| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| Zephyr's WireGuard VPN data-plane receive handler wg_process_data_message() in subsys/net/lib/wireguard/wg_crypto.c validated the anti-replay counter too late. After AEAD decryption of a MESSAGE_TRANSPORT_DATA packet succeeded, the code committed several peer-state changes — update_peer_addr() (endpoint roaming update), the keypair->last_rx/peer->last_rx liveness timers, and keypair_update() (promote next→current and destroy the previous keypair) — and only afterward called wg_check_replay(). On a replayed packet the replay check returned -EINVAL, but none of the preceding mutations were rolled back.
The AEAD tag authenticates content but not freshness, so a replayed-but-authentic transport packet decrypts correctly. An attacker who captures one valid ciphertext off the wire (an on-path or shared-medium observer) can re-inject it from an arbitrary spoofed source address. Reaching the handler requires no credentials: it is driven directly from inbound UDP datagrams via the dispatch in subsys/net/lib/wireguard/wg.c.
Because the state mutations committed before the replay check, the replay repoints the peer endpoint to the attacker-chosen source address (roaming hijack), redirecting the victim's subsequent outbound tunnel traffic until the legitimate peer's next packet re-corrects it; it also prematurely destroys the previous keypair and refreshes the RX liveness timer. The tunnel payload stays encrypted under the session keypair, so this is an integrity/availability impact (traffic redirection and session disruption), not payload disclosure. The fix moves wg_check_replay() to immediately after a successful decrypt, before any peer-state mutation, matching the WireGuard specification and the Linux reference implementation. |
| Zephyr's WireGuard implementation in subsys/net/lib/wireguard/wg_crypto.c mishandled keepalive packets. In wg_process_data_message(), any type-4 transport-data message whose payload was exactly 16 bytes (an empty plaintext plus a bare Poly1305 tag, i.e. a keepalive) was accepted and returned immediately, before wg_decrypt_packet() was ever called. The Poly1305 authentication tag was therefore never verified; the only preceding gates were a cleartext receiver-index lookup (get_peer_keypair_for_index() on the attacker-supplied data_hdr->receiver) and a non-cryptographic keypair validity/expiry check.
The path is reachable entirely from the network: inbound UDP on the WireGuard port is dispatched by wg_input() to handle_transport_data() and then wg_process_data_message(). The 32-bit receiver index is transmitted in cleartext in WireGuard handshake and data messages, so an on-path observer learns it directly and an off-path attacker can brute-force it against the UDP port. Given an active receiving-valid session for that index, an attacker could send a 16-byte garbage payload and have it accepted without possessing the session key.
On acceptance the unauthenticated message caused the management layer to observe a spoofed NET_EVENT_VPN_CONNECTED signal (setting peer->first_valid and notifying any net_mgmt listener) and incremented the keepalive-RX statistic. The impact is limited to integrity of this status signal: no plaintext is decrypted or injected, no key is disclosed, and the early-return path did not update the peer endpoint or liveness timers, so there is no traffic-injection, session-takeover, or availability consequence.
The fix removes the pre-decrypt early return so a 16-byte payload flows through wg_decrypt_packet(), which verifies the Poly1305 tag over the empty plaintext, followed by the existing anti-replay check; only an authenticated, non-replayed message is then recognised as a keepalive. Forged keepalives now fail the tag check and are counted as decrypt failures. |
| The LoRaWAN application-layer clock-synchronization service parses downlinks in clock_sync_package_callback() (subsys/lorawan/services/clock_sync.c). Its command loop only guarantees that the one-byte command id is in bounds; for the CLOCK_SYNC_CMD_APP_TIME (AppTimeAns) command the handler then reads a 4-byte time correction via sys_get_le32() plus a 1-byte token without checking that 5 bytes remain in the receive buffer (len - rx_pos). A short or crafted AppTimeAns therefore reads up to 5 bytes past the end of the decrypted payload.
The payload (rx_buf/len) is the decrypted application frame delivered to the registered downlink callback (mcps_indication->Buffer/BufferSize). Reaching the handler requires a frame on the clock-sync port that passes LoRaWAN's MAC integrity check and FRMPayload decryption, so the practical attacker is a malicious or compromised network/application server (the designated sender of AppTimeAns) or a party holding the session keys, rather than an arbitrary radio listener.
The over-read is bounded: the backing store is a fixed 255-byte static buffer, so the few stray bytes do not fault, and the read values (time_correction, token) are used only internally and never transmitted, so there is no disclosure to the attacker and no crash. The sole effect is that a stale token matching ctx.req_token can apply a garbage time_correction to the device's own clock offset (ctx.time_offset), a minor integrity impact confined to the victim's time estimate. The fix adds an explicit length check that drops a too-short AppTimeAns. Note the sibling one-byte reads in the periodicity and force-resync handlers remain unguarded with the same negligible impact. |
| The LoRaWAN TS004 Fragmented Data Block Transport handler frag_transport_package_callback() in subsys/lorawan/services/frag_transport.c parses downlink command bytes without validating that enough payload bytes remain before each access. The loop's only bound is rx_pos < len; after consuming the one-byte command id the handler cast rx_buf + rx_pos to a 10-byte struct frag_transport_setup_req, and for a DATA_FRAGMENT command passed &rx_buf[rx_pos] to the fragment decoder, which reads exactly ctx.frag_size bytes — with no remaining-length check in either case.
The fragment size is attacker-chosen in a preceding FRAG_SESSION_SETUP command (ctx.frag_size = req->frag_size, capped at CONFIG_LORAWAN_FRAG_TRANSPORT_MAX_FRAG_SIZE, default 232). rx_buf aliases the 255-byte static MacCtx.RxPayload buffer in the loramac-node MAC layer, while len is the actual decrypted payload length. By padding a downlink with mismatched-index DATA_FRAGMENT filler commands (each advancing rx_pos by three bytes without producing an answer) and appending one matching-index fragment near the end of the payload, an attacker can make the decoder read up to roughly frag_size bytes past the end of RxPayload, copying adjacent static memory into the decoder buffers and the FUOTA flash image.
The handler runs only on downlinks that have already passed the LoRaWAN frame MIC and FRMPayload decryption, so the defect is reachable only by a party holding the device's session keys (the FUOTA server or an attacker who has compromised those keys). The out-of-bounds bytes are never returned to the sender — the only uplink emitted is a status answer carrying fragment counts — so there is no direct disclosure channel, and on typical flat-memory LoRaWAN MCUs the over-read stays within mapped memory, making a crash unlikely. The impact is therefore a bounded out-of-bounds read with limited confidentiality consequence and no write or control-flow primitive. The fix adds remaining-length guards before each access. |
| The IEEE 1588 PTP management-message parser in subsys/net/lib/ptp/tlv.c mishandles the PTP_MGMT_TIME management id. In tlv_mgmt_post_recv(), the PTP_MGMT_TIME case casts mgmt_tlv->data to a 10-byte struct ptp_timestamp and reads it (then byte-swaps and writes it back) without first checking that the TLV data field is at least sizeof(struct ptp_timestamp). Every sibling management id in the same switch validates its length first; PTP_MGMT_TIME was the only case lacking that check.
The length passed in is the management data size (tlv->length - 2), and the upstream guard in ptp_tlv_post_recv() only requires tlv->length > 2, while msg_tlv_post_recv() validates only that the TLV fits within the received byte count, not a per-id minimum. A peer on the local PTP segment can therefore send a PTP_MSG_MANAGEMENT message carrying a short PTP_MGMT_TIME TLV (data as small as 2 bytes), causing the parser to read and write 8 bytes beyond the validated data. The message type and TLV contents are taken straight off the wire, so the path is reachable by any adjacent attacker when CONFIG_PTP is enabled.
The over-read and write-back stay within the struct ptp_msg allocation (mgmt_tlv->data lives in the leading mtu[NET_ETH_MTU] union member, so data + 10 lands at most a few bytes past mtu[], inside the same object), so this is an out-of-bounds read of adjacent in-object memory plus a bounded in-place corruption of the message's parsed timestamp, not past-allocation memory corruption. Impact is limited to minor information exposure of adjacent bytes and corruption of the device's parsed management TIME value; there is no crash on the access and no reachable reference-count corruption.
The fix adds if (length < sizeof(struct ptp_timestamp)) { return -EBADMSG; } before the cast, matching the other management-id cases and fully closing the receive-path defect. |
| The OCPP 1.6 client in subsys/net/lib/ocpp/ocpp_j.c contains a stack buffer overflow in parse_getconfig_msg(). When handling a GetConfiguration request from the central system, the handler copied the attacker-controlled JSON "key" string into the caller's fixed 50-byte stack buffer (skey[CISTR50], declared in subsys/net/lib/ocpp/ocpp.c) using an unbounded strcpy(). The parsed key value points directly into the receive buffer, so its length is bounded only by the message size (CONFIG_OCPP_RECV_BUFFER_SIZE, default 2048).
The GetConfiguration message is delivered over the WebSocket connection that the charge point opens to its configured central system. The reader thread ocpp_wsreader() reads the message into ui->recv_buf and dispatches it to parse_getconfig_msg() via the PDU function table. An attacker who controls the central system endpoint, or a man-in-the-middle on an unencrypted connection, can send a GetConfiguration request whose "key" field exceeds 50 bytes and overflow the reader thread's stack with attacker-chosen bytes.
The consequence is a remotely triggerable stack smash on the OCPP reader thread: at minimum a denial of service, and plausibly remote code execution depending on build-time hardening such as stack canaries and MPU configuration. The fix replaces the strcpy() with a bounded strncpy(key, payload.key[0], CISTR50 - 1) followed by explicit NUL termination, matching the bounded copies already used by the sibling handlers. |
| The Zephyr ext2 filesystem driver fails to validate the s_log_block_size field of the on-disk superblock when mounting a filesystem. ext2_verify_disk_superblock() in subsys/fs/ext2/ext2_impl.c checks the magic number, revision, inode size and group counts, but never bounds s_log_block_size. On a successful verify, subsys/fs/ext2/ext2_ops.c computes fs->block_size = 1024 << superblock.s_log_block_size from this attacker-controlled uint32_t, so a crafted value either overflows the shift (undefined behaviour) or yields a block size far larger than CONFIG_EXT2_MAX_BLOCK_SIZE.
That block size is then passed to k_mem_slab_init() by ext2_init_blocks_slab() to carve CONFIG_EXT2_MAX_BLOCK_COUNT blocks out of the fixed static buffer __ext2_block_memory_buffer, whose size is CONFIG_EXT2_MAX_BLOCK_COUNT * CONFIG_EXT2_MAX_BLOCK_SIZE. k_mem_slab_init() does not verify that the requested blocks fit the buffer, and the ext2 wrapper discards its return value, so the slab is laid out past the end of the static buffer. The mount immediately reads block-group, bitmap and inode blocks of fs->block_size bytes each into these slab blocks, producing an out-of-bounds write into adjacent static memory on the first block read.
The entire path is gated only by data read from the mounted image, making this reachable by any attacker who can present a crafted ext2 image to a device that mounts it (for example a removable SD card or storage medium). Because the ext2 driver runs in kernel mode, supplying image bytes yields a supervisor-mode memory-corruption primitive, with impact ranging from denial of service to potential code execution.
The fix rejects s_log_block_size values that overflow the shift (greater than 11) or that produce a block size exceeding CONFIG_EXT2_MAX_BLOCK_SIZE, so the block slab can no longer be initialized larger than its backing buffer. |
| The userspace syscall verifier z_vrfy_mbox_send() in drivers/mbox/mbox_handlers.c validated the nested msg->data/msg->size fields by reading them directly out of live userspace memory, and then forwarded the original, still-mutable userspace struct mbox_msg * pointer to z_impl_mbox_send() and the underlying driver. Between the access check and the driver's use of msg->data, the validated pointer could be replaced, leaving a time-of-check/time-of-use window.
On a system built with CONFIG_USERSPACE, any unprivileged userspace thread may invoke the mbox_send() system call. A second thread sharing the caller's address space can race to overwrite msg->data with a supervisor (kernel) address after the verifier's bounds check has passed but before the driver dereferences it. The driver then reads from the attacker-chosen address in supervisor context (for example memcpy(&data32, msg->data, msg->size) in the NXP mailbox driver, whose bytes are subsequently emitted to the peer mailbox endpoint).
The impact is a userspace-to-supervisor access-control bypass: disclosure of kernel memory contents (high confidentiality impact), or, for an invalid/unmapped target address, a faulting kernel read causing denial of service. The fix snapshots the entire struct mbox_msg into a kernel-stack copy with k_usermode_from_copy() and validates and forwards that immutable copy, closing the race. |
| The Zephyr ext2 filesystem driver validates the on-disk block bitmap in ext2_init_fs() (subsys/fs/ext2/ext2_impl.c) by passing fs_blocks = s_blocks_count - s_first_data_block to ext2_bitmap_count_set(). That helper (subsys/fs/ext2/ext2_bitmap.c) treats its argument as a number of bits and reads one bitmap byte per eight bits, but the bitmap buffer (BGROUP_BLOCK_BITMAP) is a single fetched block of only fs->block_size bytes (capacity fs->block_size * 8 bits). s_blocks_count and s_first_data_block are taken verbatim from the superblock and were never bounded against this single-group capacity; ext2_verify_disk_superblock() checks the magic, revision, and block-size shift but not the block count.
A crafted ext2 image with an oversized s_blocks_count (up to ~4 billion, against a maximum 4096-byte block / 32768-bit bitmap) makes ext2_bitmap_count_set() scan roughly 512 MB of memory past the bitmap block — a large out-of-bounds read of the static block slab and adjacent memory.
The defect is reached during mount: ext2_init_fs() is invoked from ext2_mount() (subsys/fs/ext2/ext2_ops.c), the registered .mount operation. Any path that mounts an attacker-supplied ext2 image (removable media, a disk/flash partition, or a downloaded image) triggers it. The kernel-privileged parser operates on attacker-controlled data, so the bug is exploitable wherever untrusted ext2 media can be mounted.
Impact is an out-of-bounds read only: the resulting bit count is compared internally and the mount is rejected, so no attacker-controlled bytes are returned (not a useful information leak). The ~512 MB over-read will almost certainly cross an unmapped or MPU-protected boundary and fault, crashing the system — a denial of service triggered by mounting a single malformed image. The fix rejects any image whose fs_blocks exceeds fs->block_size * 8 before the scan. |
| The OCPP 1.6 client in subsys/net/lib/ocpp/ocpp.c reconstructs a session handle and PDU id from the uid field of a CALLRESULT message. In ocpp_process_server_msg() the code calls atoi(strtok_r(uid, "-", &tmp)) without checking the strtok_r return value. When the server-supplied uid is empty or contains no - delimiter, strtok_r() returns NULL and atoi(NULL) dereferences a NULL pointer, which is undefined behaviour.
The uid originates from network data: parse_rpc_msg() in subsys/net/lib/ocpp/ocpp_j.c JSON-parses a frame received from the OCPP central system over TCP/WebSocket and copies the server-controlled string into the local buffer. A malicious or compromised central system, or a man-in-the-middle on a non-TLS ws:// connection, can return a malformed uid to reach the defect. No authentication beyond the existing server connection (or MITM position) is required, and the reconstructed pointer is membership-validated by ocpp_session_is_valid(), so the impact is limited to the NULL dereference rather than arbitrary pointer use.
On Zephyr targets that trap access to address 0 (MMU/MPU platforms or CONFIG_NULL_POINTER_EXCEPTION_DETECTION), the dereference faults inside the OCPP reader thread and invokes the fatal handler, producing a remote denial of service of the charge point; on bare targets where address 0 is readable the call returns 0 and is benign, so the impact is availability-only and platform-conditional.
The applied fix guards only the first atoi(); the second strtok_r(NULL, "-", &tmp) followed by pdu = atoi(buf) in the same function remains unguarded and the identical NULL dereference is still reachable from the same network input when the uid has a first token but no second --delimited token. A complete fix should validate the second token as well. |
| The virtio PCI driver (drivers/virtio/virtio_pci.c) parses a device's PCI capability list during driver initialization. In virtio_pci_read_cap() the device-supplied capability length byte cap_len (read from PCI config space via pcie_conf_read()) was only checked with assert(tmp.cap_len == cap_struct_size). That assert resolves to __ASSERT_NO_MSG(), gated by CONFIG_ASSERT, which defaults off in production builds, so the value reached the copy logic completely unvalidated.
The length then drives a loop that copies extra capability dwords into a fixed-size stack buffer supplied by the caller. A cap_len below the 24-byte base struct virtio_pci_cap underflows the unsigned extra_data_words count to a near-SIZE_MAX value, producing an effectively unbounded stack write; a cap_len above the caller's buffer (up to 255) writes up to roughly 228 bytes of device-controlled data past the buffer. Both are out-of-bounds writes of attacker-controlled content executed in kernel mode during boot-time device probe.
The input originates from the virtio device. In the common deployment where Zephyr runs as a guest under a hypervisor, the device backend is the host, which already fully outranks the guest, so the bug yields no privilege escalation. The exploitable case is a virtio device that is untrusted relative to the Zephyr kernel — an untrusted or physical/passthrough virtio PCIe device on a bare-metal system, or a confidential-computing posture where the guest must defend against the host — where a malicious device can corrupt the kernel stack and potentially achieve code execution or a crash.
The fix replaces the compiled-out assert with a runtime range check rejecting cap_len outside [sizeof(struct virtio_pci_cap), cap_struct_size] before any arithmetic or copy. |
| The Infineon Airoc Wi-Fi driver's transmit callback airoc_mgmt_send() in drivers/wifi/infineon/airoc_wifi.c allocates a net_buf from the fixed airoc_pool for every outbound packet. When whd_network_send_ethernet_data() returns a synchronous failure, the underlying WHD library does not take ownership of the buffer, but the pre-fix driver returned -EIO without releasing it. Each failed transmit therefore permanently leaks one buffer from the pool.
airoc_pool is small and fixed (AIROC_WIFI_TX_PACKET_POOL_COUNT + AIROC_WIFI_RX_PACKET_POOL_COUNT, default 20 buffers) and is shared by WHD's whd_host_buffer_get callback for both transmit and receive. Once enough send failures have leaked the pool dry, airoc_wifi_host_buffer_get() returns WHD_BUFFER_ALLOC_FAIL for all subsequent allocations, so both transmit and the WHD-driven receive path fail and Wi-Fi connectivity is lost until the device is rebooted.
The leak occurs only on the transmit error path. A Wi-Fi-adjacent attacker can influence the conditions that cause synchronous send failures (for example by deauthenticating/disassociating the station while the local stack continues to attempt transmits), and ordinary transient failures over the device's lifetime accumulate toward the same state. Reliable on-demand triggering is of high complexity and the impact is availability-only, but the resulting denial of service is permanent and non-recoverable without a reboot.
The fix releases the buffer with airoc_wifi_buffer_release() on the failure branch, returning it to the pool. The commit also removes a redundant k_sem_give() in airoc_mgmt_disconnect(); because data->sema_common is a binary semaphore (limit 1) the duplicate give merely saturated at 1 and had no security impact. |
| The UMP Stream responder library in lib/midi2/ump_stream_responder.c builds reply packets in a 16-byte struct midi_ump (uint32_t data[4]). The builders make_endpoint_info() and make_function_block_info() populate only the first two words (res.data[0] and res.data[1]) and, before this fix, declared their result as an uninitialised local (struct midi_ump res;). The remaining two words (res.data[2], res.data[3]) retain stale stack contents.
Endpoint Info and Function Block Info notifications are UMP Stream messages (UMP_MT_UMP_STREAM), which are 4 words long, so the full 16-byte packet — including the two uninitialised words — is transmitted verbatim by cfg->send(). The responder is driven by attacker-supplied UMP Stream Endpoint-Discovery / Function-Block-Discovery requests via ump_stream_respond(). In the in-tree Network MIDI 2.0 server (subsys/net/lib/midi2/netmidi2.c) these requests arrive as UDP datagrams and, with the default no-authentication endpoint, a remote peer can establish a session and trigger the responses; the same library also serves USB MIDI 2.0 hosts.
Each discovery request causes the device to disclose 8 bytes of its own uninitialised stack memory to the peer, and the request is freely repeatable. This is a confidentiality-only information leak (root cause is use of an uninitialised variable, CWE-457/CWE-908); the leaked words could include residual data or pointer values. There is no memory-corruption, integrity, or availability impact.
The fix zero-initialises both result structs (struct midi_ump res = {0};), so the trailing words are cleared before transmission. These are the only two responder builders that left trailing words unset (send_string() already zeroes its buffer), so the leak is fully closed. |
| The Hearing Access Service (HAS) GATT server in subsys/bluetooth/audio/has.c installs a connection-callback set unconditionally via BT_CONN_CB_DEFINE, so security_changed() runs for every connection that establishes security even before the application has called bt_has_register(). The service attribute pointers hearing_aid_features_attr, preset_control_point_attr, and active_preset_index_attr remain NULL until bt_has_register() resolves them and sets has.registered.
With CONFIG_BT_SETTINGS, settings_set_cb() restores each bonded client's persisted context at boot and unconditionally sets context->flags to BONDED_CLIENT_INIT_FLAGS (non-zero). When a previously bonded peer reconnects and re-establishes security during the startup window before bt_has_register() has been called, security_changed() sees the non-zero flags and schedules notify_work_handler, which calls bt_gatt_is_subscribed() with a still-NULL attribute pointer. That triggers an assertion (__ASSERT(attr, ...) in bt_gatt_is_subscribed()), or a NULL dereference of attr->uuid when assertions are compiled out.
The result is a remotely triggerable (Bluetooth, adjacent) crash of the HAS peripheral. Exploitation requires the peer to have previously bonded with the device and to reconnect within the boot-time race window before the application registers the service; a peer that reconnects persistently can prolong the outage. Impact is denial of service only, with no memory corruption or information disclosure.
The fix adds an early if (!has.registered) { return; } guard in security_changed(), so no notification work is scheduled until the GATT service is registered and its attribute pointers are valid. |
| The Zephyr virtio driver does not validate the descriptor-chain head id that the virtio device writes into the used ring. In virtio_isr() (drivers/virtio/virtio_common.c), the device-written vq->used->ring[idx].id is used directly as an index into vq->recv_cbs[] and vq->desc[], which are both allocated with exactly vq->num entries. recv_cbs[] holds {cb, opaque} callback entries, and the indexed callback pointer is then invoked as cbe.cb(cbe.opaque, used_len).
Because the id is consumed as a 16-bit value with no bound check, a malicious or compromised virtio backend (an untrusted hypervisor, or an untrusted hardware/peer-processor virtio device on a PCI or MMIO transport) can supply an id far beyond vq->num. This causes an out-of-bounds read of a {function pointer, argument} pair from heap memory beyond recv_cbs[], after which the driver calls that attacker-shaped pointer in the guest's interrupt context. No guest privileges or user interaction are required; the backend triggers it by writing the shared used ring and raising the queue interrupt.
The result is an arbitrary / attacker-influenced function-pointer call in the Zephyr guest, i.e. a control-flow-hijack primitive that can lead to code execution or, at minimum, a reliable crash. The fix rejects any used-ring id >= vq->num before indexing recv_cbs[]/desc[] or invoking the callback. This affects builds using CONFIG_VIRTIO with the PCI or MMIO transport. |
| The NVS backend of the Zephyr settings subsystem (subsys/settings/src/settings_nvs.c) reads stored setting-name entries into fixed 74-byte stack buffers and NUL-terminates them with buf[rc] = '\0', where rc is the return value of nvs_read(). Per its contract, nvs_read() returns the full stored entry length (wlk_ate.len), which can exceed the supplied buffer length — only MIN(len, stored_len) bytes are actually copied, but the return value may be much larger, bounded only by the NVS sector size. Three sites (settings_nvs_cache_match(), settings_nvs_load(), and settings_nvs_save()) used this value directly as the NUL index without clamping, so an oversized stored name entry causes a single \0 byte to be written past the end of the stack buffer at an attacker-influenced offset (CWE-787).
The oversized entry cannot arise through the normal settings API, where names are bounded by SETTINGS_MAX_NAME_LEN. It requires an actor able to write the flash that backs the settings partition — a co-resident or untrusted component sharing the flash device, a malicious settings image/restore, or offline/physical flash access (a shared-flash threat model). The malformed entry is parsed when settings_load() runs at boot or subsystem init, or during settings_save().
The out-of-bounds write is a single NUL byte at an offset equal to the crafted entry length (up to the NVS sector size), so the practical impact is a crash or denial of service and limited stack corruption rather than reliable code execution. There is no confidentiality impact, and the path is not reachable from the network through the ordinary settings interface. The fix skips any entry whose nvs_read() length is greater than or equal to the buffer size before performing the NUL store. |
| The IPv6 neighbor-discovery code in subsys/net/ip/ipv6_nbr.c processes the 6LoWPAN Context Option (6CO, RFC 6775) carried inside ICMPv6 Router Advertisements. In handle_ra_6co() the 8-bit context_len field is taken directly from the packet and was never bounded to the RFC maximum of 128. The function computes context->context_len / 8 and then performs memset(context->prefix + context_len, 0, sizeof(context->prefix) - context_len), where context->prefix is a fixed 16-byte array.
With context_len between 136 and 255 (and the option length field set to 3, which the pre-fix validation accepts), context_len / 8 evaluates to 17..31, so the memset length 16 - context_len/8 underflows the unsigned size_t argument to roughly SIZE_MAX. This produces an unbounded out-of-bounds memset that zeroes kernel memory well past the 6lo context structure.
The defect is reachable from unauthenticated, link-local input: any host on the same link can send a crafted Router Advertisement with a 6CO option. The RA handler validates only the option length field before calling handle_ra_6co(), so a single packet triggers the wild write. The code is compiled when CONFIG_NET_6LO_CONTEXT is enabled.
The impact is a reliable remote (adjacent) denial of service via memory corruption, with collateral integrity loss as the memset zeroes contiguous memory before the system faults. Router Advertisements are link-scoped and not forwarded, so the attacker must be on the same link (AV:A). The fix rejects any context_len greater than 128 before the length computation. |
| The HL7800 cellular modem driver's +CGCONTRDP: response handler on_cmd_atcmdinfo_ipaddr() in drivers/modem/vendor_standalone/hl7800.c parses the PDP-context dynamic parameters (local address, subnet mask, gateway, and DNS servers) that the cellular network assigns to the device. The response is linearized into a 256-byte stack buffer, after which each address field length is computed from comma/. delimiter positions in the network-supplied data and used directly as the length argument to strncpy() into the fixed 64-byte stack buffer temp_addr_str (and the 16-byte iface_ctx.dns_v4_string).
Because the field length is derived from attacker-controlled delimiter positions and was not bounded against the destination buffer, a single field can be far larger than 64 bytes. A malicious or impersonated cellular network (for example a rogue base station) can return a crafted +CGCONTRDP response with an overlong address field, causing strncpy() to write past temp_addr_str on the modem worker thread's stack, plus an out-of-bounds NUL write at temp_addr_str[addr_len].
No device-side privileges or user interaction are required: the device itself issues the AT+CGCONTRDP=1 query during normal network attach and parses whatever the network returns. The overflow corrupts adjacent stack memory in supervisor context, yielding at minimum a remotely triggerable crash and potentially control-flow hijacking on targets without stack protection.
The fix bounds every field length against its destination buffer (temp_addr_str and dns_v4_string) before each copy, rejecting overlong fields. |
| The Sierra Wireless HL7800 cellular modem driver (drivers/modem/vendor_standalone/hl7800.c, located at drivers/modem/hl7800.c in v4.4.0 and earlier) parses AT responses with roughly twenty handlers that call net_buf_linearize(value, sizeof(value), *buf, 0, len) into a 128-byte stack buffer and then write value[out_len] = 0. Because net_buf_linearize() (lib/net_buf/buf.c) can return a count equal to its destination-length argument, a field that exactly fills the buffer makes the terminating NUL land one byte past the end, a single-byte out-of-bounds write into adjacent stack memory.
The +KCELLMEAS cell-measurement handler on_cmd_atcmdinfo_rssi() is worse: it passed the wire length len as the destination size (net_buf_linearize(value, len, *buf, 0, len)), so a response line longer than 128 bytes overflows the value stack buffer with attacker-influenceable content. The line length comes from net_buf_findcrlf(), which accumulates bytes across the whole net_buf fragment chain and is not bounded to 128, so an over-long line reaches the defect.
The data originates from the cellular modem over UART, driven by the network: operator-scan results, +CGCONTRDP IP/DNS info, socket indications, and +KCELLMEAS neighbour-cell reports. An attacker able to shape what the modem emits — a rogue base station, a compromised modem baseband, or a remote peer feeding oversized response framing — can drive a line past 128 bytes. The handlers run in the driver's RX thread in kernel context, so the corruption is kernel-side.
The +KCELLMEAS path is a full stack buffer overflow whose worst case is code execution in kernel context and whose floor is a reliable crash; the remaining sites are single-byte NUL out-of-bounds writes. Exploitation requires the modem to emit an over-long AT response line, giving high attack complexity over an adjacent (cellular radio) vector. The fix passes sizeof(dst) - 1 (and correct explicit bounds for the IMSI and +KCELLMEAS sites) so the terminator always stays in bounds. |
| The Zephyr kernel validates the k_thread_join() and k_thread_abort() system calls (declared __syscall in include/zephyr/kernel.h) through thread_obj_validate() in kernel/thread.c. Its default switch branch is the access-denied path, taken when k_object_validate() returns -EPERM (the calling user thread was never granted access to the target thread object) or -EBADF (the supplied pointer is not a registered kernel object of the right type). That branch invoked K_OOPS(K_SYSCALL_VERIFY_MSG(ret, "access denied")), but K_SYSCALL_VERIFY_MSG treats a true expression as success; the non-zero error code ret therefore read as "verified OK", the kernel oops was never raised, and control fell through to CODE_UNREACHABLE.
Because k_thread_join() and k_thread_abort() are system calls, an unprivileged user-mode thread (under CONFIG_USERSPACE) can reach this denial path directly by calling either syscall on a thread object it does not own. Instead of the offending thread being cleanly terminated, execution reaches __builtin_unreachable() while running in supervisor mode inside the syscall handler.
On Clang builds CODE_UNREACHABLE emits an illegal-instruction trap, so a user thread can deterministically crash the kernel — a locally triggerable denial of service that escapes the userspace sandbox. On GCC builds the path is undefined behavior: the compiler may drop the return-value handling for thread_obj_validate(), so it can return an undefined bool; if that is false, the caller proceeds into the real k_thread_join()/k_thread_abort() implementation for a thread the user was never authorized to access, an access-control bypass.
The fix changes the verification expression to ret == 0, so a denied (non-zero) result now correctly raises K_OOPS and terminates the offending caller. |
