Vulnerabilities

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> usb: chipidea: udc: fix DMA and SG cleanup in _ep_nuke()<br /> <br /> The ChipIdea UDC driver can encounter "not page aligned sg buffer"<br /> errors when a USB device is reconnected after being disconnected<br /> during an active transfer. This occurs because _ep_nuke() returns<br /> requests to the gadget layer without properly unmapping DMA buffers<br /> or cleaning up scatter-gather bounce buffers.<br /> <br /> Root cause:<br /> When a disconnect happens during a multi-segment DMA transfer, the<br /> request&amp;#39;s num_mapped_sgs field and sgt.sgl pointer remain set with<br /> stale values. The request is returned to the gadget driver with status<br /> -ESHUTDOWN but still has active DMA state. If the gadget driver reuses<br /> this request on reconnect without reinitializing it, the stale DMA<br /> state causes _hardware_enqueue() to skip DMA mapping (seeing non-zero<br /> num_mapped_sgs) and attempt to use freed/invalid DMA addresses,<br /> leading to alignment errors and potential memory corruption.<br /> <br /> The normal completion path via _hardware_dequeue() properly calls<br /> usb_gadget_unmap_request_by_dev() and sglist_do_debounce() before<br /> returning the request. The _ep_nuke() path must do the same cleanup<br /> to ensure requests are returned in a clean, reusable state.<br /> <br /> Fix:<br /> Add DMA unmapping and bounce buffer cleanup to _ep_nuke() to mirror<br /> the cleanup sequence in _hardware_dequeue():<br /> - Call usb_gadget_unmap_request_by_dev() if num_mapped_sgs is set<br /> - Call sglist_do_debounce() with copy=false if bounce buffer exists<br /> <br /> This ensures that when requests are returned due to endpoint shutdown,<br /> they don&amp;#39;t retain stale DMA mappings. The &amp;#39;false&amp;#39; parameter to<br /> sglist_do_debounce() prevents copying data back (appropriate for<br /> shutdown path where transfer was aborted).

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> ntfs: -&gt;d_compare() must not block<br /> <br /> ... so don&amp;#39;t use __getname() there. Switch it (and ntfs_d_hash(), while<br /> we are at it) to kmalloc(PATH_MAX, GFP_NOWAIT). Yes, ntfs_d_hash()<br /> almost certainly can do with smaller allocations, but let ntfs folks<br /> deal with that - keep the allocation size as-is for now.<br /> <br /> Stop abusing names_cachep in ntfs, period - various uses of that thing<br /> in there have nothing to do with pathnames; just use k[mz]alloc() and<br /> be done with that. For now let&amp;#39;s keep sizes as-in, but AFAICS none of<br /> the users actually want PATH_MAX.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> kcm: fix zero-frag skb in frag_list on partial sendmsg error<br /> <br /> Syzkaller reported a warning in kcm_write_msgs() when processing a<br /> message with a zero-fragment skb in the frag_list.<br /> <br /> When kcm_sendmsg() fills MAX_SKB_FRAGS fragments in the current skb,<br /> it allocates a new skb (tskb) and links it into the frag_list before<br /> copying data. If the copy subsequently fails (e.g. -EFAULT from<br /> user memory), tskb remains in the frag_list with zero fragments:<br /> <br /> head skb (msg being assembled, NOT yet in sk_write_queue)<br /> +-----------+<br /> | frags[17] | (MAX_SKB_FRAGS, all filled with data)<br /> | frag_list-+--&gt; tskb<br /> +-----------+ +----------+<br /> | frags[0] | (empty! copy failed before filling)<br /> +----------+<br /> <br /> For SOCK_SEQPACKET with partial data already copied, the error path<br /> saves this message via partial_message for later completion. For<br /> SOCK_SEQPACKET, sock_write_iter() automatically sets MSG_EOR, so a<br /> subsequent zero-length write(fd, NULL, 0) completes the message and<br /> queues it to sk_write_queue. kcm_write_msgs() then walks the<br /> frag_list and hits:<br /> <br /> WARN_ON(!skb_shinfo(skb)-&gt;nr_frags)<br /> <br /> TCP has a similar pattern where skbs are enqueued before data copy<br /> and cleaned up on failure via tcp_remove_empty_skb(). KCM was<br /> missing the equivalent cleanup.<br /> <br /> Fix this by tracking the predecessor skb (frag_prev) when allocating<br /> a new frag_list entry. On error, if the tail skb has zero frags,<br /> use frag_prev to unlink and free it in O(1) without walking the<br /> singly-linked frag_list. frag_prev is safe to dereference because<br /> the entire message chain is only held locally (or in kcm-&gt;seq_skb)<br /> and is not added to sk_write_queue until MSG_EOR, so the send path<br /> cannot free it underneath us.<br /> <br /> Also change the WARN_ON to WARN_ON_ONCE to avoid flooding the log<br /> if the condition is somehow hit repeatedly.<br /> <br /> There are currently no KCM selftests in the kernel tree; a simple<br /> reproducer is available at [1].<br /> <br /> [1] https://gist.github.com/mrpre/a94d431c757e8d6f168f4dd1a3749daa

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> drm/amd/display: Add signal type check for dcn401 get_phyd32clk_src<br /> <br /> Trying to access link enc on a dpia link will cause a crash otherwise

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> soc: ti: k3-socinfo: Fix regmap leak on probe failure<br /> <br /> The mmio regmap allocated during probe is never freed.<br /> <br /> Switch to using the device managed allocator so that the regmap is<br /> released on probe failures (e.g. probe deferral) and on driver unbind.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> ntb: ntb_hw_switchtec: Fix array-index-out-of-bounds access<br /> <br /> Number of MW LUTs depends on NTB configuration and can be set to MAX_MWS,<br /> This patch protects against invalid index out of bounds access to mw_sizes<br /> When invalid access print message to user that configuration is not valid.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> x86/kexec: add a sanity check on previous kernel&amp;#39;s ima kexec buffer<br /> <br /> When the second-stage kernel is booted via kexec with a limiting command<br /> line such as "mem=", the physical range that contains the carried<br /> over IMA measurement list may fall outside the truncated RAM leading to a<br /> kernel panic.<br /> <br /> BUG: unable to handle page fault for address: ffff97793ff47000<br /> RIP: ima_restore_measurement_list+0xdc/0x45a<br /> #PF: error_code(0x0000) – not-present page<br /> <br /> Other architectures already validate the range with page_is_ram(), as done<br /> in commit cbf9c4b9617b ("of: check previous kernel&amp;#39;s ima-kexec-buffer<br /> against memory bounds") do a similar check on x86.<br /> <br /> Without carrying the measurement list across kexec, the attestation<br /> would fail.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> smb: client: prevent races in -&gt;query_interfaces()<br /> <br /> It was possible for two query interface works to be concurrently trying<br /> to update the interfaces.<br /> <br /> Prevent this by checking and updating iface_last_update under<br /> iface_lock.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> net/sched: act_skbedit: fix divide-by-zero in tcf_skbedit_hash()<br /> <br /> Commit 38a6f0865796 ("net: sched: support hash selecting tx queue")<br /> added SKBEDIT_F_TXQ_SKBHASH support. The inclusive range size is<br /> computed as:<br /> <br /> mapping_mod = queue_mapping_max - queue_mapping + 1;<br /> <br /> The range size can be 65536 when the requested range covers all possible<br /> u16 queue IDs (e.g. queue_mapping=0 and queue_mapping_max=U16_MAX).<br /> That value cannot be represented in a u16 and previously wrapped to 0,<br /> so tcf_skbedit_hash() could trigger a divide-by-zero:<br /> <br /> queue_mapping += skb_get_hash(skb) % params-&gt;mapping_mod;<br /> <br /> Compute mapping_mod in a wider type and reject ranges larger than U16_MAX<br /> to prevent params-&gt;mapping_mod from becoming 0 and avoid the crash.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> media: radio-keene: fix memory leak in error path<br /> <br /> Fix a memory leak in usb_keene_probe(). The v4l2 control handler is<br /> initialized and controls are added, but if v4l2_device_register() or<br /> video_register_device() fails afterward, the handler was never freed,<br /> leaking memory.<br /> <br /> Add v4l2_ctrl_handler_free() call in the err_v4l2 error path to ensure<br /> the control handler is properly freed for all error paths after it is<br /> initialized.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> drm/amdgpu: Refactor amdgpu_gem_va_ioctl for Handling Last Fence Update and Timeline Management v4<br /> <br /> This commit simplifies the amdgpu_gem_va_ioctl function, key updates<br /> include:<br /> - Moved the logic for managing the last update fence directly into<br /> amdgpu_gem_va_update_vm.<br /> - Introduced checks for the timeline point to enable conditional<br /> replacement or addition of fences.<br /> <br /> v2: Addressed review comments from Christian.<br /> v3: Updated comments (Christian).<br /> v4: The previous version selected the fence too early and did not manage its<br /> reference correctly, which could lead to stale or freed fences being used.<br /> This resulted in refcount underflows and could crash when updating GPU<br /> timelines.<br /> The fence is now chosen only after the VA mapping work is completed, and its<br /> reference is taken safely. After exporting it to the VM timeline syncobj, the<br /> driver always drops its local fence reference, ensuring balanced refcounting<br /> and avoiding use-after-free on dma_fence.<br /> <br /> Crash signature:<br /> [ 205.828135] refcount_t: underflow; use-after-free.<br /> [ 205.832963] WARNING: CPU: 30 PID: 7274 at lib/refcount.c:28 refcount_warn_saturate+0xbe/0x110<br /> ...<br /> [ 206.074014] Call Trace:<br /> [ 206.076488] <br /> [ 206.078608] amdgpu_gem_va_ioctl+0x6ea/0x740 [amdgpu]<br /> [ 206.084040] ? __pfx_amdgpu_gem_va_ioctl+0x10/0x10 [amdgpu]<br /> [ 206.089994] drm_ioctl_kernel+0x86/0xe0 [drm]<br /> [ 206.094415] drm_ioctl+0x26e/0x520 [drm]<br /> [ 206.098424] ? __pfx_amdgpu_gem_va_ioctl+0x10/0x10 [amdgpu]<br /> [ 206.104402] amdgpu_drm_ioctl+0x4b/0x80 [amdgpu]<br /> [ 206.109387] __x64_sys_ioctl+0x96/0xe0<br /> [ 206.113156] do_syscall_64+0x66/0x2d0<br /> ...<br /> [ 206.553351] BUG: unable to handle page fault for address: ffffffffc0dfde90<br /> ...<br /> [ 206.553378] RIP: 0010:dma_fence_signal_timestamp_locked+0x39/0xe0<br /> ...<br /> [ 206.553405] Call Trace:<br /> [ 206.553409] <br /> [ 206.553415] ? __pfx_drm_sched_fence_free_rcu+0x10/0x10 [gpu_sched]<br /> [ 206.553424] dma_fence_signal+0x30/0x60<br /> [ 206.553427] drm_sched_job_done.isra.0+0x123/0x150 [gpu_sched]<br /> [ 206.553434] dma_fence_signal_timestamp_locked+0x6e/0xe0<br /> [ 206.553437] dma_fence_signal+0x30/0x60<br /> [ 206.553441] amdgpu_fence_process+0xd8/0x150 [amdgpu]<br /> [ 206.553854] sdma_v4_0_process_trap_irq+0x97/0xb0 [amdgpu]<br /> [ 206.554353] edac_mce_amd(E) ee1004(E)<br /> [ 206.554270] amdgpu_irq_dispatch+0x150/0x230 [amdgpu]<br /> [ 206.554702] amdgpu_ih_process+0x6a/0x180 [amdgpu]<br /> [ 206.555101] amdgpu_irq_handler+0x23/0x60 [amdgpu]<br /> [ 206.555500] __handle_irq_event_percpu+0x4a/0x1c0<br /> [ 206.555506] handle_irq_event+0x38/0x80<br /> [ 206.555509] handle_edge_irq+0x92/0x1e0<br /> [ 206.555513] __common_interrupt+0x3e/0xb0<br /> [ 206.555519] common_interrupt+0x80/0xa0<br /> [ 206.555525] <br /> [ 206.555527] <br /> ...<br /> [ 206.555650] RIP: 0010:dma_fence_signal_timestamp_locked+0x39/0xe0<br /> ...<br /> [ 206.555667] Kernel panic - not syncing: Fatal exception in interrupt

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> drm/atmel-hlcdc: fix use-after-free of drm_crtc_commit after release<br /> <br /> The atmel_hlcdc_plane_atomic_duplicate_state() callback was copying<br /> the atmel_hlcdc_plane state structure without properly duplicating the<br /> drm_plane_state. In particular, state-&gt;commit remained set to the old<br /> state commit, which can lead to a use-after-free in the next<br /> drm_atomic_commit() call.<br /> <br /> Fix this by calling<br /> __drm_atomic_helper_duplicate_plane_state(), which correctly clones<br /> the base drm_plane_state (including the -&gt;commit pointer).<br /> <br /> It has been seen when closing and re-opening the device node while<br /> another DRM client (e.g. fbdev) is still attached:<br /> <br /> =============================================================================<br /> BUG kmalloc-64 (Not tainted): Poison overwritten<br /> -----------------------------------------------------------------------------<br /> <br /> 0xc611b344-0xc611b344 @offset=836. First byte 0x6a instead of 0x6b<br /> FIX kmalloc-64: Restoring Poison 0xc611b344-0xc611b344=0x6b<br /> Allocated in drm_atomic_helper_setup_commit+0x1e8/0x7bc age=178 cpu=0<br /> pid=29<br /> drm_atomic_helper_setup_commit+0x1e8/0x7bc<br /> drm_atomic_helper_commit+0x3c/0x15c<br /> drm_atomic_commit+0xc0/0xf4<br /> drm_framebuffer_remove+0x4cc/0x5a8<br /> drm_mode_rmfb_work_fn+0x6c/0x80<br /> process_one_work+0x12c/0x2cc<br /> worker_thread+0x2a8/0x400<br /> kthread+0xc0/0xdc<br /> ret_from_fork+0x14/0x28<br /> Freed in drm_atomic_helper_commit_hw_done+0x100/0x150 age=8 cpu=0<br /> pid=169<br /> drm_atomic_helper_commit_hw_done+0x100/0x150<br /> drm_atomic_helper_commit_tail+0x64/0x8c<br /> commit_tail+0x168/0x18c<br /> drm_atomic_helper_commit+0x138/0x15c<br /> drm_atomic_commit+0xc0/0xf4<br /> drm_atomic_helper_set_config+0x84/0xb8<br /> drm_mode_setcrtc+0x32c/0x810<br /> drm_ioctl+0x20c/0x488<br /> sys_ioctl+0x14c/0xc20<br /> ret_fast_syscall+0x0/0x54<br /> Slab 0xef8bc360 objects=21 used=16 fp=0xc611b7c0<br /> flags=0x200(workingset|zone=0)<br /> Object 0xc611b340 @offset=832 fp=0xc611b7c0