CVEs of 2026.
Every CVE published in 2026 that we hold, ranked by exploit odds.
-
CVE-2026-99134
In the Linux kernel, the following vulnerability has been resolved: mac802154: fix data race and NULL deref on local->assoc_dev local->assoc_dev is shared between the association path and the association-response worker without common synchronization. mac802154_perform_association() stores the coordinator pointer and waits for a response. Its timeout and error paths clear the pointer and return to mac802154_associate(), which may then free the coordinator object. Meanwhile, mac802154_rx_mac_cmd_worker() may observe the associating bit and enter mac802154_process_association_resp(), which dereferences assoc_dev. The worker's bit test and the handler's pointer dereference are not atomic with respect to cleanup. Cleanup can clear assoc_dev between them, causing a NULL dereference, or free the coordinator while the response handler still uses the pointer. The recorded result is exposed to the same window. assoc_status and assoc_addr are written by the handler but read by the association path while the associating bit is still set, so a second response for the same request - a malicious one, for instance - can replace them between those reads and leave the caller with an incoherent status and address pair. The response handler only needs the coordinator extended address. Replace assoc_dev with a cached address, removing the pointer lifetime dependency. Protect the cached address and the associating bit with a dedicated spinlock. A READ_ONCE()/WRITE_ONCE() pair would not guarantee an atomic __le64 access on all 32-bit architectures. wpan_dev->association_lock cannot be reused here: nl802154_associate() holds it across rdev_associate(), hence for the whole of mac802154_perform_association() including the wait for the response. A response handler taking that lock would only get it once the association has already given up. Reset the completion, publish the cached address, and set the associating bit while holding the lock. The response handler takes the lock, rechecks the bit and the cached address, records the response, clears the bit, and only then completes the waiter. Thus cleanup cannot pass the handler between its state check and completion, and the cached 64-bit value cannot tear. The handler clears the bit before completing, not the woken waiter: otherwise complete() is issued under the lock and a second (e.g. malicious) response can reacquire it before the waiter and replace the result. So a wait that returns success implies the bit is already clear, and the success and negative-response paths return directly. The transmit-error and timeout paths still clear it under assoc_lock, which serializes any racing response against the cleanup while the call returns the error it already selected. Both paths snapshot assoc_status and assoc_addr under the same lock. Both users run in process context, so a plain spinlock is sufficient. The lock is not held while waiting for the completion.
-
CVE-2026-99105
In the Linux kernel, the following vulnerability has been resolved: sched_ext: Fix stall when a task enqueued with SCX_ENQ_LAST lands back on its CPU's local DSQ With SCX_OPS_ENQ_LAST, the last runnable task on a CPU is passed to ops.enqueue() instead of being kept running. Exiting tasks without SCX_OPS_ENQ_EXITING, migration disabled tasks without SCX_OPS_ENQ_MIGRATION_DISABLED and tasks on an offline rq skip ops.enqueue() and go straight to the local DSQ. This happens in put_prev_task_scx() after the pick has settled on idle, so nothing reschedules the CPU, which idles with the task queued until an unrelated wakeup lands on it or the watchdog fires. scx_mitosis, which sets SCX_OPS_ENQ_LAST without SCX_OPS_ENQ_EXITING, hit this during mass cgroup teardown: exiting tasks stuck in exit_mmap() for over 40 seconds. A task put on the local DSQ runs, as anywhere else. If the SCX_ENQ_LAST enqueue left the task on this CPU's local DSQ, set need_resched on the task being switched in, as proxy_resched_idle() does. resched_curr() would flag the task being put instead, and __schedule() clears that right after the pick.
-
CVE-2026-112672
The issue was addressed with improved checks. This issue is fixed in macOS Golden Gate 27.2, macOS Sequoia 15.8.2, macOS Tahoe 26.7.2. A maliciously crafted ZIP archive may bypass Gatekeeper checks.
-
CVE-2026-99095
In the Linux kernel, the following vulnerability has been resolved: drm/vc4: drain the hangcheck timer and works on V3D unbind Nothing stops the hangcheck timer that submitted jobs arm, or the job_done_work that the render-done interrupt queues, at teardown: vc4_irq_disable() cancels only overflow_mem_work, and vc4_gem_destroy() runs from the drm-managed release, after vc4_v3d_unbind() has already uninstalled the V3D interrupt and cleared vc4->v3d. A timer still armed by then reads V3D registers through the NULL vc4->v3d pointer, and the late callbacks run on the vc4_dev embedding them after it has been freed. Drain them in vc4_v3d_unbind(): shut the timer down and cancel reset_work before the interrupt is taken down, because its vc4_irq_reset() re-enables it, then flush job_done_work so that completions queued at teardown still release their jobs. This issue was found by an in-house static analysis tool.
-
CVE-2026-99091
In the Linux kernel, the following vulnerability has been resolved: Bluetooth: MGMT: Fix status of pending commands flushed on power off cmd_complete_rsp() receives a struct cmd_lookup and, for a pending command that has no cmd_complete callback, falls through to cmd_status_rsp(cmd, data). cmd_status_rsp() reads data as a u8 *, so the status sent comes from the first byte of match->sk's representation, not from match->mgmt_status. commit f53e1c9c726d ("Bluetooth: MGMT: Fix possible crash on mgmt_index_removed") changed the callers' data from &status to &match and updated the cmd_complete branch to match->mgmt_status, but left the fall-through passing data unchanged. In mgmt_index_removed() match->sk is NULL, so every such command is answered with status 0x00 (Success) instead of MGMT_STATUS_INVALID_INDEX. In __mgmt_power_off() it is NULL too, unless a Set Powered command was pending: then settings_rsp() has stored that command's socket in match->sk, and the byte sent comes from the socket pointer. Either way the status the caller set, MGMT_STATUS_NOT_POWERED (or MGMT_STATUS_INVALID_INDEX when the device is being unregistered), is lost. Pass the status the callers set. This was seen on a QCA9377 laptop: bluetoothd's Start Discovery, pending when the adapter was powered off, was answered with Command Status 0x00 and no parameters, and bluetoothd 5.72 crashed on the zero-length success. BlueZ master has since added the length check on that path ("adapter: Fix crash on short start discovery reply", BlueZ a734b06059cb), but userspace should still be told Not Powered, not Success. Tested on Ubuntu 7.0.0-31 with a virtual controller (hci_vhci): power off is held in HCI Write Scan Enable while a Start Discovery is submitted, and the flush answers it with status 0x00 before this change and 0x0f (Not Powered) after.
-
CVE-2026-99062
In the Linux kernel, the following vulnerability has been resolved: vsock/virtio: account only unread bytes in read_skb() After a partial stream receive, rx_bytes tracks the unread suffix while buf_used and peer credit still cover the whole packet. virtio_transport_read_skb() dequeues that packet but passes pkt_len for both counters, underflowing rx_bytes. On a connected CID_LOCAL stream with an SK_SKB verdict, a 40-byte read from a 100-byte packet made SIOCINQ report -40, an empty recv return ELOOP, and poll report the empty socket readable. After verdict detach, receiving 40 bytes wrapped the counter to zero and hid those queued bytes until one more byte arrived. The fixed run kept the counter balanced and exposed all queued data immediately. This was found during an LLM-assisted manual source audit of VSOCK receive accounting while re-evaluating virtio_transport_read_skb() after CVE-2024-50169. Subtract only skb->len minus the VSOCK offset from rx_bytes. Retain the full packet length for buf_used and peer credit.
-
CVE-2026-99069
In the Linux kernel, the following vulnerability has been resolved: Input: ati_remote2 - fix type confusion in device attribute handlers Previously, the channel_mask and mode_mask attributes were registered on the parent usb_device via sysfs_create_group(). When commit b20d6bf8014b ("Input: ati-remote2 - use driver core to instantiate device attributes") converted the driver to let the driver core manage them via dev_groups in struct usb_driver, the attributes moved to the bound usb_interface devices, but the show and store callbacks were not updated accordingly. As a result, the callbacks still invoked to_usb_device() on a device pointer belonging to a struct usb_interface, producing an invalid usb_device pointer. Use to_usb_interface() directly in the attribute handlers to obtain the interface and its driver data.
-
CVE-2026-99055
In the Linux kernel, the following vulnerability has been resolved: tcp: reject net_iov in zerocopy receive mapping hints After copying a readable prefix, receive_fallback_to_copy() asks tcp_zerocopy_set_hint_for_skb() where page mapping can resume. If the next skb is unreadable, find_next_mappable_frag() passes its net_iov fragment to can_map_frag(). skb_frag_page() returns NULL for a net_iov, but can_map_frag() dereferences it in PageCompound(). A v7.2 KASAN run on a connected TCP socket with 64 readable bytes followed by a 4096-byte NET_IOV_DMABUF fragment reported: BUG: KASAN: null-ptr-deref in can_map_frag tcp_zerocopy_receive -> can_map_frag Kernel panic - not syncing: KASAN: panic_on_warn set The diagnostic inserted the net_iov directly because the test host has no devmem-capable NIC. Hardware end-to-end reachability remains untested and requires CONFIG_NET_DEVMEM plus a supported DMA-buf-bound RX queue. Reject all net_iov fragments before skb_frag_page(). This covers both DMABUF and IOURING net_iov types while leaving page-backed checks unchanged. With the guard, the same queue copied the readable prefix, returned a 4096-byte skip hint, and completed without a fault.
-
CVE-2026-99060
In the Linux kernel, the following vulnerability has been resolved: wifi: mt76: mt7603: initialize global station WCID mt7603 uses a driver-private global station WCID for frames that have neither a station nor a vif, including frames injected through a monitor interface. Unlike the core global WCID initialized by mt76_register_device(), this WCID is published without ever being passed to mt76_wcid_init(). Since commit 0335c034e726 ("wifi: mt76: fix race condition related to checking tx queue fill status"), mt76_tx() queues every such frame on the WCID's tx_pending or tx_offchannel queue and links its tx_list. The first injected frame therefore reaches __skb_queue_tail() with an uninitialized sk_buff_head and dereferences a NULL prev pointer. Initialize the WCID before publishing it. The fix was tested on an MT7603 PCIe device. Three wildcard probe requests were injected through a monitor interface. The hardware queue head and tail advanced from 0 to 3 and a second radio captured the third request over the air. No oops or reset occurred. After restoring AP mode, a 150 MB transfer also completed without a reset.
-
CVE-2026-99028
In the Linux kernel, the following vulnerability has been resolved: net: xilinx: axienet: Free outstanding DMA buffers on dmaengine stop In the dmaengine path the driver pre-submits RX buffers and holds in-flight TX buffers whose SKBs are DMA-mapped by the driver and freed only in the completion callbacks. On ndo_stop() dmaengine_terminate_sync() aborts these descriptors without running their callbacks, and the driver then frees only the ring shells, leaking every SKB still owned by the engine and its DMA mapping on each ifdown. With 128 RX buffers pre-posted per channel, the mapping leak can eventually exhaust a limited IOMMU aperture. Clear the slot's skb in the TX and RX callbacks so a non-NULL skb marks a slot that still owns a live, DMA-mapped buffer, and on stop unmap and free every such buffer. axienet_dma_rx_cb() runs from the DMA tasklet and re-arms the RX ring on each completion, so it can race axienet_stop(): a completion may submit a fresh buffer after dmaengine_terminate_sync() has returned, leaving the channel armed with a buffer the teardown then frees while the engine may still write into it (dma_release_channel() does not stop it either). Add a lock that axienet_dma_rx_cb() holds across the @stopping check and the resubmit, and axienet_stop() holds to set @stopping before terminating. Once @stopping is set no callback can arm a new buffer, and any armed just before is aborted by the terminate, so teardown only frees buffers the engine no longer owns.
-
CVE-2026-99020
In the Linux kernel, the following vulnerability has been resolved: regulator: tps6594-regulator: Fix n_linear_ranges for TPS65224 BUCK2-4 The TPS65224 BUCK2, BUCK3 and BUCK4 descriptors pass 4 as n_linear_ranges, but tps65224_bucks_2_3_4_ranges[] only has two entries, covering selectors 0x00-0x45. The linear range helpers loop over n_linear_ranges entries, so any lookup of a selector above 0x45 reads past the end of the array and treats whatever follows it as two more ranges. This happens on the normal registration path when the core checks the voltage constraints against every selector. TPS652G1 uses the same descriptors and is affected too. Use ARRAY_SIZE() so the count always matches the table.
-
CVE-2026-99003
In the Linux kernel, the following vulnerability has been resolved: drm/panthor: validate userspace queue count against initialized firmware slot count In panthor_drv.c:panthor_ioctl_group_create(), the queue count is validated against the hardcoded maximum: if (!args->queues.count || args->queues.count > MAX_CS_PER_CSG) return -EINVAL; However, if the firmware reports fewer slots than MAX_CS_PER_CSG, panthor_fw_init_ifaces() will only initialize that smaller number of stream interfaces. If unprivileged userspace requests more queues than the firmware reported (but no more than MAX_CS_PER_CSG), cs_slot_prog_locked() fetches an uninitialized stream interface and may crash when dereferencing cs_iface->input->ringbuf_base.
-
CVE-2026-99004
In the Linux kernel, the following vulnerability has been resolved: drm/imagination: Size page table preallocation by device address Commit 0a8224058a58 ("drm/imagination: Fix page count for page table for map() interface") passed the device address to pvr_mmu_op_context_create(), but the preallocation is still sized from device_addr + sgt_offset with an exclusive end. The offset into the object's pages has no place in a device-virtual range, and the exclusive end allocates one table too many when the range ends on a table boundary. Count the tables from device_addr and the inclusive end of the range. The new calculation does not take page table indices into account. This fixes another issue with the previous implementation, where the difference between start and end indices could underflow when a mapping crossed a page table boundary. This would lead to the driver preallocating a huge number of MMU pages, exhausting system memory and triggering an out of memory panic. [Alessio: add paragraph to point out OOM bug being fixed, cc stable]
-
CVE-2026-98996
In the Linux kernel, the following vulnerability has been resolved: fs: don't silently unmount busy mounts The propagate_mount_busy() helper exists to decide whether a synchronous can suceed. It does that by looking at the copies of the victims in the mounts its parent propagates to. One of those functions from propagation hell. It skips a copy that has child mounts except for the case of a single child mount covering the copy's root. That algorithm used to be exact. It isn't anymore. Originally only childless copies were unmounted during propagation. And 1064f874abc0 ("mnt: Tuck mounts under others instead of creating shadow/side mounts.") ended up adding covered mounts to both sides in one go. Along came 99b19d16471e ("mnt: In propgate_umount handle visiting mounts in any order") and that started to unmount more than before: [1]: A copy of a mount is unmounted when each of its children is either its overmount or a copy of the victim itself. In the former case the overmount gets reparented. In the latter case the copy of the victim will end up being unmounted. So propagate_umount() ended up being rewritten without adjusting propagate_mount_busy(). And currently propagate_mount_busy() doesn't look at copies of the victim that have two or more children nor a single child that is not the overmount. So any such copy in [1] is not checked for references. The result is that a synchronous unmount succeeds, pulling out a mount that is still in use. A container can run into this with its own mounts. The host shares a tree with the container and the host has a mount in that tree: mount -t tmpfs none /mnt mount --make-shared /mnt mount -t tmpfs none /mnt/a The container's /mnt is a slave copy of the host mount. The container now takes a detached copy of the tree and mounts that copy beneath its propagated copy of /mnt/a and keeps the file descriptor to the tree open: unshare -m --propagation unchanged mount --make-rslave / fd = open_tree(AT_FDCWD, "/mnt", OPEN_TREE_CLONE | AT_RECURSIVE) move_mount(fd, "", AT_FDCWD, "/mnt/a", MOVE_MOUNT_BENEATH) grep /mnt /proc/self/mountinfo 775 355 ... /mnt ... master:646 777 775 ... /mnt/a ... master:646 776 777 ... /mnt/a ... master:647 778 777 ... /mnt/a/a ... master:647 The now attached tree (777) is now mounted beneath the container's copy of /mnt/a. That's also where the host mount sits (as seen from the host mount namespace). So the propagated copy (776) of the host's mount is on top of its root and the copy of that (778) is inside it. Now the host runs: umount /mnt/a and it succeeds. The moved tree tree and the copy inside of it are gone from the container. Now only the propagated copy is left and it got reparented to where it was before: grep /mnt /proc/self/mountinfo 775 355 ... /mnt ... master:646 776 775 ... /mnt/a ... But the file descriptor still refers to the moved tree. So a synchronous umount detached a mount that is still in use. Teach propagate_mount_busy() the same checks that propagate_umount() does. It now checks for a single victim mount without children for which it can assume that MNT_LOCKED is cleared on every copy like propagate_umount() does. The copies form chains. IOW, when a copy itself receives propagation from the victim's parent then the mount at the same mountpoint inside that copy is a copy as well. Whether a mount receives propagation from the victim's can simply be checked by walking its masters. Walk each chain once from its top so we are sure that nested copies stay linear. Check every reference once of every copy that gets unmounted.
-
CVE-2026-98989
In the Linux kernel, the following vulnerability has been resolved: namespace: check a recursive bind mount for mount namespace loops do_loopback() refuses to bind mount the file of a mount namespace that is as old as the caller's or older because a mount namespace that holds a mount of its own file, or of an ancestor's, would cause a cycle. But it only checks the dentry the bind mount starts from. With MS_REC everything below it is copied and do_loopback() passes CL_COPY_MNT_NS_FILE so mount namespace files below the source are copied. That's fine for a source in the caller's own mount namespace. Every mount namespace file in there passed the same check when it was bind-mounted. It isn't fine for a source in another mount namespace. may_copy_tree() accepts a bind mount of any nsfs or pidfs file no matter what mount namespace it lives in so that /proc/<pid>/ns/<ns> and pidfds can be bind mounted. If we create a bind-mount stack of mount namespaces file descriptors the kernel will copy them irrespective of their ancestoral relationship to the mount namespace in question: 151 146 0:7 net:[4026531833] /tmp/nrc/y rw - nsfs nsfs rw 152 151 0:7 mnt:[4026532293] /tmp/nrc/y rw - nsfs nsfs rw Mount 152 is a mount of the mount namespace file of mount namespace 4026532293 inside mount namespace 4026532293. That namespace and every mount in it are leaked... All it takes is a process in an older mount namespace that stacks the file and repeating it leaks without limit: after control: Shmem: 380 kB child: mount(/proc/self/fd/6, MS_BIND|MS_REC): ok after cycle: Shmem: 65916 kB do_move_mount() runs check_for_nsfs_mounts() over a detached tree for exactly that reason. Do the same for the copy before it is grafted.
-
CVE-2026-98949
In the Linux kernel, the following vulnerability has been resolved: drm/imagination: Fix reference and vm_bo handling in remap() When a map overlaps part of an existing mapping, pvr_vm_gpuva_remap() splits the mapping into prev/next parts covering what the request did not take, instead of creating something new. It gets two things wrong. - A GEM reference is taken for each part but it's never needed and never dropped, so it leaks one reference per split (remap-next-in-2m in tests/imagination/pvr_vm_map.c): CRITICAL: 33558528 bytes of shmem still held after close - The parts still belong to the original object being split, but pvr_vm_gpuva_remap() links them to ctx->gpuvm_bo, the new object's vm_bo: prev_va --obj--> BO_A BO_B <--obj-- vm_bo (ctx->gpuvm_bo) \___________link___________/ (mismatch: BO_A != BO_B) drm_gpuva_link() catches the mismatch: WARNING: drivers/gpu/drm/drm_gpuvm.c:2108 at drm_gpuva_link+0x2ec/0x310 drm_WARN_ON(obj != vm_bo->obj) Call trace: drm_gpuva_link pvr_vm_gpuva_remap __drm_gpuvm_sm_map pvr_vm_map pvr_ioctl_vm_map Link them to op->remap.unmap->va->vm_bo instead. The locking of the GPUVA lists of the other objects touched by a split is not addressed here; it is handled by switching the GPUVM to immediate mode.
-
CVE-2026-98920
In the Linux kernel, the following vulnerability has been resolved: netfs: Fix missing alloc tagging of direct mempool allocations Commit 1d78d56c43ef ("netfs: Fix folio_queue ENOMEM in writeback by adding a mempool") added a mempool for the folio_queues and made the request, subrequest and folio_queue allocations distinguish between writeback and everything else. Writeback is part of memory reclaim and must not fail due to ENOMEM, so it allocates under GFP_NOFS through mempool_alloc(), which may dip into the pool's reserve and, if that runs empty, wait for elements to be returned. The GFP_KERNEL paths, which can return -ENOMEM to their callers, invoke the pool's ->alloc() callback directly instead. The direct call, however, skips the alloc_hooks() wrapper that the mempool_alloc() macro provides. The pool callbacks, mempool_alloc_slab() and mempool_kmalloc(), call kmem_cache_alloc_noprof() and kmalloc_noprof() and rely on current->alloc_tag having been set by the caller. With CONFIG_MEM_ALLOC_PROFILING_DEBUG=y this leads to current->alloc_tag not set WARNING: ./include/linux/alloc_tag.h:161 at __alloc_tagging_slab_alloc_hook alloc_tag was not set WARNING: ./include/linux/alloc_tag.h:166 at __alloc_tagging_slab_free_hook at allocation and free time respectively, as reported when reading files on a CIFS mount. The allocations are also missing from /proc/allocinfo. Wrap the direct ->alloc() invocations in alloc_hooks() with a new mempool_alloc_noreserve() helper in include/linux/mempool.h, next to the other alloc_hooks()-wrapped macros such as mempool_alloc(). The GFP_KERNEL paths keep their failable allocation semantics, they just get tagged now.
-
CVE-2026-98923
In the Linux kernel, the following vulnerability has been resolved: arm64: io: Reject non-user protection in ioremap_prot() Mapping a stack-top page via /dev/mem with PROT_NONE and then reading that process's /proc/<pid>/cmdline triggers a spurious WARN in ioremap_prot() through generic_access_phys(): WARNING: ./arch/arm64/include/asm/io.h:275 at generic_access_phys Call trace: generic_access_phys+0x1c8/0x228 (P) __access_remote_vm+0x2b4/0x398 access_remote_vm+0x14/0x30 get_mm_cmdline+0xf8/0x2a0 proc_pid_cmdline_read+0x68/0x120 generic_access_phys() passes the protection derived from the user PTE to ioremap_prot(). On arm64, a PROT_NONE mapping is represented by a present-invalid PTE, so pte_present() still returns true and the protection reaches ioremap_prot(). A PROT_NONE mapping does not have PTE_USER, causing the existing WARN_ON_ONCE() in ioremap_prot() to fire even though this is a valid user mapping. Execute-only mappings have the same issue and must not be readable through this path either. ioremap_prot() should therefore reject protection values without PTE_USER without warning. This makes the access fail cleanly for PROT_NONE and execute-only mappings while retaining the existing user-protection contract.
-
CVE-2026-98871
In the Linux kernel, the following vulnerability has been resolved: crypto: ccp - Fix possible deadlock in SEV init failure path __sev_platform_init_handle_init_ex_path() calls rmp_mark_pages_firmware() with locked=false while the parent function of init_ex_path already acquired the sev_cmd_mutex. In the case of an RMPUPDATE failure for any page after the first, the cleanup path would invoke reclaim pages which would result in a deadlock in sev_do_cmd. Pass locked=true to honor the lock status of the parent function.
-
CVE-2026-24205
NVIDIA TensorRT-LLM for any platform contains a vulnerability where a user may cause a race condition by running concurrent requests. A successful exploit of this vulnerability may lead to data tampering.
-
CVE-2026-128719
An issue in PHPGurukul Employee Leaves Management System (ELMS) Version 2.1 allows a remote attacker to escalate privileges via the password recovery functionality
-
CVE-2026-115022
Vulnerability in the Oracle Hyperion Data Relationship Management product of Oracle Hyperion (component: Access and security). The supported version that is affected is 11.2.26.0.000. Easily exploitable vulnerability allows unauthenticated attacker with network access via HTTP to compromise Oracle Hyperion Data Relationship Management. Successful attacks require human interaction from a person other than the attacker. Successful attacks of this vulnerability can result in unauthorized creation, deletion or modification access to critical data or all Oracle Hyperion Data Relationship Management accessible data. CVSS 3.1 Base Score 6.5 (Integrity impacts). CVSS Vector: (CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:N/I:H/A:N).
-
CVE-2026-118462
Update components download and execute code without proper authenticity verification, relying on ignored hashes or forgeable checksums. Some update paths also permit archive traversal or script execution.
-
CVE-2026-99211
In the Linux kernel, the following vulnerability has been resolved: netlink: reset skb headers before unicast and multicast filtering Socket filters can use SKF_NET_OFF and SKF_LL_OFF to read relative to an skb's network and MAC headers. Netlink producers may reserve headroom without resetting those headers, allowing a receiving socket's filter to read data before the initialized message. For example, IWPM HELLO replies are allocated with dev_alloc_skb(), which reserves NET_SKB_PAD bytes. An unprivileged NETLINK_RDMA user can attach a classic socket filter and disclose this headroom through the length of the delivered reply. The resets in netlink_dump() do not help because IWPM sends a separately allocated skb through netlink_unicast(). Reset both headers before running the receiver's filter in unicast and broadcast delivery, as is already done for dump skbs. This also covers other netlink producers that reserve headroom without changing their allocation or message-building code.
-
CVE-2026-99183
In the Linux kernel, the following vulnerability has been resolved: accel/rocket: search every core slot when looking up a scheduler sched_to_core() walks rdev->cores[] up to rdev->num_cores, and rocket_remove() decrements num_cores for every core it removes. Unbind a core that is not the last one and the cores behind it fall outside the search, so sched_to_core() returns NULL for a core that is still bound and still running jobs. Neither caller checks the result: rocket_job_run(): rocket_fence_create(core), core->dev rocket_job_timedout(): dev_err(core->dev, "NPU job timed out") Unbinding the middle core of the three on an RK3588 while three clients are submitting to all of them faults twice, once from the surviving core's job queue and once from its reset work: KASAN: null-ptr-deref in range [0x0000000000000220-0x0000000000000227] Workqueue: fdad0000.npu drm_sched_run_job_work [gpu_sched] pc : rocket_job_run+0x234/0x838 [rocket] Call trace: rocket_job_run+0x234/0x838 [rocket] drm_sched_run_job_work+0x2cc/0xad8 [gpu_sched] process_one_work+0x640/0x14f0 KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007] Workqueue: rocket-reset-2 drm_sched_job_timedout [gpu_sched] pc : rocket_job_timedout+0xf0/0x1e0 [rocket] Call trace: rocket_job_timedout+0xf0/0x1e0 [rocket] drm_sched_job_timedout+0x188/0x6a0 [gpu_sched] Both are the third core: the workqueue names are its device and its core->index, and it was left at slot 2 while num_cores had dropped to 2. Search all the slots that were allocated, the way find_core_for_dev() now does. A core that is still bound is then found, and the two callers get the pointer they already assume they have. This does not make unbinding one core out of several safe. An open client keeps an entity pointing at the scheduler of the core that went away: drm_sched reports it as not ready for every job that lands on it, and the client waits in dma_fence_default_wait for a fence that will never signal. Stopping the NULL dereference is what belongs in a fix; the rest wants more thought.
-
CVE-2026-99184
In the Linux kernel, the following vulnerability has been resolved: accel/rocket: search every core slot when a core is removed rocket_remove() decrements rdev->num_cores for each core it removes, while find_core_for_dev() searches slots 0 to num_cores - 1. Unbinding the cores in the order they were bound therefore loses the last one: by the time it is removed the search range has already shrunk past its slot, so find_core_for_dev() returns -1 and rocket_remove() gives up without doing anything. num_cores never reaches zero, rocket_device_fini() never runs, and the file-scoped rdev keeps pointing at a device that is going away. Binding the cores again starts from that stale count, because rocket_probe() takes rdev->num_cores as the slot to fill. On an RK3588, which describes three cores, the second round lands on slots 1, 2 and 3 while rdev->cores was allocated with room for three: rocket fdab0000.npu: Rockchip NPU core 1 version: 1179210309 rocket fdac0000.npu: Rockchip NPU core 2 version: 1179210309 rocket fdad0000.npu: Rockchip NPU core 3 version: 1179210309 The write to rdev->cores[3] is past the end of the array. Nothing in tree reads the core array often enough to notice, so the overrun is silent today. It turned up while testing a devfreq series on top of this, where a worker walks every core a few times a second, and UBSAN caught the first bool it read out of the overrun entry: UBSAN: invalid-load in drivers/accel/rocket/rocket_devfreq.c:47:10 load of value 5 is not a valid value for type '_Bool' Workqueue: devfreq_wq devfreq_monitor Record how many slots were allocated and search all of them. Every core is then found on removal, num_cores reaches zero, the device is torn down and a later bind starts from a clean rdev. This does not make unbinding a single core out of several work. probe still takes num_cores as the slot to fill, so rebinding one core while its siblings stay bound would write over a slot that is already in use, and rocket_open() still reaches for cores[0] whether or not anything is there. Both of those want more thought than a fix should carry. Found by unbinding and rebinding all three cores on an Orange Pi 5 Plus. With this applied, 25 unbind/rebind rounds and 5 module unload/reload rounds run clean there: the cores land in slots 0, 1 and 2 every time, whichever order they are bound in, and the shared supply goes back to a single user after each round.
-
CVE-2026-99137
In the Linux kernel, the following vulnerability has been resolved: cifs: don't update i_size in cifs_do_truncate without a cached handle If find_writable_file() returns null, cifs_file_flush will return 0 without issuing set_file_size, and the outer 'if (!rc)' block will set i_size to 0 before telling the server to truncate. If the cifs_open() then fails, the inode will have size 0, while the server file is unchanged. Move the netfs_resize_file() and cifs_setsize() into the 'if (cfile)', so they only run after a successful set_file_size. In the no-handle else branch, evict stale pages with truncate_inode_pages before the O_TRUNC open to dispose of old cache pages, and let the open response set the i_size.
-
CVE-2026-99111
In the Linux kernel, the following vulnerability has been resolved: smb: client: fix data corruption with concurrent writes and O_TRUNC cifs_do_truncate() flushes dirty pages with filemap_write_and_wait() and truncates the file on the server, but in the old code both operations ran without holding i_rwsem or invalidate_lock. A concurrent buffered write via netfs_perform_write() -- which only needs i_rwsem shared -- could dirty new pages after the flush but before the local truncation, and those pages would be silently discarded by cifs_setsize() -> truncate_pagecache(). Fix by acquiring inode_lock (exclusive i_rwsem) and filemap_invalidate_lock at the top of cifs_do_truncate(), so the entire flush-truncate-resize sequence is atomic with respect to: - buffered writes (blocked by exclusive i_rwsem, since netfs_start_io_write takes i_rwsem shared), - read page faults (blocked by exclusive invalidate_lock, since filemap_fault takes it shared), - writeback collection (blocked by netfs_wb_begin/netfs_wb_end around the server truncate and local resize, since netfs_writepages also acquires the wb lock).
-
CVE-2026-98929
In the Linux kernel, the following vulnerability has been resolved: macsec: initialize SecY before registering the netdevice Creating a MACsec device with MAC offload over an LRO-capable lower device triggers a warning in rtmsg_ifinfo_build_skb() when IPv4 forwarding is enabled by default. register_netdevice() invokes inetdev_init(), which disables LRO and emits a NETDEV_FEAT_CHANGE notification. This reaches macsec_fill_info() before macsec_add_dev() initializes the SecY. key_len is still zero, so macsec_fill_info() returns -EMSGSIZE and trips the WARN_ON in rtmsg_ifinfo_build_skb(), even though the skb has enough space. Even without the warning, notifications during registration can report uninitialized SecY attributes, including the SCI. This ordering has existed since the driver was introduced. Initialize the SecY and apply the new-link attributes before registration. Move MAC address inheritance into macsec_newlink() so the SCI can also be initialized before registration-time notifications report it. Move the per-CPU statistics and metadata destination allocation into ndo_init(), and release partial allocations on failure.
-
CVE-2026-99745
In the Linux kernel, the following vulnerability has been resolved: x86/bugs: Select the BHI mitigation based on the attack vector properly The BHI mitigation selection is inconsistent with the documentation (attack_vector_controls.rst), which lists BHI under both the user->kernel and guest->host attack vectors. bhi_select_mitigation() only enables it when CPU_MITIGATE_GUEST_HOST is set; otherwise it selects BHI_MITIGATION_OFF, even if CPU_MITIGATE_USER_KERNEL is set. As a result, a kernel compiled with KVM disabled gets BHI automatically disabled, since CPU_MITIGATE_GUEST_HOST defaults to off without KVM. This is a bug, given that CPU_MITIGATE_USER_KERNEL still needs it: on affected Intel CPUs these kernels lose BHI_DIS_S or the BHB clearing loop by default. Booting with mitigations=auto,no_guest_host has the same effect. Select BHI_MITIGATION_ON whenever user->kernel is mitigated, and keep BHI_MITIGATION_VMEXIT_ONLY for the guest->host-only case. [ bp: Clarify commit title. ]
-
CVE-2026-99726
In the Linux kernel, the following vulnerability has been resolved: drm/v3d: Widen cache_clean_lock over the whole L2TCACTL sequence v3d_clean_caches() and v3d_flush_l2t() both write the single L2TCACTL register and poll its status bits. The mutex cache_clean_lock exists to serialize them, but v3d_clean_caches() only took the lock around its final FLM_CLEAN write. These functions run concurrently: v3d_flush_l2t() is issued from the BIN/RENDER/CSD invalidate path while v3d_clean_caches() runs from the CACHE_CLEAN queue, and each queue's scheduler uses its own ordered workqueue, so their run_job callbacks execute in parallel. Because clean locked only its final write, a concurrent flush can write L2TCACTL during clean's unlocked phase. Both use non read-modify-write writes to the one register, so whichever lands last wins: clean's TMUWCF write can land on the flush's in-flight L2TFLS invalidate, triggering the GFXH-1897 hazard of writing L2TCACTL while a flush is pending. Hold cache_clean_lock across the entire L2TCACTL access sequence so it is fully mutually exclusive with v3d_flush_l2t(), which already takes the lock around its own write.
-
CVE-2026-99703
In the Linux kernel, the following vulnerability has been resolved: mm/hugetlb: fix mremap deadlock when PMD sharing is possible move_hugetlb_page_tables() holds the mapping's i_mmap_rwsem for write while moving hugetlb PTEs. This prevents races with truncation and protects huge_pmd_unshare(), but the move loop must not call helpers that can acquire the same rwsem for read. huge_pte_alloc() can do that when the destination can use PMD sharing: it may call huge_pmd_share(), which takes i_mmap_rwsem for read while searching for a shareable PMD table. Calling it while the move path already holds i_mmap_rwsem for write can deadlock. Avoid that path for PMD-sized hugetlb mappings that can use PMD sharing. Instead of allocating or sharing the destination page table while moving, zap the source range and let the destination fault path populate or share the PMD lazily. Reuse the hugetlb unmap logic for the source-side zap by splitting the core loop out of __unmap_hugepage_range(). The wrapper keeps the existing tlb_start_vma(), tlb_end_vma() and huge_pmd_unshare_flush() behavior, while move_hugetlb_page_tables() can reuse the unmap loop with its existing mmu_gather.
-
CVE-2026-99698
In the Linux kernel, the following vulnerability has been resolved: mm/cma: accept memory of CMA areas reserved at a fixed address CMA areas allocated from memblock have their memory accepted by memblock_alloc_range_nid(), but areas reserved at a fixed address with memblock_reserve() are not. If a fixed CMA area is requested by the cma=<size>@<base> command line parameter, it may be in regular RAM and on TDX and SEV-SNP guests it must be accepted. Accept the memory of fixed CMA areas as well, so that all CMA areas are accepted regardless of how they were reserved.
-
CVE-2026-99699
In the Linux kernel, the following vulnerability has been resolved: mm/cma: accept memory of multi-range CMA areas Patch series "mm/cma: accept memory of CMA areas reserved with memblock_reserve()", v1. Pages of allocated CMA areas are released to the buddy allocator with __free_pages(), which does not check for unaccepted memory. This usually works because memblock accepts the memory it allocates. However, multi-range CMA areas, used by hugetlb_cma, and fixed CMA areas requested with cma=<size>@<base> are reserved with memblock_reserve() and their memory is never accepted. On TDX and SEV-SNP guests the first access to such a page is fatal. To ease the backporting, each case is addressed by a separate patch: - patch 1 accepts the memory of multi-range CMA areas - patch 2 does the same for the fixed CMA areas The second patch won't apply cleanly to 6.6 and 6.12, but I believe the stable team LLMs will take care of that :) This patch (of 2): On systems with unaccepted memory, memblock allocations accept the allocated range, but when cma_declare_contiguous_multi() uses open coded allocation of multiple ranges, it reserves them with memblock_reserve() and they are never accepted. These pages are released to the buddy allocator by cma_activate_area() with __free_pages() that does not check for unaccepted memory, so the first access to such a page in a TDX or SEV-SNP guest is fatal. Accept the memory of each range when it is reserved.
-
CVE-2026-99700
In the Linux kernel, the following vulnerability has been resolved: mm/hugetlb_vmemmap: map the shared tail page writable page_ref_add_unless_zero() reads the refcount and then updates it with a cmpxchg. A lockless get, such as a page cache lookup or a PFN scanner, can read a non-zero refcount from a page that is then freed and reused as a hugetlb tail page. If HVO remaps its struct page to the shared tail page in between, the cmpxchg writes to read-only vmemmap: CPU0 (lockless get) CPU1 c = atomic_read(&page->_refcount) page freed page allocated as a hugetlb tail HVO remaps the struct page RO try_cmpxchg(&page->_refcount, &c, c + 1) -> write fault on read-only vmemmap The refcount is zero by then and the compare fails, but a compare-and-swap instruction needs write access whether or not it succeeds. The kernel oopses when hugetlb pages are allocated at runtime with HVO enabled while a lockless get races with it. Commit bd225530a4c7 ("mm/hugetlb_vmemmap: fix race with speculative PFN walkers") closed the window with RCU on both sides. The RCU was removed together with fake heads, on the wrong belief that it was only needed for them. Map the shared tail page writable instead of restoring the RCU. Every struct page on it is a tail with a zero refcount, so a late cmpxchg fails and leaves the page unchanged. RCU would put rcu_read_lock() back around every get and synchronize_rcu() back into hugetlb allocation and freeing. The cost is the protection against stray writes to tail struct pages. The shared page is shared by all hugetlb folios of the same order in the zone, so such a write now corrupts all of them instead of oopsing. Bootmem folios optimized when the memmap is set up keep their tails read-only. Nothing does a lockless get on them before hugetlb takes them over.
-
CVE-2026-99686
In the Linux kernel, the following vulnerability has been resolved: vhost: fix used ring index wrap in vhost_add_used_n_in_order vhost_add_used_n_in_order() advances start by nheads[i] but wraps it with a single subtraction. If start + nheads[i] >= 2 * vq->num, start can remain outside the ring. With used-ring logging enabled, start > vq->num makes (vq->num - start) underflow. log_write() then sets bits beyond the dirty-log bitmap, potentially corrupting adjacent userspace memory. This is reachable on the vhost-net TX path with VIRTIO_F_IN_ORDER: done_idx can reach VHOST_NET_BATCH (64), triggering the issue with a 32-entry ring. Wrap start with a mask to keep it within the ring. VHOST_SET_VRING_NUM enforces a nonzero power-of-two ring size, allowing start to be wrapped using & (vq->num - 1).
-
CVE-2026-99687
In the Linux kernel, the following vulnerability has been resolved: vhost-scsi: zero misaligned I/O bounce pages The misaligned I/O fallback copies the full bounce scatterlist to the request buffer during completion. If a data-in command transfers less than requested, part of a newly allocated bounce page can remain unwritten. Allocate bounce pages with __GFP_ZERO so all bytes copied during completion have a defined value. This path is already the slow fallback for misaligned I/O, so the added initialization cost is acceptable. Found during an LLM-assisted source audit and validated with a QEMU/KVM vhost-scsi target using misaligned data-in commands.
-
CVE-2026-99675
In the Linux kernel, the following vulnerability has been resolved: nvmet: reject keep-alive timeouts that would overflow nvmet_set_feat_kato() and nvmet_alloc_ctrl() convert the 32-bit millisecond KATO from the host to seconds with DIV_ROUND_UP(kato, 1000). The addition in DIV_ROUND_UP() is done in 32 bits, so any KATO above 4294966296 ms wraps around and rounds to 0 seconds. A Set Features command with KATO 0xffffffff then succeeds but disables the keep-alive timer: Get Features returns 0 and Keep Alive fails with Keep Alive Timeout Invalid. A Connect command with such a KATO creates a controller without a keep-alive timer. The converted value has more consumers that can overflow. Get Features multiplies it by 1000, and the timer paths multiply it by HZ in 32 bits and arm a delay that must not exceed MAX_JIFFY_OFFSET. Reject a KATO whose rounded value does not fit all of these. Return Keep Alive Timeout Invalid from both Set Features and Connect. Validate Set Features before stopping the current timer so a failed update leaves its timeout unchanged. Round in 64 bits. A KATO of 0 keeps its current meaning.
-
CVE-2026-99667
In the Linux kernel, the following vulnerability has been resolved: mm/khugepaged: always recheck the pmd in try_collapse_pte_mapped_thp() try_collapse_pte_mapped_thp() drops the PTE lock after it has cleared the page table, and takes the PMD lock and the PTE lock to remove it. In between, MADV_DONTNEED with CONFIG_PT_RECLAIM can reclaim the page table. The pmd_same() check is supposed to handle this, but it is only done with CONFIG_SPLIT_PTE_PTLOCKS. Fix it by performing the check unconditionally. Fix generated by an LLM and sent by Josef Bacik in private. [ david: adjust description ]
-
CVE-2026-99664
In the Linux kernel, the following vulnerability has been resolved: mm/gup: recheck the PUD under PTL in follow_pud_mask() follow_pud_mask() reads the PUD without a lock. If it is a leaf, it takes the lock and calls follow_huge_pud(), which reads the entry again, but only tests whether it is present. A leaf can be zapped concurrently and a page fault can fill the empty entry with a page table. This currently only applies when device-dax makes use of PUD entries, and the fault after e.g., MADV_DONTNEED would find THPs to be disabled. That requires a sequence like 1) Fault in the PUD 2) Disable THP 3) Have GUP race against concurrent MADV_DONTNEED + refault Which is rather unlikely to happen in practice, which is why we never received a report about such a problem. Before commit 1b1676180246 ("mm/gup: handle huge pud for follow_pud_mask()"), we performed that check in follow_devmap_pud() by testing for pud_devmap(). Let's fix it by just structuring follow_pud_mask() similar to follow_pmd_mask(), minus protnone stuff that does not apply and minus FOLL_SPLIT_PMD handling. That makes it much easier to spot any such inconsistencies. Perform the pud_bad() check early in follow_pmd_mask() and just forward the PUD value. We can drop the pud_present() check in follow_huge_pud(), just like we have in follow_huge_pmd(). Even though we are performing more changes than strictly necessary, the fix should be easy to backport.
-
CVE-2026-99665
In the Linux kernel, the following vulnerability has been resolved: mm/khugepaged: preserve dirty bits when collapsing PTE-mapped THPs try_collapse_pte_mapped_thp() clears PTEs without transferring their dirty bits to the folio. A shmem read fault can map a clean folio writable, so a later store only dirties the PTE. Losing that dirty bit allows reclaim to discard modified data. The assumption that regular files are read-only is also obsolete since commit 1b6444331bebb ("mm/khugepaged: enable clean pagecache folio collapse for writable files"). In any case, transferring PTE dirty bits to the folio when removing mappings (and while the folio is still mapped) is the right thing to do. Use get_and_clear_ptes() to collect the dirty bits and mark the folio dirty before unmapping it. Update the comment. This was reproduced with shmem using MADV_COLLAPSE followed by MADV_PAGEOUT. Fix generated by an LLM and sent by Chris Mason in private. [ david: adjust comment; simplify description; cover file-thp as well ]
-
CVE-2026-99666
In the Linux kernel, the following vulnerability has been resolved: mm/khugepaged: fix PMD table lock leak in try_collapse_pte_mapped_thp() For a private mapping with userfaultfd, try_collapse_pte_mapped_thp() takes the PMD table lock before it maps the page table for the second time. If that (pte_offset_map_rw_nolock) fails, because the page table vanished in the meantime due to concurrent page table reclaim, it leaves through the abort label, where the PMD table lock is unlocked if it differs from PTE table lock. In case the PMD table lock and the PTE table lock match, which is the case without CONFIG_SPLIT_PTE_PTLOCKS after pte_offset_map_rw_nolock() failed, we miss to unlock the PMD table. To fix it, forget the PTE table lock when we unmap+unlock the PTE table. Fix generated by an LLM and sent by Josef Bacik in private. [ david: move the ptl = NULL handling; rewrite description, drop wrong comments ]
-
CVE-2026-99648
In the Linux kernel, the following vulnerability has been resolved: mm/vma: call ->may_split() before a partial merge A driver can use the ->may_split() hook to forbid a VMA boundary at certain addresses. A device-dax device is a character device that gives raw access to a range of memory through mmap(). device-dax refuses any address that is not a multiple of the device's alignment through vm_ops->may_split. vma_merge_existing_range() can move boundaries, but does not call vm_ops->may_split. If a range at the edge of a VMA gets the attributes of the adjacent VMA, the function moves only that range over to the neighbour: one VMA shrinks and the other grows. This is a partial merge, and vma_modify() tries it before any split. Take a device-dax VMA that lies next to another VMA of the same file. The user calls mprotect() on the one page of the first VMA that touches the second VMA, and asks for the protection that the second VMA already has. The result is a partial merge. It puts the boundary inside a 1 GiB mapping, which is a single leaf entry at the PUD level. As part of the merge, the kernel looks for a huge PMD to split at the new boundary. That code does not expect a PUD leaf: vma_merge_existing_range() commit_merge() vma_adjust_trans_huge() split_huge_pmd_address() mm_find_pmd() treats the PUD leaf as a pointer to a PMD table, so the "table" is the device memory that the leaf maps __split_huge_pmd() reads a PMD that the user wrote to the device memory folio_put() on the folio that this PMD names Bad. While the code must at some point be taught to split PUD entries as well, the issue here is that we move VMA boundaries such that they are incompatible with the driver. Fix this by calling ->may_split() on the address that a partial merge would move the boundary to, and by abandoning the merge if the driver refuses. vma_modify() then tries the split, which fails with -EINVAL. Factor out existing split handling into check_split() and reuse it. Fix generated by an LLM and sent by Chris Mason in private. [ david: improve description; cleanup the code ]
-
CVE-2026-99644
In the Linux kernel, the following vulnerability has been resolved: mm/memory-tiers: unregister empty tiers outside memory_tier_lock When offlining a node's last memory empties its tier, clear_node_memory_tier() calls device_unregister() while holding memory_tier_lock. A concurrent sysfs read can hold a kernfs active reference and block in nodelist_show() on that mutex. device_unregister() then waits in kernfs_drain() for the read to finish and release its active reference, while the reader waits for the offlining task to release the mutex. Return the node's former tier from clear_node_memory_tier(), making the helper more similar to set_node_memory_tier(), which also returns a memory_tier pointer. Let the caller unlink an empty tier under memory_tier_lock before rebuilding demotion targets, then unregister its device after dropping the lock.
-
CVE-2026-99632
In the Linux kernel, the following vulnerability has been resolved: drm/amdkfd: use the queue's process info for the WPTR BO eviction fence amdgpu_amdkfd_map_gtt_bo_to_gart() removes the eviction fence of bo->vm_bo->vm->process_info from the WPTR BO of a new queue. bo->vm_bo is the VM which opened the BO last, not necessarily the KFD process' VM, and a render node VM has no process info. On navi31, an unprivileged process which exports its WPTR BO with AMDKFD_IOC_EXPORT_DMABUF and imports it on a second render node fd oopses in AMDKFD_IOC_CREATE_QUEUE: BUG: kernel NULL pointer dereference, address: 0000000000000060 RIP: 0010:amdgpu_amdkfd_map_gtt_bo_to_gart+0x86/0x1a0 [amdgpu] init_user_queue.part.0+0x180/0x250 [amdgpu] pqm_create_queue+0x32d/0x730 [amdgpu] kfd_ioctl_create_queue+0x583/0x9f0 [amdgpu] Pass the process info of the queue instead. With the fix, the queue is created and destroyed normally. (cherry picked from commit 6b4157498a176bd1c6083f6bb5aaf8a257a2027e)
-
CVE-2026-99592
In the Linux kernel, the following vulnerability has been resolved: x86/mm: Fix global ASID handed out to two processes On AMD CPUs with INVLPGB, two live processes can get the same MAX_ASID_AVAILABLE global ASID value. The two then hit each other's TLB entries, so each can read or corrupt the other's memory. The global ASID space is managed lazily. When a global ASID is freed, its bit is set in the global_asid_freed bitmap. Once the free ASID search reaches the end of the global_asid_used bitmap, a global flush is done to really reclaim the lazily freed global ASIDs. However, the current code can assign the same global ASID to two different processes, if one process is assigned MAX_ASID_AVAILABLE - 1, another process MAX_ASID_AVAILABLE, and both stick around (keeping their global ASID numbers in use) until the search wraps around again. At that point a third process could also get assigned MAX_ASID_AVAILABLE. the highest ASID stays in use, so it is not handed out again last_global_asid stays below it, so no reset the clear bits run out find_next_zero_bit() returns MAX_ASID_AVAILABLE global_asid_available > 0, so the failure is not caught MAX_ASID_AVAILABLE is handed out as an ASID A search only looks at the bits below MAX_ASID_AVAILABLE, so it cannot see that this value is in use. The next failed search hands it out again, even if the first process still holds it. The fix used here is to reset the global ASID space after a failed search, and to search again. If that search fails too, no global ASID is handed out. The test for !global_asid_available is removed, because use_global_asid() only calls allocate_global_asid() if global_asid_available > 0. The remaining test is for a condition that should be impossible. [ riel: simplify code, massage changelog and comment ]
-
CVE-2026-99584
In the Linux kernel, the following vulnerability has been resolved: f2fs: fix to avoid folio->mapping dereference in is_{meta, node}_folio() Otherwise, it may cause NULL pointer dereference in race condition: Thread A Thread B - f2fs_setattr - f2fs_truncate - f2fs_truncate_blocks - f2fs_do_truncate_blocks - f2fs_truncate_inode_blocks - truncate_dnode - truncate_node - invalidate_mapping_pages - folio->mapping = NULL - folio_lock - is_node_folio - F2FS_F_SB(folio): dereference on folio->mapping
-
CVE-2026-99585
In the Linux kernel, the following vulnerability has been resolved: KVM: arm64: Use kvm_s2_fault_vma_info in gmem_abort() gmem_abort() maps at s2fd->fault_ipa, which HPFAR_EL2 holds at 4K granularity whatever the page size. The generic page-table code aligns it, but pkvm_pgtable_stage2_map() looks up existing mappings over [addr, addr + size), so on a pKVM host with pages larger than 4K a guest_memfd-backed guest that faults past the first 4K of a page can find its neighbour's mapping, get -EAGAIN and take the same fault forever. The memory fault exit reports s2fd->fault_ipa too, which for a nested guest's fault is the nested IPA rather than the canonical one kvm_gmem_get_pfn() failed on. Take the addresses from kvm_s2_fault_vma_info instead, as user_mem_abort() does, and report the canonical gfn in the memory fault exit. kvm_s2_fault_get_vma_info() itself isn't called: a guest_memfd memslot's userspace_addr doesn't need to be backed by a VMA.
-
CVE-2026-99566
In the Linux kernel, the following vulnerability has been resolved: assoc_array: keep gc's cleanup out of the live tree on allocation failure assoc_array_gc() builds a copy of the tree. Where the copy reaches a shortcut, it allocates the copy of the shortcut with kmalloc_flex(), fills it with memcpy(), links it into the new tree, and only then allocates the copy of the node after it. The memcpy() leaves the new shortcut's next_node pointing at the old tree's node, so if that second allocation fails, the cleanup, assoc_array_destroy_subtree() on the new tree, follows the new shortcut into the old tree, and the BUG_ON() on the old node's back pointer fires. Allocate the node first, and the shortcut once the node exists, freeing the node again if the shortcut cannot be allocated. A copied shortcut is then only linked into the new tree when the copy of its node is there to follow it. Nothing else changes: with no allocation failing, the trees gc builds are the same as before. Found by a userspace test that makes each allocation of each operation fail in turn.
-
CVE-2026-99567
In the Linux kernel, the following vulnerability has been resolved: assoc_array: don't chain a new shortcut onto an existing one When a full node's leaves and the new key all share the node's segment, assoc_array_insert_into_terminal_node() replaces the node with a shortcut to a pair of new nodes. If the node is itself the target of a shortcut, the new shortcut is installed as that shortcut's next node, and a shortcut leading to a shortcut is something nothing expects: assoc_array_walk() indexes the second one as a node, and assoc_array_apply_edit() hits the BUG_ON() in the loop that adjusts nr_leaves_on_branch, on that same insertion. Deletion gets a node there. When the subtree below a shortcut shrinks enough, assoc_array_delete() collapses it into one node below the shortcut, and the leaves it collects can all share that node's segment. The next key that shares it too takes this path. Eighteen insertions and a deletion are enough: sixteen keys that agree on bits 0-11, one that leaves them at bit 8 and is deleted again, and one more that agrees with the sixteen. Have the new shortcut replace the one in front of the node as well. assoc_array_insert_shortcut() gives it that shortcut's back pointer and parent slot, so the caller installs it where that one was, and excises the old shortcut along with the node, which takes a second entry in excised_meta[]. The new shortcut's index key comes from the new key up to the new skip level, so it covers what the old one did. Found by a userspace test doing random insertions and deletions on keys that share long prefixes; a targeted case builds a shortcut, deletes down to the collapse and inserts one more key.
CVE records published in 2026 (ids starting CVE-2026-), sorted by how likely they are to be exploited.