CVE Vulnerability Catalog

In the Linux kernel, in sched_ext, when a sub-scheduler enable fails before scx_link_sched(), the cleanup still runs the full scx_sub_disable(), which can race with root disable and lead to use-after-free. The fix skips the cgroup ownership reset and task walk if the scheduler was never linked.

In the Linux kernel bonding driver (ALB mode), bond_alb_monitor() reads primary_is_promisc under RCU, then drops RCU and takes RTNL via rtnl_trylock() before undoing promiscuity on the active slave. In that window the active slave can change, leading to a null-pointer dereference (kernel panic) or an underflow of the promiscuity counter, leaving the interface in IFF_PROMISC.

In the Linux kernel, in the enic driver, enic_remove does not cancel the tx_hang_reset work, which can lead to use-after-free after free_netdev(). The fix uses disable_work_sync() instead of cancel_work_sync() to block re-scheduling of the work.

In the Linux kernel, in IPVS, the function ip_vs_nat_icmp can write out of bounds because the ihl field in the IPv4 header is read multiple times, and a local attacker can modify the packet during processing. The fix reads ihl only once and uses ciph->len as the offset to the transport header.

In the Linux kernel, the amdxdna_insert_pages() function in the accel/amxdna driver has two error paths that incorrectly call vma->vm_ops->close(vma) before returning an error code. This causes premature resource release, and later unmapping by the process leads to a double close and reference count underflow.

In the Linux kernel BPF subsystem, when handling scalar += pointer in adjust_ptr_min_max_vals(), copying only selected fields of pointer state is fragile because pointer provenance is tracked by several bpf_reg_state fields. The fix uses the caller's temporary offset register to preserve the scalar operand while replacing the destination with the full pointer state.

In the Linux kernel's mlx5 driver, a vulnerability exists where the firmware tracer creation function can return either NULL or ERR_PTR. The return value is stored without a check, and during core dump generation, an invalid pointer is dereferenced, causing a kernel panic.

A vulnerability in the Linux kernel's BPF conntrack kfuncs was found, where the opts argument can point to a shared map value. Inconsistent reads of the netns_id field during acquisition and release of network namespace references can lead to reference imbalance, potentially destroying a live namespace and crashing the system.

A use-after-free vulnerability was found in the Linux kernel's bpf_iter_tcp_established_batch() for TCP_NEW_SYN_RECV sockets. The request_sock is published on the ehash chain before its refcount is set, and the BPF iterator uses plain sock_hold() instead of refcount_inc_not_zero(), leading to a race that overwrites the refcount and causes use-after-free.

A vulnerability in the Linux kernel's vhost_iotlb mechanism allows IOTLB entries to be allocated beyond the configured limit, leading to uncontrolled memory consumption. Affected components include vhost, vDPA, VDUSE, and vhost-vDPA devices that may create tables with zero or no limits. The fix enforces the limit and rejects invalid values.

A type confusion vulnerability was found in the Linux kernel's hwmon driver for PMBus devices. The pmbus_notify() function incorrectly casts different attribute types to sensor_device_attribute, which may lead to an out-of-bounds read or incorrect mask matching during alerts. The fix changes the attribute structures and sets the index to -1 for attributes that do not trigger notifications.

In the Linux kernel, AF_XDP accepts a TX metadata length as small as eight bytes, but every supported request needs flags plus at least one eight-byte field. Such short metadata can let the kernel read beyond the registered area.

In the Linux kernel, AF_XDP launch-time metadata extends beyond the first 16 bytes of struct xsk_tx_metadata. The vulnerability allows rejecting the request when the registered metadata area does not contain the complete field. The fix also snapshots validated flags for the generic transmit path to avoid inconsistent decisions if user space changes flags concurrently.

In the Linux kernel, the zero-copy path validates TX metadata while obtaining the descriptor context, then reads it again later when preparing the hardware request. User space can change the metadata between those operations and bypass the original validation. The fix validates metadata in xsk_tx_metadata_request() and uses the resulting flags snapshot for every feature check.

In the Linux kernel, the sch_cake ACK filter contains a WARN_ON(1) that can be triggered by a malformed IP header. Depending on system configuration, this leads to spamming dmesg with warnings or a panic if panic_on_warn is set. The fix removes the WARN_ON because the code already correctly skips the offending packet.

In the Linux kernel, the vhost-scsi subsystem has a vulnerability due to missing validation of T10 PI scatterlist counts. A malformed request can claim protection bytes that cover or exceed the payload length, leading to a zero data SGL count reaching sg_alloc_table_chained(), which triggers BUG_ON(!nents).

A vulnerability in the Linux kernel's vhost-scsi allows feature changes after endpoint activation, leading to a kernel panic. Enabling T10-PI after setup leaves protection scatterlists uninitialized, triggering a BUG_ON in sg_alloc_table_chained().

In the Linux kernel, the openvswitch module's key_extract() function does not check the Ethernet header length before accessing it, which can lead to a kernel BUG in __skb_pull() for packets arriving on ARPHRD_NONE devices such as TUN. The fix is to call check_header() before accessing the Ethernet header.

A vulnerability was found in the Linux kernel's network filter mechanism (net/sched). The issue is a race condition during filter creation that can lead to use-after-free (UAF) in the cls_u32 class. The fix ensures rtnl_lock is always acquired when destroying locked classifiers.

In the Linux kernel, the bnxt_en driver has a vulnerability causing data corruption. When EOP (end of frame padding) is enabled on the AGG ring, zero padding may overlap with the next segment's data, especially with Relaxed Ordering (RO) enabled. On older chips (P5 and older), RO is not automatically disabled, potentially overwriting valid data.