIG_UNIX) Base Score: 7.0 HIGH (AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — the introducing commit is not in 5.19.6, and 6.18.9-hs contains the fix Affected range: Introducing commit absent from 5.19.6. Fixed in 6.7-rc5, which 6.18.9-hs contains. Upstream fix: net/unix/garbage.c

What this means for an attacker:

In net/unix/garbage.c, the Unix socket garbage collector frees orphaned socket buffers (SKBs) without coordinating with concurrent unix_stream_read_generic() operations on the socket those SKBs are queued on. The race allows unix_stream_read_generic() to access an SKB that the garbage collector has already freed, causing a use-after-free. AC:H reflects that exploitation requires precise timing between the GC sweep and a concurrent stream read.

Why the score is 0.0:

The introducing commit is not in 5.19.6. The fix is in 6.7-rc5, and 6.18.9-hs contains that fix. CONFIG_UNIX=y does not place this bug in either kernel.

What this means for you as an HS user:

5.19.6 predates this bug, and 6.18.9-hs contains the fix. The backstop on How to read the backstop sections applies to rows that are in the kernel.

CVE-2023-51043

Status: Not exploitable Component: DRM subsystem (CONFIG_DRM) Base Score: 7.0 HIGH (AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no DRM/GPU device on headless server; drm_atomic requires GPU mode-setting

In the Linux kernel before 6.4.5, drivers/gpu/drm/drm_atomic.c has a use-after-free during a race condition between a nonblocking atomic commit and a driver unload.

CONFIG_DRM=y is compiled in and 5.19.6 falls within the affected range. The drm_atomic race condition requires a process to initiate GPU mode-setting operations — specifically a nonblocking atomic commit — concurrent with driver unload. Root Lock runs on headless server hardware with no display GPU; the DRM device nodes are absent, so no mode-setting operation can be initiated. No GPU or display tool appears in the Root Lock allowlist. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-0841

Status: Not exploitable Component: hugetlbfs (CONFIG_HUGETLBFS) Base Score: 6.6 MEDIUM (AV:L/AC:L/PR:L/UI:N/S:U/C:L/I:L/A:H) Score on Root Lock: 0.0 — mount() blocked by Lockdown; hugetlbfs mount path unreachable Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: fs/hugetlbfs/inode.c

In fs/hugetlbfs/inode.c, hugetlbfs_fill_super() initialises the hugetlbfs superblock for a mount(2) call. Under certain error conditions during setup — for instance, when huge page pool allocation fails — the function dereferences a pointer that was not initialised, causing a null pointer dereference. The crash is reachable by any local user with CAP_SYS_ADMIN permission to mount hugetlbfs.

CONFIG_HUGETLBFS=y is compiled in and 5.19.6 falls within the affected range. Triggering hugetlbfs_fill_super() requires calling mount(2) with hugetlbfs as the filesystem type, which additionally requires CAP_SYS_ADMIN on Debian 11. sys_hs_lockdown_hs() sets HS_lockdown_state = 7, blocking all mount paths at kernel/namespace.c:4218, 4300, 4453 with EPERM; do_mount() returns EPERM before any hugetlbfs setup begins. In Lockdown, no approved program in the Root Lock allowlist carries a mount entry — the kernel SPF gate enforces this independently of Lockdown. The trigger cannot be reached on any Root Lock deployment.

CVE-2024-26593

Status: Not exploitable Component: Intel SMBus I2C controller (CONFIG_I2C_I801) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — no I2C/SMBus tool in Root Lock allowlist; Lockdown blocks access

In drivers/i2c/busses/i2c-i801.c, the Intel I801 SMBus driver handles block process call transactions incorrectly. Intel datasheets specify that the block buffer index must be reset twice: once before writing the outgoing data to the buffer, and once before reading the incoming response. The driver resets the index only once, causing the response to be read from the wrong buffer position and potentially returning incorrect data to callers.

CONFIG_I2C_I801=y is compiled in and 5.19.6 falls within the affected range. The Intel I2C SMBus controller is present on Intel-based servers for BMC, temperature sensor, and management bus communication. Accessing it requires root or i2c group membership and an i2c-tools or lm-sensors program — no such tool appears in the Root Lock allowlist. On a Root Lock system in Lockdown, the kernel blocks any process without an allowlist entry from executing, so a standalone exploit tool cannot reach the I2C device interface. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-38586

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: Realtek r8169 Ethernet driver (CONFIG_R8169) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low

Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: drivers/net/ethernet/realtek/r8169_main.c

What this means for an attacker:

In drivers/net/ethernet/realtek/r8169_main.c, transmitting small fragmented scatter-gather packets on an RTL8125b NIC causes the driver to populate TX ring buffer descriptors with invalid state. The NIC subsequently processes the malformed descriptors, leading to incorrect DMA operations that can corrupt memory.

Why the score is not 0.0:

CONFIG_R8169=y is compiled in and 5.19.6 falls within the affected range. The r8169 driver is active on systems with a Realtek NIC and handles all network TX traffic; the faulty path is reachable through normal network operation on affected hardware. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root. An attacker cannot execute a new exploit program to trigger this path — it has no allowlist entry and the kernel refuses to run it.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-38630

Status: Not exploitable Component: watchdog timer subsystem (CONFIG_WATCHDOG) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no watchdog daemon in Root Lock allowlist; Lockdown blocks /dev/watchdog access

When the cpu5wdt module is removing, the origin code uses del_timer() to de-activate the timer.

CONFIG_WATCHDOG=y is compiled in and 5.19.6 falls within the affected range. The cpu5wdt driver targets a PC-era ISA watchdog timer; this hardware is absent on any modern HS server deployment. Even on configurations where the hardware exists, the trigger requires a process to open and interact with /dev/watchdog — no watchdog daemon appears in the Root Lock allowlist. On a Root Lock system in Lockdown, the kernel blocks any process without an allowlist entry from executing, so a standalone exploit tool cannot reach the cpu5wdt interface. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-34777

Status: Not Affected — CONFIG_DMA_MAP_BENCHMARK not compiled Component: DMA map benchmark (CONFIG_DMA_MAP_BENCHMARK) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — CONFIG_DMA_MAP_BENCHMARK not compiled in Root Lock kernel

In kernel/dma/map_benchmark.c, map_benchmark_ioctl() passes the user-supplied NUMA node ID directly to node_possible() (line 211) without first verifying that it falls within [0, MAX_NUMNODES-1]. node_possible() uses the node ID as a bitmap index; an out-of-range value causes an out-of-bounds read in the node_possible_map bitmap.

CONFIG_DMA_MAP_BENCHMARK is not set in the HS 5.19.6 configuration. The DMA mapping benchmark module is a debug/testing facility accessible via /sys/kernel/debug/dma_map_benchmark; it is not compiled into the kernel image. The vulnerable code path does not exist on this system.

CVE-2024-39463

Status: Not exploitable Component: Plan 9 filesystem (9P) (CONFIG_9P_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — mount() blocked by Lockdown; no 9P filesystem reachable on Root Lock deployments

In fs/9p/, a use-after-free occurs on a dentry’s d_fsdata fid list when one thread looks up a fid through the dentry while another thread concurrently unlinks it. The unlinking thread frees the fid while the lookup thread still holds a reference, causing the lookup to dereference freed memory.

CONFIG_9P_FS=y is compiled in. Triggering the bug requires mounting a 9P filesystem. Lockdown categorically blocks mount() — sys_hs_lockdown_hs() sets HS_lockdown_state = 7, after which all mount paths return EPERM. No Root Lock deployment has a 9P filesystem mounted before Lockdown engages at boot. The trigger cannot be reached.

The vulnerable path never opens. The bug exists in the source — not on this system.

CVE-2024-40956

Status: Not exploitable Component: DMA engine framework (CONFIG_DMA_ENGINE) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — Intel IAX/DSA accelerator hardware absent

Use list_for_each_entry_safe() to allow iterating through the list and deleting the entry in the iteration process.

CONFIG_DMA_ENGINE=y is compiled in. idxd is the driver for Intel Data Streaming Accelerator (DSA) and Intelligence Analytics Accelerator (IAX), available in Intel Sapphire Rapids and later server CPUs. These accelerators require specific Intel hardware not present on a standard Debian 11 server.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2022-48867

Status: Not exploitable Component: DMA engine framework (CONFIG_DMA_ENGINE) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — Intel IAX/DSA accelerator hardware absent

In drivers/dma/idxd/, when the Intel Data Streaming Accelerator driver is unloaded, idxd_dmaengine_drv_remove() frees the interrupt handler while descriptor completions are still pending. Completion callbacks that fire after interrupt teardown dereference the freed interrupt state, causing a use-after-free.

CONFIG_DMA_ENGINE=y is compiled in. idxd drives Intel’s Data Streaming Accelerator hardware, present only in Intel Sapphire Rapids (and later) server CPUs. This hardware is not present on a standard Debian 11 deployment.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2024-46759

Status: Not exploitable Component: hardware monitoring subsystem (CONFIG_HWMON) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — ADC128D818 I2C ADC chip absent

DIV_ROUND_CLOSEST() after kstrtol() results in an underflow if a large negative number such as -9223372036854775808 is provided by the user.

CONFIG_HWMON=y is compiled in. adc128d818 drives the Texas Instruments ADC128D818 — a specific 8-channel I2C ADC chip used on some custom boards. This chip is not part of standard server hardware; the hwmon driver is never bound.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2024-49860

Status: Not exploitable Component: ACPI subsystem (CONFIG_ACPI) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — malformed ACPI _STR firmware absent; standard OEM server firmware returns Buffer objects as specified Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: drivers/acpi/device_sysfs.c

In the ACPI subsystem, the _STR ACPI method must return a buffer object containing a Unicode description string. description_show(), exposed via sysfs at /sys/bus/acpi/devices/*/description, calls the _STR method and dereferences the result without validating that the returned object is in fact a buffer. A crafted or malformed ACPI table that returns an integer, package, or other non-buffer object from _STR causes description_show() to access invalid memory.

CONFIG_ACPI=y is compiled in and 5.19.6 falls within the affected range. ACPI tables are loaded from OEM firmware at boot and are read-only thereafter — no userspace process can modify them without firmware-level access outside the HS adversary model. Standard OEM server firmware conforms to the ACPI specification and returns a Buffer object from _STR. On a Root Lock server deployment, no malformed _STR firmware is present; the invalid-memory path in description_show() is never reached. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2022-49029

Status: Not exploitable Component: hardware monitoring subsystem (CONFIG_HWMON) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — IBM Power Management Extension hardware absent

In drivers/hwmon/ibmpex.c, ibmpex_register_bmc() at line 509 adds a BMC device entry to the global list but does not remove it from the list on the error path. If registration fails partway through, &data->list remains linked while the containing data struct is freed, leading to a use-after-free when the list is subsequently traversed.

CONFIG_HWMON=y is compiled in. ibmpex drives the IBM Power Management Extension, specific to IBM Power Systems server hardware. This is not present on an x86 Debian 11 server.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2024-50127

Status: Not exploitable Component: network traffic scheduler (CONFIG_NET_SCHED) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — tc not in Root Lock allowlist; Lockdown blocks the exploitation trigger

In net/sched/sch_taprio.c, taprio_change() holds the admin schedule pointer while a concurrent advance_sched() call can switch or remove the schedule, making admin a dangling pointer. The critical section protected by q->current_entry_lock does not prevent this race, allowing access to freed schedule memory.

CONFIG_NET_SCHED=y is compiled in. Triggering the bug requires the tc utility (iproute2) with CAP_NET_ADMIN to install or modify a qdisc or filter. No Root Lock deployment includes tc in the Lockdown allowlist — the kernel refuses to execute it. An attacker who has already gained root cannot add it: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2024-50131

Status: Not exploitable Component: kernel tracing (CONFIG_TRACING) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — tracefs not in Root Lock allowlist; Lockdown blocks the exploitation trigger

In the kernel tracing subsystem, strlen() returns the string length excluding the null terminator. If the string length equals the maximum buffer length, the buffer has no remaining space for the null byte, and the subsequent null terminator write goes one byte past the end of the buffer — a classic off-by-one overflow.

CONFIG_TRACING=y is compiled in. Triggering the bug requires CAP_SYS_ADMIN and active access to the kernel tracing filesystem at /sys/kernel/tracing/. No Root Lock deployment permits any service to write to these paths. Without an allowlist entry covering the tracing interface, the kernel refuses access. An attacker who has already gained root cannot add it: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2024-53057

Status: Not exploitable Component: network traffic scheduler (CONFIG_NET_SCHED) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — tc not in Root Lock allowlist; Lockdown blocks the exploitation trigger

In qdisc_tree_reduce_backlog, Qdiscs with major handle ffff: are assumed to be either root or ingress.

CONFIG_NET_SCHED=y is compiled in. Triggering the bug requires the tc utility (iproute2) with CAP_NET_ADMIN to install or modify a qdisc or filter. No Root Lock deployment includes tc in the Lockdown allowlist — the kernel refuses to execute it. An attacker who has already gained root cannot add it: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2024-56606

Status: Not exploitable Component: AF_PACKET sockets (CONFIG_PACKET) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CAP_NET_RAW not in Root Lock allowlist; Lockdown blocks the exploitation trigger

After sock_init_data() the allocated sk object is attached to the provided sock object.

CONFIG_PACKET=y is compiled in. Creating an AF_PACKET raw socket requires CAP_NET_RAW. No Root Lock deployment grants CAP_NET_RAW to any service — packet capture tools such as tcpdump have no allowlist entry. Without an allowlist entry, the kernel refuses to execute them. An attacker who has already gained root cannot add one: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2025-21692

Status: Not exploitable Component: network traffic scheduler (CONFIG_NET_SCHED) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — tc not in Root Lock allowlist; Lockdown blocks the exploitation trigger

Haowei Yan g1042620637@gmail.com found that ets_class_from_arg() can index an Out-Of-Bound class in ets_class_from_arg() when passed clid of 0.

CONFIG_NET_SCHED=y is compiled in. Triggering the bug requires the tc utility (iproute2) with CAP_NET_ADMIN to install or modify a qdisc or filter. No Root Lock deployment includes tc in the Lockdown allowlist — the kernel refuses to execute it. An attacker who has already gained root cannot add it: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2022-49799

Status: Not exploitable Component: kernel tracing (CONFIG_TRACING) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — tracefs not in Root Lock allowlist; Lockdown blocks the exploitation trigger

In kernel/trace/, register_synth_event() calls trace_remove_event_call() and unregister_trace_event() on the error path when set_synth_event_print_fmt() fails. Calling both functions causes the trace event to be unregistered twice, resulting in a double-free of the trace event structure.

CONFIG_TRACING=y is compiled in. Triggering the bug requires CAP_SYS_ADMIN and active access to the kernel tracing filesystem at /sys/kernel/tracing/. No Root Lock deployment permits any service to write to these paths. Without an allowlist entry covering the tracing interface, the kernel refuses access. An attacker who has already gained root cannot add it: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2022-49892

Status: Not exploitable Component: ftrace / function tracer (CONFIG_FTRACE) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — tracefs not in Root Lock allowlist; Lockdown blocks the exploitation trigger

KASAN reported a use-after-free with ftrace ops [1]. It was found from vmcore that perf had registered two ops with the same content successively, both dynamic.

CONFIG_FTRACE=y is compiled in. Triggering the bug requires CAP_SYS_ADMIN and write access to ftrace control files under /sys/kernel/tracing/. No Root Lock deployment permits any service to access these paths. Without an allowlist entry covering the ftrace interface, the kernel refuses access. An attacker who has already gained root cannot add it: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2022-49921

Status: Not exploitable Component: network traffic scheduler (CONFIG_NET_SCHED) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — tc not in Root Lock allowlist; Lockdown blocks the exploitation trigger

We can’t use “skb” again after passing it to qdisc_enqueue(). This is basically identical to commit 2f09707d0c97 (“sch_sfb: Also store skb len before calling child enqueue”).

CONFIG_NET_SCHED=y is compiled in. Triggering the bug requires the tc utility (iproute2) with CAP_NET_ADMIN to install or modify a qdisc or filter. No Root Lock deployment includes tc in the Lockdown allowlist — the kernel refuses to execute it. An attacker who has already gained root cannot add it: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2023-53111

Status: Not exploitable Component: loop block device (CONFIG_BLK_DEV_LOOP) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — /dev/loop* access not in Root Lock allowlist; Lockdown blocks the exploitation trigger

do_req_filebacked() calls blk_mq_complete_request() synchronously or asynchronously when using asynchronous I/O unless memory allocation fails.

CONFIG_BLK_DEV_LOOP=y is compiled in. Triggering the bug requires ioctl operations on /dev/loop* with CAP_SYS_ADMIN. No Root Lock production workload uses loop devices — they are absent from the Lockdown allowlist. Without an allowlist entry, the kernel refuses access. An attacker who has already gained root cannot add one: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2025-37879

Status: Not exploitable Component: Plan 9 filesystem (9P) (CONFIG_9P_FS) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — mount() blocked by Lockdown; no 9P filesystem reachable on Root Lock deployments

In net/9p/client.c, p9_client_write() and p9_client_read_once() do not validate the count returned by the 9P server. If a misbehaving server replies with success but a negative byte count, the client treats the negative value as a large unsigned integer, potentially causing integer underflow or incorrect buffer offset calculations.

CONFIG_9P_FS=y is compiled in. Triggering the bug requires mounting a 9P filesystem. Lockdown categorically blocks mount() — sys_hs_lockdown_hs() sets HS_lockdown_state = 7, after which all mount paths return EPERM. No Root Lock deployment has a 9P filesystem mounted before Lockdown engages at boot. The trigger cannot be reached.

The vulnerable path never opens. The bug exists in the source — not on this system.

CVE-2025-37914

Status: Not exploitable Component: network traffic scheduler (CONFIG_NET_SCHED) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — tc not in Root Lock allowlist; Lockdown blocks the exploitation trigger

As described in Gerrard’s report [1], there are use cases where a netem child qdisc will make the parent qdisc’s enqueue callback reentrant.

CONFIG_NET_SCHED=y is compiled in. Triggering the bug requires the tc utility (iproute2) with CAP_NET_ADMIN to install or modify a qdisc or filter. No Root Lock deployment includes tc in the Lockdown allowlist — the kernel refuses to execute it. An attacker who has already gained root cannot add it: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2025-37915

Status: Not exploitable Component: network traffic scheduler (CONFIG_NET_SCHED) Base Score: 7.0 HIGH (AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — tc not in Root Lock allowlist; Lockdown blocks the exploitation trigger

As described in Gerrard’s report [1], there are use cases where a netem child qdisc will make the parent qdisc’s enqueue callback reentrant.

CONFIG_NET_SCHED=y is compiled in. Triggering the bug requires the tc utility (iproute2) with CAP_NET_ADMIN to install or modify a qdisc or filter. No Root Lock deployment includes tc in the Lockdown allowlist — the kernel refuses to execute it. An attacker who has already gained root cannot add it: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2025-37923

Status: Not exploitable Component: kernel tracing (CONFIG_TRACING) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — tracefs not in Root Lock allowlist; Lockdown blocks the exploitation trigger

In kernel/trace/trace.c, trace_seq_to_buffer() at line 1830 performs a slab-out-of-bounds write. syzbot reproduced a KASAN report showing that a trace sequence buffer copy operation writes beyond the allocated slab boundary, reachable through the kernel tracing filesystem interface under CAP_SYS_ADMIN.

CONFIG_TRACING=y is compiled in. Triggering the bug requires CAP_SYS_ADMIN and active access to the kernel tracing filesystem at /sys/kernel/tracing/. No Root Lock deployment permits any service to write to these paths. Without an allowlist entry covering the tracing interface, the kernel refuses access. An attacker who has already gained root cannot add it: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2025-38369

Status: Not exploitable Component: DMA engine framework (CONFIG_DMA_ENGINE) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — Intel IAX/DSA accelerator hardware absent

Running IDXD workloads in a container with the /dev directory mounted can trigger a call trace or even a kernel panic when the parent process exits while child processes are still using IDXD portal file descriptors. The portal file descriptor cleanup races with process exit, causing a use-after-free when the freed descriptor object is subsequently accessed.

CONFIG_DMA_ENGINE=y is compiled in. idxd drives Intel’s on-chip Data Streaming and Analytics Accelerator hardware. This requires specific Intel Sapphire Rapids or later CPU hardware not present on a standard server.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2025-38548

Status: Not exploitable Component: hardware monitoring subsystem (CONFIG_HWMON) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — Corsair Commander Pro hardware absent

Add buffer_recv_size to store the size of the received bytes. Validate buffer_recv_size in send_usb_cmd().

CONFIG_HWMON=y is compiled in. corsair-cpro drives the Corsair Commander Pro — a desktop PC fan/cooler controller connected via USB HID. This device is not present in a production server environment.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2022-50320

Status: Not exploitable Component: ACPI subsystem (CONFIG_ACPI) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — FPDT crash requires malformed firmware; not reachable on standard OEM hardware Affected range: Linux 5.x–5.19 (fix adds address validation before acpi_os_map_memory call) Upstream fix: drivers/acpi/acpi_fpdt.c (validate subtable->address before mapping)

In drivers/acpi/acpi_fpdt.c, acpi_init_fpdt() (line 253) passes FPDT subtable addresses from firmware-supplied ACPI tables directly to acpi_os_map_memory() without validating that the address falls within the physical memory range. On systems with buggy firmware (the Packard Bell Dot SC, Intel Atom N2600 being the reported case), FPDT entries contain addresses with high bits set outside the valid physical range. acpi_os_map_memory() then attempts to map non-existent memory, crashing the kernel. Any firmware that supplies a malformed FPDT triggers the same path.

CONFIG_ACPI=y is compiled in and 5.19.6 falls within the affected range. FPDT parsing runs at fs_initcall priority — early boot, before any user-space process is running. Triggering the invalid-address crash requires malformed FPDT entries in the system’s ACPI firmware; HeartSuite deployments use standard OEM server firmware that conforms to the ACPI specification. Injecting a crafted ACPI table requires physical or firmware-level access, which is outside the HS software-based adversary model. An adversary with firmware access has already bypassed the OS security boundary; the ACPI parsing path is therefore not a reachable software attack surface on any standard HS deployment.

CVE-2023-53395

Status: Not exploitable Component: ACPI subsystem (CONFIG_ACPI) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — AML exploit requires crafted firmware; ACPI tables read-only after boot on standard servers Affected range: Linux 5.x through affected ACPICA version Upstream fix: ACPICA commit 90310989a079 (drivers/acpi/acpica/acopcode.h)

In the ACPICA AML interpreter, the opcode table entries for the AML Timer instruction (ARGP_TIMER_OP, ARGI_TIMER_OP in drivers/acpi/acpica/acopcode.h) were inconsistent with ACPI Specification section 19.6.134, which specifies that Timer takes no arguments. The mismatch caused the AML parser to mishandle Timer opcodes in certain AML bytecode sequences, potentially treating subsequent bytecode as a spurious argument and corrupting the AML interpreter walk-state.

CONFIG_ACPI=y is compiled in and 5.19.6 falls within the affected range. AML execution runs at boot using ACPI tables supplied by the system firmware. Exploiting the walk-state corruption requires crafted AML bytecode — on a server with a reputable firmware vendor, ACPI tables are loaded from firmware storage at boot and are read-only thereafter; no userspace process can replace or modify the AML after boot without firmware-level access. This places the trigger outside the HS software-based adversary model. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2025-39869

Status: Not exploitable Component: DMA engine framework (CONFIG_DMA_ENGINE) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — Texas Instruments eDMA hardware absent

Fix a critical memory allocation bug in edma_setup_from_hw() where queue_priority_map was allocated with insufficient memory.

CONFIG_DMA_ENGINE=y is compiled in. ti-edma is the DMA controller driver for Texas Instruments Keystone/OMAP/AM embedded SoC platforms. This driver and hardware are not present on an x86 Debian 11 server.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2022-50423

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: ACPI subsystem (CONFIG_ACPI) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: Linux 5.x–5.19 Upstream fix: drivers/acpi/acpica/utdelete.c (reference count ordering fix)

What this means for an attacker:

In drivers/acpi/acpica/utdelete.c, acpi_ut_remove_reference() is called on an ACPI operand object that has already been freed by a concurrent or error-handling code path. The function reads object->common.descriptor_type (via ACPI_GET_DESCRIPTOR_TYPE, line 720) and object->common.reference_count (via acpi_ut_update_object_reference, line 740) from the already-freed memory. KASAN detects the access as a use-after-free at offset +0x3b in acpi_ut_remove_reference().

Why the score is not 0.0:

CONFIG_ACPI=y is compiled in and 5.19.6 falls within the affected range. The ACPI subsystem is active from boot; triggering this use-after-free requires manipulating the ACPI reference count lifecycle via method evaluation during device enumeration or hotplug events. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot drop and execute a new exploit program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2026-23378

Status: Not exploitable Component: network traffic scheduler (CONFIG_NET_SCHED) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — tc not in Root Lock allowlist; Lockdown blocks the exploitation trigger

Whenever an ife action replace changes the metalist, instead of replacing the old data on the metalist, the current ife code is appending the new metadata.

CONFIG_NET_SCHED=y is compiled in. Triggering the bug requires the tc utility (iproute2) with CAP_NET_ADMIN to install or modify a qdisc or filter. No Root Lock deployment includes tc in the Lockdown allowlist — the kernel refuses to execute it. An attacker who has already gained root cannot add it: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2024-36883

Status: Not exploitable Component: TCP/IP networking (CONFIG_INET) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — pernet race requires a new module load; kernel.modules_disabled=1 refuses that load after boot Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: net/core/net_namespace.c

In net/core/net_namespace.c, net_alloc_generic() reads max_gen_ptrs — the size of the generic pointers array — to determine how much memory to allocate for a new network namespace. This read occurs without holding pernet_ops_rwsem. register_pernet_operations() can increment max_gen_ptrs concurrently while holding the write side of that lock. The race can cause net_alloc_generic() to allocate an undersized array, leading to out-of-bounds access when the new namespace is subsequently populated.

CONFIG_INET=y is compiled in and 5.19.6 falls within the affected range. The race requires register_pernet_operations() to execute concurrently with net_alloc_generic(). register_pernet_operations() is invoked exclusively from module initialization (module_init routines). Early in every Root Lock kernel boot, before sshd starts, heartsuite-kernel-latch.service sets kernel.modules_disabled=1. Until the next reboot, the kernel refuses init_module and finit_module, so no new module can call register_pernet_operations(). See How to read the backstop sections. Built-in pernet operations register during kernel init, before the latch, and max_gen_ptrs is stable after that. The race cannot be triggered by loading a new module after the latch.

CVE-2024-36971

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: TCP/IP destination cache (CONFIG_INET) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: 5.19.6 falls within the affected range Upstream fix: net/core/dst.c — RCU locking in __dst_negative_advice()

What this means for an attacker:

This CVE was actively exploited in the wild (Google Threat Analysis Group, 2024). It describes a use-after-free in net/core/dst.c. __dst_negative_advice() clears sk->dst_cache when a cached destination entry is marked invalid — reading the entry, determining it should be dropped, then calling sk_dst_reset() — without proper RCU locking across this sequence. A concurrent operation can free the destination entry between the initial read and the reset, producing a use-after-free on the freed dst entry. The result is local privilege escalation to root; attack vector is local (AV:L), not remote.

Why the score is not 0.0:

CONFIG_INET=y is compiled in and 5.19.6 falls within the affected range. __dst_negative_advice() is invoked whenever a cached destination becomes invalid, reachable through normal network activity or by triggering ICMP unreachable messages from a local process. There is no hardware dependency and no special configuration required to reach the code path. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot drop and execute a new exploit program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-38577

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: RCU tasks subsystem (CONFIG_TASKS_RCU) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: kernel/rcu/tasks.h

What this means for an attacker:

In kernel/rcu/tasks.h, show_rcu_tasks_trace_gp_kthread() formats diagnostic counters for the RCU tasks trace grace-period kthread into a fixed-size buffer using sprintf(). The function does not bound the number of characters written; if individual counter values are sufficiently large, the formatted output overflows the buffer. The sysfs interface exposing this data is readable by any local user via /sys/kernel/rcu_tasks_kthread_status or equivalent debugfs entries.

Why the score is not 0.0:

CONFIG_TASKS_RCU=y is compiled in and 5.19.6 falls within the affected range. RCU tasks is a core kernel synchronisation mechanism active at all times; the overflow condition requires unusually large counter values, making reliable exploitation difficult on a production system. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root; an attacker cannot execute a non-allowlisted program without an allowlist entry.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-40958

Status: Not exploitable Component: network namespaces (CONFIG_NET_NS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CLONE_NEWNET not in Root Lock allowlist; Lockdown blocks the exploitation trigger

In the network namespace subsystem, a use-after-free occurs through a refcount underflow. syzkaller triggered a refcount_t: addition on 0 warning at lib/refcount.c:25, indicating that a network namespace object’s reference count reached zero while still being accessed, with a subsequent attempt to increment the freed object’s refcount in refcount_warn_saturate().

CONFIG_NET_NS=y is compiled in. Creating a network namespace requires CLONE_NEWNET with CAP_NET_ADMIN. User namespaces (which would bypass the capability requirement) are disabled on the Root Lock kernel. No Root Lock production service creates network namespaces — they are absent from the Lockdown allowlist. Without an allowlist entry, the kernel refuses access. An attacker who has already gained root cannot add one: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

CVE-2024-41039

Status: Not exploitable Component: ALSA sound subsystem (CONFIG_SND) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no audio hardware present

Fix the checking that firmware file buffer is large enough for the wmfw header, to prevent overrunning the buffer.

CONFIG_SND=y is compiled in. No audio hardware is present on a headless Debian 11 server. The ALSA subsystem does not create /dev/snd device nodes without an audio card. The ioctl path that exposes this bug is never instantiated.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2024-46713

Status: Not exploitable Component: perf events subsystem (CONFIG_PERF_EVENTS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — perf_event_paranoid=3 restricts perf_event_open(); no profiling tool in Root Lock allowlist

Ole reported that event->mmap_mutex is strictly insufficient to serialize the AUX buffer, add a per RB mutex to fully serialize it.

CONFIG_PERF_EVENTS=y is compiled in and 5.19.6 falls within the affected range. On a Root Lock system, perf_event_paranoid=3 restricts perf_event_open() to processes with CAP_SYS_ADMIN; no profiling or performance analysis tool appears in the Root Lock allowlist. The exploitation path — loading and executing a non-allowlisted program — is blocked at the kernel execution gate before any perf subsystem interaction is possible. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-46852

Status: Not exploitable Component: DMA-BUF shared buffer (CONFIG_DMA_SHARED_BUFFER) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — no DRM device on headless HS server; DMA-BUF operations unreachable

Until VM_DONTEXPAND was added in commit 1c1914d6e8c6 (“dma-buf: heaps: Don’t track CMA dma-buf pages under RssFile”) it was possible to obtain a mapping larger than the buffer by calling mremap() on a DMA-BUF heap allocation. The DMA-BUF heap mmap handler did not set VM_DONTEXPAND, allowing the VMA to be extended beyond the original allocation size and enabling out-of-bounds access to adjacent memory.

CONFIG_DMA_SHARED_BUFFER=y is compiled in and 5.19.6 falls within the affected range. DMA-BUF buffer sharing requires access to a DRM or V4L2 device. Root Lock runs on headless server hardware with no GPU or video capture device; the DRM and V4L2 device nodes are absent, so the exploitation path — opening a DRM device and issuing mmap() on its DMA-BUF — is hardware-unreachable. No GPU or multimedia tool appears in the Root Lock allowlist. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2022-48950

Status: Not exploitable Component: perf events subsystem (CONFIG_PERF_EVENTS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — perf_event_paranoid=3 restricts perf_event_open(); no profiling tool in Root Lock allowlist

In kernel/events/core.c, perf_pending_task() can execute after the associated perf_event object has been freed. When a task exits and its pending perf events are processed, a race allows the task-work callback to fire after the event is released, causing a use-after-free.

CONFIG_PERF_EVENTS=y is compiled in and 5.19.6 falls within the affected range. On a Root Lock system, perf_event_paranoid=3 restricts perf_event_open() to processes with CAP_SYS_ADMIN; no profiling or performance analysis tool appears in the Root Lock allowlist. The exploitation path — loading and executing a non-allowlisted program — is blocked at the kernel execution gate before any perf subsystem interaction is possible. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2022-49026

Status: Not exploitable Component: Intel e100 Fast Ethernet driver (CONFIG_E100) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — Intel Pro/100 NIC absent on any modern HS server deployment

In e100_xmit_prepare(), if we can’t map the skb, then return -ENOMEM, so e100_xmit_frame() will return NETDEV_TX_BUSY and the upper layer will resend the skb.

CONFIG_E100=y is compiled in and 5.19.6 falls within the affected range. The Intel e100 driver supports legacy Intel Pro/100 Fast Ethernet cards, a line discontinued in the early 2000s. No modern server or datacenter hardware ships with or supports this NIC; the driver code is compiled in but the hardware is universally absent on any HS deployment. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-50055

Status: Not in this kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: core kernel (CONFIG_BASE_FULL) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — introduced in Linux 6.3 and fixed in 6.11.4, so 5.19.6 and 6.18.9-hs are outside that range Affected range: Linux 6.3 through 6.6.57, and 6.7 through 6.11.4. 5.19.6 and 6.18.9-hs are outside both ranges. Upstream fix: drivers/base/bus.c

What this means for an attacker:

In drivers/base/bus.c, bus_register() allocates a subsys_private struct (@priv) and calls kset_register() to publish the bus kobject. If a subsequent step in bus_register() fails — for example, during sysfs attribute file creation — the error path calls kset_unregister(), which frees @priv through its kobject release callback. bus_register() then also frees @priv directly in its own error path, causing a double-free.

Why the score is 0.0:

The bug arrived in Linux 6.3. One fix is in 6.6.57, and the later copy is fixed in 6.11.4. 5.19.6 is older than the bug. 6.18.9-hs is past that second fix. CONFIG_BASE_FULL=y does not place the bug in either kernel.

What this means for you as an HS user:

5.19.6 predates this bug, and 6.18.9-hs contains the fix. The backstop on How to read the backstop sections applies to rows that are in the kernel.

CVE-2024-50112

Status: Not Affected — LAM not implemented in Linux 5.19.x Component: x86_64 architecture (CONFIG_X86_64) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — LAM infrastructure absent from Linux 5.19.x (introduced in 6.2)

Linear Address Masking (LAM) is an x86_64 feature that allows software to store metadata in the upper bits of a canonical virtual address; it requires explicit kernel support — arch_prctl LAM commands, CR3 tag bit management, and associated data structures — to activate. The SLAM transient execution attack exploits an interaction between LAM tag bits and the speculative address-translation pipeline when a LAM-enabled process is running. This LAM kernel infrastructure was introduced upstream in Linux 6.2. The 5.19.6 kernel contains no LAM code paths; no process can enable LAM regardless of privilege level, and the transient execution oracle the SLAM paper describes does not exist in this kernel.

CVE-2024-50193

Status: Not exploitable Component: x86_64 architecture (CONFIG_X86_64) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — perf_event_open() blocked by perf_event_paranoid=3; no perf tool in Root Lock allowlist Affected range: Linux 5.x–6.11 Upstream fix: arch/x86/kernel/nmi.c (CPU buffer flush ordering fix)

On x86_64, the MDS/MD_CLEAR mitigation (VERW-based CPU buffer flush) is applied after exc_nmi() completes but before IRET restores register state. This ordering leaves a window in which speculative execution can observe uninitialised microarchitectural buffer contents from the interrupted context — a same-CPU information disclosure in the MDS (Microarchitectural Data Sampling) class.

CONFIG_X86_64=y is compiled in and 5.19.6 falls within the affected range. Triggering NMIs from ring-3 requires perf_event_open() or hardware performance counters. On a Root Lock system, perf_event_paranoid=3 restricts perf_event_open() to processes with CAP_SYS_ADMIN; no profiling or performance analysis tool appears in the Root Lock allowlist. The exploitation path — loading and executing a non-allowlisted program — is blocked at the kernel execution gate before any perf subsystem interaction is possible. After gaining root through any avenue, Lockdown’s allowlist refuses new code and blocks allowlist modification — no persistence, no backdoors, no cross-reboot survival.

CVE-2024-56600

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: IPv6 networking stack (CONFIG_IPV6) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low

Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: net/ipv6/af_inet6.c

What this means for an attacker:

In net/ipv6/af_inet6.c, sock_init_data() attaches the newly allocated sk pointer to sock->sk before inet6_create() completes setup. If inet6_create() fails at a later step and frees the sk, sock->sk retains the dangling pointer. The socket cleanup path subsequently calls sock->sk->sk_prot->close() on the freed sk, causing a use-after-free.

Why the score is not 0.0:

CONFIG_IPV6=y is compiled in and 5.19.6 falls within the affected range. IPv6 socket creation is triggered whenever a process opens an IPv6 socket — a common operation on any networked system. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root. An attacker cannot execute a new exploit program to reach this path — it has no allowlist entry and the kernel refuses to run it.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-56601

Status: In the kernel on 5.19.6. Fixed upstream on 6.18.9-hs. Component: TCP/IP networking (CONFIG_INET) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 7.3 HIGH — Lockdown reduces MI: High→Low

Affected range: Linux 5.x–6.x; 5.19.6 falls within range
Upstream fix: net/ipv4/af_inet.c

What this means for an attacker:

In net/ipv4/af_inet.c, sock_init_data() attaches the newly allocated sk pointer to sock->sk before inet_create() completes setup. If inet_create() fails at a later step and frees the sk, sock->sk retains the dangling pointer. The socket cleanup path subsequently calls sock->sk->sk_prot->close() on the freed sk, causing a use-after-free.

Why the score is not 0.0:

CONFIG_INET=y is compiled in and 5.19.6 falls within the affected range. The TCP/IP stack is always active; INET socket creation occurs on every TCP or UDP connection. In Lockdown, hs_sandbox_caching.c enforces the SPF allowlist against all processes including root. An attacker cannot execute a new exploit program — it has no allowlist entry and the kernel refuses to run it.

What this means for you as an HS user:

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2024-56616

Status: Not exploitable Component: DRM subsystem (CONFIG_DRM) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — DisplayPort MST display hardware absent

Fix the MST sideband message body length check, which must be at least 1 byte accounting for the message body CRC (aka message data CRC) at the end of the message.

CONFIG_DRM=y is compiled in. DisplayPort Multi-Stream Transport (DP MST) is used for daisy-chaining multiple monitors via DisplayPort. A headless server has no display output hardware; the DP MST sideband message path is never reached.

The attack vector has no path to execution on a standard Debian 11 server deployment. Lockdown provides a backstop regardless: root cannot modify the allowlist, install persistent backdoors, or survive a reboot.

CVE-2026-53341

Status: Not exploitable
Component: file handles / fhandle (CONFIG_FHANDLE)
Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) — NVD base pending; UAF class aligned with peer kernel entries
Score on Root Lock: 0.0 — CAP_DAC_READ_SEARCH required to pass may_decode_fh; unprivileged callers receive -EPERM; Lockdown blocks allowlist modification
Affected range: Linux 6.11 through 6.18.y prior to 6.18.36; production 6.18.9-hs still lacks the upstream fix until a newer 6.18.y base
Upstream fix: 6.18.36+ (32138633e51e — fhandle / may_decode_fh)

may_decode_fh() in the file-handle open path reads mount::mnt_ns without holding a lock. A concurrent unmount can free the mount namespace after an RCU grace period, so a racing open_by_handle_at can use the pointer after free. Upstream notes the race requires CONFIG_PREEMPTION or CONFIG_RCU_STRICT_GRACE_PERIOD and rates practical impact as limited (integer-comparison leak, hang, or crash) rather than a polished privilege-escalation chain.

CONFIG_FHANDLE=y is compiled in on 6.18.9-hs. Entering the vulnerable helper requires open_by_handle_at and CAP_DAC_READ_SEARCH (or the admin + DAC branch in that helper). Without those capabilities, may_decode_fh returns -EPERM before the unlocked mnt_ns read matters for an unprivileged attacker. No Root Lock deployment places a dedicated file-handle exploit tool in the allowlist. An attacker who has already gained root cannot add one: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot. Same product model as other capability-gated local paths (for example AF_PACKET / CAP_NET_RAW).

The trigger cannot be reached on any default Root Lock deployment.

CVE-2026-53223

Status: Not exploitable
Component: AF_PACKET timestamp cmsgs (CONFIG_PACKET)
Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H)
Score on Root Lock: 0.0 — CAP_NET_RAW not granted to services; packet-capture tools absent from Root Lock allowlist; Lockdown blocks the exploitation trigger
Affected range: kernels prior to the AF_PACKET timestamp cmsg guard; production 6.18.9-hs lacks the upstream fix until base ≥ 6.18.36-class
Upstream fix: 3dde4fb941fa (guard timestamp cmsgs to real error-queue skbs)

skb_is_err_queue() treated PACKET_OUTGOING as proof that an skb came from the socket error queue. AF_PACKET also delivers legitimate outgoing taps with that pkt_type while skb->cb holds packet control state, not sock_exterr_skb. With socket timestamping enabled, the generic timestamp cmsg path can emit SCM_TIMESTAMPING_OPT_STATS from the wrong buffer and disclose heap contents or trip hardened usercopy.

CONFIG_PACKET=y is compiled in. Creating an AF_PACKET socket requires CAP_NET_RAW. No Root Lock deployment grants CAP_NET_RAW to any service — packet capture tools such as tcpdump have no allowlist entry. Without an allowlist entry, the kernel refuses to execute them. An attacker who has already gained root cannot add one: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot. HeartSuite network hooks cover connect and destination sendto only; they do not cover socket / setsockopt / recvmsg on this path — reachability is closed by capability and allowlist composition, not by the net hook.

The trigger cannot be reached on any default Root Lock deployment.

CVE-2026-46300

Status: Not exploitable Component: skbuff coalescing and ESP-in-TCP (CONFIG_NET, CONFIG_INET_ESPINTCP) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) — CNA (kernel.org) Score on Root Lock: 0.0 — ESP-in-TCP is not compiled; the TCP-coalesce-then-ESP write path does not exist Affected range: Linux 3.9 through 6.18.32 (and listed LTS windows). 5.19.6 and production 6.18.9-hs are in range. Upstream fix: 6.18.33 (6.18 stable); 5.19 branch is EOL

This CVE describes a local privilege escalation (Fragnesia) in which skb_try_coalesce() attaches paged fragments from one socket buffer to another and drops the shared-fragment marker. ESP input then treats the coalesced buffer as privately owned, skips copy-on-write, and decrypts in place over page-cache-backed fragments. The published local-root path splices a privileged file into a TCP stream and switches that socket into ESP-in-TCP.

CONFIG_NET=y is compiled on both fielded kernels, so skb_try_coalesce() is present. The write that turns the lost marker into a page-cache overwrite is ESP input after TCP receive coalescing. That composition is ESP-in-TCP (CONFIG_INET_ESPINTCP / CONFIG_INET6_ESPINTCP). Both options are not set on 5.19.6-HeartSuite-2.0 and on 6.18.9-hs. IPv4 ESP is also not compiled on 5.19.6. 6.18.9-hs builds IPv4/IPv6 ESP as modules and still leaves both ESP-in-TCP options unset. The running System.map on both kernels exports skb_try_coalesce and contains no espintcp symbol.

A second independent stop matches CVE-2026-43284. Establishing any XFRM security association requires XFRM management tooling (ip xfrm, setkey, strongSwan, libreswan, or an equivalent IKE daemon). None of those programs are in the Root Lock default allowlist. Under Lockdown the allowlist is chattr +i immutable and FS_IOC_SETFLAGS returns EPERM for all callers.

The trigger cannot be reached on any Root Lock deployment.

If a custom kernel is built with CONFIG_INET_ESPINTCP=y or CONFIG_INET6_ESPINTCP=y and XFRM management tooling is added to the allowlist, treat this CVE as Affected at 7.8 HIGH and apply the standard backstop.

CVE-2026-45920

Status: Not exploitable Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — filesystem-shutdown injection (EXT4_IOC_SHUTDOWN) is not reachable from the allowlist Affected range: 2.6.29–5.10.252; 5.11–5.15.202; 5.16–6.1.166 (includes 5.19.6); 6.2–6.6.129; 6.7–6.12.74; 6.13–6.18.13 (includes 6.18.9); 6.19–6.19.3 Upstream fix: 5.10.253, 5.15.203, 6.1.167, 6.6.130, 6.12.75, 6.18.14, 6.19.4 (5.19 branch is EOL; no backport)

This CVE describes a double decrement of the ext4 dirty-clusters counter on the multi-block allocator error path. After the filesystem is placed in forced-shutdown state, ext4_mb_mark_diskspace_used() decrements s_dirtyclusters_counter and then ext4_handle_dirty_metadata() returns an error; the caller ext4_mb_new_blocks() decrements the same counter again in the failed-allocation cleanup path. The counter reaches −1 and ext4_put_super() emits a kernel WARNING. The documented reproduction is fstests generic/388 (allocator stress plus filesystem-shutdown injection).

CONFIG_EXT4_FS=y on 5.19.6 and CONFIG_EXT4_FS=m on 6.18.9-hs; both kernels fall inside the NVD range. Ordinary reads and writes on the mounted ext4 volume do not take this error path. The path requires EXT4_FLAGS_SHUTDOWN first. On 5.19.6 that bit is set only by ext4_shutdown(), reached solely through EXT4_IOC_SHUTDOWN, which returns -EPERM unless the caller has CAP_SYS_ADMIN. The userspace programs that inject that ioctl (xfs_io) and that drive the concurrent allocator stress (fsstress) are not in the HeartSuite allowlist. The kernel refuses to execute any program without an allowlist record. Lockdown also returns -EPERM on mount(), fsmount(), and move_mount(), so a test image cannot be mounted to host the same sequence.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-46094

Status: Not Affected on 5.19.6; Not exploitable on 6.18.9-hs Component: ext4 filesystem (CONFIG_EXT4_FS) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — 5.19.6 is outside the affected range; on 6.18.9-hs Lockdown returns -EPERM from mount() / fsmount() / move_mount(), so a crafted xattr image cannot be attached Affected range: 6.3 through 6.6.139; 6.7 through 6.12.85; 6.13 through 6.18.26; 6.19 through 7.0.3. 5.19.6 is not in range. Production 6.18.9-hs remains in range until base ≥ 6.18.27 Upstream fix: 6.6.140, 6.12.86, 6.18.27, 7.0.4; mainline eceafc31ea7b

This CVE describes an out-of-bounds read in fs/ext4/xattr.c. check_xattrs() walks on-disk xattr entries and tests (void *)next >= end. That test lets next land inside the last four bytes of the xattr region. The next loop iteration calls IS_LAST_ENTRY(), which reads a u32 and overruns the valid region (CWE-125). Integrity impact is None.

CONFIG_EXT4_FS=y on 5.19.6 and CONFIG_EXT4_FS=m on 6.18.9-hs. The check_xattrs() helper was introduced in 6.3; 5.19.6 predates it and is outside the NVD window. 6.18.9 sits inside 6.13–6.18.26. The v6.18.9 check_xattrs() still uses the unfixed (void *)next >= end test. The function is static (absent from System.map) and is compiled into ext4.ko.

The overrun requires a corrupted xattr block on a mounted ext4 image. Kernel-formed setxattr on the already-mounted root does not produce that layout. Attaching a crafted image requires mount(), fsmount(), or move_mount(). mount is in the shipped command list. Under Lockdown those three syscalls return -EPERM in kernel/namespace.c. losetup, setfattr, getfattr, and mkfs.ext4 are not in the allowlist; that is not the binding stop.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-46020

Status: Not Affected on 5.19.6; Not exploitable on 6.18.9-hs
Component: DAMON core — damos_quota_goal->nid for node_mem_{used,free}_bp (CONFIG_DAMON, CONFIG_DAMON_SYSFS)
Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H)
Score on Root Lock: 0.0 — 5.19.6 is outside the affected range and DAMON is not compiled; on 6.18.9-hs the DAMON sysfs trigger is not reachable from the allowlist
Affected range: Linux 6.16 through 6.18.26; 6.19 through 7.0.3. 5.19.6 is not in range. Production 6.18.9-hs remains in range until base ≥ 6.18.27
Upstream fix: 40250b2dded0604a112be605f3828700d80ad7c2 (mainline 7.1-rc1); stable 6.18.27, 7.0.4

This CVE describes an out-of-bounds read in DAMON core. Users can set damos_quota_goal->nid to an arbitrary node id for the node_mem_{used,free}_bp quota-goal metrics. DAMON then calls si_meminfo_node() without validating that id. An invalid nid produces a kernel NULL-pointer dereference and an out-of-bounds read of node data. The kernel interface is DAMON_SYSFS. The documented trigger is the damo userspace tool.

# CONFIG_DAMON is not set on 5.19.6-HeartSuite-2.0. The introducing commit is 6.16. 5.19.6 predates the feature. The 5.19.6 System.map has no damos_get_node_mem_bp symbol.

On 6.18.9-hs, CONFIG_DAMON=y and CONFIG_DAMON_SYSFS=y. damos_get_node_mem_bp is present in the 6.18.9-hs System.map. That is not enough. Reaching the path requires writing DAMON sysfs quota-goal attributes under /sys/kernel/mm/damon/. damo is not in the shipped allowlist. Lockdown refuses to execute it. The allowlisted writers that exist (echo, tee, printf, bash) receive write grants only for /usr/lib and /etc from the default record seed. Opening /sys/kernel/mm/damon/ for write returns EACCES. Under Lockdown the allowlist is immutable: FS_IOC_SETFLAGS returns EPERM, so root cannot add damo or a DAMON sysfs write grant.

The trigger cannot be reached on any default Root Lock deployment.

If your 6.18.9-hs deployment adds damo or grants write access to /sys/kernel/mm/damon/ to an allowlisted program, treat this CVE as Affected at 7.1 HIGH and apply the I:N infoleak backstop.

CVE-2026-46121

Status: Not Affected on 5.19.6; Not exploitable on 6.18.9-hs Component: DAMON sysfs schemes (CONFIG_DAMON, CONFIG_DAMON_SYSFS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — 5.19.6 is outside the affected range and does not compile DAMON; on 6.18.9-hs the memcg_path race is not reachable from the allowlist Affected range: 6.6.96–6.6.139; 6.12.36–6.12.87; 6.15.5–6.15.x; 6.16; 6.16.1 through 6.18.29; 6.19 through 7.0.6. 5.19.6 is not in range. Production 6.18.9-hs remains in range until base ≥ 6.18.30 Upstream fix: 6.18.30; 6.12.88; 6.6.140; 7.0.7

This CVE describes a use-after-free in mm/damon/sysfs-schemes.c. Direct reads of the DAMON sysfs memcg_path and path files race with writes that free the backing buffer. The commit path that copies those strings into DAMON is already under damon_sysfs_lock; the user-facing show/store path was not. Two open files on the same attribute can therefore read a buffer after it has been freed.

On 5.19.6, # CONFIG_DAMON is not set. The 5.19.6 DAMON tree has sysfs.c and dbgfs.c only — no sysfs-schemes.c and no memcg_path attribute. The vulnerable code is not present.

On 6.18.9-hs, CONFIG_DAMON=y and CONFIG_DAMON_SYSFS=y. memcg_path_show, memcg_path_store, and damon_sysfs_scheme_filter_memcg_path_attr are in the running System.map. That is not enough to reach the race. The memcg_path and path files exist only after userspace creates the scheme-filter hierarchy under /sys/kernel/mm/damon/admin. The kernel documents that directory as a privileged admin interface and names damo as the userspace tool. damo is not in the HeartSuite allowlist. Default allowlist records grant /usr/lib and /etc only; writes to /sys/kernel/mm/damon/ are denied. CONFIG_DAMON_RECLAIM and CONFIG_DAMON_LRU_SORT run in-kernel and do not create those scheme-filter files. Under Lockdown, FS_IOC_SETFLAGS returns -EPERM, so root cannot add damo or that sysfs path.

The trigger cannot be reached on any default Root Lock deployment.

If a 6.18.9-hs deployment adds damo or write access to /sys/kernel/mm/damon/admin to the allowlist, treat this CVE as Affected at 7.8 HIGH and apply the standard backstop.

CVE-2026-46279

Status: Not exploitable — feature not compiled Component: mm/alloc_tag (CONFIG_MEM_ALLOC_PROFILING) Base Score: 7.8 HIGH (CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_MEM_ALLOC_PROFILING is not compiled; alloc_tag/codetag is not in the kernel.

The bug is an uninitialized alloc_tag/codetag on pages allocated before page_ext is ready during boot. That path is compiled only when CONFIG_MEM_ALLOC_PROFILING is enabled. The published warning requires CONFIG_MEM_ALLOC_PROFILING_DEBUG. HeartSuite 6.18.9-hs is in the NVD range (6.10 through 6.18.26) and has # CONFIG_MEM_ALLOC_PROFILING is not set. HeartSuite 5.19.6 is outside the NVD range (starts at 6.10) and the option is not present in that kernel.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-46281

State on 6.18.9-hs: Affected. Base score 7.8 HIGH. This row stays on the patch date.

Status: Not in this kernel on 5.19.6. In the kernel on 6.18.9-hs. Component: vmalloc — virtually contiguous allocator (CONFIG_MMU) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 on 5.19.6 (outside the affected range). 7.1 HIGH on 6.18.9-hs (CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H/MC:H/MI:N/MA:H) — Modified Integrity is None; Modified Confidentiality stays High on a worst-case allowlist; Modified Availability stays High because the bug can panic the kernel Affected range: 6.18 through 6.18.26; also 6.19 through 7.0.3. 5.19.6 is not in range. Production 6.18.9-hs remains in range until base ≥ 6.18.27 Upstream fix: e9b057a44def (mainline); stable 6.18.27+

This CVE describes an out-of-bounds write in vrealloc_node_align(). When the helper forces a new allocation while shrinking the object, it allocates size bytes and copies old_size bytes into that buffer.

CONFIG_MMU=y compiles mm/vmalloc.c on both fielded kernels. That is not enough for 5.19.6: the helper was added in 6.18. The 5.19.6 kernel has kvrealloc only. The 5.19.6 vrealloc_node_align path does not exist.

On 6.18.9-hs the helper is in the running kernel. vrealloc_node_align_noprof is present and kvrealloc_node_align_noprof is exported. The overflow is a kernel memory-corruption bug in a core allocator. Lockdown does not remove that code path from an already-running process.

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

These constraints are why Modified Integrity is None. Modified Confidentiality stays High: the live session can still read files and send them off the host through an allowlisted outbound utility. Modified Availability stays High because the bug can panic the kernel. The score on 6.18.9-hs is 7.1 HIGH. It stays above 0.0 because the helper is compiled in.

CVE-2026-52968

Status: Not exploitable Component: KVM s390 PCI (CONFIG_KVM_S390) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — CONFIG_KVM_S390 not compiled; s390 PCI GAIT path absent from the x86_64 image Affected range: Linux 6.0–6.1.174, 6.2–6.6.140, 6.7–6.12.90, 6.13–6.18.32, 6.19–7.0.9, and 7.1-rc1–rc3 Upstream fix: 6.1.175, 6.6.141, 6.12.91, 6.18.33, 7.0.10, 7.1-rc4

In arch/s390/kvm/pci.c and arch/s390/kvm/interrupt.c, kvm_s390_pci_aif_enable(), kvm_s390_pci_aif_disable(), and aen_host_forward() index the GAIT by multiplying the AISB index by sizeof(struct zpci_gaite) on a pointer that is already typed as struct zpci_gaite *. The offset is double-scaled and the access lands at element aisb*16 instead of aisb, which is out of bounds when aisb >= 32 (ZPCI_NR_DEVICES=512).

5.19.6 is outside the affected range (the code landed in 6.0) and # CONFIG_KVM is not set. 6.18.9-hs is inside 6.13–6.18.32 and builds x86 KVM (CONFIG_KVM=m, CONFIG_KVM_X86=m). Both kernels are CONFIG_X86_64=y. CONFIG_S390 and CONFIG_KVM_S390 do not appear in either production config. The 6.18.9-hs System.map contains none of kvm_s390_pci_aif_enable, kvm_s390_pci_aif_disable, aen_host_forward, or zpci_gaite. The compiled KVM module is Intel/AMD x86; it does not contain the s390 PCI GAIT indexer.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-52969

Status: Not exploitable — feature not compiled on 5.19.6; Not exploitable — tool not in the program allowlist on 6.18.9-hs Component: KVM dirty ring (CONFIG_KVM) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — 5.19.6 does not compile host KVM; on 6.18.9-hs the dirty-ring reset path is reached only by a KVM userspace program, and none is in the allowlist Affected range: 5.11–5.15.208; 5.16–6.1.174; 6.2–6.6.140; 6.7–6.12.90; 6.13–6.18.32; 6.19–7.0.9; 7.1-rc1–rc3. Both 5.19.6 and 6.18.9-hs are in range. Fixed in 6.18.33+ and 7.0.10+ Upstream fix: 577a8d3bae05 (mainline); stable 6.18.33+

This CVE describes a wrapped u64 offset in kvm_reset_dirty_gfn(). The bounds check adds offset to __fls(mask) without first rejecting a wrap. A process that holds /dev/kvm and uses the dirty ring can rewrite slot and offset fields so the check passes. The wrapped offset then indexes a near-U64_MAX gfn and performs an out-of-bounds load.

On 5.19.6, CONFIG_KVM is not set. Host KVM is not in the running kernel.

On 6.18.9-hs, CONFIG_KVM=m. The trigger requires a loaded kvm module and a userspace program that opens /dev/kvm, enables the dirty ring, and issues KVM_RESET_DIRTY_RINGS. No KVM userspace program is in the Root Lock allowlist. After boot, kernel.modules_disabled=1 refuses a later load of kvm. See How to read the backstop sections. Under Lockdown, the allowlist cannot be modified. The dirty-ring reset path is never reached.

The trigger cannot be reached on any Root Lock deployment.

If kvm was already loaded when the latch ran, and the allowlist includes a KVM userspace program (qemu, libvirt, firecracker, or equivalent), this CVE applies at its base score of 7.8 HIGH. Treat it as Affected and apply the standard backstop logic.

CVE-2026-53004

Status: Not exploitable Component: SCTP (CONFIG_IP_SCTP) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — 5.19.6 does not compile SCTP; 6.18.9 builds SCTP as a module, and kernel.modules_disabled=1 refuses a later load Affected range: Linux through 7.0.9 (Ubuntu 5.15–7.0 and Debian bullseye 5.10 needed the fix) Upstream fix: 7.1-rc1 / 7.0.10

sctp_getsockopt_peer_auth_chunks() checks if (len < num_chunks) and then writes num_chunks bytes to p->gauth_chunks, which sits eight bytes inside the caller’s optval. The check omits the sctp_authchunks header. When the caller passes len == num_chunks, copy_to_user() writes eight bytes past the declared buffer. Those bytes land in the caller’s own userspace; this is not kernel memory corruption.

On 5.19.6, # CONFIG_IP_SCTP is not set. The function is absent from the vmlinux System.map. On 6.18.9-hs, CONFIG_IP_SCTP=m. Reaching the function requires the SCTP module to be loaded, an SCTP association with AUTH enabled, and getsockopt(SCTP_PEER_AUTH_CHUNKS). The shipped allowlist has no SCTP tools. After boot, kernel.modules_disabled=1 refuses a later load of sctp. See How to read the backstop sections. Under Lockdown, FS_IOC_SETFLAGS returns EPERM, so those programs cannot be added. The SCTP stack is not loaded.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-53264

Status: Not exploitable
Component: net/sched action API (CONFIG_NET_SCHED, CONFIG_NET_CLS_ACT)
Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) — CNA (kernel.org)
Score on Root Lock: 0.0 — no tc action can be instantiated; tc is absent from the allowlist, and kernel.modules_disabled=1 refuses a later load of the action modules
Affected range: 4.14 through unfixed stables including 6.18 before 6.18.36; 5.19.6 and 6.18.9-hs are in range
Upstream fix: stable 6.18.36+ (5.19 branch is EOL; no backport)

This CVE describes a use-after-free in the traffic-control action lifecycle. Concurrent NEWTFILTER and DELFILTER netlink operations race on a tc_action: the delete path frees the object immediately while a lookup still holds the pointer, so the increment of the action refcount touches freed memory.

CONFIG_NET_SCHED=y and CONFIG_NET_CLS_ACT=y are compiled in on both fielded kernels. The race requires an instantiated tc action attached to a filter.

On 5.19.6 every CONFIG_NET_ACT_* option is not set. No action kind is registered in the running image, so a NEWTFILTER that names an action cannot create one.

On 6.18.9-hs every CONFIG_NET_ACT_* option is a module and those kinds are not in vmlinux. After boot, kernel.modules_disabled=1 refuses a later load. See How to read the backstop sections. The tc program that issues NEWTFILTER and DELFILTER is also not on the allowlist. Under Lockdown the allowlist cannot be extended.

The trigger cannot be reached on any Root Lock deployment.

If the action modules were already loaded when the latch ran, and the allowlist includes tc, treat this CVE as Affected at 7.8 HIGH and apply the standard backstop.

CVE-2026-53359

Status: Not exploitable Component: KVM x86 shadow MMU (CONFIG_KVM) Base Score: 8.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H) — CNA (kernel.org); NVD assessment pending Score on Root Lock: 0.0 — 5.19.6 does not compile CONFIG_KVM; 6.18.9 compiles CONFIG_KVM=m but no QEMU or libvirt program is in the allowlist Affected range: Linux 2.6.36 through 6.1.176; 6.2 through 6.6.143; 6.7 through 6.12.94; 6.13 through 6.18.37; 6.19 through 7.1.2 Upstream fix: 6.1.177, 6.6.144, 6.12.95, 6.18.38, 7.1.3, 7.2

This CVE is a use-after-free in the KVM x86 shadow MMU. The host reuses a cached shadow page when the guest frame number matches even though the page role does not. A later reverse-map walk then dereferences a freed shadow page.

On 5.19.6, # CONFIG_KVM is not set. Host KVM is not compiled. CONFIG_KVM_GUEST=y only enables paravirtual guest support; it does not compile the host shadow MMU.

On 6.18.9, CONFIG_KVM=m with CONFIG_KVM_INTEL=m and CONFIG_KVM_AMD=m. Reaching the bug requires a running KVM guest: a userspace hypervisor (QEMU, libvirt, or equivalent) must create a VM and drive the shadow MMU. No such program is in the HeartSuite allowlist. The kernel refuses to execute it. An attacker who has already gained root cannot add one: Lockdown prevents allowlist modification, backdoor installation, and persistence across reboot.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-63794

Status: Not Affected on 5.19.6; Not exploitable on 6.18.9-hs Component: KVM AMD SVM — SEV debug crypt (CONFIG_KVM, CONFIG_KVM_AMD, CONFIG_KVM_AMD_SEV) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — 5.19.6 does not compile host KVM; on 6.18.9-hs the SEV debug-encrypt ioctl is not reachable from the allowlist Affected range: Linux 4.16 through 6.18.37 (plus the other listed stable windows). 5.19.6 is in the 5.16–<6.1.177 window but host KVM is not compiled. Production 6.18.9-hs remains in range until base ≥ 6.18.38 Upstream fix: 6.18.38, 6.12.95, 6.6.144, 6.1.177, 5.15.211, 5.10.260

This CVE describes a page overflow in sev_dbg_crypt() on the ENCRYPT path in arch/x86/kvm/svm/sev.c. The per-iteration copy length is clipped to the source page remainder (PAGE_SIZE - s_off) but not the destination page remainder (PAGE_SIZE - d_off). When d_off > s_off, __sev_dbg_encrypt_user issues a PSP command and a memcpy into a single-page bounce buffer that overflows. The entry is kvm_vm_ioctl → sev_mem_enc_ioctl → sev_dbg_crypt (KVM SEV debug encrypt).

On 5.19.6, # CONFIG_KVM is not set. CONFIG_KVM_GUEST=y is guest-side paravirt only. There is no CONFIG_KVM_AMD / CONFIG_KVM_AMD_SEV, and sev_dbg_crypt is not in the 5.19.6 image.

On 6.18.9-hs, CONFIG_KVM=m, CONFIG_KVM_AMD=m, and CONFIG_KVM_AMD_SEV=y. Reaching the overflow requires a loaded kvm_amd module, an SEV guest, and the SEV debug-encrypt ioctl. No qemu, libvirt, virsh, or KVM/SEV userspace appears in the Root Lock allowlist. After boot, kernel.modules_disabled=1 refuses a later load of kvm_amd. See How to read the backstop sections. The kernel refuses to run a dropped program with no allowlist entry. After gaining root through any other avenue, Lockdown still blocks allowlist modification, so those tools cannot be added for the life of the boot.

The trigger cannot be reached on any Root Lock deployment.

If kvm_amd was already loaded when the latch ran, and the allowlist includes qemu-system, libvirt, or any other program that issues KVM SEV debug-encrypt ioctls, treat this CVE as Affected at 7.8 HIGH and apply the standard backstop.

CVE-2026-63804

Status: Not exploitable Component: GFS2 clustered filesystem (CONFIG_GFS2_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — GFS2 cluster tools are not on the allowlist; the unmount path never runs Affected range: Linux 6.6 through 6.6.143, 6.7 through 6.12.94, 6.13 through 6.18.37, and 6.19 through 7.1.2. HeartSuite 5.19.6 is outside this range. HeartSuite 6.18.9 is inside it. Upstream fix: rcu_barrier() before free_sbd() in gfs2_put_super(); stable backports through 6.18.38

During GFS2 unmount, gfs2_qd_dealloc() runs as an RCU callback and touches the superblock after that superblock is already freed. The bug is a use-after-free in the quota-object teardown path.

HeartSuite 5.19.6 does not compile GFS2 (CONFIG_GFS2_FS is not set) and sits below the NVD floor of 6.6.

HeartSuite 6.18.9 compiles GFS2 as a module (CONFIG_GFS2_FS=m, CONFIG_GFS2_FS_LOCKING_DLM=y, CONFIG_DLM=m). The trigger is unmount of an already-mounted GFS2 volume. That state requires the GFS2 and DLM modules plus the cluster userspace (mount.gfs2 and the DLM/corosync stack). Those programs are not on the HeartSuite allowlist, so they cannot execute. No Root Lock deployment mounts GFS2 as a live filesystem.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-64121

Status: Not exploitable Component: IFB intermediate functional block (CONFIG_IFB) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) — NVD Score on Root Lock: 0.0 — CONFIG_IFB is not compiled on 5.19.6; on 6.18.9-hs the trigger requires ip and ethtool, which are absent from the allowlist, and kernel.modules_disabled=1 refuses a later load of ifb Affected range: 5.17 through 6.1.174; 6.2 through 6.6.141; 6.7 through 6.12.91; 6.13 through 6.18.33; 6.19 through 7.0.10; plus 7.1-rc1–rc4. Both production kernels (5.19.6 and 6.18.9-hs) sit in range. Upstream fix: ethtool stats walk dev->num_tx_queues; stable 6.18.34+

This CVE describes a slab out-of-bounds read in the IFB ethtool stats path. ifb_dev_init() allocates dp->tx_private with dev->num_tx_queues entries. ifb_get_ethtool_stats() instead walks dev->real_num_rx_queues. On an IFB device created with more RX queues than TX queues, the walk indexes past the allocation and copies adjacent slab data out through ETHTOOL_GSTATS. Integrity impact is none. The bug does not grant root.

On 5.19.6, CONFIG_IFB is not compiled. The Kconfig depends on NET_ACT_MIRRED || NFT_FWD_NETDEV. Both parents are unset (# CONFIG_NET_ACT_MIRRED is not set, # CONFIG_NF_TABLES is not set), so the CONFIG_IFB symbol is not offered. The 5.19.6 System.map has no ifb_get_ethtool_stats symbol. The callbacks in drivers/net/ifb.c are not in the running image.

On 6.18.9-hs, CONFIG_IFB=m. The module is built, not builtin. Reaching the bug requires a loaded ifb, an asymmetric IFB device, and an ethtool stats query. After boot, kernel.modules_disabled=1 refuses a later load of ifb. See How to read the backstop sections. ip and ethtool are absent from the Root Lock allowlist. Under Lockdown the allowlist cannot be changed: FS_IOC_SETFLAGS returns -EPERM, and mount(), fsmount(), and move_mount() return -EPERM. Root cannot add those programs.

The trigger cannot be reached on any Root Lock deployment.

If ifb was already loaded when the latch ran, and the allowlist includes ip and ethtool, treat this CVE as Affected at 7.1 HIGH for confidentiality and availability only.

CVE-2026-64600

State on 6.18.9-hs: Affected. Base score 7.8 HIGH. This row stays on the patch date.

Status: Not in this kernel on 5.19.6. In the kernel on 6.18.9-hs. Component: XFS reflink / copy-on-write (CONFIG_XFS_FS) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 on 5.19.6 (XFS is not compiled). 7.1 HIGH on 6.18.9-hs (CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H/MC:H/MI:N/MA:H) — Modified Integrity is None; Modified Confidentiality stays High on a worst-case allowlist; Modified Availability stays High because the bug can panic the kernel Affected ranges: NVD: 4.11 through 5.15.211; 5.16 through 6.1.177; 6.2 through 6.6.144; 6.7 through 6.12.95; 6.13 through 6.18.38; 6.19 through 7.1.3. HeartSuite 5.19.6 is in range with XFS not compiled. HeartSuite 6.18.9-hs is in range with CONFIG_XFS_FS=m. Upstream fix: 5.15.212, 6.1.178, 6.6.145, 6.12.96, 6.18.39, 7.1.4

The bug is a race in the XFS reflink copy-on-write path. After xfs_reflink_fill_cow_hole and xfs_reflink_fill_delalloc drop and re-take ILOCK to start a transaction, they refresh the CoW fork mapping and leave the data-fork mapping stale. A concurrent aligned O_DIRECT writer can finish a CoW cycle in that window. The first writer then operates on the wrong physical block, including a block that now solely backs the reflink source. That is a local privilege-escalation primitive.

On 5.19.6 HeartSuite, # CONFIG_XFS_FS is not set. The helpers are not compiled. That kernel is Not Affected.

On 6.18.9-hs, CONFIG_XFS_FS=m and CONFIG_MODULES=y. The unfixed fill helpers are in the 6.18.9 XFS tree. The installer unpacks the full module tarball into /lib/modules and, on Amazon Linux, forces xfs into the initrd. Amazon Linux 2023, Rocky 9, and RHEL 9 use XFS as the root filesystem. cp, dd, and python3 are allowlisted. Creating a new XFS image requires mkfs.xfs, which is not in the allowlist, and Lockdown returns -EPERM from mount(), fsmount(), and move_mount(). That closes a late mount of an attacker-supplied XFS volume. It does not close the path when XFS is already mounted: aligned O_DIRECT writes and reflink clones are ordinary file I/O.

Even with this CVE exploited to root, the attacker cannot run new code on this system. Lockdown’s allowlist refuses every non-allowlisted program at execve, including in the worst case where the attacker has cleared Lockdown. No persistence, no backdoors, no cross-reboot survival. (How.)

A reboot is a clean slate. The attack does not survive it.

CVE-2026-45837

CVE-2026-45839

Status: Not exploitable — feature not compiled Component: BPF CO-RE relocation parser (CONFIG_BPF_SYSCALL) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_BPF_SYSCALL is not compiled; bpf() returns ENOSYS; bpf_core_parse_spec() is unreachable

CVE-2026-45839 is a signed-index out-of-bounds read in bpf_core_parse_spec(). CO-RE accessor strings are colon-separated field indices parsed with sscanf("%d"). A negative index passes the upper-bound check (access_idx >= btf_vlen(t)) and is cast to u32 0xffffffff inside btf_member_bit_offset(), reading far past the BTF members array. A local caller with CAP_BPF triggers it on BPF_PROG_LOAD.

# CONFIG_BPF_SYSCALL is not set on the Root Lock kernel (bpf() returns ENOSYS). There is no verifier, no CO-RE relocation path, and no bpf_core_parse_spec() in the running kernel. Both lines also ship CONFIG_DEBUG_INFO_NONE=y with no vmlinux BTF.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-45851

Status: Not Affected on 5.19.6; Not exploitable — hardware absent on 6.18.9 Component: EFI unaccepted memory table (CONFIG_UNACCEPTED_MEMORY absent on 5.19.6; CONFIG_UNACCEPTED_MEMORY=y on 6.18.9) Base Score: 7.1 HIGH (CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — 5.19.6 is outside the NVD range and does not compile unaccepted memory; 6.18.9 compiles the path but the trigger requires Intel TDX guest firmware that presents an unaccepted memory table.

reserve_unaccepted() under-reserves the EFI unaccepted memory table when the table start address is not page-aligned. The tail of the table is left unreserved, the table is overwritten or becomes inaccessible, and the kernel panics in accept_memory(). Upstream recorded that failure when starting Intel TDX guests with specific memory sizes (for example greater than 64 GB).

NVD marks Linux before 6.6 unaffected. HeartSuite 5.19.6 has no CONFIG_UNACCEPTED_MEMORY option and no accept_memory symbol. HeartSuite 6.18.9 sits in 6.13–6.18.13 and builds CONFIG_UNACCEPTED_MEMORY=y, but a standard Root Lock deployment is not an Intel TDX guest and does not receive that firmware table.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-45853

Status: Not Affected on 5.19.6; Not exploitable — hardware absent on 6.18.9 Component: AMDGPU DRM driver (CONFIG_DRM_AMDGPU) Base Score: 7.8 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — # CONFIG_DRM_AMDGPU is not set on 5.19.6 and the NVD range starts at 6.11; on 6.18.9 the driver is compiled as a module but no AMD GPU is present Affected range: 6.11 through 6.12.74; 6.13 through 6.18.13; 6.19 through 6.19.3 Upstream fix: 6.12.75, 6.18.14, 6.19.4

amdgpu_gmc_get_nps_memranges() in drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c frees a range table with kfree() after amdgpu_discovery_get_nps_info() allocated it with kvcalloc(). When that allocation comes from vmalloc, kfree() corrupts kernel memory.

On 5.19.6 the AMDGPU driver is not compiled and the function does not exist. On 6.18.9 CONFIG_DRM_AMDGPU=m sits inside the NVD window, but the path runs only when an AMD GPU is present and the driver binds. No AMD GPU is present on a Root Lock server deployment.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-45893

Status: Not exploitable — feature not compiled on 5.19.6; Not exploitable — tool not in the program allowlist on 6.18.9-hs Component: AppArmor DFA table unpack (CONFIG_SECURITY_APPARMOR) Base Score: 7.1 HIGH (CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — 5.19.6 does not compile AppArmor; on 6.18.9-hs the policy-unpack path is not reachable from the allowlist Affected range: Linux 4.11 through 6.12.74; 6.13 through 6.18.13; 6.19 through 6.19.3. Both HeartSuite production kernels sit in that window until the config and allowlist gates apply. Upstream fix: 6.12.75; 6.18.14; 6.19.4

This CVE describes an unaligned memory access in unpack_table() (security/apparmor/match.c). AppArmor copies DFA match tables out of a userspace policy blob with aligned be16_to_cpu / be32_to_cpu. A blob that is not naturally aligned produces an out-of-bounds read. The function runs only when a profile is loaded or replaced through the AppArmor securityfs interface.

On 5.19.6, # CONFIG_SECURITY_APPARMOR is not set. The AppArmor code is not compiled. The CONFIG_LSM string still names apparmor; that string does not compile the module when the Kconfig option is off.

On 6.18.9-hs, CONFIG_SECURITY_APPARMOR=y and CONFIG_DEFAULT_SECURITY_APPARMOR=y. AppArmor is a live LSM. That is not enough to reach unpack_table. Reaching the bug requires writing a packed profile to /sys/kernel/security/apparmor/.load or .replace. apparmor_parser and the aa-* policy tools are not on the allowlist. Default allowlist records grant /usr/lib and /etc only, read-only; writes to /sys/kernel/security/apparmor/ are denied. Under Lockdown, FS_IOC_SETFLAGS returns -EPERM, so the allowlist cannot be extended to add those programs or that path.

The trigger cannot be reached on any default Root Lock deployment.

If a 6.18.9-hs deployment adds apparmor_parser or write access to /sys/kernel/security/apparmor/ to the allowlist, treat this CVE as Affected at 7.1 HIGH with Confidentiality and Availability High and Integrity None.

CVE-2026-45903

Status: Not Affected Component: BPF syscall / verifier helper prototypes (CONFIG_BPF_SYSCALL) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — CONFIG_BPF_SYSCALL is not compiled; bpf() returns ENOSYS

After a verifier refactor, several BPF helper prototypes omitted MEM_RDONLY or MEM_WRITE on ARG_PTR_TO_MEM arguments. The verifier then treated helper buffers as unchanged and dropped later reads, producing incorrect memory access.

NVD lists Linux 6.14 through 6.18.13 and 6.19 through 6.19.3. 5.19.6 is outside that range. On both released pins the bpf() syscall is not compiled. There is no verifier and no helper-prototype path.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-45943

Status: Not exploitable — feature not compiled on 5.19.6; Not exploitable — tool not in the program allowlist on 6.18.9-hs Component: EROFS filesystem (CONFIG_EROFS_FS, CONFIG_EROFS_FS_ZIP) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — 5.19.6 does not compile EROFS; on 6.18.9-hs the erofs module cannot be loaded and a ztailpacking image cannot be attached Affected range: 5.17 through 6.12.77; 6.13 through 6.18.13; 6.19 through 6.19.3. Both 5.19.6 and 6.18.9-hs are in range. Fixed in 6.12.78, 6.18.14, 6.19.4, and 7.0. Upstream fix: 6.12.78, 6.18.14, 6.19.4, 7.0; stable commits 5de1aa0bf3a5, 92088bd9aa2a, ad07ea069f92, c134a40f86ef

This CVE describes a NULL pointer dereference in z_erofs_decompress_pcluster(). Compressed folios for ztailpacking pclusters must be valid before those pclusters are added to the I/O chain. If a fatal signal interrupts read_mapping_folio() while the inline tail is being fetched, z_erofs_decompress_queue() treats the folio as valid and dereferences a NULL page. Integrity impact is None. The entry is a read() of a file on a mounted EROFS volume that uses ztailpacking.

On 5.19.6, # CONFIG_EROFS_FS is not set. The EROFS decompression path is not in the running kernel.

On 6.18.9-hs, CONFIG_EROFS_FS=m and CONFIG_EROFS_FS_ZIP=y. Reaching the path requires the erofs module to be loaded and a ztailpacking EROFS volume to be mounted, then a read() of a compressed file. After boot, kernel.modules_disabled=1 refuses a later load of erofs. See How to read the backstop sections. mkfs.erofs, dump.erofs, and fsck.erofs are not on the allowlist. mount is on the allowlist; under Lockdown, mount(), fsmount(), and move_mount() return -EPERM, so a brought-in image cannot be attached. HeartSuite startup and setup do not mount EROFS.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-45957

Status: Not exploitable Component: RCU preemptible tree (CONFIG_PREEMPT_RCU, CONFIG_TREE_RCU); documented trigger is ftrace softirq stack tracing (CONFIG_FTRACE, CONFIG_TRACING) Base Score: 7.1 HIGH (AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) — NVD Score on Root Lock: 0.0 — the deadloop requires ftrace softirq events with stack traces; no tracing program and no tracefs write grant are in the allowlist Affected range: Linux 5.8 through 6.6.127, 6.7–6.12.74, 6.13–6.18.13, and 6.19–6.19.3. 5.19.6 and 6.18.9-hs are both in range. Fixed in 6.6.128, 6.12.75, 6.18.14, 6.19.4, and 7.0 Upstream fix: d41e37f26b31 (mainline); stable 6.18.14+

This CVE describes an infinite recursion in rcu_read_unlock_special(). After commit 5f5fa7ea89dc removed recursion protection from __rcu_read_unlock(), that function can call raise_softirq_irqoff(RCU_SOFTIRQ) without a pending-work flag. When ftrace is recording the softirq raise and walking the stack, the unwind re-enters rcu_read_unlock_special() and the CPU deadloops. The impact is a local denial of service. The path does not escalate to root.

CONFIG_PREEMPT_RCU=y, CONFIG_TREE_RCU=y, CONFIG_FTRACE=y, and CONFIG_TRACING=y are compiled in on both fielded kernels. That is not enough. The deadloop is not the ordinary RCU unlock path. It requires the irq/softirq tracepoints to be armed with stack traces. Those controls live under /sys/kernel/tracing/. Enabling them takes a tracing program (trace-cmd, perf) or a write to that filesystem. None of those programs appear in the allowlist. Default allowlist directory grants are /usr/lib and /etc only. File-open write checks refuse any other path. An attacker who has already gained root cannot add a tracer: Lockdown returns EPERM for FS_IOC_SETFLAGS, mount(), fsmount(), and move_mount(), so the allowlist cannot be expanded and a bind-mount cannot cover /sys/kernel/tracing/.

The trigger cannot be reached on any default Root Lock deployment.

If your deployment adds trace-cmd, perf, or a program that writes /sys/kernel/tracing/ to the allowlist, treat this CVE as Affected at 7.1 HIGH for availability.

CVE-2026-46033

Status: 5.19.6 Not exploitable — feature not compiled; 6.18.9-hs Not exploitable — tool not in the program allowlist Component: IPsec authencesn (CONFIG_CRYPTO_AUTHENC) Base Score: 7.1 HIGH (CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H) Score on Root Lock: 0.0 — 5.19.6 does not compile the AF_ALG AEAD interface the bug uses; 6.18.9-hs compiles that interface as a module, and kernel.modules_disabled=1 refuses a later load. ip, setkey, and IPsec daemons are also absent from the allowlist

authencesn requires a zero authsize or an authsize of at least 4 bytes because the ESN encrypt and decrypt paths always move 4 bytes of high-order sequence number at the end of the authenticated data. Instance creation copied the inner ahash digest size into the default authsize without rejecting the invalid 1..3 range. Binding that instance through AF_ALG then ran the ESN tail handling with a too-short tag and hit an out-of-bounds read.

Both Root Lock kernels are in the NVD range (Linux 4.11 through 6.18.26) and compile crypto/authencesn.c via CONFIG_CRYPTO_AUTHENC. The unprivileged trigger is AF_ALG (CONFIG_CRYPTO_USER_API_AEAD) after instantiating authencesn with a 1..3-byte ahash such as cbcmac(cipher_null) from the CCM template. 5.19.6 has CONFIG_CRYPTO_USER_API_AEAD and CONFIG_CRYPTO_USER not set, so that userspace crypto path is not present. 6.18.9-hs builds AF_ALG AEAD, authenc, CCM, and IPsec ESP as modules. Reaching the path requires those modules to be loaded. After boot, kernel.modules_disabled=1 refuses a later load. See How to read the backstop sections. The program allowlist does not include ip, setkey, or IPsec daemons.

The trigger cannot be reached on any Root Lock deployment.

CVE-2026-46045

Status: Not Affected on 5.19.6; Not exploitable — feature not compiled on 6.18.9-hs Component: MD last-level bitmap (CONFIG_MD_LLBITMAP) Base Score: 7.8 HIGH (CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) Score on Root Lock: 0.0 — CONFIG_MD_LLBITMAP is not compiled; 5.19.6 is outside the affected range

The bug is in md-llbitmap page reads: the code picks the first assigned non-faulty rdev without checking In_sync, so bitmap pages can be read from a spare still being rebuilt. That path exists only in drivers/md/md-llbitmap.c under CONFIG_MD_LLBITMAP.

NVD lists Linux 6.18 through 6.18.26 and 6.19 through 7.0.3. HeartSuite 5.19.6 is outside that range and the option is not present. HeartSuite 6.18.9-hs is in range and has # CONFIG_MD_LLBITMAP is not set. Classic MD bitmap (CONFIG_MD_BITMAP) is a different file and is not this bug.