CVE-20XX-?
agent?

CVEs of 2026.

Every CVE published in 2026 that we hold, ranked by exploit odds.

3,319 CVEs · updated 2026-11-20 20:55 UTC

Reset all
  1. CVE-2026-99776

    In the Linux kernel, the following vulnerability has been resolved: net/9p/usbg: clear stale client pointer on close p9_usbg_close() tears down the client transport, but usb9pfs keeps using usb9pfs->client from asynchronous TX and RX completion handlers. A late completion can therefore dereference a client that has already been freed during mount teardown. Clear usb9pfs->client under usb9pfs->lock when closing the transport, detach any pending TX request from in_req->context, and make the TX/RX completion handlers bail out once the transport has been detached. This keeps late completions from touching a freed or rebound p9_client.

    Memory safety OS & kernel Linux

    unscored No fix listed
  2. CVE-2026-99778

    In the Linux kernel, the following vulnerability has been resolved: net/9p/usbg: fix out_req buffer leak in disable_usb9pfs usb9pfs->out_req is allocated with alloc_ep_req(), which also allocates the request's data buffer (req->buf). disable_usb9pfs() frees it with plain usb_ep_free_request(), which only frees the usb_request structure itself and leaks req->buf on every close. Use free_ep_req() instead, matching the allocator, as documented by its own comment ("Requests allocated via alloc_ep_req() *must* be freed by free_ep_req()").

    Memory safety OS & kernel Linux

    unscored No fix listed
  3. CVE-2026-99765

    In the Linux kernel, the following vulnerability has been resolved: scsi: target: Fix demo-mode dynamic ACL dev-entry UAF vs LUN teardown A demo-mode (generate_node_acls=1) login that allocates a dynamic NodeACL populates that ACL's per-LUN dev entries in core_tpg_add_node_to_devs() *before* the ACL is linked onto tpg->acl_node_list by target_add_node_acl(): core_tpg_check_initiator_node_acl(): ... core_tpg_add_node_to_devs(acl, tpg, NULL); /* builds deves */ target_add_node_acl(acl); /* only now listed */ The LUN teardown sweep, core_clear_lun_from_tpg(), enumerates only tpg->acl_node_list, so a dev entry whose ACL is still in the window above is invisible to the sweep and is never disabled. It is left on the se_lun->lun_deve_list after core_tpg_remove_lun() frees the se_lun; when the same unpacked_lun is later re-added and the se_lun is reallocated at the same address, the stale dev entry is rediscovered by target_nacl_find_deve() and core_enable_device_list_for_node() corrupts lun_deve_list: BUG: KASAN: slab-use-after-free ... core_enable_device_list_for_node+0x5ce/0x880 core_tpg_add_node_to_devs+0x169/0x190 core_dev_add_lun+0x146/0x190 target_fabric_port_link+0xe9/0x1b0 configfs_symlink+0x381/0xae0 __x64_sys_symlink+0x55/0x80 The demo-mode login path that drives the ACL/dev-entry creation is reachable by an unauthenticated initiator; the colliding LUN add/remove is a local configfs operation. Commit 49cb77e297dc ("target: Avoid mappedlun symlink creation during lun shutdown") introduced se_lun->lun_shutdown and guarded the *explicit* ACL MappedLUN path (target_fabric_mappedlun_link()), but the demo-mode automatic path, core_tpg_add_node_to_devs() -> core_enable_device_list_for_node(), was never given the equivalent guard, and the publish-after-populate ordering in core_tpg_check_initiator_node_acl() was left unchanged. Fix it in three cooperating parts: 1) Publish the dynamic NodeACL onto acl_node_list *before* creating its dev entries, so the teardown sweep can observe them. 2) Refuse to attach a dev entry to an se_lun that is shutting down, checked under nacl->lun_entry_mutex -- the same lock the sweep takes -- so once the sweep has passed an ACL any later enable for that LUN observes lun_shutdown and skips it. 3) In core_enable_device_list_for_node(), insert the new dev entry onto lun->lun_deve_list while still holding nacl->lun_entry_mutex, in both the replace and the new-entry branch. Previously the entry was added to nacl->lun_entry_hlist under the mutex but to lun->lun_deve_list only after dropping it; with the ACL now published early, the sweep could take the mutex in that gap, find the entry on the hlist, core_disable_device_list_for_node() it (list_del() on the not-yet- linked lun_link succeeds and the entry is freed), and then the enable would list_add_tail() the freed entry back onto lun_deve_list. Linking both lists under the mutex makes enable and the sweep mutually exclusive, closing that window. core_tpg_add_node_to_devs() already ignores the return value of core_enable_device_list_for_node(), so a shutting-down LUN is simply skipped.

    Memory safety OS & kernel Linux

    unscored No fix listed
  4. CVE-2026-99767

    In the Linux kernel, the following vulnerability has been resolved: xfrm4: use a full socket for local errors xfrm4_local_error() passes skb->sk to inet_sk() and ip_local_error() without ensuring it points to a full socket. TCP SYN-ACK skbs can be owned by TCP_NEW_SYN_RECV request sockets. xfrm4_tunnel_check_size() already checks sk_fullsock() since commit 1620c88887b1 ("xfrm: Fix the usage of skb->sk"), so iptfs_is_too_big() is the only remaining IPv4 caller passing a request socket here: the sk it tests is the listener, while skb->sk is the request socket. When IP-TFS rejects such an skb because it exceeds its payload MTU, the local error path treats the request socket as a struct inet_sock. KASAN reports a slab-out-of-bounds read in ip_local_error() beyond the request_sock_TCP allocation. Resolve the skb owner with skb_to_full_sk() before accessing inet_sock state or reporting the error. This maps a request socket to its listener and rejects sockets that cannot be used as full sockets, matching what commit 6973a21ee73c ("ipv6: xfrm: use full sockets in local error paths") did for xfrm6_local_error().

    Memory safety OS & kernel Linux

    unscored No fix listed
  5. CVE-2026-99771

    In the Linux kernel, the following vulnerability has been resolved: wifi: mac80211: reload TDLS header pointer after skb_linearize() ieee80211_process_tdls_channel_switch() sets tf = skb->data before it calls skb_linearize(), and reads tf->action_code afterwards. ieee80211_rx_h_data() queues TDLS channel switch requests and responses to the interface after pulling only offsetof(struct ieee80211_tdls_data, u) bytes, so the skb can still be non-linear when the interface work handles it. Drivers that build paged RX skbs do this for every frame that does not fit into the skb head, e.g. iwlwifi (mvm), which is also the only mac80211 driver that sets NL80211_FEATURE_TDLS_CHANNEL_SWITCH. If the remaining data does not fit into the tailroom, skb_linearize() moves the data to a new head and frees the old one, and the action code is then read from freed memory. Depending on what that memory holds by then, the frame is handled as before or ends in the WARN_ON_ONCE() of the default case. The frame only has to be a data frame with the TDLS ethertype from an authorized peer, e.g. relayed by the AP from another station in the BSS; the read happens before the sender is checked to be a TDLS peer. Load tf after skb_linearize(), as the request and response handlers already do. Found by static analysis tool CodeQL.

    Memory safety OS & kernel Linux

    unscored No fix listed
  6. CVE-2026-99752

    In the Linux kernel, the following vulnerability has been resolved: drm/loongson: Handle buffer mapping failures when clearing a BO lsdc_bo_clear() ignores the return value from lsdc_bo_kmap() and writes to lbo->kptr unconditionally. When the first mapping of a newly allocated buffer fails, kptr is still NULL and the subsequent memset accesses it. Return the mapping error without touching the buffer and propagate it from lsdc_gem_object_create(). Install the GEM object functions before clearing so that drm_gem_object_put() can release the buffer correctly on failure. Do not add the failed object to the tracking list or return an uncleared buffer to the caller. Imported buffers still skip clearing, and successfully mapped buffers are cleared and unmapped as before. Tested on LS7A2000 with a one-shot range error in ttm_bo_kmap(). Its -EINVAL return reached userspace, the object was destroyed once, and no GEM handle was published. Normal create/map/zero/write/read/close cycles passed, with the tracked BO count and VRAM usage unchanged. AI assistance was used for the lifetime analysis, fix, fault-injection tools, build and test execution.

    Memory safety OS & kernel Linux

    unscored No fix listed
  7. CVE-2026-99753

    In the Linux kernel, the following vulnerability has been resolved: drm/loongson: Fix double free on BO initialization failure If ttm_bo_init_validate() fails, it drops the buffer object's reference and cleans it up through the supplied destroy callback. lsdc_bo_destroy() releases the GEM object and frees the enclosing lsdc_bo, so freeing it again in lsdc_bo_create() causes a double free on a synchronous failure path. Let TTM own the cleanup after initialization has started and return the error directly. Keep the explicit free on drm_gem_object_init() failure, which occurs before ownership is passed to TTM. Tested on LS7A2000 by making drm_vma_offset_add() return -ENOSPC for one selected dumb-buffer creation. The error reached userspace, the destroy callback ran once, and no handle was published or tracked BO retained. Normal buffer creation, zeroing, mapping, readback and release passed. AI assistance was used for the ownership analysis, fix, fault-injection tools, build and test execution.

    Memory safety OS & kernel Linux

    unscored No fix listed
  8. CVE-2026-99754

    In the Linux kernel, the following vulnerability has been resolved: scsi: target: Verify the iSCSI ISID is null-terminated iscsi_parse_pr_out_transport_id() returns the ISID of a FORMAT CODE 01b TransportID as a pointer into the PERSISTENT RESERVE OUT parameter list, lowercased in place and never null-terminated. spc4r37 Table 508 requires the field to be null-terminated, but a non-conforming initiator can omit the NULL; when the ISID ends at the end of the received buffer, the pointer is not a C string. The registration code still consumes it as one: strcmp() in __core_scsi3_locate_pr_reg() and in the REGISTER AND MOVE path, and snprintf("%s") in __core_scsi3_do_alloc_registration(), which also reads its first 8 bytes unconditionally with get_unaligned_be64() - a NUL check does not bound that read. When the parameter list spans more than one page, transport_kmap_data_sg() maps it with vmap(), and the over-read walks off the end of the mapping into the vmalloc() guard page and oopses. Reproduced with a PR OUT REGISTER + SPEC_I_PT parameter list of exactly 8192 bytes whose final TransportID carries a 12-byte ISID without a terminating NUL: BUG: unable to handle page fault for address: ffffc9000138c000 RIP: 0010:strcmp+0xc/0x30 Call Trace: __core_scsi3_locate_pr_reg+0xd3/0x120 [target_core_mod] core_scsi3_decode_spec_i_port+0x8fc/0x13a0 [target_core_mod] core_scsi3_emulate_pro_register+0x73c/0x900 [target_core_mod] ... iscsi_target_rx_thread+0xef/0x1d0 [iscsi_target_mod] With a single-page parameter list the buffer is a kmap() direct mapping, so the over-read continues into the physically adjacent page instead of faulting. snprintf("%s") then copies whatever follows the ISID in memory into pr_reg_isid, and PR IN READ FULL STATUS reports the registration back to any initiator, turning the over-read into an info leak of kernel heap contents. Reproduced with a 4096-byte parameter list whose last TransportID ends at the end of the buffer with a 4-byte ISID and no NUL, interleaved with I/O to populate adjacent pages: the READ FULL STATUS response returned the ISID as "0xleak" followed by raw bytes that were not part of the parameter list. The snprintf() into pr_reg_isid[PR_REG_ISID_LEN] caps the echoed ISID at 15 characters, so at most 11 bytes past the buffer are leaked per command, but the PR OUT command is not rate limited. Any initiator that can send PR OUT to an iSCSI-exported LUN can trigger either variant, including a guest via vhost-scsi. Require 8 bytes to remain after the ",i,0x" separator, and a NUL inside the rest of the descriptor, before the ISID pointer is returned. REGISTER AND MOVE fails with TCM_INVALID_PARAMETER_LIST; on SPEC_I_PT no target port claims the malformed TransportID, and the command fails the same way.

    Memory safety OS & kernel Linux

    unscored No fix listed
  9. CVE-2026-99755

    In the Linux kernel, the following vulnerability has been resolved: ptp: fix use-after-free in extts_fifo_show() reading tsevqs unlocked extts_fifo_show() walks ptp->tsevqs with list_count_nodes() and then list_first_entry() without holding ptp->tsevqs_lock, the spinlock that "protects tsevqs from concurrent access". Every other accessor of the list honours it: ptp_open() (list_add_tail), ptp_release() (list_del followed by kfree of the per-fd queue), ptp_clock_release() and ptp_clock_event() all take tsevqs_lock. extts_fifo_show() is the only one that does not. A per-fd timestamp queue is appended to tsevqs on open("/dev/ptpN") and removed and freed on close(), both under tsevqs_lock: scoped_guard(spinlock_irq, &ptp->tsevqs_lock) list_del(&queue->qlist); bitmap_free(queue->mask); kfree(queue); So a read of /sys/class/ptp/ptpN/fifo running concurrently with a close() of a /dev/ptpN fd can have list_count_nodes() walk into the queue that close() is removing and freeing, dereferencing freed memory (use-after-free) or a poisoned list pointer. Take tsevqs_lock around the list walk. The fifo only ever draws from the first queue, which is the registration-time queue created in ptp_clock_register() and freed only in ptp_clock_release(); that cannot run while the sysfs attribute is being read, so the dequeue from queue->buf under queue->lock may stay outside tsevqs_lock and the lock order is unchanged.

    Memory safety OS & kernel Linux

    unscored No fix listed
  10. CVE-2026-99724

    In the Linux kernel, the following vulnerability has been resolved: ntfs: don't allow compression on volumes with clusters larger than 4K On a volume made with "mkntfs -c 8192", a user can set FILE_ATTR_COMPRESSED on an empty file through system.ntfs_attrib. Writing to the file and reading it back after dropping the page cache then crashes: BUG: KASAN: slab-out-of-bounds in ntfs_read_compressed_block+0xa2c/0x2040 Read of size 8192 at addr ffff88804355a000 by task dd/106 The read path only handles clusters up to 4K. It copies each cluster out of a single page, and a compression block of 16 8K clusters does not fit in the 64K ntfs_compression_buffer. Loading an inode already rejects compressed data on such volumes. Reject the flag change the same way.

    Memory safety OS & kernel Linux

    unscored No fix listed
  11. CVE-2026-99725

    In the Linux kernel, the following vulnerability has been resolved: cipso: fix optlen corruption in cipso_v4_delopt() cipso_v4_delopt() recomputes the option length by walking the option buffer after the CIPSO option has been moved out. The walk then reads the stale bytes at the tail of the buffer as option data, and a crafted setsockopt(IP_OPTIONS) buffer can make it treat one of them as an option length, increasing opt->opt.optlen beyond the valid option data. The corrupted length stays on the socket and is later used as the copy length by getsockopt(IP_OPTIONS) and by the udp/raw/ping sendmsg paths, where it exceeds the size of the buffers they copy into. Compute the new length from the intact options before the move and subtract the CIPSO length, and fill the padding with IPOPT_END.

    Memory safety OS & kernel Linux

    unscored No fix listed
  12. CVE-2026-99717

    In the Linux kernel, the following vulnerability has been resolved: hwmon: (pmbus/q54sj108a2) Add missing NULL pointer check When CONFIG_DEBUG_FS is enabled and per-client PMBus debugfs directory creation fails during pmbus_do_probe(), q54sj108a2_probe() passes the NULL pointer returned by pmbus_get_debugfs_dir() to debugfs_create_dir() without validation, creating its debugfs directory at the debugfs root instead of under the client directory. If pmbus_init_debugfs() fails to create the client directory, data->debugfs is left as NULL or an ERR_PTR, which pmbus_get_debugfs_dir() converts to NULL. Passing a NULL parent to debugfs_create_dir() places q54sj108a2_dir in the root of the debugfs filesystem and populates it with attribute files whose private data pointers reference psu->debugfs_entries. Because this root-level directory is outside the client debugfs hierarchy, PMBus core cleanup does not remove it on device unbind, leaking the debugfs entries and causing a use-after-free of the devm-allocated struct q54sj108a2_data if the orphaned files are accessed via q54sj108a2_debugfs_read() or q54sj108a2_debugfs_write() after device removal. Fix the problem by adding the missing NULL pointer check.

    Memory safety OS & kernel Linux

    unscored No fix listed
  13. CVE-2026-99718

    In the Linux kernel, the following vulnerability has been resolved: hwmon: (peci/dimmtemp) Fix devres leak and UAF on concurrent unbind When a peci_dimmtemp device is unbound concurrently with its delayed initialization work (detect_work), newly allocated devm resources and the hwmon device are leaked, leading to a use-after-free when reading the lingering hwmon sysfs attributes. In peci_dimmtemp_probe(), detect_work is registered for cancellation via devm_delayed_work_autocancel(). During driver detach, devres_release_all() first moves all existing entries from dev->devres_head onto a local todo list before invoking their release callbacks in reverse order. If create_dimm_temp_info_delayed() is already running when devm_delayed_work_drop() calls cancel_delayed_work_sync(), create_dimm_temp_info() can allocate dimmtemp_label strings via devm_kzalloc()/devm_kasprintf() and register the hwmon device via devm_hwmon_device_register_with_info(), appending new devres entries to the now-empty dev->devres_head. Once the work finishes, devres_release_all() frees priv and priv->name from the local todo list without re-evaluating dev->devres_head, leaving the late devres allocations and registered hwmon device behind with a dangling priv pointer. Cancel detect_work synchronously in a driver .remove() callback instead of relying on devm_delayed_work_autocancel() so that the work is fully drained before devres_release_all() detaches dev->devres_head, allowing any devm resources registered by the delayed work to be released in proper reverse order. Also only schedule detect_work when check_populated_dimms() returns -EAGAIN so that probe failure paths do not leave delayed work pending.

    Memory safety OS & kernel Linux

    unscored No fix listed
  14. CVE-2026-99721

    In the Linux kernel, the following vulnerability has been resolved: hwmon: (aquacomputer_d5next) reject short status reports The HID raw_event callback runs before HID core validates the received report length. aqc_raw_event() reads device-specific status fields at fixed offsets, including multi-byte values, without first ensuring that the received buffer contains every field. A truncated report can therefore cause out-of-bounds reads and update hwmon state with unrelated data. The required length differs between supported devices and cannot be derived from the HID report descriptor. Define the minimum length needed by each asynchronous status format, select it while configuring the device, and reject reports that do not contain the last field accessed by that format. Keep the published length zero until probe configuration is complete, which also excludes legacy devices whose status is retrieved through feature reports.

    Memory safety OS & kernel Linux

    unscored No fix listed
  15. CVE-2026-99714

    In the Linux kernel, the following vulnerability has been resolved: nvme: Do not drop the cdev reference twice when cdev_device_add() fails Since commit 9952f3882ecb ("nvme: fix cdev lifetime"), nvme_add_ns_cdev() and nvme_add_ns_head_cdev() take a reference to the namespace (or namespace head) before calling nvme_cdev_add(), to be dropped by the char device's release, nvme_cdev_rel(), and drop it themselves when nvme_cdev_add() fails. But when cdev_device_add() fails, nvme_cdev_add() has already put the device, and nvme_cdev_rel() has already dropped that reference: the caller drops it a second time. With one allocation failure injected into cdev_device_add() for a namespace on an NVMe over loop controller, deleting the controller gives, with native multipath: BUG: KASAN: slab-use-after-free in nvme_put_ns_head+0x21/0x140 Write of size 4 at addr ff1100000a4ea078 by task init/1 nvme_put_ns_head nvme_put_ns nvme_remove_namespaces nvme_do_delete_ctrl nvme_delete_ctrl_sync nvme_sysfs_delete Freed by task 1: kfree nvme_ns_remove nvme_remove_namespaces refcount_t: underflow; use-after-free. Take the reference in nvme_cdev_add() instead, right before cdev_device_add(), and let the callers neither take nor drop it. Add nvme_is_ns_cdev() for the check that nvme_cdev_rel() and nvme_cdev_add() now share.

    Memory safety OS & kernel Linux

    unscored No fix listed
  16. CVE-2026-99705

    In the Linux kernel, the following vulnerability has been resolved: mm: memory: fix truncation removing mapped folios Ayush reported an issue that shmem/tmpfs folios are still mapped when truncating them on production: " BUG: Bad page cache in process ... pfn:... page dumped because: still mapped when deleted " I reproduced the issue with Ayush's reproducer [1] with the latest mm-unstable branch, and I got the following crash info. From the dump message, we can see that truncate_inode_folio() is really trying to remove mapped folios, which is incorrect. " [ 189.171977] page: refcount:3 mapcount:1 mapping:00000000fe168ee1 index:0x3880 pfn:0x191647 [ 189.171995] memcg:ffff0000cc073d40 [ 189.171997] aops:shmem_aops ino:3c01 dentry name(?):"memfd:runsc-memory" [ 189.172006] flags: 0x17fffef0002022d(locked|referenced|uptodate|lru| workingset|swapbacked|node=0|zone=2|lastcpupid=0x3ffff) ....... [ 189.172017] page dumped because: VM_BUG_ON_FOLIO(folio_mapped(folio)) [ 189.172026] ------------[ cut here ]------------ [ 189.172027] kernel BUG at mm/filemap.c:155! [ 189.172057] Internal error: Oops - BUG: 00000000f2000800 [#1] SMP ....... [ 189.177721] CPU: 3 UID: 0 PID: 5521 Comm: repro Kdump: loaded Tainted: 7.3.0-rc4+ #277 PREEMPT(lazy) ....... [ 189.183986] Call trace: [ 189.184124] filemap_unaccount_folio+0xf0/0x1e8 (P) [ 189.184391] __filemap_remove_folio+0x34/0x160 [ 189.184633] filemap_remove_folio+0x4c/0xb0 [ 189.184859] truncate_inode_folio+0x34/0x58 [ 189.185087] shmem_undo_range+0x220/0x658 [ 189.185308] shmem_fallocate+0x2f0/0x470 [ 189.185527] vfs_fallocate+0x128/0x328 ...... [ 189.187078] el0t_64_sync+0x184/0x188 " After analysis, I believe the race exists between truncation and MADV_DONTNEED, and shmem's ->map_pages() merely makes the issue easier to reproduce. Since MADV_DONTNEED synchronously releases the pagetable page before calling tlb_flush_rmaps(), this could cause another thread's truncation to skip zap_pte_range() (due to pmd is none) but still observe the folio's mapcount as non-zero. A possible race scenario is as follows: CPU 0 CPU 1 shmem_fallocate madvise_dontneed_single_vma unmap_mapping_range ....... ...... (filemap_map_pages() remap folios) zap_pte_range truncate_inode_folio zap_empty_pte_table(pmd clear) unmap_mapping_folio ...... zap_pmd_range (saw pmd none, skip) filemap_remove_folio BUG_ON(folio_mapped) tlb_flush_rmaps(remove rmaps) To fix this issue, we should move the pmd clear operation (via zap_empty_pte_table()) to after tlb_flush_rmaps() and use sync_with_folio_pmd_zap() to sync with concurrent zapping, so that when we find a pmd_none() while unmapping a folio without holding the PTL, the folio's mappings are guaranteed to have been removed via tlb_flush_rmaps(). Meanwhile move the second pmd_none() check to where it belongs: if splitting zapped the PMD.

    Memory safety OS & kernel Linux

    unscored No fix listed
  17. CVE-2026-99709

    In the Linux kernel, the following vulnerability has been resolved: inet: frags: limit the truesize of a reassembly queue The truesize of fragments in a reassembly queue is only bounded by the per netns ipfrag_high_thresh sysctl, which has no upper limit. inet_frag_reasm_finish() sums the truesize of all fragments in an int. If this sum reaches 2^32, it wraps, and the final sk_wmem_alloc adjustment of the socket owning the reassembled skb can cancel its own reference, leading to a use-after-free of this socket. This can be triggered from an unprivileged user/network namespace, raising ipfrag_high_thresh and using padded TUN packets, which keep a large truesize after being trimmed by ip_rcv_core(). Track the sum of the truesize of fragments in each queue, and refuse a fragment that would make it bigger than INT_MAX / 2. Callers kill the queue on this error. This bound is far above the needs of legitimate traffic.

    Memory safety OS & kernel Linux

    unscored No fix listed
  18. CVE-2026-99697

    In the Linux kernel, the following vulnerability has been resolved: mm: memcontrol: fix NULL pointer deref and UAF from high_work_func() on dying memcg When memory.high is exceeded in non-task context, try_charge_memcg() schedules memcg->high_work without taking a reference on the memcg, and high_work_func() invokes reclaim_high() without pinning the memcg. With obj_cgroup reparenting, introduced in commit bf4f059954dc ("mm: memcg/slab: obj_cgroup API"), objects charged to a child memcg are reparented to its parent in mem_cgroup_css_offline() instead of holding a reference on the child until uncharged. Subsequent uncharges of those objects go to the parent, which can leave the child's memory use permanently above its high limit, so any later charge on the offlined memcg in non-task context schedules high_work. This was later widened from kernel memory to LRU folios when commit f1cf8d2f36dc ("mm: memcontrol: eliminate the problem of dying memory cgroup for LRU folios") converted LRU folios to use obj_cgroup reparenting. cancel_work_sync(&memcg->high_work) only runs at the end of teardown in mem_cgroup_css_free(), so high_work_func() can start or still be running reclaim_high() after mem_cgroup_css_released() or shrink_many() has called lru_gen_release_memcg() and unhashed lruvec->lrugen.list. If reclaim tries to rotate the memcg, lru_gen_rotate_memcg() dereferences a NULL pprev pointer. This issue is specific to MGLRU and was introduced with the memcg LRU in commit e4dde56cd208 ("mm: multi-gen LRU: per-node lru_gen_folio lists"). With either traditional LRU or MGLRU, css_free_rwork_fn() calls css_rstat_exit() which frees per-CPU rstat data before calling mem_cgroup_css_free(). If reclaim_high() is still running, rstat updates via mod_lruvec_state() -> __css_rstat_updated() cause a use-after-free on the per-CPU rstat data. This was introduced when commit 5da3bfa029d6 ("cgroup: use separate rstat trees for each subsystem") moved css_rstat_exit() ahead of ss->css_free(). Fix both problems by acquiring a reference with css_tryget_online() in high_work_func() before calling reclaim_high(), and by resetting memory.high and swap.high to PAGE_COUNTER_MAX in mem_cgroup_css_offline() so that charges via references on an offlined memcg do not trigger high_work on the dead child and skip checking live ancestors in try_charge_memcg(). Acquiring the reference in high_work_func() rather than before scheduling high_work is safe because mem_cgroup_css_free() calls irq_work_sync() and cancel_work_sync() which cancels pending work and waits for an already running high_work_func() to complete before mem_cgroup_free() frees the memcg. css_tryget_online() uses percpu_ref_tryget_live() and is safe to call after percpu_ref_exit() before the cgroup is freed unlike css_tryget(). Oops: general protection fault, probably for non-canonical address 0xdffffc0000000000: 0000 [#1] SMP KASAN NOPTI KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007] CPU: 1 UID: 0 PID: 27 Comm: kworker/1:0 Tainted: G B W 7.3.0-rc5-g2bba28e98578 #284 PREEMPT(full) Tainted: [B]=BAD_PAGE, [W]=WARN <-- Taint from lru_gen_rotate_memcg() VM_WARN_ON_ONCE Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.17.0-debian-1.17.0-1 04/01/2014 Workqueue: events high_work_func RIP: 0010:lru_gen_rotate_memcg+0x1a3/0xe90 Call Trace: <TASK> ? read_ctrl_pos+0x22f/0x3f0 shrink_lruvec+0x16cb/0x1b80 ? ret_from_fork_asm+0x1a/0x30 ? lapic_next_event+0x15/0x20 ? clockevents_program_event+0x2bb/0x6f0 ? lapic_next_event+0x15/0x20 ? __pfx_shrink_lruvec+0x10/0x10 ? __pfx_clockevents_program_event+0x10/0x10 ? _raw_spin_lock+0x84/0xe0 ? __pfx__raw_spin_lock+0x10/0x10 ? kasan_save_stack+0x30/0x50 ? psi_group_change+0x381/0x740 ? __pfx_sched_clock_cpu+0x10/0x10 ? save_fpregs_to_fpstate+0xc4/0x2c0 ? finish_task_switch.isra.0+0x1de/0xac0 ? __schedule+0xf62/0x42c0 ? __pfx___schedule+0x10/0x10 ? __pfx___irq_work_queue_lo ---truncated---

    Memory safety OS & kernel Linux

    unscored No fix listed
  19. CVE-2026-99691

    In the Linux kernel, the following vulnerability has been resolved: ocfs2/cluster: unlink socket containers before releasing resources A socket container's node and socket must remain alive while the container is on sock_containers. The debugfs formatter reads sc_node and sc_sock under o2net_debug_lock without taking their references, but sc_kref_release() releases both before taking that lock to unlink the container. With CONFIG_DEBUG_FS enabled, a synchronous sender can retain its last socket-container reference after o2net_disconnect_node() has detached the connection and flushed o2net_wq. The caller runs outside that workqueue. After the sender removes its configfs dependency, node removal can proceed without waiting for the sender's final sc_put(): Sender (final sc_put) Configfs rmdir Debugfs reader -------------------- -------------- -------------- o2nm_undepend_item() Drop node item refs o2nm_node_put() [node can be freed] sc_seq_show() Lock debug lock Read sc_node->nd_name Unlock debug lock sc_node = NULL o2net_debug_del_sc() The earlier socket release has a similar window. The container is still selectable while its socket has been released and sc_sock has not yet been cleared: Sender (final sc_put) Debugfs reader -------------------- -------------- sock_release(sc->sc_sock) [socket can be freed] sc_seq_show() Lock debug lock Read sc_sock->sk fields Unlock debug lock sc_sock = NULL ... o2net_debug_del_sc() Move o2net_debug_del_sc() before releasing either resource. Unlinking under the existing debug lock waits for an active formatter to finish and prevents later readers from selecting the container. This covers both socket fields and node fields, including nd_num in the stats formatter. KASAN report as below: BUG: KASAN: slab-use-after-free in string+0x42b/0x470 Read of size 1 at addr ffff88810f56fc90 by task python3/506 CPU: 0 UID: 0 PID: 506 Comm: python3 Not tainted 7.3.0-rc4-next-20260921-pmb-ocfs2-functional-v1+ #1 PREEMPT(lazy) Hardware name: QEMU Ubuntu 24.04 PC v2 (i440FX + PIIX, arch_caps fix, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 Call Trace: <TASK> dump_stack_lvl+0x93/0xd0 print_report+0xce/0x630 kasan_report+0xe0/0x110 string+0x42b/0x470 vsnprintf+0x330/0x1110 seq_printf+0x1a8/0x260 sc_seq_show+0x604/0x9f0 seq_read_iter+0x480/0x1110 seq_read+0x362/0x5a0 full_proxy_read+0x11f/0x180 vfs_read+0x1a0/0xc20 ksys_read+0x111/0x200 do_syscall_64+0x114/0x620 entry_SYSCALL_64_after_hwframe+0x77/0x7f (User register dump omitted.) </TASK> Allocated by task 497: kasan_save_stack+0x33/0x60 kasan_save_track+0x14/0x30 __kasan_kmalloc+0xaa/0xb0 __kmalloc_cache_noprof+0x28d/0x610 o2nm_node_group_make_item+0xea/0x1a0 configfs_mkdir+0x762/0xca0 vfs_mkdir+0x399/0x7e0 filename_mkdirat+0x129/0x450 __x64_sys_mkdir+0x70/0xa0 do_syscall_64+0x114/0x620 entry_SYSCALL_64_after_hwframe+0x77/0x7f Freed by task 505: kasan_save_stack+0x33/0x60 kasan_save_track+0x14/0x30 kasan_save_free_info+0x3b/0x60 __kasan_slab_free+0x5f/0x80 kfree+0x308/0x580 config_item_cleanup+0x14b/0x230 config_item_put+0x79/0x90 sc_put+0x454/0x6e0 o2net_send_message_vec+0x63d/0x1560 o2net_send_message+0xa6/0xe0 validation_send_write+0x17f/0x1d0 full_proxy_write+0x11f/0x180 vfs_write+0x25a/0x1010 ksys_write+ ---truncated---

    Memory safety OS & kernel Linux

    unscored No fix listed
  20. CVE-2026-99694

    In the Linux kernel, the following vulnerability has been resolved: ocfs2: do not index before the entry array when dividing an empty bucket ocfs2_divide_xattr_bucket() appends an empty bucket carrying a sentinel hash when the source bucket has no divide position, taking that hash from xh_entries[start - 1]. ocfs2_xattr_find_divide_pos() returns count when every entry shares one hash, and its loop does not run at all for count < 2, so an empty bucket comes back with start == count == 0 and the read is xh_entries[-1]. That is outside the __counted_by_le(xh_count) bounds of xh_entries[] and lands on the header, so the value picked up is xh_count | xh_free_start << 16. One above it is written into the new bucket's xh_entries[0] and committed, which leaves a bucket whose first hash does not follow the bucket before it. ocfs2_xattr_bucket_find() binary searches on xh_entries[0], so the broken key hides the range the sentinel was meant to cover: getxattr() answers -ENODATA for an attribute that is stored, and an insert can go to the wrong bucket and leave two copies of a name. The cross cluster divide hands the same value back as a tree record's e_cpos, which then claims a hash range it does not cover. Read xh_entries[0] instead. The branch is only reached when every entry shares one hash, so for a non-empty bucket it holds the same value, and for an empty one it is the slot that carries the bucket key. An empty bucket reaches the divide from ocfs2_check_xattr_bucket_collision(), which reports no collision for one, through ocfs2_add_new_xattr_bucket() and ocfs2_extend_xattr_bucket(). It does not take a failed set to get there: a divide that finds no position is itself what leaves an empty sentinel bucket behind, and a later extend can pick it as the target.

    Memory safety OS & kernel Linux

    unscored No fix listed
  21. CVE-2026-99684

    In the Linux kernel, the following vulnerability has been resolved: hwrng: virtio - prevent use-after-free after a failed restore remove_common() frees the driver data during freeze, but does not clear vdev->priv, which points to freed memory. If probe_common() then fails during restore, a later remove or freeze can dereference the dangling pointer. Clear vdev->priv on the probe_common() error cleanup path and before freeing the driver data in remove_common(). Also reset the device and return early from remove_common() if vdev->priv is already NULL for a safe teardown after a failed restore.

    Memory safety OS & kernel Linux

    unscored No fix listed
  22. CVE-2026-99688

    In the Linux kernel, the following vulnerability has been resolved: net/sched: sch_qfq: prevent aggregate self-replacement qfq_change_class() snapshots the current aggregate settings while holding the qdisc tree lock, but drops the lock before selecting the destination aggregate. During that gap, qfq_enqueue() can move the class to an aggregate matching the requested settings. When qfq_change_class() resumes, qfq_find_agg() then returns cl->agg. If it is a singleton, qfq_deact_rm_from_agg() frees the aggregate before qfq_add_to_agg() immediately accesses the same pointer, causing a use-after-free. While holding the tree lock, skip the replacement when the destination is already the current aggregate. Any estimator replacement has already completed, so the class change can finish normally.

    Memory safety OS & kernel Linux

    unscored No fix listed
  23. CVE-2026-99689

    In the Linux kernel, the following vulnerability has been resolved: xfs: fix temporary cursor leak in btree shift corruption paths In xfs_btree_lshift and xfs_btree_rshift, when XFS_IS_CORRUPT detects an invalid firstrec/lastrec result after xfs_btree_dup_cursor has successfully allocated a temporary cursor, the error path jumps to error0 which returns without freeing tcur. This leaks the cursor memory, the xfs_group reference, and the per-level buffer references held by the duplicated cursor. Jump to error1 instead, which calls xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR) to properly release all resources.

    Memory safety OS & kernel Linux

    unscored No fix listed
  24. CVE-2026-99674

    In the Linux kernel, the following vulnerability has been resolved: can: dev: can_dropped_invalid_skb: drop CAN XL frames on non-CAN XL devices Sending a PF_PACKET bypasses the CAN framework logic and can directly reach a CAN driver's xmit() function. The PF_PACKET framework only checks that skb->len does not exceed the net_device MTU. For a CAN device that is not CAN XL capable, anything above CANFD_MTU (72 bytes) is therefore dropped before it reaches the driver. However, CAN XL frames are variable length. can_is_canxl_skb() accepts lengths in the range CANXL_HDR_SIZE + CANXL_MIN_DLEN up to CANXL_MTU, i.e. 13 to 2060 bytes. As a result, an ETH_P_CANXL skb with a length between 13 and 72 bytes can pass both the MTU and the can_dropped_invalid_skb() checks. A driver that does not support CAN XL will interpret canxl_frame->flags as a length because of the overlap with can_frame->len. And because CANXL_XLF is set, the resulting length is between 128 and 255. For drivers that do not check can_frame->len before copying can_frame->data, as most drivers do, this results in a buffer overflow of up to 247 bytes. Drop ETH_P_CANXL skbs if the device does not have the CAN_CAP_XL capability. Keep can_is_canxl_skb() for the validation of CAN XL skbs.

    Memory safety OS & kernel Linux

    unscored No fix listed
  25. CVE-2026-99681

    In the Linux kernel, the following vulnerability has been resolved: blk-mq: fix out-of-bounds read in blk_mq_free_rqs A KASAN slab-out-of-bounds read can occur in blk_mq_free_rqs() when there is a mismatch between the number of hardware queues allocated for the IO scheduler (et->nr_hw_queues) and the number of hardware queues in the block tag set (set->nr_hw_queues). This mismatch can happen if blk_mq_update_nr_hw_queues() fails halfway through (e.g., due to memory pressure during blk_mq_prealloc_tag_set_tags()). In this error path, the elevator is restored with a larger number of hardware queues than the block tag set actually has. When the elevator is later freed, blk_mq_free_sched_tags() iterates up to the new et->nr_hw_queues and calls blk_mq_free_rqs(). In blk_mq_free_rqs(), it attempts to access set->tags[hctx_idx] to get the driver tags. Because set->nr_hw_queues was not updated due to the earlier failure, set->tags still has the old (smaller) size, leading to an out-of-bounds read. BUG: KASAN: slab-out-of-bounds in blk_mq_free_rqs+0xde/0x680 Read of size 8 at addr ffff88818d67fc28 by task syz-executor117/5839 Call Trace: blk_mq_free_rqs+0xde/0x680 blk_mq_free_map_and_rqs+0x40/0xf0 blk_mq_free_sched_tags blk_mq_free_sched_res+0xeb/0x280 elevator_change_done+0x1d2/0x5c0 elevator_change+0x34f/0x480 elv_iosched_store+0x504/0x630 queue_attr_store+0x207/0x2b0 To fix this, explicitly check if hctx_idx < set->nr_hw_queues before accessing set->tags[hctx_idx]. If hctx_idx >= set->nr_hw_queues, the hardware queue doesn't exist in the tag set, meaning there are no driver tags to clear mappings from. In this case, safely set drv_tags to NULL. The subsequent call to blk_mq_clear_rq_mapping() already handles a NULL drv_tags pointer and will safely return.

    Memory safety OS & kernel Linux

    unscored No fix listed
  26. CVE-2026-99661

    In the Linux kernel, the following vulnerability has been resolved: mm/gup: check for overflow when converting nr_pages to bytes in GUP-fast The overflow check in gup_fast_fallback() only covers the addition to start. The preceding nr_pages << PAGE_SHIFT can already have truncated the length on 32bit. Users such as RapidIO's rio_dma_transfer() can derive such page counts from a 64-bit userspace length. Let's just check_shl_overflow() to reject any such overflow with -EOVERFLOW early.

    Memory safety OS & kernel Linux

    unscored No fix listed
  27. CVE-2026-99662

    In the Linux kernel, the following vulnerability has been resolved: mm/gup: fix out-of-bounds write for nr_pages == 0 in GUP-fast With nr_pages == 0, GUP-fast still enters its page table walk and can write past the caller's page array. The old end <= start check caught empty ranges, but was replaced with an overflow check. Users such as OrangeFS can pass a zero page count derived from a userspace-supplied length. So let's return zero for nr_pages == 0 before starting the walk, just like GUP-slow. Fix generated by an LLM and sent by Chris Mason in private. [ david: simplify description ]

    Memory safety OS & kernel Linux

    unscored No fix listed
  28. 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 ]

    Memory safety OS & kernel Linux

    unscored No fix listed
  29. 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.

    Memory safety OS & kernel Linux

    unscored No fix listed
  30. CVE-2026-99655

    In the Linux kernel, the following vulnerability has been resolved: mm/memory: reject remote COW faults on PFN mappings When handling a COW fault without a normal source page (PFN mappings), __wp_page_copy_user() copies from the faulting user address. For remote faults, this accesses the wrong address space, resulting in wrong data getting copied. For example, user_events updates an enable bit in the registering process's memory. If that memory needs to be faulted in, user_event_enabler_fault_fixup() calls fixup_user_fault() from a work item. Placing the enable bit in a private PFN mapping can trigger this remote COW path. Reject these faults with VM_FAULT_SIGBUS when the target mm differs from current->mm: remote faults. Ordinary (non-PFN) mappings are not affected. The user-address copy dates back to commit 6aab341e0a28 ("mm: re-architect the VM_UNPAGED logic"); likely, actual ways to get remote faults in there on COW PFN mappings appeared afterwards. But the fix is simple and straight forward. Fix generated by an LLM and sent by Chris Mason in private. [ david: simplify comment and description ]

    Memory safety OS & kernel Linux

    unscored No fix listed
  31. CVE-2026-99645

    In the Linux kernel, the following vulnerability has been resolved: mm/cma: don't release CMA pages still in use When a driver calls dma_free_contiguous() before quiescing DMA, the page might still have a reference from the device. put_page_testzero() there returns false, WARN fires, but the code proceeds to free_contig_frozen_range() putting a live page onto buddy and clearing the bitmap. A subsequent allocation can hand the same PFN to a new owner while the original holder still references it. A concurrent put_page() that drops the last reference between the testzero loop and free_contig_frozen_range() can double-queue the page via page->lru, corrupting buddy lists. Avoid the corruption by not freeing a CMA region if any of the pages inside are still in use. Make it explicit in the warning that the driver allowed a leak. While at it, use page_ref_freeze() to make the semantics clearer and switch to pr_warn_ratelimited(). [ david: use page_ref_freeze() + pr_warn_ratelimited() ]

    Memory safety OS & kernel Linux

    unscored No fix listed
  32. CVE-2026-99650

    In the Linux kernel, the following vulnerability has been resolved: mm/ksm: fix use of a freed anon_vma at swap-in A folio read from a swap slot previously shared by KSM can remain in the swap cache after the process that first mapped it exits. Its anon_vma can then be freed, so ksm_might_need_to_copy() must not dereference it to compare roots. Doing so can result in some false-positives, resulting in the reverse mapping missing the new folio mapping, preventing migration and reclaim. SLAB_TYPESAFE_BY_RCU allows individual anon_vma objects to be reused immediately after freeing. A stale pointer can then refer to another anon_vma whose root matches, even though the target VMA is not linked to that anon_vma. Move the anonymous rmap compatibility checks into folio_anon_rmap_compatible(). Keep the page-index check and compare the folio's anon_vma pointer against the VMA's own anon_vma and root first. If neither matches, bite the bullet and walk the VMA's anon_vma_chain. The VMA keeps these anon_vmas alive, and finding a match guarantees that reverse mapping can find the new mapping. Lorenzo Stoakes says in a private discussion: "This isn't quite equivalent to the existing check. Every anon_vma in an anon_vma_chain shares the same root, but if you split a VMA then fork both of them like this: VMA A / \ <split> VMA B VMA C \ / root <- anon_vma 1 ----^ ^---- <fork> / parent \ anon_vma 2 anon_vma 3 | | VMA D VMA E VMA's D and E both get their own new anon_vma's such that: VMA D: anon_vma 2 -> anon_vma 1 -> root VMA E: anon_vma 3 -> anon_vma 1 -> root And if you found a folio referencing anon_vma 2, the existing check: anon_vma->root == vma->anon-vma->root Would be true for VMA's B, C, D & E. However after your change it'd be true for VMA D only, as it's the only one with anon_vma 2 in its anon_vma_chain. Whether or not that could actually happen in practice... 🙂 I don't know. [...] after your change it'd be true for VMA D only, as it's the only one with anon_vma 2 in its anon_vma_chain. [...] I would [...] say that you are making the test stricter but more correct. " If this affects performance when KSM was never even involved, we could just track whether KSM ever merged anything and just skip these checks. There isn't really a lot we can do otherwise: for mapped folios we might be able to use folio_get_anon_vma(), but for unmapped folios there really is no easy way to see if the anon_vma is still valid. Document the (confusing) lifetime requirements of folio_anon_vma().

    Memory safety OS & kernel Linux

    unscored No fix listed
  33. CVE-2026-99638

    In the Linux kernel, the following vulnerability has been resolved: net/sched: clamp skb->queue_mapping before indexing the tx queue This issue was caught by Sashiko (nipa). A producer-side path cap is already merged (ea4d4b5dddb5); this patch adds the matching consumer-side caps. The issue: The tx queue is picked and capped in netdev_core_pick_tx() before the __dev_xmit_skb() qdisc enqueue path. A tc BPF program running inside an egress qdisc (not clsact) can then change __sk_buff->queue_mapping; that store is never re-checked. bpf_convert_ctx_access() bounds it only by NO_QUEUE_MAPPING, not by the device's queue count. Two consumers read the stale value: - On dequeue, qdisc_restart() calls netdev_get_tx_queue(), which indexes dev->_tx[]. netdev_get_tx_queue() guards it with just a DEBUG_NET_WARN_ON_ONCE, so sch_direct_xmit() takes the out-of-bounds txq->_xmit_lock and writes txq->xmit_lock_owner past the end of the allocation on non-lltx devices such as ifb. Drivers that index their rings from skb_get_queue_mapping() (ex: ifb) go out of range the same way. - On enqueue, taprio_enqueue() indexes q->qdiscs[] with the same value; the array is sized dev->num_tx_queues. A duplicating netem on a child qdisc clones the skb and re-enqueues the clone at the taprio root (qdisc_root_bh()) through a direct rootq->enqueue() call, so the value reaches taprio_enqueue() before any dequeue cap runs. The Fix: Cap the mapping at the consumers and write the capped value back, so the qdisc's _tx[] lookup, the taprio q->qdiscs[] lookup and any later driver read stay in range. On the dequeue side the cap folds into the dequeue path instead of a separate walk over the returned list: dequeue_skb() caps the head, and each bulk helper caps the element it appends. For a qdisc that is not TCQ_F_ONETXQUEUE, netdev_cap_txqueue() logs a ratelimited notice and selects queue 0. A TCQ_F_ONETXQUEUE qdisc is bound to a single tx queue: dequeue_skb() uses q->dev_queue for the frozen check, the BQL limit and the trace point, while qdisc_restart() transmits on the skb's queue. Capping an out-of-range mapping to queue 0 would therefore send a per-queue child's skb through a different queue than the one it was dequeued for. On mq, mqprio, taprio and htb offload children, a stopped queue 0 could then park the list in q->gso_skb, and the later wake would schedule queue 0's own child instead of this qdisc. Pin a TCQ_F_ONETXQUEUE mapping to the qdisc's own queue to keep the documented invariant. The pin covers every mapping, not only an out-of-range one: an in-range mapping that a filter sets on one of those per-queue children is rewritten to the child's own queue, and this path does not call netdev_cap_txqueue(), so it logs no notice. On a single-queue device qdisc_create() sets TCQ_F_ONETXQUEUE, so the one-queue trigger described below also takes this branch and is corrected without a notice. That also keeps a bulk list on one queue by construction: every element of a TCQ_F_ONETXQUEUE qdisc, including a follower changed by a filter, is pinned to q->dev_queue before it is appended. The fast bulk helper therefore needs no head comparison and no skb_bad_txq requeue; the slow helper, used only by non-ONETXQUEUE qdiscs, still compares and requeues a mismatched follower, where a wake on any queue reschedules q. On the enqueue side taprio_enqueue() caps the mapping against real_num_tx_queues and writes it back before selecting q->qdiscs[queue], so a duplicating child cannot steer the clone out of the array. Conditions to recreate the bug: The dequeue path: load a SCHED_CLS or SCHED_ACT program that stores __sk_buff->queue_mapping (a tc BPF action is enough), attach it to a transmit qdisc (for example a matchall filter on a prio root), and send a packet. With a 1-queue device a store of 1 indexes one past dev->_tx[]. Under KASAN with panic_on_warn=1 the unfixed kernel faults with "BUG: KASAN: slab-out- ---truncated---

    Memory safety OS & kernel Linux

    unscored No fix listed
  34. 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.

    Memory safety OS & kernel Linux

    unscored No fix listed
  35. CVE-2026-99623

    In the Linux kernel, the following vulnerability has been resolved: drm/amdgpu: Clamp the ATIF backlight curve to its 99 points amdgpu_atif_query_backlight_caps() stores the firmware's number_of_points as data_points but copies only the 99 points that luminance_data[] holds. number_of_points is a u8, so a firmware table that claims more makes the custom curve code in amdgpu_dm index past the array, and dm_query_extended_brightness_caps() copy past the end of the data_points[] buffer it fills Clamp the count to ATIF_QBTC_MAX_DATA_POINTS and warn, since a table that claims more is a firmware bug (cherry picked from commit af8a1c2b7d3de4960e10f94e8ba2fc0ccd168072)

    Memory safety OS & kernel Linux

    unscored No fix listed
  36. CVE-2026-99617

    In the Linux kernel, the following vulnerability has been resolved: arm64: mm: validate /dev/mem mmap ranges without overflow An arm64 syzkaller reproducer creates a FUSE character device with major:minor 1:1, so opening it uses the /dev/mem file operations. It maps one page from offset 0xfffffffffffff000 at user address zero. The next openat() with a NULL pathname triggers a synchronous external abort (SEA, ESR 0x96000010) in strncpy_from_user() via do_getname(), panicking the kernel. valid_mmap_phys_addr_range() checks the exclusive end by adding the mapping size to the starting address. For this offset, the sum wraps to zero and passes the PHYS_MASK test, allowing an out-of-range PFN to reach remap_pfn_range(). The inclusive-end overflow check in mmap_mem_prepare() does not catch it because the end is ULONG_MAX. Reject starting PFNs outside PHYS_MASK and compare the inclusive length against the remaining physical range without overflowing. This also allows mappings that end exactly at PHYS_MASK, which the old check incorrectly rejects. [[email protected]: changed the check to look more like __access_ok()]

    Memory safety OS & kernel Linux

    unscored No fix listed
  37. CVE-2026-99618

    In the Linux kernel, the following vulnerability has been resolved: nvmet: don't return memory contents beyond valid data when reading ANA log page The host side does not know how much valid data will be returned by the target so it will typically provide a buffer larger than the amount of valid data in the ANA log page. The memory allocation used by the target does not guarantee that prior kernel memory contents will not be returned past the end of the valid page data. Fix this to ensure that the SGL is zeroed before copying the data.

    Memory safety OS & kernel Linux

    unscored No fix listed
  38. CVE-2026-99621

    In the Linux kernel, the following vulnerability has been resolved: drm/amdgpu: unlock the VM and BO when attaching a KFD BO fails kfd_mem_attach() locks the page directory and the BO with a drm_exec transaction on its stack, but leaves through the unwind label without drm_exec_fini() when locking fails, e.g. when MAP_MEMORY_TO_GPU gets a signal while the page directory or the BO is busy. Reservations already acquired then remain locked with their ww_acquire_ctx pointing into the expired stack frame. With one thread doing GEM_VA map/unmap on the KFD VM and another one mapping KFD BOs while getting SIGUSR1 every 20 us: BUG: KASAN: stack-out-of-bounds in __ww_mutex_lock.constprop.0+0x34f5/0x3da0 ww_mutex_lock_interruptible+0x2f/0x150 drm_exec_lock_obj+0x107/0x2d0 [drm_exec] kfd_mem_attach+0x4dd/0x1b60 [amdgpu] and the process ends up in D state. Finish the transaction before unwinding. (cherry picked from commit 7b885fa27e4b9fa5cbdc64e66fc35872aee296f5)

    Memory safety OS & kernel Linux

    unscored No fix listed
  39. CVE-2026-99594

    In the Linux kernel, the following vulnerability has been resolved: arm64: kexec_file: Fix elf_headers memory leak in retry loop A retry allocates new elf headers and overwrites image->elf_headers, leaking the previous buffer. Free it before assigning. [[email protected]: drop the redundant NULL check before vfree()]

    Memory safety OS & kernel Linux

    unscored No fix listed
  40. 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

    Memory safety OS & kernel Linux

    unscored No fix listed
  41. 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.

    Memory safety OS & kernel Linux

    unscored No fix listed
  42. CVE-2026-99590

    In the Linux kernel, the following vulnerability has been resolved: ceph: clamp the inline data length in ceph_fill_inline_data() The MDS decides how much inline data to attach to a reply, and the client parses inline_len without any upper bound: ceph_decode_need() only verifies that the message actually carries that many bytes. ceph_fill_inline_data() then copies the data into a single folio with no length check, so a malicious or buggy MDS returning more inline data than the folio can hold writes past its end. Clamp the length to folio_size() so both callers, handle_cap_grant() and fill_inode(), are covered.

    Memory safety OS & kernel Linux

    unscored No fix listed
  43. 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.

    Memory safety OS & kernel Linux

    unscored No fix listed
  44. CVE-2026-99574

    In the Linux kernel, the following vulnerability has been resolved: f2fs: bound the SIT/NAT bitmaps by the checkpoint pack sit_ver_bitmap_bytesize and nat_ver_bitmap_bytesize are used as memcpy lengths into and out of sbi->ckpt, which is sized from cp_payload alone, and nothing compares either length against that allocation. sanity_check_ckpt() has checked these two fields since c77ec61ca0a4, which added them after a crafted image overran build_sit_info(). That check ties each length to a segment count; it never compares either against the allocation, which is the half this adds. Mounting over-reads in build_sit_info() and build_node_manager(), and the first checkpoint writes the same lengths back through the same pointers, so the over-read's contents reach the on-disk checkpoint. An unmount is enough; the filesystem does not have to be used. Bound both ends the way __bitmap_ptr() lays them out. The three branches differ, so a single sum of the two lengths would still admit a small overflow when cp_payload is non-zero.

    Memory safety OS & kernel Linux

    unscored No fix listed
  45. 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.

    Memory safety OS & kernel Linux

    unscored No fix listed
  46. CVE-2026-133528

    BeamNG.drive before 0.40.6 contains an out-of-bounds write vulnerability because JIT-compiled LuaJIT traces omit FFI array bounds checks, letting untrusted Lua escape the sandbox. Attackers supplying a malicious mod can JIT-compile an FFI store loop, write out of range using a tostring() address leak, re-enable ffi.load, and run commands as the game user.

    Memory safety BeamNG GmbH BeamNG.drive

    9.3 CRITICAL Fix listed
  47. CVE-2026-131868

    Convert::ASN1 versions before 0.37 for Perl allow memory exhaustion via a declared element length passed unchecked to sysread in asn_read and asn_get. The length an element declares in its BER header is passed to sysread as the read size. Perl grows the target buffer to that size before reading, so memory is allocated from the declared length, not from the data received. A 6-byte header can declare a length of up to 4GB. A single header makes the reading process allocate the declared size. Where the allocation fails, perl exits with "Out of memory", which eval cannot catch.

    Memory safety Library

    7.5 HIGH Fix listed
  48. CVE-2025-71474

    lwext4 through 1.0.0 contains an out-of-bounds write vulnerability in ext4_fs_init_block_bitmap() that allows attackers to corrupt heap memory via crafted superblock geometry fields. Attackers can supply an ext4 image with oversized s_reserved_gdt_blocks and BLOCK_UNINIT groups to write past the bitmap buffer, crashing applications that mount it.

    Memory safety OS & kernel Gkostka lwext4

    6.9 MEDIUM Fix listed
  49. CVE-2025-71473

    lwext4 through 1.0.0 contains a heap buffer overflow in ext4_fs_init_inode_bitmap() in src/ext4_fs.c because it does not validate s_inodes_per_group against the inode bitmap size. Attackers can supply a crafted ext4 image with an oversized s_inodes_per_group so memset writes past the heap buffer, corrupting memory and crashing the mounting application.

    Memory safety OS & kernel Gkostka lwext4

    6.9 MEDIUM Fix listed
  50. CVE-2026-133519

    FFmpeg 7.0 through 9.0.2 contains an out-of-bounds read vulnerability in dvdvideo_menu_next_ps_block() in libavformat/dvdvideodec.c when demuxing DVD menu domains. Attackers can craft a DVD image whose last cell sector and NAV next-VOBU pointer bypass the end-of-cell check, overreading pgc->cell_playback and crashing the process.

    Memory safety FFmpeg

    6.8 MEDIUM Fix listed

CVE records published in 2026 (ids starting CVE-2026-), sorted by how likely they are to be exploited.