CVE Vulnerability Catalog

In the Linux kernel, the drm/amdgpu driver changed the handling of the vm->task_info object. The object is now dynamically allocated and reference-counted, improving safety and stability.

In the Linux kernel's networking subsystem, a dangling sk pointer is left when socket creation fails, potentially leading to use-after-free when __sock_release() is called with an fentry probe.

In the Linux kernel, in the sc16is7xx serial driver, a TX FIFO corruption was fixed that could occur when a packet was received on channel A at almost the same time as a packet was about to be transmitted on channel B.

In the Linux kernel, in the Btrfs filesystem, an issue with readahead on the relocation inode was fixed when the filesystem uses a RAID stripe tree (RST), which could cause ENOENT errors.

In the Linux kernel, in the L2TP subsystem, a possible tunnel refcount underflow was fixed, which could occur during session deletion when the tunnel pointer was invalid.

In the Linux kernel, in the Bluetooth subsystem (MGMT), possible deadlocks were fixed that could occur due to hci_cmd_sync_dequeue, causing threads to block for extended periods.

In the Linux kernel, in the DRM/XE driver (guc_submit), a race condition around the suspend_pending field was fixed, which could cause assertions and warnings in certain test scenarios.

In the Linux kernel, in the Bluetooth subsystem (hci_conn), the function disable_delayed_work_sync is used instead of cancel_delayed_work_sync, which not only cancels ongoing work but also disables new submissions, which is desirable.

In the Linux kernel's F2FS filesystem, there is potential corruption when moving a directory, similar to an issue in ext4. The fix addresses this problem.

In the Linux kernel, in the NFS server (nfsd), a race between the laundromat thread and the free_stateid operation was fixed, which could lead to incorrect handling of expired delegations.

In the Linux kernel, an overflow in the ring-buffer's __rb_map_vma function was fixed, which occurred when calculating the number of pages (nr_pages). A check was added before the calculation to avoid the issue.

In the Linux kernel, in the RDMA/hns driver, a soft lockup issue under heavy CEQE load was fixed by moving CEQE handling to the BH (bottom half) context.