Search Results (4513 CVEs found)

CVE Vendors Products Updated CVSS v3.1
CVE-2026-74378 1 Linux 1 Linux Kernel 2026-08-17 7.8 High
In the Linux kernel, the following vulnerability has been resolved: RDMA/rxe: Fix TOCTOU heap overflow in get_srq_wqe get_srq_wqe() reads wqe->dma.num_sge from the shared receive queue buffer, which is mapped into userspace. It validates num_sge against max_sge, but then re-reads the same field to calculate the memcpy size. A concurrent userspace thread can modify num_sge between validation and use, causing a heap buffer overflow when copying the WQE into qp->resp.srq_wqe. Read num_sge into a local variable and use it for both the bounds check and the size calculation.
CVE-2026-74387 1 Linux 1 Linux Kernel 2026-08-17 7.8 High
In the Linux kernel, the following vulnerability has been resolved: ALSA: seq: midi: Serialize output teardown with event_input event_process_midi() borrows msynth->output_rfile.output and then passes the substream to dump_midi() and snd_rawmidi_kernel_write() without synchronizing with the output open/close transition. midisynth_use() also publishes output_rfile before snd_rawmidi_output_params() has finished. The last midisynth_unuse() can therefore release the same rawmidi file and free substream->runtime before snd_rawmidi_kernel_write1() takes its runtime buffer reference. That leaves the event_input path using a stale substream or runtime and can end in a NULL-deref or use-after-free. Fix this with two pieces of synchronization. Keep a short IRQ-safe spinlock only for publishing or clearing output_rfile and for pairing the output snapshot with an snd_use_lock_t reference. Once event_process_midi() has taken that in-flight reference, it drops the spinlock before calling snd_seq_dump_var_event(), dump_midi(), or snd_rawmidi_kernel_write(). midisynth_unuse() now detaches the visible rawmidi file under the same spinlock, waits for the in-flight writers to drain, and only then drains and releases the saved file. midisynth_use() likewise opens into a local snd_rawmidi_file and publishes it only after snd_rawmidi_output_params() succeeds. The buggy scenario involves two paths, with each column showing the order within that path: event_input path: last unuse path: 1. event_process_midi() snapshots 1. midisynth_unuse() starts output_rfile.output. tearing down output_rfile. 2. dump_midi() reaches 2. snd_rawmidi_kernel_release() snd_rawmidi_kernel_write() closes the output file. before runtime is pinned. 3. close_substream() frees 3. The callback keeps using substream->runtime. the borrowed substream. Validation reproduced this kernel report: KASAN null-ptr-deref in snd_rawmidi_kernel_write1+0x56/0x360 RIP: 0033:0x7fde7dd0837f RIP: 0010:snd_rawmidi_kernel_write1+0x56/0x360
CVE-2026-74438 1 Linux 1 Linux Kernel 2026-08-17 7.8 High
In the Linux kernel, the following vulnerability has been resolved: crypto: sun4i-ss - Remove insecure and unused rng_alg Remove sun4i_ss_rng, as it is insecure and unused: - It has multiple vulnerabilities. sun4i_ss_prng_seed() is missing locking and has a buffer overflow. sun4i_ss_prng_generate() fails to fill the entire buffer with cryptographic random bytes, because it rounds the destination length down and also doesn't actually wait for the hardware to be ready before pulling bytes from it. - No user of this code is known. It's usable only theoretically via the "rng" algorithm type of AF_ALG. But userspace actually just uses the actual Linux RNG (/dev/random etc) instead. And rng_algs don't contribute entropy to the actual Linux RNG either. (This may have been confused with hwrng, which does contribute entropy.) The sun4i_ss_prng_seed() buffer overflow was reported by Tianchu Chen and discovered by Atuin - Automated Vulnerability Discovery Engine There's no point in fixing all these vulnerabilities individually when this is unused code, so let's just remove it.
CVE-2026-74513 1 Linux 1 Linux Kernel 2026-08-17 7.8 High
In the Linux kernel, the following vulnerability has been resolved: dibs: fix use-after-free of dmb_node in loopback attach/detach/unregister dibs_lo_attach_dmb(), dibs_lo_detach_dmb() and dibs_lo_unregister_dmb() look up the dmb_node under dmb_ht_lock, drop the lock and only then operate on the node's refcount. Nothing keeps the node alive across that window: __dibs_lo_unregister_dmb() removes the node from the hash table under the write lock and immediately frees it. A concurrent final put can therefore free the node between the lookup and the refcount operation: CPU0 (attach) CPU1 (owner unregisters) read_lock_bh(&dmb_ht_lock) find dmb_node (refcnt == 1) read_unlock_bh(&dmb_ht_lock) refcount_dec_and_test() 1 -> 0 write_lock_bh(&dmb_ht_lock) hash_del(&dmb_node->list) write_unlock_bh(&dmb_ht_lock) kfree(dmb_node) refcount_inc_not_zero(&dmb_node->refcnt) <-- use-after-free The same window exists for the refcount_dec_and_test() calls in the detach and unregister paths. Close the race structurally by making hash table membership and the refcount transitions atomic with respect to each other: - Perform the final refcount_dec_and_test() and hash_del() in a single dmb_ht_lock write-side critical section, in both the unregister and the detach path. Freeing the node still happens after the lock is dropped, which is safe because a node whose refcount reached zero has left the hash table and can no longer be found. - This establishes the invariant that any node found in the hash table holds at least one reference, and that the final reference can only be dropped under the write lock. dibs_lo_attach_dmb() can thus take its reference with a plain refcount_inc() while still holding the read lock; refcount_inc_not_zero() is no longer needed. __dibs_lo_unregister_dmb() no longer touches the hash table and is renamed to dibs_lo_free_dmb() accordingly. Note: commit cc21191b584c ("dibs: Move data path to dibs layer") moved the code to its current location; the race was introduced earlier by commit c3a910f2380f ("net/smc: implement DMB-merged operations of loopback-ism"). Tested SMC-D via ISM and dibs loopback.
CVE-2026-74538 1 Linux 1 Linux Kernel 2026-08-17 8.8 High
In the Linux kernel, the following vulnerability has been resolved: Bluetooth: ISO: lock sk in iso_connect_ind Accessing iso_pi(sk)->conn requires lock_sock, which is not taken in the "ev3" part of iso_connect_ind. It may also be NULL if socket has transitioned away from the LISTEN/CONNECT states before locking. Fix by adding lock/release. Recheck hcon is valid after lock acquire where needed.
CVE-2026-74262 1 Linux 1 Linux Kernel 2026-08-17 7.8 High
In the Linux kernel, the following vulnerability has been resolved: kcm: use WRITE_ONCE() when changing lower socket callbacks kcm_attach() replaces a live lower TCP socket's sk_data_ready and sk_write_space callbacks with KCM handlers, and kcm_unattach() restores them later. Those callback-pointer updates are still plain stores even though the same fields can be read and invoked concurrently on other CPUs. If another CPU observes an older callback snapshot after the live field has already been restored, callback execution can run with a mismatched target and sk_user_data state, leading to stale or misdirected wakeups. Use WRITE_ONCE() for the callback replacement and restore operations so these shared callback fields follow the same visibility contract already established by the earlier 4022 fixes.
CVE-2026-74407 1 Linux 1 Linux Kernel 2026-08-17 8.8 High
In the Linux kernel, the following vulnerability has been resolved: wifi: ath11k: cancel SSR work items during PCI shutdown A reboot can crash the kernel if it overlaps with WLAN firmware crash recovery (SSR). The crash is a NULL pointer dereference in the MHI teardown path while freeing DMA-backed MHI contexts. Simplified trace: dma_free_attrs mhi_deinit_dev_ctxt [mhi] ath11k_pci_power_down [ath11k_pci] ath11k_pci_shutdown [ath11k_pci] device_shutdown kernel_restart On the host side, SSR is driven by the MHI RDDM callback, which queues reset_work to perform device recovery. reset_work power-cycles the device by calling ath11k_hif_power_down() followed by ath11k_hif_power_up(). The power-down phase deinitializes MHI and frees DMA resources. Shutdown/reboot runs fully asynchronously with this RDDM-driven SSR recovery flow. As a result, the shutdown path (ath11k_pci_shutdown() -> ath11k_pci_power_down()) can race with the SSR recovery sequence. Fix this by canceling SSR-related work items during PCI shutdown, marking the device as unregistering, and serializing the RDDM callback path that checks and queues reset_work. This ensures that no new SSR recovery work can be queued once teardown has started, and that any in-flight recovery work is fully synchronized before device power-down, preventing MHI teardown and DMA resource freeing from running more than once. Note: This issue only affects PCI/MHI-based devices. AHB-based ath11k devices do not queue reset_work in normal SSR flows. Tested-on: WCN6855 hw2.1 PCI WLAN.HSP.1.1-04866.5-QCAHSPSWPL_V1_V2_SILICONZ_IOE-1
CVE-2026-74489 1 Linux 1 Linux Kernel 2026-08-17 8.8 High
In the Linux kernel, the following vulnerability has been resolved: wifi: mac80211: fix tid_tx use-after-free on BA session stop ieee80211_stop_tx_ba_cb() hands tid_tx to kfree_rcu() through ieee80211_remove_tid_tx(), and then reads tid_tx->ndp after dropping sta->lock: ieee80211_remove_tid_tx(sta, tid); /* kfree_rcu(tid_tx, rcu_head) */ ... spin_unlock_bh(&sta->lock); if (start_txq) ieee80211_agg_start_txq(sta, tid, false); if (send_delba) ieee80211_send_delba(..., tid_tx->ndp); That read is not covered by an RCU read-side critical section, and it runs in preemptible process context: both callers hold the wiphy mutex, reaching it either from the ieee80211_ba_session_work() wiphy work or from ieee80211_sta_tear_down_BA_sessions() during station teardown. Softirqs can run in that window too, both from the local_bh_enable() that ends ieee80211_agg_start_txq() and from any interrupt exit, so the RCU callback can free tid_tx before the read. Driving the function from a test module with the grace period forced into that window, KASAN reports the read, and the free arrives on the ordinary RCU softirq path: BUG: KASAN: slab-use-after-free in ieee80211_stop_tx_ba_cb+0x3cd/0x400 Read of size 1 at addr ffff888002b9f52e by task kworker/0:1/10 [...] Freed by task 57: __kasan_slab_free+0x47/0x70 __rcu_free_sheaf_prepare+0x70/0x250 rcu_free_sheaf_nobarn+0x18/0x40 rcu_core+0x426/0x1310 handle_softirqs+0x144/0x590 __irq_exit_rcu+0xea/0x150 irq_exit_rcu+0x9/0x20 sysvec_apic_timer_interrupt+0x6b/0x80 asm_sysvec_apic_timer_interrupt+0x1a/0x20 send_delba is only set when tx_stop is set, which happens for AGG_STOP_LOCAL_REQUEST alone, so this is reached on local teardown - session idle timeout, PTK rekey, suspend, HW reconfig - and not from a peer's DELBA. Read ndp into a local before the session is freed, while sta->lock is still held. tid_tx->ndp has a single writer, in ieee80211_tx_ba_session_handle_start(), which cannot run concurrently here: both paths are serialised by the wiphy mutex, and the session is already marked HT_AGG_STATE_STOPPING at this point. tid_tx->ndp is also the only tid_tx dereference left after ieee80211_remove_tid_tx() in this function. [move/change the comment a bit to be more general not just on ndp, initialize ndp directly]
CVE-2026-74574 1 Linux 1 Linux Kernel 2026-08-17 7.8 High
In the Linux kernel, the following vulnerability has been resolved: dmaengine: idxd: fix fdev setup failure cleanup in idxd_cdev_open() The failed_dev_add and failed_dev_name paths drop the file-device reference while wq->wq_lock is still held. If put_device(fdev) drops the last reference, idxd_file_dev_release() runs synchronously and tries to take wq->wq_lock again, deadlocking. Those paths also fall through into the later ctx cleanup labels even though idxd_file_dev_release() owns that cleanup and frees ctx. This can make idxd_xa_pasid_remove(ctx) and kfree(ctx) operate on a freed context. Move idxd_wq_get() before file-device setup can fail, since the release callback always calls idxd_wq_put(). Then unlock wq->wq_lock before put_device(fdev) and return directly from the file-device setup failure path, leaving ctx cleanup to the release callback.
CVE-2026-74506 1 Linux 1 Linux Kernel 2026-08-17 7.8 High
In the Linux kernel, the following vulnerability has been resolved: afs: Fix UAF when sending a message In afs_make_call(), there's a race with async call reception and destruction. If a call is dispatched that doesn't have call->write_iter set (used to specify the data content for FS.StoreData), then the first rxrpc_kernel_send_data() will not set MSG_MORE in the msghdr. Once rxrpc_send_data() queues the last request packet, the response could come in at any time and cause the call to be completed and put. However, afs_make_call() will look at the call again to see it ->write_iter should be handled - something it's only allowed to do if it has its own ref on the call. Whilst this is the case for synchronous calls, it isn't true for async calls such as FS.FetchData. There's also a potential UAF in afs_make_call() in the event that an asynchronous call is being sent, but the call fails in some way (e.g. it gets aborted from the server). The problem there is that afs_make_call() tries to abort a call if the rxrpc send fails, but the asynchronous notification from rxrpc may have caused the afs_call to be torn down. generic/650 plays games with randomly taking CPUs offline, and can interject a significant delay such that the call is deallocated before afs_make_call() gets to check call->write_iter - and a UAF ensues (caught by KASAN). BUG: KASAN: slab-use-after-free in afs_make_call+0x1c90/0x2210 [kafs] Read of size 8 at addr ffff888035e050e8 by task fsstress/1409 Fix this by making afs_make_op_call() give the op->call its own ref rather than transferring the caller's ref to it and then dropping the ref when afs_make_call() returns. This also means that the afs_make_call() func never loses its ref on the call now.
CVE-2026-74523 1 Linux 1 Linux Kernel 2026-08-17 7.5 High
In the Linux kernel, the following vulnerability has been resolved: qede: sync udp_tunnel ports outside qede_lock in the recovery path A TX timeout on a qede NIC that has VXLAN/GENEVE tunnel ports configured wedges the rtnetlink control plane of the whole machine: NETDEV WATCHDOG: ens6f1 (qede): transmit queue 2 timed out 10226 ms [qede_tx_timeout:586(ens6f1)]TX timeout on queue 2! [qede_recovery_handler:2665(ens6f0)]Starting a recovery process The recovery path deadlocks on the driver's own mutex: qede_sp_task rtnl_lock() mutex_lock(&edev->qede_lock) <- taken qede_recovery_handler qede_load udp_tunnel_nic_reset_ntf __udp_tunnel_nic_device_sync info->sync_table == qede_udp_tunnel_sync mutex_lock(&edev->qede_lock) <- same task: deadlock The mutex is not recursive, so the kworker blocks on itself with rtnl_lock held, and neither lock is ever released. Every task that calls rtnl_lock() afterwards (ip, ovs-vswitchd, lldpad, IPv6 addrconf, sshd) blocks forever while the node still answers ping. In a vmcore from an affected production node rtnl_mutex.owner decodes to the very kworker blocked at the innermost mutex_lock() above. Re-sync the tunnel ports from qede_sp_task() after the internal lock is dropped, still under rtnl_lock as the udp_tunnel API requires. This mirrors qede_open(), which calls udp_tunnel_nic_reset_ntf() under rtnl without the internal lock. qede_recovery_handler() now returns whether it has successfully reloaded an open device, and the caller re-syncs the ports only in that case. This keeps the old gating exactly: a device that was down or a failed recovery returns false, as those paths never reached the udp_tunnel_nic_reset_ntf() call before either. This was the only user of the qede_lock()/qede_unlock() helpers, so remove them.
CVE-2026-74563 1 Linux 1 Linux Kernel 2026-08-17 7.8 High
In the Linux kernel, the following vulnerability has been resolved: rds: tcp: hold the RCU lock across ipv6_chk_addr() in rds_tcp_laddr_check() rds_tcp_laddr_check() looks up a scoped IPv6 interface with dev_get_by_index_rcu(), drops the RCU read-side lock, and only then passes the bare struct net_device * into ipv6_chk_addr(). dev_get_by_index_rcu() only keeps the device alive within the same RCU read-side section. After rcu_read_unlock(), a concurrent RTM_DELLINK can free the net_device; ipv6_chk_addr() then dereferences the stale pointer in __ipv6_chk_addr_and_flags() (e.g. l3mdev_master_dev_rcu(dev)), reading freed memory. Keep the RCU read-side lock held across the ipv6_chk_addr() call instead of dropping it right after the lookup, so the device cannot be freed while it is in use. BUG: KASAN: slab-use-after-free in __ipv6_chk_addr_and_flags (... net/ipv6/addrconf.c:1998) Read of size 8 at addr ffff8880106ec000 by task exploit/153 Call Trace: ... kasan_report (mm/kasan/report.c:595) __ipv6_chk_addr_and_flags (... net/ipv6/addrconf.c:1998) ipv6_chk_addr (net/ipv6/addrconf.c:2031 net/ipv6/addrconf.c:1972) rds_tcp_laddr_check (net/rds/tcp.c:370) rds_bind (net/rds/bind.c:248) __sys_bind (net/socket.c:1920) __x64_sys_bind (net/socket.c:1956) do_syscall_64 (arch/x86/entry/syscall_64.c:63) entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:121)
CVE-2026-74398 1 Linux 1 Linux Kernel 2026-08-17 9.8 Critical
In the Linux kernel, the following vulnerability has been resolved: ipv6: addrconf: bail out of dad_failure when state is no longer POSTDAD addrconf_dad_failure() transitions ifp->state from DAD to POSTDAD via addrconf_dad_end(), which drops ifp->lock on return. The lock is re-acquired after net_info_ratelimited(). A concurrent ipv6_del_addr() can take the lock in that window, set ifp->state to DEAD and run list_del_rcu(&ifp->if_list). addrconf_dad_failure() then overwrites DEAD with ERRDAD at errdad: and schedules a new dad_work. The work calls ipv6_del_addr() again, hitting the already-poisoned list entry: general protection fault: 0000 [#1] SMP NOPTI CPU: 4 PID: 217 Comm: kworker/4:1 Workqueue: ipv6_addrconf addrconf_dad_work RIP: 0010:ipv6_del_addr+0xe9/0x280 RAX: dead000000000122 Call Trace: addrconf_dad_stop+0x113/0x140 addrconf_dad_work+0x28c/0x430 process_one_work+0x1eb/0x3b0 worker_thread+0x4d/0x400 kthread+0x104/0x140 ret_from_fork+0x35/0x40 Fold the addrconf_dad_end() logic into addrconf_dad_failure() under a single ifp->lock critical section. The STABLE_PRIVACY branch temporarily drops ifp->lock around address regeneration, so at lock_errdad: verify the state is still POSTDAD before transitioning to ERRDAD; bail out otherwise to avoid overwriting a state set by another path while the lock was released.
CVE-2026-74436 1 Linux 1 Linux Kernel 2026-08-17 9.8 Critical
In the Linux kernel, the following vulnerability has been resolved: rxrpc: serialize kernel accept preallocation with socket teardown rxrpc_kernel_charge_accept() reads rx->backlog without any socket/backlog synchronization and passes that raw pointer into rxrpc_service_prealloc_one(). A concurrent rxrpc_discard_prealloc() sets rx->backlog = NULL and frees the backlog rings, so a kernel preallocation worker can keep using a freed struct rxrpc_backlog while updating *_backlog_head/tail and array slots. Serialize the state check and backlog lookup with the socket lock, and reject kernel preallocation once teardown has disabled listening or discarded the service backlog.
CVE-2026-72494 1 Linux 1 Linux Kernel 2026-08-17 9.8 Critical
In the Linux kernel, the following vulnerability has been resolved: RDMA/irdma: Replace waitqueue and flag with completion The driver previously used a waitqueue along with an explicit request_done flag, but without proper barriers around request_done. An earlier patch by Gui-Dong Han <hanguidong02@gmail.com> attempted to fix this by adding the missing memory barriers. Rather than adding the barriers, this patch replaces the waitqueue+flag with a completion, which is designed for this exact purpose.
CVE-2026-74374 1 Linux 1 Linux Kernel 2026-08-17 7.5 High
In the Linux kernel, the following vulnerability has been resolved: md/raid1,raid10: fix error-path detection with md_cloned_bio() Detect the error path using md_cloned_bio() instead of relying on r1_bio in raid1 or r10_bio->read_slot in raid10, which may be NULL or -1 after splitting and resubmitting a failed bio. As a result, the error path may not be recognized and memory allocations can incorrectly use GFP_NOIO instead of (GFP_NOIO | __GFP_HIGH), which can lead to a deadlock under memory pressure.
CVE-2026-74453 1 Linux 1 Linux Kernel 2026-08-17 7.8 High
In the Linux kernel, the following vulnerability has been resolved: drm/vc4: Zero the tile state data array before each BIN job The binner BO is a single 16MB buffer split into 512KB slots that are handed out to jobs at submission time and recycled as jobs complete, without ever being cleared. Each slot holds the job's Tile State Data Array (TSDA) at its start, followed by the tile allocation pool. While the tile allocation pool is only walked by the render thread through branches the binner generated during the current job, the TSDA is the PTB's own per-tile bookkeeping and is consumed by the hardware itself. Although the kernel sets the "Auto-initialise Tile State Data Array" flag in the tile binning mode configuration, the PTB demonstrably still acts on stale tile state left by the slot's previous user: the binner ends up creating invalid command streams with invalid primitive streams and branches, which can cause GPU hangs as observed in [1][2]. Zero the TSDA when the job's binning slot is configured. This clears 48 bytes per tile (~24KB for a 1080p frame) in the submission path, and guarantees the PTB never sees another job's tile state. The tile count is only checked for being non-zero today, so the 8-bit fields it comes from can describe a tile state array almost six times larger than the slot it has to live in. Bound it before the slot is handed out, since such size decides how much of the slot is left for the tile alloc pool.
CVE-2026-74575 1 Linux 1 Linux Kernel 2026-08-17 8.8 High
In the Linux kernel, the following vulnerability has been resolved: thunderbolt: Prevent XDomain delayed work use-after-free on disconnect tb_xdp_handle_request() runs on system_wq and queues xd->state_work via queue_delayed_work() in three request handlers: PROPERTIES_CHANGED_REQUEST, UUID_REQUEST (via start_handshake), and LINK_STATE_CHANGE_REQUEST. Similarly, update_xdomain() queues xd->properties_changed_work when local properties change. Concurrently, tb_xdomain_remove() calls stop_handshake() which does cancel_delayed_work_sync() on both delayed works. Later, tb_xdomain_unregister() calls device_unregister() which eventually frees the xdomain. Since commit 559c1e1e0134 ("thunderbolt: Run tb_xdp_handle_request() in system workqueue") moved the request handler off tb->wq, the handler and the remove path are no longer serialized. If queue_delayed_work() executes after cancel_delayed_work_sync() but before the xdomain is freed, the delayed work fires on a freed object. Add xd->removing that tb_xdomain_remove() sets under xd->lock before calling stop_handshake(). Each external queue site holds the same lock and checks removing before calling queue_delayed_work(). This provides the mutual exclusion needed: either the queue site acquires the lock first and queues work that the subsequent cancel will see, or the remove path acquires the lock first and the queue site observes removing == true and skips the queue.
CVE-2026-74440 1 Linux 1 Linux Kernel 2026-08-17 7.8 High
In the Linux kernel, the following vulnerability has been resolved: drm/xe: Wait on external BO kernel fences in exec IOCTL Before arming a user job, xe_exec_ioctl() only added the VM's dma-resv KERNEL slot as a dependency. That slot covers rebinds and the kernel operations of the VM's private BOs, but not external BOs (bo->vm == NULL), which carry their kernel operations (evictions, moves, ...) in their own dma-resv KERNEL slot. The DMA_RESV_USAGE_KERNEL slot is the cross-driver contract for memory management operations that must complete before the BO or its backing store may be used: any accessor is required to wait on the KERNEL fences before touching the resv. By skipping the external BOs' KERNEL slots, the exec path violated that contract and could schedule a user job while a kernel operation on an external BO mapped by the VM was still in flight, racing against it and potentially reading or writing memory that was being moved. Replace the VM-only dependency with an iteration over every object locked by the exec, adding each object's KERNEL slot as a job dependency. This covers the VM resv (rebinds and private BOs) as well as every external BO, mirroring the drm_gpuvm_resv_add_fence() call that later publishes the job fence to the same set of objects. Long-running mode continues to skip this, as before. (cherry picked from commit a6b842acf3ddd1efc53a56de9260cfa718fb35e7)
CVE-2026-74446 1 Linux 1 Linux Kernel 2026-08-17 7.8 High
In the Linux kernel, the following vulnerability has been resolved: drm/amdkfd: hold event_mutex while checkpointing CRIU events kfd_criu_checkpoint_events() counts the entries in p->event_idr via kfd_get_num_events(), allocates an array sized to that count, and then walks the same IDR to fill it. Neither the count nor the walk holds p->event_mutex. The CRIU checkpoint caller holds only p->mutex. Event create and destroy (kfd_event_create()/kfd_event_destroy()) take p->event_mutex and do not take p->mutex, so a second thread in the same process can insert or remove events between the count and the walk. If an event is inserted, the walk iterates more entries than were counted and writes past the end of the ev_privs allocation; if an event is removed, the walk dereferences an entry that is being freed. Hold p->event_mutex across the count and the walk so both observe a consistent view of p->event_idr. The lock is released before copy_to_user(), which only touches the local buffer. The caller already holds p->mutex and the create/destroy paths never take p->mutex, so the p->mutex -> p->event_mutex order is not inverted and no deadlock is introduced. (cherry picked from commit ff57e223ab105795b05d3ef3f3c35a5a441bcbaa)