No fix yet.
CVEs with no patch, advisory or fixed version linked, ranked by severity and exploit odds.
-
CVE-2026-99531
In the Linux kernel, the following vulnerability has been resolved: ksmbd: verify transform SessionId matches the decrypted header When an encrypted request comes in, the decryption key is picked by the SessionId in the transform header, but the request is then authorized under whatever session the decrypted inner header names. Nothing ever compares the two, so on a connection with more than one session a client can get a request decrypted with one session's key and run it as another session. Encrypted requests are exempt from signing too, so a valid AEAD tag is the only proof of session identity, and it is checked against the wrong session. MS-SMB2 says the server must check that the transform SessionId matches the one in the decrypted SMB2 header and treat a mismatch as a protocol error. Add that check in __handle_ksmbd_work() after decryption, before dispatching: the message has to be at least one fixed SMB2 header, the first operation must not set SMB2_FLAGS_RELATED_OPERATIONS and its SessionId must match the transform SessionId, and every following non-related operation in a compound chain must carry the same SessionId. NextCommand offsets must be 8-byte aligned and stay inside the message; the bounds check is done by subtraction so it cannot wrap. When the decrypted payload is a compression transform, the existing decompression step runs first and the decompressed SMB2 message is checked instead, since a compression transform header carries no SessionId.
-
CVE-2026-99516
In the Linux kernel, the following vulnerability has been resolved: nvmet-fc: fix use-after-free in nvmet_fc_ls_create_connection() nvmet_fc_ls_create_connection() looks up the association with nvmet_fc_find_target_assoc(), which returns it with a reference held, and drops that reference as soon as the target queue has been allocated. iod->assoc is then dereferenced again by nvmet_fc_makeconnid() while building the connection id. The queue only borrows the pointer and takes no reference of its own. Once the lookup reference is dropped the association is kept alive solely by the reference taken when it was created, so a disconnect LS or a remote port unregister running concurrently can free it before nvmet_fc_makeconnid() reads it: BUG: KASAN: slab-use-after-free in nvmet_fc_handle_ls_rqst_work (drivers/nvme/target/fc.c:193 drivers/nvme/target/fc.c:1841 drivers/nvme/target/fc.c:1995 drivers/nvme/target/fc.c:2021) Read of size 8 at addr ffff888179f1b000 by task kworker/u9:2/1167 Hardware name: QEMU Ubuntu 24.04 PC v2 (i440FX + PIIX, arch_caps fix, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 Workqueue: nvmet-wq nvmet_fc_handle_ls_rqst_work Call Trace: <TASK> dump_stack_lvl (lib/dump_stack.c:94 lib/dump_stack.c:120) print_report (mm/kasan/report.c:378 mm/kasan/report.c:482) kasan_report (mm/kasan/report.c:595) nvmet_fc_handle_ls_rqst_work (drivers/nvme/target/fc.c:193 drivers/nvme/target/fc.c:1841 drivers/nvme/target/fc.c:1995 drivers/nvme/target/fc.c:2021) process_scheduled_works (kernel/workqueue.c:3322 kernel/workqueue.c:3405) worker_thread (kernel/workqueue.c:3486) kthread (kernel/kthread.c:436) ret_from_fork (archkernel/process.c:158) ret_from_fork_asm (arch/x86/entry/entry_64.S:245) </TASK> Freed by task 1167 on cpu 0 at 405.998150s: kasan_save_track (mm/kasan/common.c:57 mm/kasan/common.c:78) kasan_save_free_info (mm/kasan/generic.c:584) __kasan_slab_free (mm/kasan/common.c:253 mm/kasan/common.c:285) kfree (include/linux/kasan.h:235 mm/slub.c:2705 mm/slub.c:6405 mm/slub.c:6720) nvmet_fc_handle_ls_rqst_work (drivers/nvme/target/fc.c:1808 drivers/nvme/target/fc.c:1995 drivers/nvme/target/fc.c:2021) process_scheduled_works (kernel/workqueue.c:3322 kernel/workqueue.c:3405) worker_thread (kernel/workqueue.c:3486) kthread (kernel/kthread.c:436) ret_from_fork (archkernel/process.c:158) ret_from_fork_asm (arch/x86/entry/entry_64.S:245) The buggy address belongs to the object at ffff888179f1b000 which belongs to the cache kmalloc-2k of size 2048 The buggy address is located 0 bytes inside of freed 2048-byte region [ffff888179f1b000, ffff888179f1b800) Hold the lookup reference until the response has been built, and drop it on the error path as well.
-
CVE-2026-99515
In the Linux kernel, the following vulnerability has been resolved: iommufd: Fix event queue release use-after-free iommufd_eventq_fops_release() drops the file's users reference before reading eventq->ictx. Once that reference is dropped, IOMMU_DESTROY in another thread can remove and free the event queue, so the later load accesses freed memory. Read ictx while the users reference still pins the event queue, then drop the reference and put the saved context. The context reference taken by iommufd_eventq_init() keeps ictx valid until the put. Fault and vEVENTQ file descriptors share this release path.
-
CVE-2026-99503
In the Linux kernel, the following vulnerability has been resolved: ata: sata_fsl: Fix use-after-free of host_priv on probe() failure sata_fsl_host_stop() releases hcr_base and host_priv: static void sata_fsl_host_stop(struct ata_host *host) { struct sata_fsl_host_priv *host_priv = host->private_data; iounmap(host_priv->hcr_base); kfree(host_priv); } ->host_stop() is called by the driver core through the ata_host_stop() devres action which ata_host_start() registers, so it is called both when the device is unbound and when probe() fails after the host has been started. The error_exit_with_cleanup label of sata_fsl_probe() releases hcr_base and host_priv as well, and it is reachable from the two device_create_file() calls which are done after the host has been activated. In that case sata_fsl_host_stop() runs on an already freed host_priv, resulting in a use-after-free and a double iounmap()/kfree(). Add a separate error label for the failures happening after the host has been activated, which only detaches the host and leaves hcr_base and host_priv to sata_fsl_host_stop(). Note that ata_host_detach() is dropped from error_exit_with_cleanup, as that label is now only reachable while host is still NULL.
-
CVE-2026-99505
In the Linux kernel, the following vulnerability has been resolved: apparmor: fix a use-after-free in aa_lookupn_ns() aa_lookupn_ns() takes a reference on the namespace it finds with aa_get_ns(), which is kref_get() on the reference count that the namespace is released through, without checking whether the namespace is still in use. A concurrent removal of a ":"-prefixed namespace looks the same namespace up and only then takes the parent lock to remove it, so the last reference can be dropped between the lookup and the aa_get_ns() call: the reference count then reports the increment on a zero count and the kernel classifies the following use of the object as a use-after-free. A user with CAP_MAC_ADMIN can trigger this by writing a ":<name>" removal and a ":<name>/<profile>" removal to /sys/kernel/security/apparmor/.remove concurrently. refcount_t: addition on 0; use-after-free. WARNING: lib/refcount.c:25 at refcount_warn_saturate+0xc0/0xe0, CPU#6 Call Trace: <TASK> aa_lookupn_ns+0xa9/0xc0 aa_remove_profiles+0x4fb/0x730 ? __kvmalloc_node_noprof+0x219/0x5c0 ? __pfx_aa_remove_profiles+0x10/0x10 ? aa_loaddata_alloc+0x65/0x1f0 ? _copy_from_user+0x2d/0x80 ? aa_loaddata_alloc+0x111/0x1f0 profile_remove+0x2ee/0x4a0 vfs_write+0x21e/0xcf0 ? rcu_core+0x29c/0x1740 ? timerqueue_linked_add+0x1f4/0x3d0 ? __pfx_vfs_write+0x10/0x10 ? mutex_lock+0x81/0xe0 ? __pfx_mutex_lock+0x10/0x10 ? __pfx_rcu_core+0x10/0x10 ? __pfx_hrtimer_update_next_event+0x10/0x10 ? mutex_unlock+0x7b/0xd0 ? fdget_pos+0x24d/0x4b0 ksys_write+0xf7/0x1c0 ? __pfx_ksys_write+0x10/0x10 ? restore_fpregs_from_fpstate+0x37/0xc0 do_syscall_64+0xf9/0x540 entry_SYSCALL_64_after_hwframe+0x77/0x7f </TASK> Fix by rejecting a namespace whose reference count is zero.
-
CVE-2026-99506
In the Linux kernel, the following vulnerability has been resolved: mm/hugetlb: fix use-after-free of a new anon folio For a fault in a private mapping with nothing in the page cache, hugetlb_no_page() allocates a folio, locks it and maps it. The folio has one reference, which belongs to the page table entry from then on. After dropping the page table lock the function still sets the migratable flag of the folio and unlocks it. Whoever zaps the entry in between frees the folio. The fault holds the fault mutex and the lock of the vma. MADV_DONTNEED and the first unmap of a hole punch, hugetlb_vmdelete_list(), do not take the fault mutex, and a private vma only has a lock if it owns a reservation map: not with MAP_NORESERVE, and not in the child after fork(). Then the flag is set in, and folio_unlock() is called on, a folio that is free or has a new owner. A surplus page goes back to the page allocator while it is locked: BUG: Bad page state in process init pfn:108e00 flags: 0x200000000000059(locked|uptodate|dirty|head|node=0|zone=2) page dumped because: PAGE_FLAGS_CHECK_AT_FREE flag(s) set A page of the pool can be given to the next user, of any file, who locks it and loses the lock to the late folio_unlock(): kernel BUG at mm/filemap.c:1532! hugetlb_no_page+0x111b/0x23e0 hugetlb_fault+0x3a9/0x14a0 An unprivileged process is enough: threads that write to a MAP_PRIVATE | MAP_NORESERVE hugetlb mapping, threads that call MADV_DONTNEED on it, and one that takes the freed pages for another file. Until commit 3a47c54f09c4 ("hugetlbfs: revert use i_mmap_rwsem for more pmd sharing synchronization") the fault held i_mmap_rwsem for reading, which everybody who zaps takes for writing. The lock that commit bf4916922c60 ("hugetlbfs: extend hugetlb_vma_lock to private VMAs") gave to private vmas closed the window again, but only for those with a reservation map. Be done with the folio before the page table lock is dropped, as hugetlb_wp() is with its copy. The function already unlocks page cache folios under that lock. Original fix generated by an LLM and sent in private by Josef. [[email protected]: simplify+move comment]
-
CVE-2026-99492
In the Linux kernel, the following vulnerability has been resolved: HID: wacom: Advertise SW_MUTE_DEVICE capability prior to registration Input subsystem guidelines require that device capabilities are advertised before the input device is registered. The Wacom driver was violating this by advertising the SW_MUTE_DEVICE capability post-registration in wacom_set_shared_values() (and duplicating it in device-specific setup cases). Resolve this by moving the SW_MUTE_DEVICE capability setup to wacom_setup_touch_input_capabilities() for all touch devices that support it. For generic touch devices whose capabilities depend on mute switch usages parsed from a sibling Pen/Pad interface, defer registration with -EPROBE_DEFER until the sibling has parsed its descriptors and initialized shared capabilities.
-
CVE-2026-99494
In the Linux kernel, the following vulnerability has been resolved: s390/zcrypt: Fix out-of-bounds ptr advance in cca_query_crypto_facility() The FQ reply parser code blindly advanced the walk pointer by a length value read directly from the hardware reply payload without checking that the advance stayed within the allocated reply buffer. A corrupt or malicious device response could push ptr beyond the cprbmem region, causing an out-of-bounds dereference or kernel memory exposure via the subsequent memcpy(). Fix this by tracking the remaining reply buffer space in a variable and validating each device-supplied length field against it before advancing or dereferencing the pointer.
-
CVE-2026-99497
In the Linux kernel, the following vulnerability has been resolved: s390/probes: Fix wrong opcode for cgrb fixup The last byte of the cgrb instruction is 0xe4, however the probes code compares against 0xe6. Fix it.
-
CVE-2026-99483
In the Linux kernel, the following vulnerability has been resolved: ALSA: pcmtest: handle multiple wraps in random capture fill The random capture helpers assume that the data generated for one timer tick wraps the DMA ring at most once. However, b_rw describes 200 ms of audio because the timer runs at 5 Hz, so it can be several times larger than a small ring. For 48 kHz, S16_LE, four-channel capture with a 512-frame buffer, the DMA ring is 4096 bytes while one tick generates 76800 bytes. After filling the 4096-byte tail, the existing code issues a second 72704-byte write from the ring base, writing 68608 bytes beyond the ring. Both interleaved and non-interleaved capture are affected. KASAN reports the first write into the adjacent free page as a use-after-free in the random-byte generator. The underlying defect is an out-of-bounds write from the live DMA ring, not a stale DMA allocation. Split each random write repeatedly at the interleaved ring or per-channel block boundary. Keep the hardware-pointer update exactly once after the complete logical write. The issue reproduced on three of three clean boots. With this change, the same vulnerable geometry completed six times without a sanitizer finding. Boundary, pattern-fill, no-timer, open/close, and module-reload controls also completed normally. snd-pcmtest is a test driver, and selecting random fill requires an administrator-controlled module parameter. The written values come from the kernel random generator; no privilege escalation or attacker-controlled write primitive was demonstrated.
-
CVE-2026-99488
In the Linux kernel, the following vulnerability has been resolved: klist: avoid accesses after waking klist_remove() klist_remove() waits until a node is unreferenced so that its caller can free the containing object. klist_release() currently publishes waiter->woken and wakes the waiter before its final accesses to the waiter and node. klist_remove() can then return, allowing its stack waiter and the containing object to be freed or reused while klist_release() is still running. In particular, bus_remove_driver() can free drv->p while __device_attach() walks the same bus klist with bus_for_each_drv(). On an arm64 Cortex-A72 system, an unpatched 7.2.0-rc3 kernel with CONFIG_PREEMPT_RT=y and CONFIG_KASAN=y reproduced the bug through the in-tree I2C/at24 path. KASAN reported a use-after-free in klist_dec_and_del() reached from klist_next()/bus_for_each_drv() while at24 was being unregistered. Clear n_klist and take a task reference before publishing woken. Use release/acquire accesses for that publication and wake the referenced task. The task reference keeps the waiter task alive if it returns and exits before wake_up_process().
-
CVE-2026-99489
In the Linux kernel, the following vulnerability has been resolved: HID: wacom: Redesign shared sibling data lifecycle The Wacom driver coordinates state between sibling interfaces of the same physical device using a shared structure 'wacom_shared' inside 'wacom_hdev_data'. The driver kept a volatile representative pointer 'data->dev' pointing to a sibling 'hid_device' for physical path comparisons during sibling matching. This pointer management is fragile. When the representative device is disconnected, wacom_remove_shared_data() failed to clear/update 'data->dev', leading to a Use-After-Free vulnerability when subsequent sibling probes dereference the dangling 'data->dev' pointer. Resolve this issue by redesigning the sibling data lifecycle: - Eliminate the volatile 'data->dev' representative pointer completely - Redesign 'wacom_hdev_data' to store stable static copies of the required attributes upon first allocation: 'phys' path string, 'vendor', 'product' IDs and the sibling's 'device_type' - Use these static attributes for stable sibling matching in wacom_are_sibling() and wacom_get_hdev_data() This ensures sibling matching remains safe and stable even if individual siblings are dynamically added or removed. To secure the lifecycle against concurrent probe/disconnect races: - Switch kref_put() to kref_put_mutex() in wacom_remove_shared_data() to serialize refcount drops with list traversal and lookup - Modify wacom_release_shared_data() to assume the list lock is already held Also, do not accumulate the 'device_type' capability flag during subsequent sibling probes. Keeping only the first probed sibling's device_type exactly preserves the original sibling matching behavior without introducing side effects.
-
CVE-2026-99490
In the Linux kernel, the following vulnerability has been resolved: HID: wacom: Fix Use-After-Free in wacom_bamboo_pad wacom_bamboo_pad_pen_event() accesses wacom->shared->pen locklessly relative to wacom_remove_shared_data() which nullifies it. This can lead to a Use-After-Free if the sibling device is removed while events are being processed. Resolve this by introducing RCU protection for pen and touch pointers: - Annotate 'pen' and 'touch' in wacom_shared struct with __rcu. - Wrap lockless readers in wacom_bamboo_pad_pen_event() with rcu_read_lock() and rcu_dereference(). - Update writers in wacom_sys.c using rcu_assign_pointer(). - Use rcu_dereference_protected for comparisons under wacom_udev_list_lock. - Also use rcu_access_pointer in wacom_sibling_pending() and wacom_mode_change_work() to avoid warnings (while lockless access in the latter remains a pre-existing issue).
-
CVE-2026-99491
In the Linux kernel, the following vulnerability has been resolved: HID: wacom: Fix Use-After-Free in wacom_intuos_pad wacom_intuos_pad() accesses wacom->shared->touch_input locklessly inside the interrupt handler context. If the Touch sibling device is disconnected, wacom_remove_shared_data() clears 'touch_input' outside any lock, creating a Time-of-Check to Time-of-Use (TOCTOU) race condition where a preempted reader in interrupt context dereferences the freed pointer, leading to a Use-After-Free. Resolve this by introducing RCU protection for the touch_input pointer: - Annotate 'touch_input' in wacom_shared struct with __rcu - Wrap all lockless readers in wacom_wac.c with guard(rcu)() and rcu_dereference() using a unified wacom_report_touch_mute() helper - Update writers in wacom_sys.c using rcu_assign_pointer() - Call synchronize_rcu() in wacom_remove_shared_data() to ensure all active RCU readers have finished before the input device is freed Also wrap wacom_set_shared_values() and touch/pen assignments in wacom_add_shared_data() inside the wacom_udev_list_lock to serialize concurrent probe assignments, and verify that 'shared->touch == hdev' before setting touch_input to prevent concurrent sibling probe state desynchronization.
-
CVE-2026-99475
In the Linux kernel, the following vulnerability has been resolved: genirq: Defer kfree(irqaction) until irq_thread_dtor() finishes When an interrupt thread exits abnormally via do_exit() (e.g. the interrupt handler oopses) without returning from irq_thread(), a concurrent call to free_irq() could free the irqaction before the irq_thread_dtor() exit task work runs. irq thread (oopsed) free_irq() ------------------- ---------- do_exit() exit_mm() complete_vfork_done() -------> kthread_stop() returns exit_files() ... kfree(action) exit_task_work() irq_thread_dtor() action->... <- use-after-free Track the pending exit work with the IRQTF_EXIT_WORK thread flag, and have __free_irq() wait for the bit to clear before kfree(). irq_thread_dtor() only runs on irq_thread abnormal exit. If that dtor also faults, free_irq() will hang indefinitely, but double IRQ thread faults require a reboot anyways. [ tglx: Add missing include ]
-
CVE-2026-99477
In the Linux kernel, the following vulnerability has been resolved: fs/proc/task_mmu: pin mm_users in do_pagemap_scan() The PAGEMAP_SCAN ioctl takes mmap_read_lock() on an mm obtained from /proc/<pid>/pagemap without holding a reference to mm_users. proc_mem_open() deliberately grabs only mm_count, so a consumer that touches the address space has to call mmget_not_zero() first. Every other consumer in this file does -- m_start(), do_procmap_query(), show_smaps_rollup() and pagemap_read() -- but do_pagemap_scan() does not. Without that reference the target can reach exit_mmap() concurrently, and exit_mmap() calls unmap_vmas() while holding mmap_lock only for reading, which does not exclude the scanner. For ordinary VMAs this is harmless, as free_pgtables() and remove_vma() run afterwards under mmap_write_lock. A shared hugetlb VMA is different: __hugetlb_zap_end() frees the VMA's hugetlb_vma_lock inside the read-locked phase. walk_hugetlb_range() can therefore load vma->vm_private_data, have the lock freed underneath it, and then down_read() a freed rw_semaphore. The load happens before the freeing side clears the pointer, so this is a use-after-free and not a NULL dereference. Any unprivileged user can reach it with no huge pages configured, as hugetlb_reserve_pages() allocates the lock before the VM_NORESERVE early return. Pin mm_users across the scan, the way the other consumers do. mmput() from the exiting task can then no longer reach exit_mmap() while a walk is in progress. [[email protected]: remove unneeded & undesirable test for !mm]
-
CVE-2026-99479
In the Linux kernel, the following vulnerability has been resolved: fbdev: atmel_lcdfb: Prevent FIFO work after removal The LCD controller IRQ schedules sinfo->task to recover from a FIFO underflow. Probe currently requests the IRQ before initializing the work, so an early interrupt can schedule an uninitialized work item. The probe error path and remove path cancel the work before freeing the IRQ. A FIFO-underflow interrupt can schedule the work again after that cancellation. framebuffer_release() may then free sinfo while the work is still pending. KASAN reported a use-after-free in atmel_lcdfb_task() when an underflow interrupt occurred during removal. With the IRQ freed before the work was cancelled, the same test completed without a kernel diagnostic. Initialize the work before requesting the IRQ. In both cleanup paths, free the IRQ before cancelling the work so that it cannot be scheduled again after cancellation.
-
CVE-2026-99480
In the Linux kernel, the following vulnerability has been resolved: drm/panthor: Fix TOCTOU race between GROUP_REGISTERED check and erase panthor_group_destroy() checks the GROUP_REGISTERED mark and then calls xa_erase() without holding the xarray lock across both. If the handle is destroyed by another thread in between, and a concurrent GROUP_CREATE reuses the freed ID, the first destroy erases and releases the new, still-initializing group, leading to a use-after-free in panthor_group_create(). Do the mark check and erase atomically under xa_lock().
-
CVE-2026-99481
In the Linux kernel, the following vulnerability has been resolved: Bluetooth: hci_vhci: Pin module for debugfs file operations An open debugfs file retains its file_operations pointer in the debugfs proxy state. However, force_suspend_fops, force_wakeup_fops, aosp_capable_fops, and force_devcoredump_fops have no .owner set to THIS_MODULE. As a result, full_proxy_open_regular() calls fops_get() which succeeds without taking a reference to the hci_vhci module. When userspace opens any of these debugfs files and later closes the /dev/vhci file descriptor (triggering vhci_release() and debugfs file removal), the open debugfs file descriptor remains valid in userspace. Because the module refcount was not incremented, the hci_vhci module can be unloaded via delete_module (rmmod hci_vhci). When the userspace process subsequently closes the debugfs file descriptor, full_proxy_release() accesses the already unmapped file_operations structure, resulting in a kernel page fault / oops: BUG: unable to handle page fault for address: fffffbfff846f18b #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page Oops: Oops: 0000 [#1] SMP KASAN NOPTI RIP: 0010:full_proxy_release+0x8e/0x140 Call Trace: <TASK> __fput+0x39f/0xa60 fput_close_sync+0xff/0x200 __x64_sys_close+0x8c/0xf0 do_syscall_64+0xc3/0x590 entry_SYSCALL_64_after_hwframe+0x76/0x7e </TASK> Modules linked in: ... [last unloaded: hci_vhci(E)] The issue was discovered via manual code audit of drivers/bluetooth/hci_vhci.c and verified on a Linux 7.3-rc6 test VM with CONFIG_BT_VHCI=m by keeping a debugfs file open across controller teardown and module unload. Set .owner = THIS_MODULE on all four file_operations structures so that debugfs pins the module until all file descriptors are closed.
-
CVE-2026-99469
In the Linux kernel, the following vulnerability has been resolved: media: gspca: use vb2_video_unregister_device() on disconnect gspca uses vb2_fop_release(), but gspca_disconnect() only unregisters the video device. If the queue owner closes the device after USB disconnect, vb2_fop_release() releases the queue and invokes gspca_stop_streaming() at that point. gspca_stream_off() calls subdriver stop callbacks and usb_set_interface() through gspca_dev->dev. By the time the late close runs, the USB core can have freed the usb_device, resulting in a use-after-free. This was observed as a KASAN use-after-free in sd_stopN(). Use vb2_video_unregister_device() so the queue is released and streaming is stopped synchronously during disconnect, while the usb_device is still valid. Since the helper takes the queue lock, which is usb_lock, call it after dropping that lock.
-
CVE-2026-99470
In the Linux kernel, the following vulnerability has been resolved: media: cx23885: unregister DVB bus when SP2 CI setup fails The DVBSKY S950C, DVBSKY T980C and TechnoTrend CT2-4500 CI paths register the DVB bus before attaching their SP2 CI client. If that client cannot be created, bound, or pinned, the current error path removes the I2C frontend components and only deallocates the frontend list, leaving the registered DVB adapter behind. Removing an I2C demod before unregistering the DVB bus can also expose the use-after-free ordering fixed for the normal remove path. Unregister the DVB bus immediately when SP2 CI setup fails, before the existing I2C client cleanup. The bus helper empties the frontend list, so the later frontend deallocation has nothing left to release. Keep failures from vb2_dvb_register_bus() on the existing cleanup path. Also release an unbound SP2 client and propagate the actual I2C creation or CI registration error. This issue was identified during our ongoing static-analysis research while reviewing kernel code.
-
CVE-2026-99473
In the Linux kernel, the following vulnerability has been resolved: KVM: guest_memfd: Gracefully handle xarray errors when binding a memslot If inserting a memslot into a guest_memfd's bindings xarray fails, propagate the error back to the caller, i.e. fail memslot creation as well. Signalling success and continuing on with memslot creation results in use-after-free, as the guest_memfd instance will remain reachable via the memslot after the file is freed (kvm_gmem_release() won't nullify the file pointer due to lack of a valid binding). Opportunistically WARN and reject binding if KVM_MEMSLOT_GMEM_ONLY is already set, partly to guard against goofs elsewhere, but mostly so that KVM doesn't need to worry about clobbering flags when unwinding on failure. Regarding the unwind, the slot must be fully prepared before inserting it into the bindings, at which point the slot becomes reachable. I.e. waiting to update the slot in order to avoid the ugly unwind isn't an option. And as part of the unwind, explicitly nullify the relevant bindings, as xarray can store a subset of entries when populating a range.
-
CVE-2026-99474
In the Linux kernel, the following vulnerability has been resolved: HID: hiddev: keep state alive through disconnect unlock mutex_unlock() clears the mutex owner before taking its wait lock. A thread spinning in the final hiddev file release can acquire existancelock after hiddev_disconnect() clears the owner, then free hiddev before the disconnecting thread reaches the mutex wait lock. This causes a use-after-free in the mutex slow unlock path. Give the connection and each open file an explicit reference. Drop each reference only after its existancelock critical section has completed, so neither unlock path can free the mutex while the other is still using it.
-
CVE-2026-99460
In the Linux kernel, the following vulnerability has been resolved: mptcp: hold MP_JOIN msk ref when cloning reqsk TCP request migration clones pending request sockets with inet_reqsk_clone(). For MPTCP MP_JOIN requests this byte-copies subflow_req->msk, but the clone does not own a reference. The original and cloned requests can consequently drop the same msk reference, leaving one request with a dangling pointer. This manifests as a KASAN slab-use-after-free in subflow_req_destructor(): BUG: KASAN: slab-use-after-free in subflow_req_destructor (net/mptcp/subflow.c:45) Write of size 4 at addr ff1100010e008d40 by task swapper/0/0 CPU: 0 UID: 0 PID: 0 Comm: swapper/0 Not tainted 7.2.0-rc5-00353-gc27e36054537 #10 PREEMPT(full) Call Trace: subflow_req_destructor (net/mptcp/subflow.c:45) subflow_v4_req_destructor (net/mptcp/subflow.c:694) __reqsk_free (net/ipv4/inet_connection_sock.c:906) reqsk_timer_handler (include/net/request_sock.h:137 net/ipv4/inet_connection_sock.c:1147) Allocated by task 10014: sk_prot_alloc (net/core/sock.c:2246) sk_clone (net/core/sock.c:2488) mptcp_sk_clone_init (include/net/sock.h:1848 net/mptcp/protocol.c:3564) subflow_syn_recv_sock (net/mptcp/subflow.c:883) tcp_check_req (net/ipv4/tcp_minisocks.c:934) Freed by task 0: slab_free_after_rcu_debug (include/linux/kasan.h:235 mm/slub.c:2677 mm/slub.c:6439) rcu_core (kernel/rcu/tree.c:2645 kernel/rcu/tree.c:2897) Last potentially related work creation: kmem_cache_free (mm/slub.c:2638 mm/slub.c:6377 mm/slub.c:6504) __sk_destruct (net/core/sock.c:2289 net/core/sock.c:2391) sk_destruct (net/core/sock.c:2419) __sk_free (net/core/sock.c:2430) sk_free (net/core/sock.c:2441) mptcp_close (include/net/sock.h:2020 net/mptcp/protocol.c:3399) inet_release (net/ipv4/af_inet.c:442) The buggy address belongs to the cache MPTCP of size 2968 The buggy address is 128 bytes inside a freed 2968-byte region A non-NULL subflow_req->msk means that the request owns one reference. The third ACK can concurrently transfer the original request reference to the child and release it, so taking an unconditional hold on the copied pointer is unsafe. MPTCP sockets use SLAB_TYPESAFE_BY_RCU and all current inet_reqsk_clone() callers run in an RCU read-side critical section. Document and lockdep-check that calling-context requirement. Read the pointer from the original request and acquire a reference only if it is still live. Use refcount_inc_not_zero_acquire() so the subsequent ownership validation cannot be reordered before the try-get, then re-read the original request to verify that it still owns the same msk. If either check fails, clear the clone pointer; otherwise its normal destructor balances the new reference. Mark the ownership-transfer store with WRITE_ONCE() to match the lockless reads. The following patch completes the clone fixup for MP_CAPABLE token ownership. Both patches carry the same Fixes tag and are required for stable backports.
-
CVE-2026-99461
In the Linux kernel, the following vulnerability has been resolved: mm: khugepaged: fix swap entry value to folio_pfn() Patch series "mm: khugepaged: fix tracepoint UAF", v5. The khugepaged tracepoints take a folio pointer and call folio_pfn(), but by then the folio may no longer be valid: freed after folio_put(), folio_unlock() or pte_unmap_unlock(), or not a folio at all but an xarray-encoded swap entry. On classic SPARSEMEM, dereferencing it oopses khugepaged as soon as the trace event is enabled; on other memory models it merely prints a bogus pfn. Pass the pfn to the tracepoints directly, captured while the folio is still pinned, closing the use-after-free windows in mm_khugepaged_scan_file(), mm_khugepaged_scan_pmd() and mm_khugepaged_collapse_file(). This patch (of 3): When the swap entries found exceed max_ptes_swap, the loop is left via break with folio still holding the xarray value that encodes the swap entry, not valid folio pointer. That value is passed to trace_mm_khugepaged_scan_file(), which feeds it to folio_pfn(). On FLATMEM and SPARSEMEM_VMEMMAP, the page_to_pfn() is plain pointer arithmetic, so the trace event merely prints bogus scan_pfn. On classic SPARSEMEM, the page_to_pfn() reads page->flags, dereferencing the tiny encoded integer and oopsing khugepaged whenever the trace event is enabled. So when folio is the swap entry value, simply set pfn to -1, just like exhausted scan naturally. And the folio_put() has maybe dropped the last reference of folio. The trace_mm_khugepaged_scan_file() is left with a dangling folio pointer. so using the folio_pfn() before dropping the reference, closing use-after-free window.
-
CVE-2026-99463
In the Linux kernel, the following vulnerability has been resolved: mm/rmap: don't discard lazyfree THPs at PMD level __discard_anon_folio_pmd_locked() clears a lazyfree THP's PMD before it determines whether the folio can be discarded, and restores it if the folio is dirty or has extra references. It is done under only the PMD lock and the anon_vma read lock inside try_to_unmap_one() and no mmap lock is held. The temporarily none PMD causes UAF. A parallel munmap() or mremap() is supposed to serialize with it on the PMD lock, but zap_pmd_range() (called by munmap()) and move_page_tables() (called by mremap()) do not take the lock when they see the temporarily none PMD and skip the PMD. There is no more page table walk after the none PMD is restored: - munmap(): free_pgtables() waits in unlink_anon_vmas() for the anon_vma lock, then free_pmd_range() drops the restored PMD as bad without removing the rmap, so the folio keeps a stale mapcount. - whole-VMA MREMAP_DONTUNMAP: dontunmap_complete() unlinks the source VMA from its anon_vma while the restored PMD still maps the folio. - plain mremap(): the THP is not moved but unmapped, so the THP's data is lost. In the first two cases, the folio stays mapped, but no VMA that maps it is linked to its anon_vma. When the anon_vma is freed, it is still reachable when one rmap walks the folio. Racing lazyfree reclaim against MREMAP_DONTUNMAP as an unprivileged user reproduces a KASAN use-after-free in folio_lock_anon_vma_read() as reported in the Closes link. One solution is to use an invalidated PMD instead of a cleared one until the discard is committed, but it needs arch fixes first. 1. GUP-fast still follows an invalidated PMD on riscv, LoongArch and s390. 2. sparc64 miscounts thp_pte_count when an invalidated PMD is set again or cleared. To keep the fix simple for backport, stop discarding lazyfree THPs at PMD level instead. try_to_unmap_one() splits the PMD and discards the folio at PTE level. unmap_huge_pmd_locked() and __discard_anon_folio_pmd_locked() are left in place, unused, to minimize this fix for backport and the invalidated PMD approach is planned as a follow-up series to use them again.
-
CVE-2026-99464
In the Linux kernel, the following vulnerability has been resolved: media: redrat3: fix UAF in probe error path leaving rc device registered redrat3 stores its device name and phys path inside the main rr3 structure, and the rc device's device_name and input_phys point into rr3. When the RC device is registered, it keeps those pointers. KASAN: slab-use-after-free in string_nocheck lib/vsprintf.c:648 Read of size 1 at addr ffff888051fda758 by task udevd/7464 Call Trace: string_nocheck lib/vsprintf.c:648 [inline] string+0x216/0x2d0 lib/vsprintf.c:730 vsnprintf+0x74a/0xef0 lib/vsprintf.c:2945 vscnprintf+0x41/0x90 lib/vsprintf.c:3014 sysfs_emit+0x10e/0x180 fs/sysfs/file.c:761 input_dev_show_name+0x58/0x70 drivers/input/input.c:1282 Allocated by task 10: redrat3_dev_probe+0x477/0x2570 drivers/media/rc/redrat3.c:1023 Freed by task 10: redrat3_delete drivers/media/rc/redrat3.c:466 [inline] redrat3_dev_probe+0x1bf4/0x2570 drivers/media/rc/redrat3.c:1124 Syzkaller triggers this via usb probing. It probes the RedRat3 USB interface, which calls redrat3_dev_probe() in redrat3.c. That function allocates rr3, then builds an rc device whose name points into rr3 via redrat3_init_rc_dev() in redrat3.c, and registers it with rc_register_device(). When the detector enable fails after the RC device is registered, the probe jumps to the led_free path. That path unregisters the LED but does not unregister the RC device before freeing rr3 via redrat3_delete() in redrat3.c. The RC device still holds device_name = rr3->name which is now freed. Later udevd reads /sys/.../input device name via sysfs_emit() in file.c, which calls input_dev_show_name() in input.c, which emits dev->name which is the freed rr3->name. When the detector enable fails after the RC device is registered, the error path must not leave the RC device registered. Make the probe error path clean up the same way redrat3_dev_disconnect() does. This change should be safe because the RC device is fully registered at this point and its teardown via rc_unregister_device() is protected by the input device mutex, and rr3 is still alive during unregister so device_name remains valid until after unregister. The URBs are killed before rc_free, so no completion can run after rc is freed. No new lock ordering is introduced.
-
CVE-2026-99467
In the Linux kernel, the following vulnerability has been resolved: media: marvell: mcam: stop DMA and cancel s_bh_work before freeing DMA buffers In vmalloc mode the frame-completion IRQ queues mcam_frame_work() on cam->s_bh_work, which memcpy()s from the dma_bufs[] that mcam_free_dma_bufs() frees with dma_free_coherent(). mccic_shutdown() frees those buffers without stopping the controller, so while streaming a late frame IRQ can re-arm the work after the buffers are gone, causing a use-after-free. Stop the controller in mccic_shutdown() when streaming is still active (gated on an open fd, which holds a runtime-PM reference, so the device is powered), cancel s_bh_work in mcam_free_dma_bufs(), and move INIT_WORK() into mccic_register() before the device can be published.
-
CVE-2026-99455
In the Linux kernel, the following vulnerability has been resolved: serial: core: shut down initialized port on removal When a serial port configured as a console is opened and subsequently closed, tty_port_shutdown() skips invoking port->ops->shutdown() because port->console is true. As a result, tty_port_initialized() remains true and the port's interrupt handler stays registered in the irq subsystem. If the underlying device is later unbound or removed (e.g. via sysfs driver unbind or device hot-unplug), serial_core_remove_one_port() unregisters the console, frees uport->name via kfree(), and tears down the port without shutting it down or freeing its IRQ. Consequently, the irqaction descriptor remains registered in genirq with action->name pointing to freed memory. This triggers a use-after-free read whenever genirq accesses action->name, such as when reading /proc/interrupts: BUG: KASAN: slab-use-after-free in string+0x476/0x570 Call Trace: string+0x476/0x570 vsnprintf+0x422/0x1300 seq_printf+0x... show_interrupts+... Fix this by acquiring port->mutex in serial_core_remove_one_port() after console unregistration, clearing port->console under the mutex to prevent data races with concurrent tty_port_shutdown(), and invoking uart_shutdown(NULL, state) if the port remains initialized. This ensures that port->ops->shutdown() is called and the IRQ is freed before kfree(uport->name).
-
CVE-2026-99456
In the Linux kernel, the following vulnerability has been resolved: scsi: dc395x: sync the waiting_timer before freeing the host The waiting_timer callback re-arms itself and takes the host lock. adapter_uninit() currently stops it with timer_delete(), which does not wait for a callback already running on another CPU. The final scsi_host_put() that frees the host (and the host lock the callback dereferences) runs right after adapter_uninit(), so a concurrent or just re-armed callback can fire after the free, leaving a potential use-after-free window. Replace timer_delete() with timer_shutdown_sync() at the top of adapter_uninit(), before the host-lock section: the timer is dequeued, a running callback is waited for and rearming is blocked before the chip is halted, so it cannot fire at all once the adapter has ceased to function. Waiting outside the lock avoids the self-deadlock that would result from syncing a callback blocked on the lock we still hold. No shutdown is needed for selto_timer because it has no reachable armer or callback. This issue was found by an in-house static analysis tool.
-
CVE-2026-99522
In the Linux kernel, the following vulnerability has been resolved: RDMA/irdma: avoid use-after-free in icrdma_remove icrdma_remove() dereferences iwdev after calling irdma_ib_unregister_device(): irdma_ib_unregister_device(iwdev) -> ib_unregister_device(&iwdev->ibdev) -> __ib_unregister_device() -> ib_dealloc_device() irdma registers irdma_ib_dealloc_device() as the dealloc_driver callback in its ib_device_ops. The RDMA core explicitly documents ib_unregister_device() that when ops.dealloc_driver is used, ib_dev will be freed upon return from the function. The ib_device is embedded in struct irdma_device and iwdev itself was allocated by ib_alloc_device(irdma_device, ibdev). Consequently, once irdma_ib_unregister_device() returns, iwdev must no longer be dereferenced. However, icrdma_remove() currently accesses iwdev->rf afterwards when deinitializing interrupts, destroying ah_tbl_lock, and freeing rf. Although the final ib_device storage is released with kfree_rcu(), the object lifetime has already ended and callers must not rely on the RCU grace period to continue dereferencing iwdev. Save iwdev->rf before unregistering the RDMA device and use the saved pointer for the remaining cleanup. This issue was found by manual code inspection.
-
CVE-2026-99527
In the Linux kernel, the following vulnerability has been resolved: spmi: Fix potential use-after-free by grabbing of_node reference As noticed by Sashiko during a review run of an unrelated patch, in of_spmi_register_devices(), for_each_available_child_of_node() is used to loop through children, and to also assign a node to a newly created SPMI child device. Problem is that the refcount is dropped at every iteration so, in the specific case of DT overlays, a use-after-free may occur when an overlay is dynamically unloaded! To resolve this, increase the of_node refcount when assigning (in function of_spmi_register_devices) and release the reference in spmi_device_remove().
-
CVE-2026-99517
In the Linux kernel, the following vulnerability has been resolved: nvmet: fix use-after-free of the old registrant in nvmet_pr_preempt() nvmet_pr_preempt() handles a preempt sent by the reservation holder by changing the reservation type with nvmet_pr_update_reg_attr(), unless the reservation is of an All Registrants type. For the holder that helper does not modify the registrant in place: it puts a copy on the registrant list, points pr->holder to the copy and passes the old registrant to kfree_rcu(). If the reservation type changed, nvmet_pr_preempt() then hands ®->hostid of that old registrant to nvmet_pr_resv_released(). Unless the reservation released notification is masked, that function takes subsys->lock and compares the host id against every controller of the subsystem. The caller only holds pr->pr_sem and is not in an RCU read-side critical section, so nothing keeps the old registrant alive while it is read. The result is a read of 16 freed bytes that decides which controllers get the notification. Found by code inspection. kfree_rcu() defers the free, so it usually happens after the read; when the pending free is forced to run before the read, KASAN reports the host id as freed memory. Pass the host id of the controller that sent the command instead. The registrant was looked up by that host id, and the controller outlives the request.
-
CVE-2026-99498
In the Linux kernel, the following vulnerability has been resolved: s390/kprobes: Prevent kprobes on instructions with exception table entry The mvcos exception handler ex_handler_ua_mvcos() decodes the faulting instruction, assuming it is an mvcos instruction. In case the instruction is kprobed the decoded instruction is a breakpoint instruction instead, which leads to incorrect instruction decoding and potential register corruption. Fix this by simply preventing to set a kprobe on such instructions, similar like arm64 is doing it.
-
CVE-2026-99486
In the Linux kernel, the following vulnerability has been resolved: rust: configfs: fix data offset calculation for subsystem callbacks `get_group_data()` chooses between `Group::<Parent>::container_of()` and `Subsystem::<Parent>::container_of()` based on whether `this` represents the root group of a configfs subsystem. It detects this by checking `(*this).cg_subsys.is_null()`, but `link_group()` in `fs/configfs/dir.c` unconditionally sets `cg_subsys` for every `config_group` attached anywhere in a registered subsystem, including the subsystem's own `su_group`. The only `config_group` with a NULL `cg_subsys` is the configfs root, on which userspace cannot trigger callbacks. The check is therefore always false at runtime, and the `Subsystem` branch is dead. Subsystem-level callbacks reach the `Group` branch and may read `data` at the wrong offset. Thus change the `is_root` check to correctly identify whether a group is a root group by comparing `this` to `subsys.su_group`. [ Andreas: Fix clippy warning on 1.85.0 by moving safety comment. ]
-
CVE-2026-99487
In the Linux kernel, the following vulnerability has been resolved: rust: bug: pass TAINT_WARN to warn_slowpath_fmt() on UML On UML, `warn_flags!` calls `warn_slowpath_fmt()` directly. Its third argument is a taint number, but `warn_on!` passes its bug flags there. As a result, `add_taint()` gets 2304 instead of 9 (`TAINT_WARN`). The warning is not recorded: `/proc/sys/kernel/tainted` stays 0 after `warn_on!` fires, where it should be 512. It also causes an out-of-bounds write in `add_taint()`. Pass `TAINT_WARN` directly, as C does on UML.
-
CVE-2026-99462
In the Linux kernel, the following vulnerability has been resolved: mm/truncate: look up the end-edge straddler by index In truncate_inode_partial_folio(), after the first split at the start edge, folio_split() unlocks and drops the refcount of the after-split sub-folios. The sub-folio straddling the end of the truncation range is therefore unlocked and only transiently ref'd in the page cache while the code still derives it from a page pointer inside the original folio. Between the first split finishing and page_folio() resolving split_at2, that tail page can be reclaimed, freed and reallocated as a new large folio in the same mapping at a different file offset. folio2 then points at a folio that does not cover the end boundary, yet folio2->mapping == folio->mapping still holds, so the stale pointer passes the mapping check and folio_split_or_unmap() splits a folio at a wrong position (or, with a transient refcount, a use-after-free window opens between try_get and the split). __folio_split()'s own folio != page_folio(split_at) check cannot catch this either since split_at2 has been reallocated as part of the new folio, so page_folio(split_at2) resolves back to folio2. Look the straddler up by its page index instead. __filemap_get_folio() returns the folio currently covering the boundary, ref'd and locked, with the mapping validated under the lock, so the split target is always the real folio at the end edge.
-
CVE-2026-99542
In the Linux kernel, the following vulnerability has been resolved: ksmbd: deny access for an empty DACL instead of granting the request smb_check_perm_dacl() returns the access mask it computes to the SMB2 CREATE handler, which uses it as the granted access for the open. For a security descriptor whose DACL is present but contains no ACEs, the function is supposed to deny access: an empty (non-NULL) DACL grants no access to anyone except the object owner's implicit READ_CONTROL and WRITE_DAC (MS-DTYP 2.4.5). The empty-DACL branch only denies when both - the DACL has no trailing bytes (pdacl_size == sizeof(struct smb_acl)), and - the request contains bits beyond READ_CONTROL / WRITE_DAC. In every other case it takes the shared "goto err_out" with rc still 0, which returns success without modifying *pdaccess, so the caller grants the full requested access. Three problems follow: - The READ_CONTROL / WRITE_DAC carve-out models the owner's implicit rights, but the check runs before is_owner is computed, so those rights are handed to *every* caller, not just the owner. For READ_CONTROL the SMB2 CREATE path does not fall back to inode_permission(), so any user can read the security descriptor of an object whose empty DACL is meant to deny them. - A request of exactly READ_CONTROL and/or WRITE_DAC is granted to any caller. WRITE_DAC lets the handle rewrite the DACL. - A DACL that declares zero ACEs but carries trailing bytes (pdacl_size > sizeof(struct smb_acl)) skips the deny entirely and is granted whatever was requested. On shares where the NT ACL is the authoritative access control -- e.g. "force user" shares, where every client maps to one POSIX identity so POSIX cannot distinguish users -- this is an access-control bypass. Compute is_owner before the empty-DACL branch and deny access from an empty DACL unless the caller is the owner requesting no more than READ_CONTROL / WRITE_DAC.
-
CVE-2026-99472
In the Linux kernel, the following vulnerability has been resolved: KVM: pfncache: use a dedicated invalidation sequence for cache refresh The gfn_to_pfn_cache refresh path guards against mmu notifier invalidations which complete while it has dropped gpc->lock for the HVA->PFN lookup: hva_to_pfn_retry() samples kvm->mmu_invalidate_seq and retries if it changed, or if mn_active_invalidate_count is still elevated. That is insufficient for HVA-based caches. mmu_invalidate_seq is only advanced by kvm_mmu_invalidate_end() when the invalidated range overlaps a memslot, and an HVA-based cache (e.g. the Xen shared_info page mapped with KVM_XEN_ATTR_TYPE_SHARED_INFO_HVA) need not be backed by any memslot at all. An invalidation of the cached HVA which starts and ends entirely within the lookup window is thus invisible to the retry check: mn_active_invalidate_count is back to zero and the sequence never moved. The refresh then publishes a mapping of a page which has already been freed, and the next reader dereferences it: BUG: KASAN: use-after-free in kvm_xen_shared_info_init+0x3c6/0x440 Read of size 4 at addr ffff8880599c2900 by task syz.2.383/7257 Since gfn_to_pfn_cache_invalidate_start() deliberately skips caches which are not currently valid (including one whose refresh is in progress, as the refresh clears the valid flag before dropping the lock), the retry check is the only line of defence, and it must fire for *any* invalidation, not just those hitting a memslot. Add a dedicated kvm->gpc_invalidate_seq, incremented by every kvm_mmu_notifier_invalidate_range_end() under mn_invalidate_lock before mn_active_invalidate_count is decremented, and check it in hva_to_pfn_retry() instead of mmu_invalidate_seq. Incrementing in range_end() in the same critical section as the in-progress count also closes the variant where the cache is activated with the contested HVA only after invalidate_range_start() has run. The same bug is also reachable through the per-vCPU vcpu_info cache (KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO_HVA), where the stale mapping is then dereferenced by kvm_setup_guest_pvclock() on the next KVM_RUN: BUG: KASAN: use-after-free in kvm_setup_guest_pvclock+0x5bf/0x660 This intentionally makes refresh retry on *unrelated* mmu notifier events; restoring precision (and reworking the GPC locking more generally) is left for a subsequent series. Reproducers: https://david.woodhou.se/xen_shinfo_race.c https://david.woodhou.se/vcpu_info_race.c
-
CVE-2026-99445
In the Linux kernel, the following vulnerability has been resolved: power: supply: pf1550: drain IRQ work before unregistering supplies The VBUS work calls power_supply_changed(), but its managed cancellation is registered before the supplies. Cleanup can therefore unregister a supply while IRQ-triggered work is still pending. Commit 838767f50747 ("power: supply: pf1550: Fix use-after-free in power_supply_changed()") moved IRQ requests after supply registration, but left work cancellation before the supplies. That change protects the direct IRQ callbacks, while this fix drains the work they queue. Register the three work items after both supplies and before the IRQs. Cleanup then frees the IRQ producers, drains the work, and unregisters the supplies. The teardown ordering issue was found by static analysis.
-
CVE-2026-99450
In the Linux kernel, the following vulnerability has been resolved: vfio/nvgrace-gpu: Use a core-only open_device in fallback mode When the ACPI memory properties are missing or the C2C link is down, the device is exposed through nvgrace_gpu_pci_core_ops. That table opens through nvgrace_gpu_open_device() but closes through vfio_pci_core_close_device(), so open and close are not symmetric. nvgrace_gpu_open_device() registers nvdev->usemem for poison handling unconditionally. In fallback mode usemem.memlength is zero, and register_pfn_address_space() inserts a node covering [0, ULONG_MAX]. Nothing removes it: a second open fails with -EBUSY, and once the device is unbound the node outlives the nvdev that contains it. The next open after a rebind reports: BUG: KASAN: slab-use-after-free in interval_tree_iter_first+0x209/0x310 Read of size 8 at addr ffff88800894aa30 by task nvgrace_repro/203 Call Trace: <TASK> dump_stack_lvl+0x5d/0x80 print_report+0x153/0x4c6 kasan_report+0xe4/0x1a0 interval_tree_iter_first+0x209/0x310 register_pfn_address_space+0x97/0x100 nvgrace_gpu_open_device+0x21c/0x730 vfio_df_open+0x213/0x520 vfio_df_ioctl_bind_iommufd+0x3e8/0x6e0 vfio_device_fops_unl_ioctl+0x782/0x1570 __x64_sys_ioctl+0x13c/0x1c0 do_syscall_64+0x106/0x630 entry_SYSCALL_64_after_hwframe+0x77/0x7f </TASK> Allocated by task 188: ... _vfio_alloc_device+0x3a/0x500 nvgrace_gpu_probe+0xcb/0x660 local_pci_probe+0xcf/0x170 ... Freed by task 188: ... kfree+0x20d/0x600 device_release+0xcf/0x250 kobject_put+0x152/0x2f0 pci_device_remove+0xb0/0x1c0 device_release_driver_internal+0x38d/0x540 unbind_store+0xde/0x100 ... Give the fallback table its own open_device that only calls vfio_pci_core_enable() and vfio_pci_core_finish_enable(), the counterpart of vfio_pci_core_close_device(). Fallback mode then never touches the nvgrace memory regions, as Alex suggested. This should be safe because the fallback table uses the vfio-pci core read, write and mmap handlers, so nothing there uses the state nvgrace_gpu_open_device() sets up. The one visible difference is that a fallback open no longer fails when BAR0 cannot be mapped, which matches vfio-pci.
-
CVE-2026-99565
In the Linux kernel, the following vulnerability has been resolved: ksmbd: wait for negotiation before receiving another PDU The receive thread queues a request and can read the next PDU before the negotiation worker finishes. While need_neg is set, the next PDU can cause ksmbd_init_smb_server() to replace conn->ops. A pipelined SMB2 request can therefore make the worker handle an earlier, shorter SMB1 request with SMB2 operations and read beyond its buffer. An SMB1 multi-protocol negotiate may be followed by an SMB2 NEGOTIATE that changes conn->ops again. Wait for work queued while the connection is new or needs negotiation to complete before reading another PDU. This keeps the operations stable for both stages of negotiation without changing when need_neg is cleared.
-
CVE-2026-99556
In the Linux kernel, the following vulnerability has been resolved: ksmbd: serialize directory emptiness checks with readdir SMB2 QUERY_DIRECTORY holds fp->readdir_lock while it calls iterate_dir() on fp->filp. An SMB2 SET_INFO request with FILE_DISPOSITION_INFORMATION can resolve the same FileId and call ksmbd_vfs_empty_dir() on the same filp without that lock. iterate_dir()'s inode read lock permits concurrent readers and does not serialize per-file iterator state. ext4 stores its dx readdir cursor and fname RB tree in file->private_data. Concurrent reset or refill can therefore free a fname node while the other worker still traverses it or retains its RB-tree link, resulting in use-after-free accesses. An authenticated SMB client with list and delete rights on a directory in a writable KSMBD share backed by ext4 can trigger this when the directory takes the dx readdir path. The client overlaps an SMB2 QUERY_DIRECTORY request with an SMB2 SET_INFO request carrying FILE_DISPOSITION_INFORMATION on one live FileId. In one x86_64 SMP KASAN multi-shot guest using ksmbd-for-next at 171df27077fb, the unmodified kernel reported a cross-worker use-after-free before a panic. The first report's access and free stacks show the two request paths directly (trimmed): BUG: KASAN: slab-use-after-free in ext4_htree_store_dirent+0x488/0x550 Read of size 8 by task kworker/2:0/29 Workqueue: ksmbd-io handle_ksmbd_work Call Trace: ext4_htree_store_dirent+0x488/0x550 htree_dirblock_to_tree+0x535/0x990 ext4_htree_fill_tree+0x319/0xba0 ext4_readdir+0x175a/0x2f40 iterate_dir+0x1b6/0x630 smb2_query_dir+0x1003/0x1cc0 Freed by task 279: free_rb_tree_fname+0x67/0xc0 ext4_readdir+0x13a9/0x2f40 iterate_dir+0x1b6/0x630 ksmbd_vfs_empty_dir+0xa2/0x100 smb2_set_info+0x2ffa/0x45d0 Take fp->readdir_lock across the entire emptiness check. This places both iterators over that handle's filp in the same lock domain while preserving the disposition status handling. The exact code change completed a warning-free x86_64 KASAN bzImage build. Tracing three request pairs showed that the two iterate_dir() calls no longer overlapped and the expected statuses were preserved. A further 1,600 reproducer iterations completed with expected statuses and no KASAN report, Oops, or panic. The reproducer and complete guest configuration are available privately to maintainers on request.
-
CVE-2026-99557
In the Linux kernel, the following vulnerability has been resolved: ceph: fail readdir when the MDS reply carries no dirfrag A ceph mount oopses in ceph_readdir() when the MDS answers a READDIR request with a reply whose extra blob is empty. parse_reply_info() skips reply blobs whose length is zero, so parse_reply_info_readdir() never runs, r_reply_info.dir_dir stays NULL, and the reply still parses as successful. ceph_mdsc_do_request() returns 0 and ceph_readdir() then reads rinfo->dir_dir->frag with no NULL check: /* extra */ ceph_decode_32_safe(&p, end, len, bad); if (len > 0) { ... err = parse_reply_info_extra(&p, p + len, req, features, s); ... } The oops on v7.3.0-rc4-00537-ga3ff15db6820, one getdents64(2) on the mount root after an ms_mode=legacy mount, trimmed: BUG: KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007] Oops: general protection fault, probably for non-canonical address 0xdffffc0000000000: 0000 [#1] SMP KASAN NOPTI CPU: 1 UID: 0 PID: 266 Comm: ls Tainted: G N 7.3.0-rc4-00537-ga3ff15db6820 #1 PREEMPT(lazy) RIP: 0010:ceph_readdir+0xbff/0x3ea0 Call Trace: <TASK> ? __pfx_ceph_readdir+0x10/0x10 wrap_directory_iterator+0xa2/0xe0 iterate_dir+0x1da/0x5d0 __x64_sys_getdents64+0x13b/0x2a0 do_syscall_64+0xdd/0x4a0 entry_SYSCALL_64_after_hwframe+0x77/0x7f </TASK> The unconditional frag read came in with 81c6aea5275ea ("ceph: handle frag mismatch between readdir request and reply"), so the bug is old. Treat a reply with no dirfrag as a protocol violation: warn once, drop the request and return -EIO, the same way ceph_readdir() already handles the two malformed-reply cases later in the function. Tested on v7.3.0-rc4-00537-ga3ff15db6820 (KASAN). Unpatched, the getdents64(2) above takes the machine down (oops=panic) on the first READDIR reply. With the patch it warns once and getdents64(2) returns -EIO, no oops, no hang. A conformant reply for an empty directory — dirfrag present, zero entries, FRAG_END|FRAG_COMPLETE — still lists "." and ".." and returns EOF on both kernels, so normal readdir behavior is unchanged. The oops is still present in v7.3-rc5 (72d3fcf802c45).
-
CVE-2026-99561
In the Linux kernel, the following vulnerability has been resolved: fork: don't create io threads once PF_POSTCOREDUMP is set zap_process() skips every thread that already has PF_POSTCOREDUMP set. Such a thread is past synchronize_group_exit(), so a coredump can't catch it anymore. It isn't counted in core_state->threads_remaining and it isn't sent SIGKILL. do_exit() sets PF_POSTCOREDUMP in synchronize_group_exit() and calls io_uring_files_cancel() right after that. The cancellation runs task work and a create_worker_cb() that io-wq queued before the exit creates a new io-wq worker from there. If another thread started a coredump in the meantime that worker joins a thread-group which is already being dumped. zap_process() never saw it so it was never counted. But when it exits it sees signal->core_state in synchronize_group_exit() and decrements threads_remaining like any other thread. So the count reaches zero one thread early and the dumper leaves coredump_wait_inactive() while a thread it counted is still running. The task has no fatal signal pending because zap_process() deliberately didn't send it one, and it never went through get_signal() so it doesn't have PF_SIGNALED either. Refuse to create an io thread when the creator has PF_POSTCOREDUMP. That costs nothing. io_uring_files_cancel() raises IO_WQ_BIT_EXIT before it runs any task work, so a worker created from there only ever gets to exit again, and io_should_retry_thread() doesn't retry -EINTR. Clearing PF_POSTCOREDUMP for the new thread in copy_process() was the other option. It makes the worker a thread like any other, but it only helps while the dump hasn't started. A worker born after zap_process() has run is invisible to it whatever its flags say and still decrements threads_remaining on the way out. Refusing to create it covers both, and then no task is ever born with the flag, so there is nothing to clear. Moving io_uring_files_cancel() ahead of synchronize_group_exit() closes the same window. It runs the cancellation, and whatever that can block on, before the thread announces itself to the dumper. A thread that blocks there never reaches coredump_task_exit() at all, so the dumper ends up waiting for a thread that never parks. Refusing the creation leaves the cancellation where it is. But it's very ugly to run io_uring work even before we did all the generic exit work.
-
CVE-2026-99548
In the Linux kernel, the following vulnerability has been resolved: smb: client: fix early procfs exposure race Currently, init_cifs() calls cifs_proc_init() near the beginning of the function, before INIT_LIST_HEAD, leading to a momentarily uninitialized, but exposed, procfs. If a userspace process reads this empty procfs, iterating over a zeroed list_head leads to a NULL pointer dereference and a kernel oops/panic. For example, if a delay is added between cifs_proc_init() and INIT_LIST_HEAD, and a userspace process is polling /proc/fs/cifs/DebugData, the following oops occurs: BUG: kernel NULL pointer dereference, address: 0000000000000608 Oops: Oops: 0000 [#1] SMP PTI CPU: 0 UID: 0 PID: 3477 Comm: cat Not tainted 7.3.0-02212-g4df4d2a577db-dirty #6 PREEMPT(lazy) RIP: 0010:cifs_debug_data_proc_show+0x103/0xd80 [cifs] Call Trace: <TASK> ? seq_read_iter+0x86/0x400 ? seq_read_iter+0x86/0x400 seq_read_iter+0x183/0x400 proc_reg_read_iter+0x62/0x90 copy_splice_read+0x215/0x310 splice_file_to_pipe+0xec/0x1c0 do_splice+0x4b8/0x780 __se_sys_splice+0x18f/0x250 do_syscall_64+0x109/0x330 entry_SYSCALL_64_after_hwframe+0x77/0x7f </TASK> Fix this by moving cifs_proc_init() down in init_cifs(), ensuring it runs only after all internal lists, locks, counters, and workqueues are fully ready, but before registering the filesystem. Update the error cleanup ladder and exit_cifs() to maintain proper LIFO order.
-
CVE-2026-99549
In the Linux kernel, the following vulnerability has been resolved: smb: client: do not pass a NULL buf to is_oplock_break() receive_encrypted_read() sets *num_mids = 1 and never writes the caller's bufs[]. Offload returns -1 with *num_mids = 0, so the demultiplex thread skips the frame. The inline decrypt path returns 0 even when smb2_find_mid() fails. cifs_demultiplex_thread() then treats mids[i] == NULL as an oplock break and calls is_oplock_break(bufs[i], server) with bufs[i] still NULL. smb2_is_valid_oplock_break() reads hdr.Command at offset 12 and oopses. A large encrypted READ whose MessageId does not match is enough to panic cifsd. Oops: general protection fault, probably for non-canonical address 0xdffffc0000000001: 0000 [#1] SMP KASAN PTI KASAN: null-ptr-deref in range [0x0000000000000008-0x000000000000000f] CPU: 0 PID: 100 Comm: cifsd RIP: smb2_is_valid_oplock_break+0xc8/0x2270 [cifs] cifs_demultiplex_thread Kernel panic - not syncing: Fatal exception Return -1 when inline decrypt finds no mid, and skip is_oplock_break() plus the unknown-frame dump when bufs[i] is NULL. receive_encrypted_standard() already assigns bufs[*num_mids].
-
CVE-2026-99550
In the Linux kernel, the following vulnerability has been resolved: ocfs2: pin owning modules for open debugfs files The OCFS2 debugfs file operation tables do not set owner. The full debugfs proxy obtains real_fops with fops_get() at open and keeps it until final release. Without an owner, this does not acquire a reference to the module containing the table and its callbacks. Removing the debugfs entries does not drain retained open files. An open connected_nodes file can therefore outlive ocfs2_nodemanager and make full_proxy_release() read real_fops->release from freed module memory. Heming also reproduced the issue with fs_state and asked for the other OCFS2 debugfs operation tables to be checked. Set owner to THIS_MODULE in all eleven explicit debugfs operation tables: five in ocfs2_nodemanager, four in ocfs2_dlm, and two in ocfs2. The existing debugfs get/put pair then holds the appropriate module until final release. Leave the open, read and release callbacks unchanged. The blockcheck table already sets owner through DEFINE_DEBUGFS_ATTRIBUTE(), so it needs no change. The original controlled test used an unloadable OCFS2 module and case-local probe filters without changing the candidate open, unload or release logic. Closing a retained connected_nodes file after normal unload produced: BUG: unable to handle page fault for address: fffffbfff804cfef Oops: Oops: 0000 [#1] SMP KASAN NOPTI RIP: 0010:full_proxy_release+0xbf/0x190 Call Trace: <TASK> __fput+0x39a/0xa60 fput_close_sync+0xff/0x200 __x64_sys_close+0x8c/0xf0 do_syscall_64+0x114/0x620 entry_SYSCALL_64_after_hwframe+0x77/0x7f </TASK> Modules linked in: [last unloaded: ocfs2_nodemanager]
-
CVE-2026-99551
In the Linux kernel, the following vulnerability has been resolved: ocfs2: drain quorum work before removing the cluster The configured cluster must remain available until an in-flight quorum decision has finished using its fencing method. o2quo_make_decision() drops qs_lock before calling o2quo_fence_self(), which marks heartbeat regions for unclean stop and then reads o2nm_single_cluster->cl_fence_method. Cluster removal clears the singleton without waiting for this independent system-workqueue item. When a failed-quorum decision overlaps removal of the heartbeat regions, nodes and cluster through configfs, the worker can be preempted after marking the regions and resume after the singleton has been cleared: Quorum worker Configfs removal select fence; drop qs_lock o2hb_stop_all_regions() remove regions; join heartbeat threads remove nodes; stop o2net clear singleton; release cluster read cl_fence_method The resumed read dereferences NULL instead of applying the configured fencing method. The heartbeat thread's dependency protects the local node only until the thread exits, and draining o2net does not wait for qs_work. Add o2quo_flush_work() to drain pending or running quorum decisions before clearing the singleton in o2nm_cluster_group_drop_item(). By this point the heartbeat regions and nodes have been removed, so their work publishers are stopped. The worker can finish while the cluster and its configured fencing method remain available. It does not acquire the configfs mutex held by the removal path. Keep o2quo_exit() as the counterpart of o2quo_init() for module teardown and initialization failure, and have it call the same drain helper. Cluster removal uses only the explicitly named drain operation. KASAN report as below: Oops: general protection fault, probably for non-canonical address 0xdffffc000000011b: 0000 [#1] SMP KASAN NOPTI KASAN: null-ptr-deref in range [0x00000000000008d8-0x00000000000008df] CPU: 0 UID: 0 PID: 72 Comm: kworker/0:2 Not tainted 7.3.0-rc4-next-20260921-pmb-ocfs2-functional-v1+ #1 PREEMPT(lazy) Hardware name: QEMU Ubuntu 24.04 PC v2 (i440FX + PIIX, arch_caps fix, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 Workqueue: events o2quo_make_decision RIP: 0010:o2quo_fence_self+0x65/0x200 Register dump omitted. Call Trace: <TASK> o2quo_make_decision+0x345/0x3a0 ? __pfx_o2quo_make_decision+0x10/0x10 ? process_one_work+0x834/0x1b40 ? srso_alias_return_thunk+0x5/0xfbef5 ? lock_release+0xc8/0x290 ? srso_alias_return_thunk+0x5/0xfbef5 process_one_work+0x9a8/0x1b40 ? __pfx_process_one_work+0x10/0x10 ? lock_acquire+0x190/0x300 ? lock_is_held_type+0x8f/0x100 ? srso_alias_return_thunk+0x5/0xfbef5 worker_thread+0x65c/0xe40 ? srso_alias_return_thunk+0x5/0xfbef5 ? __kthread_parkme+0x177/0x220 ? srso_alias_return_thunk+0x5/0xfbef5 ? __pfx_worker_thread+0x10/0x10 kthread+0x351/0x460 ? srso_alias_return_thunk+0x5/0xfbef5 ? __pfx_kthread+0x10/0x10 ret_from_fork+0x659/0x940 ? __pfx_ret_from_fork+0x10/0x10 ? srso_alias_return_thunk+0x5/0xfbef5 ? __switch_to+0x74f/0xf70 ? __pfx_kthread+0x10/0x10 ret_from_fork_asm+0x1a/0x30 </TASK> Modules linked in: ---[ end trace 0000000000000000 ]---
-
CVE-2026-99552
In the Linux kernel, the following vulnerability has been resolved: ocfs2/dlm: drain join handlers before releasing module text The code for a registered DLM handler must remain resident until its last invocation finishes. dlm_unregister_net_handlers() only removes dlm_join_handlers from the o2net registry. o2net_handler_get() takes a kref on the wrapper, but that reference does not pin ocfs2_dlm text. o2net_process_message() calls nh_func after dropping o2net_handler_lock. With CONFIG_OCFS2_FS_O2CB=m, ocfs2_dlm can be loaded independently of ocfs2_stack_o2cb while nodemanager continues handling messages from a configured, heartbeating peer. If no dependent module or module reference prevents normal unload, the following ordering is possible: Receive worker Module unload -------------- ------------- o2net_handler_get() take wrapper kref drop handler lock <preempted> dlm_exit() unregister handlers release DLM module text call saved nh_func in released module text Use o2net_unregister_and_flush_handler_list() in the shared DLM unregister helper. It first prevents new lookups, then waits for the ordered o2net workqueue to drain, including a concurrent destruction of that queue. The acquired invocation therefore finishes before dlm_exit() returns and the loader releases the callback code. The same helper covers partial registration and dlm_init() error unwind before module memory is released. Skip the drain when the private join-handler list is empty. Early dlm_init() failures can reach this helper before any handler has been published. Partial-registration failure also reaches it a second time after the first call has already unregistered and drained the list. Neither case needs to take o2net_wq_mutex or drain unrelated work. The list is changed only by serialized module init/exit paths; a nonempty list still uses the full unregister-and-flush helper. Oops report as below: Oops: int3: 0000 [#1] SMP KASAN NOPTI CPU: 2 UID: 0 PID: 54 Comm: kworker/u16:3 Not tainted 7.3.0-rc4-next-20260921-pmb-ocfs2-functional-v1+ #1 PREEMPT(lazy) Hardware identification omitted. Workqueue: o2net o2net_rx_until_empty [ocfs2_nodemanager] RIP: 0010:0xffffffffc0423871 [Instruction byte dump omitted.] Register dump omitted. Call Trace: <TASK> ? o2net_rx_until_empty+0x19b1/0x3300 [ocfs2_nodemanager] ? __pfx_o2net_rx_until_empty+0x10/0x10 [ocfs2_nodemanager] ? lock_acquire+0x190/0x300 ? process_one_work+0x935/0x1b40 ? process_one_work+0x834/0x1b40 ? srso_alias_return_thunk+0x5/0xfbef5 ? lock_release+0xc8/0x290 ? srso_alias_return_thunk+0x5/0xfbef5 ? process_one_work+0x9a8/0x1b40 ? __pfx_process_one_work+0x10/0x10 ? lock_acquire+0x190/0x300 ? lock_is_held_type+0x8f/0x100 ? srso_alias_return_thunk+0x5/0xfbef5 ? worker_thread+0x65c/0xe40 ? srso_alias_return_thunk+0x5/0xfbef5 ? __kthread_parkme+0x177/0x220 ? srso_alias_return_thunk+0x5/0xfbef5 ? __pfx_worker_thread+0x10/0x10 ? kthread+0x351/0x460 ? srso_alias_return_thunk+0x5/0xfbef5 ? __pfx_kthread+0x10/0x10 ? ret_from_fork+0x659/0x940 ? __pfx_ret_from_fork+0x10/0x10 ? srso_alias_return_thunk+0x5/0xfbef5 ? __switch_to+0x74f/0xf70 ? __pfx_kthread+0x10/0x10 ? ret_from_fork_asm+0x1a/0x30 </TASK> Modules linked in: ocfs2_nodemanager [last unloaded: ocfs2_dlm] ---[ end trace 0000000000000000 ]---
CVEs for which no patch, fixed version or vendor advisory is linked in the CVE.org or NVD record. Treat these as exposed until the vendor says otherwise.