CVEs of 2026.
Every CVE published in 2026 that we hold, ranked by exploit odds.
-
CVE-2026-99663
In the Linux kernel, the following vulnerability has been resolved: mm/gup: fix clearing folios in memfd_pin_folios() memfd_pin_folios() pins folios obtained directly from the page cache without checking whether they are uptodate. However, shmem fallocate() can leave newly allocated folios uncleared and not uptodate. Callers such as udmabuf can then expose uninitialized memory to userspace. Clear non-uptodate folios and mark them uptodate under the folio lock before returning them to the caller. Fix generated by an LLM and sent by Chris Mason in private. [ david: simplify description and comment ]
-
CVE-2026-99653
In the Linux kernel, the following vulnerability has been resolved: mm: fix unkillable fault loop on MAP_DROPPABLE memory handle_mm_fault() suppresses VM_FAULT_OOM for VM_DROPPABLE mappings. For user-mode faults, pending signals are processed before retrying the access. Kernel-mode faults on user buffers instead return directly to the faulting instruction. With persistent allocation failures (esp. memcg limits), a syscall, such as write(), can therefore loop while not dropping locks. Only suppress VM_FAULT_OOM when FAULT_FLAG_USER is set, allowing kernel-mode faults and GUP to fail and signal allocation errors. Fix generated by an LLM and sent by Chris Mason in private. [ david: simplify description ]
-
CVE-2026-99654
In the Linux kernel, the following vulnerability has been resolved: mm: fix unmap_mapping_range() beyond 4 GiB on 32-bit Commit 9eab0421fa94 ("mm: fix unmap_mapping_range high bits shift bug") made holebegin and holelen unsigned before they are shifted, by casting them to pgoff_t. That is an unsigned long: on a 32-bit kernel the cast cuts off the upper half of the two loff_t values, so a hole which begins at 4 GiB or beyond is taken for one in the first 4 GiB of the file. truncate_pagecache() passes the new size of the file and asks for private COWed pages to be unmapped as well. If a file of 6 GiB is truncated to 5 GiB, everything from 1 GiB on is unmapped, and what a process has written to a private mapping of the range between 1 GiB and 5 GiB is lost: it reads the contents of the file again. Where the range has an end, the wrong range is unmapped instead of the right one. The page cache copes with that, as it unmaps every folio that it removes. The callers which only have unmap_mapping_range() to get rid of user mappings do not, for example drm_vma_node_unmap(), whose fake offsets reach beyond 4 GiB on 32-bit, and revoke_iomem() with physical addresses of more than 32 bits. Cast to u64 instead. Nothing changes for 64-bit kernels. Fix generated by an LLM and sent by Josef Bacik in private.
-
CVE-2026-99657
In the Linux kernel, the following vulnerability has been resolved: mm/mmu_notifier: set ->mm only after mmgrab() Radeon uses a non-NULL notifier->mm to decide whether to call mmu_interval_notifier_remove() when freeing a buffer object. __mmu_interval_notifier_insert() sets ->mm before validating the range, so a wrapping userptr range can leave ->mm set after insertion fails. Cleanup then tries to remove an unregistered notifier for which no mm reference was acquired. Set ->mm only after mmgrab(), once insertion can no longer fail. While we could fix radeon's code instead, setting ->mm on success only makes the code more robust. Fix generated by an LLM and sent by Chris Mason in private. [ david: simplify description ]
-
CVE-2026-99658
In the Linux kernel, the following vulnerability has been resolved: mm/mmu_notifier: finish non-blocking interval invalidation on error When an ordinary MMU notifier rejects a non-blocking invalidation, the interval invalidation started beforehand is left unfinished: the caller skips mmu_notifier_invalidate_range_end(). Subsequent interval notifier operations can then wait forever. The only in-tree trigger is nouveau SVM, which uses both notifier types and rejects non-blocking invalidations. Reaping via the OOM reaper or process_mrelease() can therefore leave the GPU fault worker stuck, holding a reference that prevents the dying process's address space and memory from being freed. End the interval invalidation when an ordinary notifier fails a non-blocking invalidation, like mn_itree_invalidate() already does when an interval notifier fails. Make that handling clearer in the documentation of mmu_notifier_invalidate_range_start_nonblock(). Fix generated by an LLM and sent by Josef Bacik in private. [ david: simplify description, update function doc ]
-
CVE-2026-99659
In the Linux kernel, the following vulnerability has been resolved: mm/mmu_gather: fix flushing ranges with 4 GiB or larger __tlb_adjust_range() takes the size of the range which has to be flushed as an unsigned int. Its callers, tlb_flush_{pte,pmd,pud,p4d}_range(), get and pass an unsigned long. Consequently, a range that is 4 GiB or larger gets truncated, possibly to 0. On arm64 with hugetlb, a 64k kernel can have 16 GiB hugetlb folios mapped contiguously at the PMD level and flush them when unmapping, and a 16k kernel can flush 64 GiB PUDs through tlb_unshare_pmd_ptdesc(). So at least these two scenarios are affected. Fix it by flushing the proper range instead of the truncated one by properly using "unsigned long" for range_size. Fix generated by an LLM and sent by Josef Bacik in private. [ david: use two-tab indent, simplify description ]
-
CVE-2026-99647
In the Linux kernel, the following vulnerability has been resolved: mm/huge_memory: split huge PUDs when adjusting VMA boundaries PFN mappings such as VFIO PCI BAR mappings can use PUD leaves while allowing VMA splits at page granularity. However, vma_adjust_trans_huge() only handles PMDs. After faulting a writable PUD-sized VFIO BAR mapping, changing its first page to read-only with mprotect() makes the PMD walk treat the PUD leaf as a PMD table. Applying MADV_DONTFORK to the first PMD-sized chunk leaves the PUD spanning two VMAs: the PMD-aligned boundary skips splitting. Walk the page tables once at each adjusted boundary and split a huge PUD or PMD leaf as needed. Handle PUD leaves even when the boundary is PMD-aligned.
-
CVE-2026-99651
In the Linux kernel, the following vulnerability has been resolved: mm/ksm: reset the unstable trees of all nodes Each full scan resets ksm_nr_node_ids unstable tree roots. Enabling merge_across_nodes resets this count to 1, even if per-node roots were previously allocated. So when toggling "merge_across_nodes" 0 -> 1 -> 0, a scan that starts while the setting is 1 and continues after switching back to 0 can encounter stale unstable tree roots, resulting in UAF. Only root is allowed to change "merge_across_nodes", and the documented sequence is: "merge_across_nodes setting can be changed only when there are no ksm shared pages in the system: set run 2 to unmerge pages first, then to 1 after changing merge_across_nodes, to remerge according to the new setting." Let's fix it by tracking in ksm_nr_node_ids the actual number of allocated tree roots. remove_all_stable_nodes() and ksm_check_stable_tree() will now scan more tree roots (skipping the empty ones), which is harmless.
-
CVE-2026-99642
In the Linux kernel, the following vulnerability has been resolved: mm/gup: check folio->mapping under folio lock in memfd_pin_folios() memfd_pin_folios() pins folios from the page cache without taking the folio lock. khugepaged or MADV_COLLAPSE can therefore replace a folio after checking its refcount while memfd_pin_folios() still pins the old folio. The caller then receives a folio which is no longer part of the memfd. For udmabuf and iommufd, this can cause data corruption: DMA accesses the detached folio rather than the live memfd contents. DMA reads can return stale data, while DMA writes update the detached copy and are lost from the live memfd. The detached folio remains pinned, so these accesses do not constitute a use-after-free or target another application's memory. For memfd live update, the stale contents can reappear in the memfd after handover. Lock the folio and recheck folio->mapping before pinning it, like other page cache users do. If the folio was removed, retry the folio lookup from the index that was removed.
-
CVE-2026-99633
In the Linux kernel, the following vulnerability has been resolved: drm/amdgpu: keep the CS user fence BO in GTT The CS ioctl hands the GPU the GART address of the user fence BO, but since the BO keeps its own preferred and allowed domains, it no longer makes sure the BO is in GTT. A user fence in a VRAM BO, which only needs to be a page in size, gets its address from amdgpu_bo_gpu_offset(), which warns about non-contiguous VRAM. On navi31, a CS with a user fence in a 4K VRAM BO, or in one created with AMDGPU_GEM_CREATE_VM_ALWAYS_VALID: WARNING: amdgpu_object.c:1531 at amdgpu_bo_gpu_offset+0x86/0xb0 [amdgpu], CPU#11: cs_gem_checks/18393 RIP: 0010:amdgpu_bo_gpu_offset+0x86/0xb0 [amdgpu] Call Trace: <TASK> amdgpu_cs_parser_bos.isra.0+0x4eb/0x850 [amdgpu] amdgpu_cs_ioctl+0x97e/0x2510 [amdgpu] drm_ioctl_kernel+0xac/0x110 [drm] drm_ioctl+0x2c6/0x5b0 [drm] __x64_sys_ioctl+0xa0/0xf0 do_syscall_64+0xa3/0x4d0 entry_SYSCALL_64_after_hwframe+0x76/0x7e and BOs which can't be in GTT at all, like GWS or OA ones of a page in size, pass the checks as well. Only take BOs which can't be anywhere else than in GTT as user fence, which the CS then validates there along with its other BOs. While at it, detect userptrs with amdgpu_ttm_tt_is_userptr(); amdgpu_ttm_tt_get_usermm() returns NULL once the task which created the userptr is gone. v2: only take BOs which can only be in GTT instead of special casing the user fence BO in the validation, use amdgpu_ttm_tt_is_userptr() (Christian) (cherry picked from commit f16989def9561d04d0f2e77216f5319112d29bf1)
-
CVE-2026-99634
In the Linux kernel, the following vulnerability has been resolved: mptcp: do not use Fast Open on MP_JOIN subflows tcp_fastopen_create_child() hands the SYN packet itself to subflow_syn_recv_sock(). For an MP_JOIN request this takes the fatal fallback: the cloned child is destroyed and handed back with drop_req flagged, but tcp_fastopen_create_child() does not check the flag and queues it, so accept() can expose the freed child. MP_JOIN cannot use Fast Open: data on a subflow requires the completed HMAC exchange (RFC 8684, sec. 3.2). Refuse it (only the TFO path passes a SYN skb here) and let tcp_conn_request() fall back to the regular MP_JOIN handshake; also strip the Fast Open cookie from MP_JOIN SYN/ACKs. [ Moving code to avoid clearing TS and consolidating checks. ]
-
CVE-2026-99635
In the Linux kernel, the following vulnerability has been resolved: vhost/vsock: size receive SKBs from declared payload vhost_vsock_alloc_skb() sizes the skb from iov_length(), the total guest descriptor length. virtio_transport_inc_rx_pkt() instead adds hdr->len to rx_bytes and buf_used, and budgets each queued skb as SKB_TRUESIZE(0). The guest controls the descriptor length and hdr->len independently. A guest can therefore provide a 64 KiB descriptor while declaring a zero payload. The descriptor-sized skb remains queued, rx_bytes and buf_used remain unchanged, and only the fixed queue-entry budget limits retention. This is not a lifetime leak: the skb is freed when dequeued or at socket teardown. Repeating this across connections can nevertheless exhaust host memory. On Linux v7.2, 455 such descriptors retained 30,255,680 bytes for a 262,144-byte receive buffer; a full-payload control retained 265,984 bytes. Vhost previously read hdr->len and allocated pkt->buf from that value. The conversion from virtio_vsock_pkt to sk_buff instead allocated receive skbs from the descriptor length, introducing the allocation-accounting mismatch. Copy the header into a stack object and validate hdr->len before allocating the skb. Size the allocation as sizeof(hdr) + payload_len so descriptor capacity no longer determines queued memory. Check payload_len against len - sizeof(hdr) to avoid addition overflow on 32-bit hosts.
-
CVE-2026-99624
In the Linux kernel, the following vulnerability has been resolved: drm/amdgpu: Report MES ring wait timeouts. Return -ETIMEDOUT when the fence wait before submission expires instead of reporting success without submitting the packet. (cherry picked from commit 34341594a87b4a3f28be41e3e70a4787f02b3d9e)
-
CVE-2026-99625
In the Linux kernel, the following vulnerability has been resolved: drm/amdgpu: Fix duplicate dependency fence releases drm_sched_job_add_dependency() consumes its fence reference on success and failure. Do not release the fence again if drm_sched_job_add_dependency() fails. (cherry picked from commit e04f3dfb2071451633e6440a3911a9bb65df9144)
-
CVE-2026-99627
In the Linux kernel, the following vulnerability has been resolved: drm/amdgpu/gfx12: disallow GFXOFF while updating the SPM VMID gfx_v12_0_update_spm_vmid() does a read-modify-write of RLC_SPM_MC_CNTL over MMIO without keeping GFX powered, unlike gfx11. When GFX is in GFXOFF at that point the update doesn't stick, and with the SPM VMID programmed on submission again, the job of a process with a reserved VMID can still run with the SPM writing through VMID 0. On navi48, after a few other submissions and a VM fault on the same VM, a COPY_DATA of RLC_SPM_MC_CNTL from such a job reads 0 although the update was called for it: $ amd_uapi_regression --run-subtest cs-secure-ib-compute,gem-mapping-info-high-va,gpuvm-fault-high-va,reserved-vmid-spm-vmid amdgpu 0000:03:00.0: [gfxhub] page fault (src_id:0 ring:40 vmid:7 pasid:185) amdgpu 0000:03:00.0: Faulty UTCL2 client ID: SDMA1 (0xe) RLC_SPM_MC_CNTL in the job: 0x00000000 With GFXOFF disallowed around the update it reads the reserved VMID 15. (cherry picked from commit 907aebc5578ae09fa5c0e2f0784201295837e60b)
-
CVE-2026-99629
In the Linux kernel, the following vulnerability has been resolved: drm/amdgpu: clear RB_OVERFLOW bit when enabling interrupts for navi10_ih Port this change to navi10_ih.c: commit afbf7955ff01 ("drm/amdgpu: clear RB_OVERFLOW bit when enabling interrupts") Setting IH_RB_WPTR to 0 does not clear RB_OVERFLOW if RB_ENABLE is not set. Pulse WPTR_OVERFLOW_CLEAR after RB_ENABLE (and with WPTR_OVERFLOW_ENABLE) so overflow is cleared when interrupts are enabled. (cherry picked from commit 713fb4ab04f4d2af6120f8eb607c5673b2759b38)
-
CVE-2026-99622
In the Linux kernel, the following vulnerability has been resolved: drm/amdgpu: drop the dma-buf reference when a KFD import fails import_obj_create() takes a reference on the dma-buf for the new kgd_mem, but the error path after validating the BO doesn't drop it. Since the validation reserves the BO interruptibly, an import failing with -ERESTARTSYS leaks the exported BO with its memory, and keeps the module in use. Importing a 64 MiB VRAM BO in a loop while getting SIGUSR1 every 20 us and another thread changes the BO's mappings: INFO: 171821 imports, VRAM used before 65 MiB, after 129 MiB modprobe: FATAL: Module amdgpu is in use. Drop the reference, and free the sync object and the mutex of the kgd_mem as well. With that VRAM usage returns to 65 MiB after the same test. (cherry picked from commit b4d1eca2729840148ff2ea957724ce764ca5a94e)
-
CVE-2026-99609
In the Linux kernel, the following vulnerability has been resolved: ALSA: usx2y: Require a buffer of two URBs usx2y_urb_capt_retire() wraps the ring buffer once per packet and usx2y_urb_play_prepare() once per URB of nr_of_packs() packets of up to 50 frames. Both assume that the data fits in the buffer, and the hw constraints only require about two 1 ms periods, so a packet larger than the buffer on capture, or a URB larger than the buffer on playback, violates that assumption. Constrain SNDRV_PCM_HW_PARAM_BUFFER_SIZE in snd_usx2y_pcm_open() to at least two URBs, 2 * nr_of_packs() * 50 frames, for both directions. The bound is in frames and derived from the driver's own packet limit, so it does not read the endpoint, whose altsetting is only selected in prepare. Two URBs rather than one, because the position advances by a URB on playback and only moves when a URB is less than the whole buffer. Keep a guard in usx2y_urb_capt_retire() that drops a packet larger than the buffer, for the case the constraint makes unreachable. With nrpacks above about 163 (S16_LE) or 109 (S24_3LE) the bound exceeds the 64 KiB buffer the substream preallocates (twice that with the preallocation raised to its 128 KiB maximum), so hw_params fails for those module settings. That is intended: the driver cannot be safe with a URB larger than half the largest buffer. Reproduced on 7.3-rc7 with a USB/IP device presenting the US-122 interface layout and 1024-byte packets into a 512-byte capture buffer: BUG: KASAN: out-of-bounds in i_usx2y_urb_complete+0x1285/0x1ac0 Read of size 18446744073709551104 at addr ffff88800e448800 by task vhci_rx/83 Tested on 7.3-rc7 with the same device, S16_LE at 44100 Hz, nrpacks 4, so the bound is 400 frames. A 128-frame capture buffer and a 200-frame playback buffer are refused at hw_params with EINVAL. A 400-frame buffer streams for 30 s in each direction with the position advancing inside the buffer and no KASAN report; the capture guard did not fire. The same 400-frame capture buffer with S24_3LE, which selects altsetting 2 in prepare, streams for 30 s with the same result.
-
CVE-2026-99614
In the Linux kernel, the following vulnerability has been resolved: drm/nouveau/nvkm: clear use.enabled when subdev init fails On Kepler boards without the video decode falcon firmware installed (msvld/mspdec/msppp), every other VA-API decode attempt binds to an engine with no ucode loaded. The submitted work never completes, which leads to a context switch timeout, a channel recovery that cannot reset the engine, and eviction waits on fences that never signal. The process ends up stuck in D state, or killed by SIGBUS when it touches a buffer evicted after the wait times out: msvld: unable to load firmware data msvld: init failed, -19 fifo: SCHED_ERROR 0a [CTXSW_TIMEOUT] vo[5444]: channel 14 killed! WARNING: drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.c:117 at nvkm_runl_rc+0x2a4/0x3b0 [nouveau] [TTM] Buffer eviction failed WARNING: drivers/gpu/drm/nouveau/nouveau_bo.c:1092 at nouveau_bo_move_ntfy.part.0+0x9b/0x100 [nouveau] Tracing nvkm_subdev_ref() and nvkm_subdev_init_() shows the pattern: a reference taken with enabled=0 runs nvkm_falcon_init(), which returns -ENODEV; the next reference sees enabled=1 and succeeds without running it. Clear use.enabled when init() fails, so that every later reference retries init and fails cleanly. The other readers of use.enabled are fini() handlers that use it to skip teardown, which they now also do after a failed init(). On a GK107 (Quadro K1100M) without video firmware, playing H.264 with mpv --hwdec=auto: without this patch the GPU channel is killed within a minute and mpv hangs; with it, VA-API decoder creation fails, mpv falls back to software decoding, and no faults are logged.
-
CVE-2026-99615
In the Linux kernel, the following vulnerability has been resolved: l2tp: only return sessions of the given tunnel in l2tp_session_get() Commit 5f77c18ea556 ("l2tp: use IDR for all session lookups") made pppol2tp_connect() and the netlink session commands find sessions with l2tp_session_get(), passing the socket of the tunnel they name. L2TPv3 session IDs are unique per network namespace, except that sessions carried over UDP are only unique per UDP socket, the socket of their tunnel. However, l2tp_v3_session_get() only checks the socket when another session has the same session ID. Otherwise it returns the session, whichever tunnel it belongs to. As of now, a session command that names the wrong tunnel acts on a session of another tunnel, and a PPPoL2TP socket for a second L2TPv3 tunnel fails with EEXIST, because it finds session 0 of the first tunnel. The issue can be reproduced with: ip l2tp add tunnel tunnel_id 3 peer_tunnel_id 3 encap udp \ local 127.0.0.1 remote 127.0.0.1 udp_sport 5003 udp_dport 6003 ip l2tp add tunnel tunnel_id 4 peer_tunnel_id 4 encap udp \ local 127.0.0.1 remote 127.0.0.1 udp_sport 5004 udp_dport 6004 ip l2tp add session tunnel_id 3 session_id 7 peer_session_id 7 ip l2tp del session tunnel_id 4 session_id 7 ip l2tp show session Before that commit, the delete fails with "No such device", and session 7 of tunnel 3 is still listed. After that commit, the delete succeeds, and session 7 of tunnel 3 is gone. This patch will make l2tp_session_get() return NULL when the L2TPv3 session it finds does not belong to the given socket. The receive path calls l2tp_v3_session_get() directly and is not changed.
-
CVE-2026-99602
In the Linux kernel, the following vulnerability has been resolved: udf: don't let the extent type hide an exhausted free-space table extent udf_table_new_block() takes the first block of the free-space table extent closest to the goal. It keeps that extent's type and length together in goal_elen, subtracts one block, and deletes the extent only if goal_elen then reaches zero. UDF 2.60 section 2.3.7.1 requires the free-space extents of an Unallocated Space Entry to be of type 1 (allocated but not recorded), and mkudffs writes them that way. For such an extent the type bits keep goal_elen non-zero, so taking its last block leaves a type-1 extent of length zero behind, starting at the block after the extent, which is in use. The next allocation that picks this empty extent returns that in-use block. Subtracting a block from 0x40000000 then borrows from the type bits, and the extent becomes type 0 with a length of 2^30 - blocksize: about a gigabyte of "free" space overlapping live metadata, the other table extents and whatever lies behind the partition. From then on blocks are handed out twice, or past the end of the partition. Extents that udf_table_free_blocks() adds are type 0, where the check works, so only tables written by mkudffs or by another implementation are affected. Nothing more than filling such a filesystem is needed: on a fresh 1 MiB "mkudffs --space=unalloctable" image, creating empty files until the partition is full writes file entries over the reserve volume descriptor sequence and the backup anchor, and df then reports a negative amount of used space. On syzbot's images the same double allocation is what the two reports below trip over. In the first, ftruncate() extends a new file with enough hole extents to need a chain of allocation extent descriptors; one AED is placed on another inode's file entry, and a later one is placed on the block of the AED currently being filled, which udf_setup_indirect_aext() zeroes, so __udf_add_aext() finds lengthAllocDescs out of step with its cursor. In the second, an AED is placed past the end of the device, sb_getblk() fails, and the error path of udf_do_extend_file() calls udf_truncate_extents() on an extent list that no longer covers i_size. Keep the type separately from the length, as udf_table_prealloc_blocks() already does.
-
CVE-2026-99603
In the Linux kernel, the following vulnerability has been resolved: mptcp: pm: kernel: use RCU to free endpoint listeners mptcp_pm_nl_create_listen_socket() starts the TCP listener directly instead of going through mptcp_listen(). It consequently misses setting SOCK_RCU_FREE on the parent MPTCP socket. tcp_child_process() can invoke the listener's subflow_data_ready() callback without holding the MPTCP parent lock or a reference to the parent. An authenticated MP_JOIN racing with endpoint teardown can therefore access the parent after its last reference has been dropped. The ADD_ADDR error path can also close an already published listener. Set SOCK_RCU_FREE before __inet_listen_sk() publishes the TCP listener, matching mptcp_listen(). This protects the parent on every subsequent close path, including a failed endpoint addition.
-
CVE-2026-99604
In the Linux kernel, the following vulnerability has been resolved: drm/etnaviv: Flush the tile status cache through both the PE and BLT Since commit f232d9ec029c ("drm/etnaviv: fix TS cache flushing on GPUs with BLT engine"), jobs on GPUs with a BLT engine use a BLT command to flush the tile status cache instead of TS_FLUSH_CACHE, which is broken on most of these GPUs. GC7000 r6008/r6009 and GC8200 r6401 are the other way around: TS_FLUSH_CACHE works, but the BLT command does not flush the cache. The vendor driver handles these GPUs using TS_FLUSH_CACHE, but the etnaviv kernel driver has no way to identify them. As a result, dirty tile status cache lines survive the end of a job. They are written back by the next cache flush, using the page table of the job issuing it. If that job belongs to another process, the write goes to the wrong address space. On GC7000 r6009 this causes MMU faults and GPU recoveries blamed on an innocent process. I see around 1100 per run of a multisample depth/stencil test running alongside two other GPU clients. Emit both flushes at the end of every job on GPUs with a BLT engine. Whichever does not work on a given GPU is a no-op. Since dirty lines are only written back once, this costs one additional state load per job.
-
CVE-2026-99606
In the Linux kernel, the following vulnerability has been resolved: net/smc: don't spin in smc_rx_recvmsg() while a pipe holds spliced data While bytes spliced off the RMB are still in a pipe, recvmsg() waits for the pipe to release them before it copies the data behind them. The path to that wait skips the timeout and signal checks, and smc_rx_wait() returns at once on RCV_SHUTDOWN. With a signal pending, after SHUT_RD, or once the peer closes or shuts down writing, recvmsg() spins on a CPU until the pipe is drained, and SIGKILL cannot stop it. MSG_DONTWAIT and an expired SO_RCVTIMEO wait there too instead of returning -EAGAIN. Check the timeout and signals before that wait, and end it when the pipe releases the data or the connection is killed, not on RCV_SHUTDOWN. A zero-length read returns 0 without waiting for the pipe; a non-blocking one returns -EAGAIN while poll() still reports the socket readable (splice() can still take those bytes). The wait sets SOCKWQ_ASYNC_WAITDATA like smc_rx_wait(), so a byte arriving behind the pipe raises no SIGIO on a reader already asleep for it, and a full barrier orders its wait queue entry before its read of splice_pending, so no wake is lost.
-
CVE-2026-99607
In the Linux kernel, the following vulnerability has been resolved: ext2: preserve IOMAP_F_DIRTY for extending direct writes ext2_iomap_begin() sets IOMAP_F_DIRTY for writes that may extend i_size, but clears the flag after ext2_get_blocks(). When the write uses an existing mapped block, iomap can then use FUA and skip generic_write_sync(), even though ext2_dio_write_end_io() updates the inode size. An O_DSYNC write can therefore return without persisting the new size. For example, on a FUA-capable device, a 512-byte O_DIRECT|O_DSYNC write at offset 512 into a fsynced one-byte file returns successfully and extends the in-memory size to 1024, while an immediate read of the on-disk inode still shows size 1. An explicit fsync persists the new size. Move setting of IOMAP_F_DIRTY after the initialization.
-
CVE-2026-99608
In the Linux kernel, the following vulnerability has been resolved: ALSA: usx2y: Guard the playback copy usx2y_urb_play_prepare() sums the packet lengths of a whole URB, copies that many frames with a single wrap, and sizes each playback packet from the capture packet's actual_length. Three things are unchecked: that the URB fits in the ring buffer, that a packet fits in the playback endpoint's maxpacksize, from which tmpbuf is allocated, and that actual_length is a whole number of frames. The ISO descriptors take the raw actual_length while transfer_buffer_length and the copies take count * stride, so a packet that is not a multiple of the stride makes the descriptors add up to more than the buffer, by up to stride - 1 bytes per packet. Return -EPIPE, as the existing packet-size check does, when the URB is larger than the buffer or a packet is larger than maxpacksize, and size each descriptor from whole frames, so the descriptors add up to transfer_buffer_length and each stays within maxpacksize. The buffer case is prevented by the previous patch and this is the guard for it; the other two depend only on the device. The first case is reachable on the unpatched driver with ordinary packets: on 7.3-rc7, a USB/IP device presenting the US-122 interface layout and sending 48-frame packets, with a 100-frame playback buffer (two 50-frame periods, inside the stock constraints), KASAN reports: BUG: KASAN: out-of-bounds in i_usx2y_urb_complete+0x79b/0x1ac0 Read of size 18446744073709551280 at addr ffff88800e6b02e0 by task vhci_rx/80 With both patches, the same device and a 200-frame playback buffer is refused at hw_params with EINVAL, and a 400-frame buffer plays for 30 s with the position advancing inside the buffer, no KASAN report and neither guard message. The maxpacksize check is not exercised: the emulated device reports the same wMaxPacketSize on both endpoints. The descriptor case is visible with capture packets of 195 bytes at stride 4 (48 frames and 3 bytes): without this change the four descriptors add up to 780 bytes for 768 bytes of data, and USB/IP's completion check refuses the URB ("total length of iso packets 780 not equal to actual length of buffer 768", -EPROTO), after which the driver stops the streams. With this patch every descriptor is 192 bytes, the descriptors add up to the data in every URB for 30 s, the position advances inside the buffer and KASAN reports nothing. The 48-frame playback run above is unchanged.
-
CVE-2026-99596
In the Linux kernel, the following vulnerability has been resolved: nvme: wait for SRCU readers before splicing the namespace list Readers walk ctrl->namespaces under ctrl->srcu, and some of them sleep there, e.g. nvme_wait_freeze() waiting for I/O to drain. nvme_remove_namespaces() moves the list onto its on-stack ns_list with list_splice_init_rcu(), which only waits for RCU readers before relinking the entries. A reader still on the last namespace then follows its next pointer to ns_list instead of ctrl->namespaces and treats the stack as a namespace. Deleting a controller while a Format NVM passthrough waits for a delayed read on the last namespace gives: BUG: KASAN: stack-out-of-bounds in nvme_wait_freeze+0x102/0x120 Read of size 8 at addr ff110001d2f3fc38 by task nvme/4413 Call Trace: nvme_wait_freeze nvme_passthru_start nvme_submit_user_cmd nvme_user_cmd nvme_dev_ioctl BUG: unable to handle page fault for address: ffffffff8af04850 Use list_splice_init_srcu() instead, which waits for the SRCU readers. This also takes the RCU grace period out of every controller deletion. Deleting 64 tcp controllers with one namespace each, median of 3 runs on each of two boots per kernel: before after nvme disconnect-all (ms) 3491 / 3509 2996 / 3165 nvme_remove_namespaces() avg (ms) 39.0 / 40.5 31.9 / 34.7
-
CVE-2026-99597
In the Linux kernel, the following vulnerability has been resolved: srcu: Add list_splice_init_srcu() list_splice_init_rcu() makes a list look empty to new readers, waits for the current ones with the given sync function, and only then links the entries into the new list. The sync function takes no argument, so it can wait for RCU readers but not for the readers of an srcu_struct. Passing synchronize_rcu() for a list read under SRCU still builds, but leaves SRCU readers on the entries while the last one's next pointer is moved onto the new list head. nvme_remove_namespaces() does this today. Add list_splice_init_srcu(), which waits for the readers of the given srcu_struct instead.
-
CVE-2026-99598
In the Linux kernel, the following vulnerability has been resolved: MIPS: sgi-ip32: Leave MACE PCI control alone in config accesses mace_pci_read_config() clears the master abort interrupt enable in the MACE PCI control register around each config read. A register read followed by two register writes can hang CRIME 1.1 under DMA load. Ack master aborts in the error register instead, and follow every write with a read so back-to-back accesses cannot form that sequence.
-
CVE-2026-99599
In the Linux kernel, the following vulnerability has been resolved: udf: don't use eloc after the end of the unallocated space table When udf_next_aext() returns 0, its eloc, elen and etype outputs hold no defined value. Two walks of the unallocated space table still use eloc after the walk has reached the end of the table. udf_table_prealloc_blocks() looks for a free extent starting at first_block and decides whether it found one by comparing eloc with first_block after the walk. If the table's list ends in a continuation descriptor that leads to an empty allocation extent, eloc is then the block of that allocation extent, which is in use, and if that is first_block, the block is handed out as preallocation. Once udf_discard_prealloc() no longer frees the wrong block, this is what still makes fsx in generic/091 and generic/263 read back bad data on a table, in about one run in four. Only take the extent if the walk stopped at it. When udf_table_free_blocks() cannot merge the freed blocks into an existing extent, it appends a new one and sets only its block number, so the partition reference comes from whatever eloc held after the walk, and with long allocation descriptors it is written to disk. For a table without extents, eloc was never set. Take the partition reference from the freed blocks instead. Short allocation descriptors do not store the partition reference, which is why this half remains a latent long-descriptor fix rather than a reproduced failure.
-
CVE-2026-99600
In the Linux kernel, the following vulnerability has been resolved: udf: fix preallocation discarding before an empty allocation extent udf_discard_prealloc() finds the last extent of a file by peeking at the next one with udf_next_aext() and stopping when that returns 0. The peek reads into eloc and elen, and the code then uses them as the last extent. But when udf_next_aext() returns 0 they hold no defined value: if the extent list ends in a continuation descriptor that leads to an empty allocation extent, they describe that continuation, i.e. the block of the empty allocation extent, one block long. The kernel creates such lists itself: when udf_delete_aext() removes the only descriptor of the last allocation extent in a list, it leaves that allocation extent empty and the continuation to it in place. If the last extent is a preallocation, udf_discard_prealloc() deletes it with udf_delete_aext(), which finds the following allocation extent empty, so it also removes the continuation and frees that allocation extent's block. udf_discard_prealloc() then frees eloc and elen: that block a second time, instead of the preallocated blocks. On a space bitmap the preallocated blocks leak and the free block count drifts. On an unallocated space table the second free adds a second free extent for the same block, which is later handed out twice: fsx as run by generic/091 and generic/263 ends up with two parts of its test file in one block and reads back bad data. Peek into separate variables, so that eloc and elen keep the last extent that udf_next_aext() returned with 1.
-
CVE-2026-99587
In the Linux kernel, the following vulnerability has been resolved: ata: libata-scsi: reject data-in commands with a data-out buffer SG_IO can submit a data-in SCSI command with DMA_TO_DEVICE. The scatterlist can then contain pages pinned without write access. ata_scsi_rbuf_fill() writes simulated replies to that scatterlist; in the reported case, the copy tried to write the shared zero page and faulted in the kernel. The same mismatch can affect SG_IO commands translated by libata. ZBC IN and GET PHYSICAL ELEMENT STATUS completion handlers can modify the receive buffer even if the drive aborts the command. For ATAPI, the successful INQUIRY completion also rewrites the buffer in atapi_fixup_inquiry(). Reject these mismatched commands before the buffer is written or the command is issued. Report ABORTED COMMAND / DATA IN BUFFER ERROR because the CDB itself can be valid; the supplied transfer direction is wrong.
-
CVE-2026-99588
In the Linux kernel, the following vulnerability has been resolved: batman-adv: tt: only update flags for TT events In the first implementation of the translation table events, it was only possible to announce: * ADD: new locally detected client * DEL: disappeared local client If an ADD event was queued and then a DEL event was queued in the same timeslot, it is valid to destroy both events because nothing actually changed for remode nodes But with commit 3c4f7ab60cc2 ("batman-adv: improve the TT component to support runtime flag changes") and subsequent changes, it could also happen that there are extra ADD events: * ADD: _NEW_ locally detected client * ADD: already existing client with new flags * DEL: disappeared local client The first and the last type of event could in theory still be removed together from the (to be announced) events queue. But if the "update flags" ADD is dropped together with the DEL, it would still leave global translation table of a remote node in the incorrect state - believing that the client didn't disappear. If a detected client was previously announced, a DEL event must only overwrite the "update flags" ADD but the DEL itself must still continue to exist in the (to be anounced) events queue. For now, adjustment of the announced flags is used for everything instead of dropping already queued up events completely. Similarly, a DEL followed by an ADD caused both entries to be deleted. But the new ADD could announce flags changes which were not yet present for the client before it was marked for DELetion. The new ADD must therefore only overwrite the DEL - but still be kept.
-
CVE-2026-99589
In the Linux kernel, the following vulnerability has been resolved: batman-adv: tt: clear ROAM flag when re-adding a pending local client When a ROAMING_ADV is received for an announced local client, batadv_tt_local_mark_removed() sets BATADV_TT_CLIENT_ROAM, queues a DEL|ROAM change and marks the entry as BATADV_TT_CLIENT_PENDING. If the client sends a frame again before the next TTVN commit, batadv_tt_local_add_existing() only clears the PENDING flag and returns early. The ADD cancels the queued DEL, but the entry keeps the ROAM flag: * batadv_is_my_client() reports the client as not served by this node * batadv_tt_local_client_is_roaming() reroutes unicast frames for it to the node which sent the ROAMING_ADV The ROAM flag is only removed with the next frame of the client. Since BATADV_TT_CLIENT_ROAM is set together with BATADV_TT_CLIENT_PENDING, it has to be cleared also at the same time with BATADV_TT_CLIENT_PENDING when the client roamed back.
-
CVE-2026-99591
In the Linux kernel, the following vulnerability has been resolved: iommu/amd: Force device table to maximum size IOMMU can perform out-of-bounds read of the device table entry (DTE) that is adjacent to the last DTE and falls outside the device table. For example, if the device table can hold 0 to N-1 entries, then the IOMMU can be made to fetch the Nth entry, which falls outside the device table. The device table is owned by the hypervisor (HV) and is assigned in 4KB pages that are contiguous in system physical address (SPA). If the out-of-bound DTE falls on a guest-owned page, then the guest can exploit this bug to setup malicious table walks to corrupt HV pages and/or bring down the system. Note that this bug does not manifest when the device table size is set to its maximum size of 2MB. Therefore, force the device table size to the maximum 2MB on affected processors. These are CPU family 0x17 (Zen1/Zen2) through family 0x1a (Zen5/BRH). This is a mitigation for CVE-2024-21963. kdump reuses the previous kernel's device table only when the size encoded in the IOMMU device-table register matches pci_seg->dev_table_size. A mixed rollout disagrees: the unpatched kernel uses the IVRS-derived size and the patched kernel uses 2MB, so __reuse_device_table() rejects the table. On SNP-capable IOMMUs that rejection panics the kdump kernel. Kernels that both have this patch agree on 2MB and reuse succeeds. Update the production kernel and the crash kernel together.
-
CVE-2026-99580
In the Linux kernel, the following vulnerability has been resolved: mm/userfaultfd: only allocate anon_vma in CoW mappings find_vma_and_prepare_anon() can end up allocating an anon_vma in VMAs that are not CoW mappings, which is wrong. If we have a MAP_SHARED R/O file mapped, we lack VM_SHARED (but do have VM_MAYSHARE) and lack VM_MAYWRITE. While uffd later rejects such mappings (uffd registration requires VM_MAYWRITE, see comment in validate_dst_vma) find_vma_and_prepare_anon() still wrongly allocates an anon vma before it validates the vma. This in turn later confuses copy_page_range()->vma_needs_copy() such that we end up calling copy_hugetlb_page_range() with ranges that actually do not contain *any* per-pagetable information that would require copying, which is the basis for PMD-table sharing in the first place. copy_hugetlb_page_range() should not have to worry about shared PMD tables at all: it doesn't make any sense to copy page tables that are shared, as they can be fully recreated later. Ending up in the function with shareable page tables is the real issue. If a destination PMD table is shared in copy_hugetlb_page_range(), it might get concurrently unshared as we are not holding relevant locks, resulting in us using a page table that is no longer mapped and possibly freed after calling: dst_pte = huge_pte_alloc(dst, dst_vma, addr, sz); To fix it, do not allocate an anon_vma in non-CoW mappings and add a safety guard to copy_hugetlb_page_range(), making sure that we never end up messing with shared PMD tables in there. Before commit 867a43a34ff8a ("userfaultfd: use per-vma locks in userfaultfd operations") we would always validate the VMA before allocating an anon_vma, and consequently never wrongly allocate an anon_vma. As uffd_lock_vma() is part of the same lookup machinery and also contains a misleading VM_SHARED check, change that one as well directly for consistency. A follow-up cleanup will stop taking the hugetlb_vma_lock_read(src_vma) lock in copy_hugetlb_page_range() entirely. Further, the usage of VM_SHARED in userfaultfd code should be cleaned up later to focus on CoW vs. non-CoW mappings.
-
CVE-2026-99576
In the Linux kernel, the following vulnerability has been resolved: f2fs: zone: fix to avoid bio split on sequential zone In the block layer, bio_split_to_limits() splits a bio if it exceeds any underlying queue limits: 1. queue_max_sectors: Splits bios larger than the transfer limit. This limit is set by the device driver/DMA controller (max_hw_sectors) and device-mapper targets. When flushing dirty pages (governed by VFS writeback_chunk_size() and f2fs nr_pages_to_skip()), f2fs merges contiguous pages into bios up to 512KB (1024 sectors) or more without checking queue_max_sectors, which will be split if the device or DM target enforces a smaller limit. 2. queue_max_segments: Splits bios exceeding scatter-gather segment limits. f2fs allocates bios with up to 256 pages, exceeding the 128-segment limit of DM targets (dm-default-key/dm-crypt) if pages are discontiguous. 3. chunk_sectors: Splits bios crossing chunk/zone boundaries. In f2fs, section_size strictly matches zone_size, and block allocation never spans across section boundaries, so bios never cross chunk_sectors. Crucially, when bio_split_to_limits() splits a bio, bio_submit_split() submits the tail half via submit_bio_noacct() before returning the head half to the caller. On a sequential zoned device, this causes the tail bio to arrive at the zone write plug before the head: dm-3: zone 7: prepare_bio wp mismatch: bio sector 30720 offset 2048 != wp_offset 248, len 248 The write plug rejects the out-of-order tail with BLK_STS_IOERR, causing f2fs_stop_checkpoint() and remounting the filesystem read-only: 2026-09-30T22:09:23.725678+08:00 localhost kernel: CPU: 9 UID: 0 PID: 243 Comm: kworker/9:1H Not tainted 6.12.81-gb3ef518fb26d #5 2026-09-30T22:09:23.725750+08:00 localhost kernel: Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.17.0-debian-1.17.0-1 04/01/2014 2026-09-30T22:09:23.725753+08:00 localhost kernel: Workqueue: dm-3_zwplugs blk_zone_wplug_bio_work 2026-09-30T22:09:23.725756+08:00 localhost kernel: Call Trace: 2026-09-30T22:09:23.725758+08:00 localhost kernel: <TASK> 2026-09-30T22:09:23.725758+08:00 localhost kernel: dump_stack_lvl+0x4d/0x70 2026-09-30T22:09:23.725759+08:00 localhost kernel: f2fs_stop_checkpoint+0x1df/0x5d0 2026-09-30T22:09:23.725759+08:00 localhost kernel: ? blk_zone_wplug_bio_work+0x4eb/0x740 2026-09-30T22:09:23.725760+08:00 localhost kernel: f2fs_cache_write_end_io+0x545/0x6e0 2026-09-30T22:09:23.725761+08:00 localhost kernel: blk_zone_wplug_bio_work+0x4eb/0x740 2026-09-30T22:09:23.725762+08:00 localhost kernel: process_one_work+0x5c6/0xde0 2026-09-30T22:09:23.725764+08:00 localhost kernel: ? assign_work+0x124/0x4e0 2026-09-30T22:09:23.725765+08:00 localhost kernel: worker_thread+0x403/0xb70 2026-09-30T22:09:23.725765+08:00 localhost kernel: ? __kthread_parkme+0x82/0x140 2026-09-30T22:09:23.725766+08:00 localhost kernel: ? __pfx_worker_thread+0x10/0x10 2026-09-30T22:09:23.725766+08:00 localhost kernel: kthread+0x235/0x2e0 2026-09-30T22:09:23.725767+08:00 localhost kernel: ? recalc_sigpending+0x128/0x1d0 2026-09-30T22:09:23.725767+08:00 localhost kernel: ? __pfx_kthread+0x10/0x10 2026-09-30T22:09:23.725769+08:00 localhost kernel: ret_from_fork+0x2f/0x70 2026-09-30T22:09:23.725769+08:00 localhost kernel: ? __pfx_kthread+0x10/0x10 2026-09-30T22:09:23.725770+08:00 localhost kernel: ret_from_fork_asm+0x1a/0x30 2026-09-30T22:09:23.725771+08:00 localhost kernel: </TASK> 2026-09-30T22:09:23.725771+08:00 localhost kernel: F2FS-fs (dm-0): Stopped filesystem due to reason: 3 To prevent write bios destined for sequential zones from being split: 1. In __bio_alloc(), bound npages to bdev_max_segments(bdev) for writes to sequential zone areas. 2. In page_is_mergeable(), check queue_max_segments() and queue_max_sectors() for sequential zone areas, so f2fs will not merge beyond the queue limits.
-
CVE-2026-99568
In the Linux kernel, the following vulnerability has been resolved: iommufd/iova_bitmap: Ensure that iova_bitmap_set() handles the full IOVA Generally userspace is going to step the dirty bitmap through the IOVA space. However if the bitmap ends in the middle of a large PTE this is problematic. The first step will clear the dirty bit and report dirty to the end of its bitmap while the second step will see the clear'd dirty and report clean for the start of the bitmap. If it happens this ends up losing dirties. Further this function is missing bounds checking so if this does happen ther is a good chance it will write memory beyond what the userspace was expecting. Return an error code if this illegal condition happens.
-
CVE-2026-99569
In the Linux kernel, the following vulnerability has been resolved: f2fs: release a removed xattr node in roll-forward recovery When xattrs shrink enough to fit in the inline xattr area, write_all_xattrs() truncates the xattr node and clears i_xattr_nid. If the file is then fsynced without a checkpoint, only the inode block is written, with i_xattr_nid of 0. Roll-forward recovery starts from the checkpointed inode, whose i_xattr_nid still points to the old xattr node. f2fs_recover_inline_xattr() copies only the inline xattr area, and f2fs_update_inode() writes the old i_xattr_nid back. So the inode keeps the stale xattr node: - The node block is never released, and fsck reports "nonzero bytes in xattr space after end of list". - If the inline xattr list ends exactly at the end of the inline area, its terminator is read from the stale node, and the removed xattrs come back after recovery. When the fsynced inode has no xattr node but the checkpointed one has, release the old xattr node as truncate_node() does: check its block address, invalidate the block and drop its node cache, which may still be dirty.
-
CVE-2026-99570
In the Linux kernel, the following vulnerability has been resolved: f2fs: fix to invalidate curseg if change_curseg() fails Our consistency stress test (fsstress + fio + atomic write workers + "f2fs_io shutdown 1") triggers below kernel BUG: F2FS-fs (vdd): Stopped filesystem due to reason: 0 ------------[ cut here ]------------ kernel BUG at fs/f2fs/segment.c:3989! Oops: invalid opcode: 0000 [#1] SMP KASAN PTI Workqueue: writeback wb_workfn (flush-253:48) RIP: 0010:f2fs_allocate_data_block+0x257a/0x38e0 Call Trace: <TASK> do_write_block+0x155/0xc10 f2fs_outplace_write_data+0x192/0x340 f2fs_do_write_data_page+0xc57/0x1880 f2fs_write_single_data_page+0xda9/0x15a0 f2fs_write_cache_pages+0x973/0x1f20 __f2fs_write_data_pages+0x6c4/0x1890 do_writepages+0x21e/0x560 __writeback_single_inode+0x119/0x11b0 writeback_sb_inodes+0x67d/0x17a0 wb_writeback+0x207/0x820 wb_workfn+0x128/0xc30 process_one_work+0x925/0x1600 worker_thread+0x60a/0xff0 kthread+0x35f/0x460 ret_from_fork+0x474/0x7e0 ret_from_fork_asm+0x1a/0x30 </TASK> The BUG is f2fs_bug_on(sbi, curseg->next_blkoff >= BLKS_PER_SEG(sbi)) in f2fs_allocate_data_block(), next_blkoff was 512. The root cause is a race in between atomic write commit and shutdown as below: Thread A (commit atomic write) Thread B (shutdown) Thread C (writeback) - f2fs_ioc_commit_atomic_write - f2fs_commit_atomic_write - __f2fs_commit_atomic_write - __replace_atomic_write_block - f2fs_replace_block - __f2fs_replace_block - f2fs_ioc_shutdown - f2fs_do_shutdown : F2FS_GOING_DOWN_METASYNC - f2fs_sync_fs - f2fs_write_single_data_page : pass f2fs_cp_error() check - f2fs_stop_checkpoint : set CP_ERROR_FLAG - change_curseg - reset_curseg : curseg->segno = new_segno : curseg->alloc_type = SSR : curseg->next_blkoff = __next_free_blkoff() which is BLKS_PER_SEG if there is no free slot in new_segno - f2fs_get_sum_cache : return -EIO due to f2fs_cp_error() : return error w/o restoring curseg : goto out_unlock, skip restoring old segno, blkoff and alloc_type - f2fs_do_write_data_page - f2fs_outplace_write_data - do_write_block - f2fs_allocate_data_block : curseg->segno is valid : f2fs_bug_on(next_blkoff >= BLKS_PER_SEG) Since commit 86f33603f8c5 ("f2fs: handle errors of f2fs_get_meta_page_nofail"), change_curseg() can fail after the curseg has already been switched to the new segment, then the curseg is left in an inconsistent state: it points to a segment which may have no free slot, its summary block is zeroed, and the caller can not restore it. Subsequent block allocation on the curseg will trigger the BUG. Moreover, the writeback thread dies with curseg_lock, curseg_mutex and sentry_lock held, so the commit, shutdown and other I/O threads hang in D state. Since change_curseg() only fails once CP_ERROR_FLAG is set, no checkpoint will be committed afterward, let's invalidate the curseg via reset_curseg_fields() on failure, as new_curseg() does for -ENOSPC, then f2fs_allocate_data_block() will fail with -ENOSPC on the curseg instead of triggering BUG, and do_write_block() can handle the error since CP_ERROR_FLA ---truncated---
-
CVE-2026-99540
In the Linux kernel, the following vulnerability has been resolved: ksmbd: fix parent dentry lifetime in directory enumeration ksmbd_populate_dot_dotdot_entries() builds the ".." entry from the open directory's d_parent and passes that pointer to ksmbd_vfs_fill_dentry_attrs() without taking a reference. That pointer is only valid while the directory stays under this parent. If a concurrent rename moves the directory elsewhere, the old parent loses its last reference; a following rmdir then turns it into a negative dentry with d_inode cleared, and after the RCU grace period the dentry itself is freed. The worker still holds the raw pointer and dereferences d_inode through vfs_getattr() and vfs_getxattr(). A remote SMB client, authenticated or guest, triggers this by listing a directory while that directory is moved under a new parent and the old parent is deleted. It can be done by the client itself with write access or by other concurrent writers, triggering null-ptr-deref and slab-use-after-free on a KASAN build. KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007] Workqueue: ksmbd-io handle_ksmbd_work RIP: 0010:xattr_permission+0x9d/0x160 Call Trace: vfs_getxattr+0x9c/0x1b0 ksmbd_vfs_getxattr+0x82/0x100 ksmbd_vfs_get_dos_attrib_xattr+0x7f/0x130 ksmbd_vfs_fill_dentry_attrs+0x435/0x650 ksmbd_populate_dot_dotdot_entries+0x25e/0x630 smb2_query_dir+0xcf9/0x1a20 handle_ksmbd_work+0x5f1/0x1c80 process_one_work+0x90b/0x19d0 Fix by using the dentry reference counting APIs: take the parent with dget_parent(), take a matching dget() on ".", and dput() the dentry once ksmbd_vfs_fill_dentry_attrs() has returned or the pattern does not match.
-
CVE-2026-99541
In the Linux kernel, the following vulnerability has been resolved: ksmbd: fix OOB read and cross-share confusion in ksmbd_validate_name_reconnect() ksmbd_validate_name_reconnect() indexes ab_pathname[share->path_sz + 1] without checking that the path is longer than share->path_sz + 1 (d_path() fills the buffer from the end). A durable fp is not bound to its original share, so a client can preserve a file from a short-path share and reconnect it on a long-path share, making strcmp() read past the PATH_MAX buffer - with a long enough share path this oopses the kernel, and on a mapped page it also gives a byte-equality oracle on adjacent heap (the reconnect succeeds exactly when the out-of-bounds string equals the requested name). A same-share variant with the durable fp at the share root reads 1 byte past the buffer (strlen(ab_pathname) == share->path_sz). Check before comparing: the path must be longer than share->path_sz, it must start with share->path, and the next character must be '/'. The share root itself is accepted only for an empty name.
-
CVE-2026-99532
In the Linux kernel, the following vulnerability has been resolved: ksmbd: validate security descriptor offsets before pointer arithmetic parse_sec_desc() constructs SID and DACL pointers by adding untrusted 32-bit offsets to the security descriptor before checking their bounds. The existing checks reject offsets into the descriptor header, but do not reject offsets beyond the descriptor's length. On a 32-bit kernel, a large offset can wrap the pointer before the descriptor. The end-only checks in parse_sid() and parse_dacl() can then accept it, allowing an authenticated SMB2 SET_INFO request to access memory outside the security descriptor and fault the kernel. Check each nonzero offset against acl_len before constructing its pointer. Initialize the optional owner and group SID pointers to NULL so an omitted SID remains absent when parsing the DACL.
-
CVE-2026-99534
In the Linux kernel, the following vulnerability has been resolved: ksmbd: prevent stale data from being written by SMB2 WRITE requests Initialize the compatibility byte in request buffers without zeroing the entire allocation. Keep the one-byte compatibility exception for SMB2 messages so valid zero-length WRITE requests remain compatible, and validate the data range in the WRITE handlers against the current compound request before accessing it. This prevents stale request-buffer data from being written to files or sent to IPC pipes.
-
CVE-2026-99535
In the Linux kernel, the following vulnerability has been resolved: ksmbd: prevent share root removal through empty paths A FileLinkInformation request can supply "/" as the new link name. smb2_get_name() strips the trailing slash and passes an empty string to ksmbd_vfs_kern_path_start_removing(). ksmbd_vfs_path_lookup() resolves an empty pathname as the configured share path, without LOOKUP_BENEATH. With ReplaceIfExists set, smb2_create_link() can therefore remove the share root from its parent before link creation fails. An authenticated client with a writable share and DELETE access to a file can trigger this if the share root is empty and the mapped user has write and search permission on its parent directory. Return -EINVAL for empty pathnames when for_remove is set, before resolving the share path or acquiring removal resources. Ordinary lookups continue to accept empty pathnames for opening the share root.
-
CVE-2026-99537
In the Linux kernel, the following vulnerability has been resolved: ksmbd: preserve unreadable named streams on open A write-only client can open an existing named stream with FILE_OPEN_IF without permission to read the underlying file. The case-insensitive lookup finds the xattr name, but querying its value length then fails with -EACCES. smb2_set_stream_name_xattr() treats every negative result as a missing stream and replaces the existing xattr with an empty value. A failed xattr list is not proof that the stream is absent either: for example, -ENOMEM can lead to the same empty replacement. Preserve listing errors and create a stream only for -ENOENT or -ENODATA. The latter can occur if the xattr is removed between listing its name and querying its value. Return other lookup errors without changing the stream. A write-only FILE_OPEN now reports access denied instead of name not found. FILE_OVERWRITE_IF and FILE_SUPERSEDE on an unreadable stream also fail instead of appearing to replace it. Stream writes and EOF changes already need to read the old value, so such handles cannot update the stream after opening either.
-
CVE-2026-99538
In the Linux kernel, the following vulnerability has been resolved: ksmbd: ignore network devices of other namespaces With no interface list configured, ksmbd binds every interface that comes up. ksmbd_netdev_event() does not look at the device's namespace, and create_socket() opens the listener in the namespace of the task raising NETDEV_UP. A container with its own address on a bridge brings up its interface and finds port 445 in its namespace already taken by ksmbd, so an SMB server running inside the container fails with "Address already in use". Interfaces are also matched by name only, so a container's eth0 going down would stop the listener on the host's eth0. Only handle netdev events for devices in init_net.
-
CVE-2026-99539
In the Linux kernel, the following vulnerability has been resolved: ksmbd: fix session reference leak on compounded SESSION_SETUP A compound request can place SESSION_SETUP after another command. That earlier command leaves a reference to its session in work->sess. smb2_sess_setup() then overwrites work->sess with the session it looks up or creates, and the dispatcher later puts work->sess only once. The inherited reference is never dropped. That leaks the session, together with its setup connection and socket, and every repeat of the compound leaks another. unreferenced object (size 512): __kmalloc_cache_noprof+0x32c/0x3a0 ksmbd_smb2_session_create+0xf6/0x9c0 [ksmbd] smb2_sess_setup+0xda4/0x6280 [ksmbd] handle_ksmbd_work+0x88c/0x1cb0 [ksmbd] Save the inherited session on entry. After the request has been processed, release it once, but only when smb2_sess_setup() has replaced work->sess with its own session or cleared it. Do this after ksmbd_conn_unlock() so the put does not run under srv_mutex. An error path may instead leave the inherited session in work->sess. Keep it there, so the dispatcher still puts it and the error response is still signed or encrypted with it as before.
-
CVE-2026-99525
In the Linux kernel, the following vulnerability has been resolved: smb: client: fix use-after-free in cifs_ses_add_channel() The error path of cifs_ses_add_channel() calls cifs_put_tcp_session() before cifs_chan_clear_need_reconnect(). The latter reaches cifs_ses_get_chan_index(), which dereferences server->terminate; if the put was the final reference, TCP_Server_Info is already freed and this is a use-after-free. Same put-then-use pattern as the recently fixed sibling in commit 717e0a25036b ("cifs: Fix server use-after-free in cifs_chan_skip_or_disable()"). Move the put last, but do not read chan->server after dropping chan_lock: chan points into ses->chans[], and once chan_count is decremented a concurrent add_channel can reuse the slot and overwrite the pointer, so the put would act on the new channel's server. Cache the pointer under chan_lock, clear the reconnect bit and decrement chan_count while still holding the reference, and release it after the unlock. This issue was discovered by Tencent CodeBuddy Security.
-
CVE-2026-99526
In the Linux kernel, the following vulnerability has been resolved: smb: client: fix use-after-free infoleak via the readdir resume name CIFSFindNext() copies the resume name that cifs_save_resume_key() recorded from the last entry of the previous response into its request; psrch_inf->presume_name points into the response buffer that entry came from, and nothing ever clears it. find_cifs_entry() records a new name only from a response with a usable last entry, and cifs_readdir() only from an entry it walks to, so a response that carries no entries and leaves last_entry NULL, because CIFSFindFirst() or CIFSFindNext() rejected LastNameOffset, refreshes neither and presume_name keeps pointing into the previous response. That buffer has already been released: by CIFSFindNext() before it installs the new response, or by find_cifs_entry() when it rewinds the search, which does not drop the name either and can then fail in initiate_cifs_search() before a new response replaces it. The next CIFSFindNext() then copies up to PATH_MAX - 1 bytes of the resume name out of the released buffer into its request and sends them to the server: a use-after-free read that leaks kernel memory to a malicious or compromised server, which can provoke it with a single FindNext response that carries no entries and a LastNameOffset above CIFSMaxBufSize, which is all the existing check rejects. With KASAN enabled, a server that answers a listing that way gives: CIFS: VFS: ignoring corrupt resume name ================================================================== BUG: KASAN: slab-use-after-free in CIFSFindNext+0x8ca/0x14c0 Read of size 4080 at addr ffff88807c823f98 by task ls/83 CPU: 0 UID: 0 PID: 83 Comm: ls Tainted: G B 7.3.0-rc4-00457-gf14572c203d5 #69 PREEMPT(lazy) Tainted: [B]=BAD_PAGE Call Trace: <TASK> kasan_report+0xdf/0x1a0 kasan_check_range+0x10f/0x1e0 __asan_memcpy+0x23/0x60 CIFSFindNext+0x8ca/0x14c0 cifs_readdir+0xf51/0x29c0 iterate_dir+0x1c0/0x570 __x64_sys_getdents64+0x133/0x270 do_syscall_64+0x109/0x5d0 entry_SYSCALL_64_after_hwframe+0x77/0x7f </TASK> Allocated by task 83 on cpu 1 at 13.497119s: cifs_buf_get+0x36/0x90 smb_init+0x4f/0x110 CIFSFindNext+0xf7/0x14c0 cifs_readdir+0xf51/0x29c0 Freed by task 83 on cpu 0 at 13.584018s: cifs_buf_release+0x41/0x80 CIFSFindNext+0xfce/0x14c0 cifs_readdir+0xf51/0x29c0 The buggy address is located 16280 bytes inside of freed 16588-byte region [ffff88807c820000, ffff88807c8240cc) ================================================================== The same memcpy() first reads past the end of that object while it is still live, a slab-out-of-bounds read that the following patches bound; that is the taint the report above carries. SMB1 is not negotiated by default; reaching this code requires an explicit vers=1.0 mount. Clear presume_name and resume_name_len whenever a new response is installed and when the rewind path releases the search buffer, so that the name never outlives the buffer it points into, and reset resume_key with them, since it was taken from the same entry and would otherwise be sent with an empty name. A response without a usable last entry then continues the search with an empty resume name rather than a stale one; CIFSFindNext() skips the copy of a zero-length name so that it never runs on the NULL pointer. The next patch in the series leaves last_entry NULL for a response that carries no entries, and bounds LastNameOffset against the received response rather than against CIFSMaxBufSize. Responses that are given a last entry today then take the path fixed here, so that patch must not be applied without this one.
CVE records published in 2026 (ids starting CVE-2026-), sorted by how likely they are to be exploited.