The kernel CNA published 516 Linux kernel CVE records from 15 to 21 November 2026. All 516 are live and none was rejected. 17 records carry a CNA score: 3 are CRITICAL and 14 are HIGH. The 3 CRITICAL records are 2 in the Ceph client and 1 in the NFS server. Every record names a fixed release, and 514 of the 516 name a fix in mainline. There is no 0-day this week. No record says that a bug is being exploited, and no record contains exploit code. The action is to update to the fixed version for your branch. Minimum safe versions: 6.12.118, 6.18.64 and 7.3.5. On 5.10, 5.15, 6.1 and 6.6 at most 2 records list a fix, and the newest releases (5.10.274, 5.15.225, 6.1.192 and 6.6.162) already contain those fixes. 7.2 is end of life, so move to 7.3.5.
This week has 516 Linux kernel CVEs. That is a very high count. The previous edition had 4. Most records name fixes in the 6.12.117, 6.18.61 and 7.3.2 releases or in the 17 November releases. Most fixes are narrow. Which records matter to you depends on what your product uses. 88 records are in network file system code (SMB, NFS, SUNRPC and Ceph), 62 are in memory management and 23 are in KVM. Only 17 of the 516 records have a CNA score, so a record with no score is not a record with no risk.
The one action: update to the fixed version for your branch
Run uname -r and find your branch in the table. If your version is lower than the value in the second column, plan the update. Each value is the highest fixed version that any record this week names for that branch. The values for 6.12, 6.18 and 7.3 are equal to the newest releases on kernel.org on 21 November 2026.
raghu@techveda.org:~$ uname -r
6.12.117The output above is an example. 6.12.117 is lower than 6.12.118, so that system needs the update.
| Stable branch | Update to at least | Notes |
|---|---|---|
| 5.10 | 5.10.271 (newest release is 5.10.274) | 1 record lists a fix for this branch, and 5.10.274 already contains it. 242 records affect 5.10 and list no fix for it. These include 2 of the 3 CRITICAL records. kernel.org lists December 2026 as the projected end of life for this branch. |
| 5.15 | 5.15.222 (newest release is 5.15.225) | 1 record lists a fix for this branch, and 5.15.225 already contains it. 287 records affect 5.15 and list no fix for it. These include all 3 CRITICAL records. kernel.org lists December 2026 as the projected end of life for this branch. |
| 6.1 | 6.1.189 (newest release is 6.1.192) | 1 record lists a fix for this branch, and 6.1.192 already contains it. 314 records affect 6.1 and list no fix for it. These include all 3 CRITICAL records. |
| 6.6 | 6.6.158 (newest release is 6.6.162) | 2 records list a fix for this branch, and 6.6.162 already contains them. 349 records affect 6.6 and list no fix for it. These include all 3 CRITICAL records. |
| 6.12 | 6.12.118 | 370 records list a fix for this branch. 6.12.118 clears all of them and is the newest release. 46 records affect 6.12 and list no fix for it. None of these has a CNA score. |
| 6.18 | 6.18.64 | 427 records list a fix for this branch. 6.18.64 clears all of them and is the newest release. 35 records affect 6.18 and list no fix for it. None of these has a CNA score. |
| 7.2 | Move to 7.3.5 | kernel.org marks 7.2 as end of life at 7.2.15. 4 records list a fix in 7.2.15. 496 records affect 7.2 and list no fix for it. |
| 7.3 | 7.3.5 | 514 records list a fix for this branch. 7.3.5 clears all of them and is the newest release. The other 2 records were fixed in mainline 7.1, so 7.3 already has those fixes. |
| Mainline | 7.4-rc3 | 514 records name a mainline fix, from 7.1 to 7.4-rc3. 7.4-rc3 is the newest mainline release candidate. 2 records (CVE-2026-99566 and CVE-2026-99567) name stable fixes only. |
Branches that are not listed were not affected, or are not maintained. The counts of records with no fix use the first affected version and the affected stable releases that each record lists. 9 records name no first affected version, so those counts can be a little low. On 5.10, 5.15, 6.1 and 6.6 the newest releases are from 6 November, and they do not contain most of this week’s fixes. kernel.org lists December 2026 as the projected end of life for 5.10 and 5.15. 7.2 is already end of life.
Mobile and automotive
Phones, head units and telematics units use Bluetooth, Wi-Fi and CAN. This week has 3 Bluetooth records, 9 Wi-Fi records and 5 CAN records. None of them has a CNA score. The four records below are the ones to check first.
CVE-2026-99793: Bluetooth L2CAP accepts a connection response too early
A Bluetooth peer can send an L2CAP connection response before the local side has sent its connection request. The channel then finishes configuration and becomes connected before security has completed. If security fails later, the channel stays up on a link that did not reach the required security level. The gate is CONFIG_BT. The record has no CNA score. The first affected version is 2.6.12. The record lists fixes in 6.12.118, 6.18.64, 7.3.5 and mainline 7.4-rc1. It affects 5.10, 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99793 on cve.org.
CVE-2026-99794: Bluetooth L2CAP accepts signalling responses in the wrong state
A Bluetooth peer can send configuration messages while an incoming channel still waits for security or for a deferred accept. A configuration response can then move the channel to the connected state. Configuration responses can also act on channels that are closing. The fix accepts each signalling response only in the state that expects it. The gate is CONFIG_BT. The record has no CNA score. The first affected version is 2.6.12. The record lists fixes in 6.12.118, 6.18.64, 7.3.5 and mainline 7.4-rc1. It affects 5.10, 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99794 on cve.org.
CVE-2026-99672: CAN receive lists keep a pointer to a freed socket
The receive lists of a CAN device were protected by a lock that belongs to the network namespace. If a CAN device is moved to another network namespace while an ISO-TP or BCM socket is closing, the socket does not remove its filters. The remaining receivers then point to a freed ISO-TP socket or to a freed BCM operation. The trigger is a namespace move that happens while the socket closes. This is relevant to products that move CAN devices into containers. The gate is CONFIG_CAN. The record has no CNA score. The first affected version is 4.12. The record lists fixes in 6.12.118, 6.18.64, 7.3.5 and mainline 7.4-rc2. It affects 5.10, 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99672 on cve.org.
CVE-2026-99674: CAN XL frame reaches a driver that does not support CAN XL
A program can send a frame through a PF_PACKET socket directly to the transmit function of a CAN driver. A CAN XL frame of 13 to 72 bytes passes the size checks on a device that does not support CAN XL. The driver then reads a flags field as a length. The record says that most drivers copy the data without a length check, and that this gives a buffer overflow of up to 247 bytes. The sender is a program on the same system. The gate is CONFIG_CAN_DEV. The fixed code is in the common CAN driver module. The record has no CNA score. The first affected version is 6.1. The record lists fixes in 7.3.5 and mainline 7.4-rc2. It affects 6.1, 6.6, 6.12 and 6.18 and lists no fix for those branches. See CVE-2026-99674 on cve.org.
The Logitech record CVE-2026-99313 also applies to head units that pair with Logitech HID++ devices over Bluetooth. The xHCI record CVE-2026-99308 also applies to head units with USB ports. Both are described in the Embedded and IoT section.
Embedded and IoT
Embedded products are affected through USB ports, input devices, boot-time drivers and architecture code. This week also has 27 records in display, GPU and accelerator drivers, and none of those has a CNA score. The four records below apply to many boards.
CVE-2026-99308: xHCI bounce buffer overflow with an out-of-spec USB device
Some high-speed USB devices have bulk endpoints with a 1024-byte packet size. This is outside the USB specification, but the USB core accepts it. The xHCI driver allocates a 512-byte bounce buffer for such an endpoint. It can write up to 1024 bytes into the buffer when certain scatter-gather transfers are used. The record says a malicious device can try this. The gate is CONFIG_USB_XHCI_HCD. The record has no CNA score. The first affected version is 4.8. The record lists fixes in 6.12.118, 6.18.63, 7.3.4 and mainline 7.4-rc1. It affects 5.10, 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99308 on cve.org.
CVE-2026-99313: Logitech HID++ driver reads a stale stack pointer
The hid-logitech-hidpp driver keeps a pointer to a buffer on the kernel stack after the command function returns. An event that is handled on another thread can then use the stale pointer. The result is an out-of-bounds access on the kernel stack of another thread, or a NULL pointer dereference. The CNA vector is adjacent network (AV:A). The CNA rationale says the report that starts the bug arrives over Bluetooth from a bonded Logitech HID++ device. The record is new this week, but the fixed releases are old. The newest release of every listed branch already has the fix. The gate is CONFIG_HID_LOGITECH_HIDPP. The CNA score is 8.8 (HIGH). The first affected version is 3.19. The record lists fixes in 5.10.271, 5.15.222, 6.1.189, 6.6.158, 6.12.112, 6.18.54 and mainline 7.1. See CVE-2026-99313 on cve.org.
CVE-2026-99442: nvmem cell read goes past the end of the buffer
When an nvmem cell has a bit offset of 8 bits or more, the read loop reads more bytes than the cell holds. The record says the extra data does not leak directly, because the wrongly read bits are cleared. The record says the bug was found with KASAN. Many embedded boards keep values such as MAC addresses and calibration data in nvmem cells. The gate is CONFIG_NVMEM. The record has no CNA score. The first affected version is 6.15. The change that caused the bug was backported, so the record also lists 6.12.31 as affected. Later releases of that branch are affected too. The record lists fixes in 6.12.117, 6.18.61, 7.3.2 and mainline 7.4-rc1. See CVE-2026-99442 on cve.org.
CVE-2026-99617: arm64 /dev/mem mapping check overflows
This record applies to arm64 only. The range check for /dev/mem mappings adds the mapping size to the start address. For a start offset near the top of the address range the sum wraps to zero, and the check passes. A page frame number that is out of range then reaches remap_pfn_range(). The test program in the record caused a kernel panic. It used a FUSE character device with device number 1:1 to reach the /dev/mem file operations. The gate is an arm64 kernel. The record has no CNA score. The first affected version is 3.7. The record lists fixes in 6.12.118, 6.18.64, 7.3.5 and mainline 7.4-rc2. It affects 5.10, 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99617 on cve.org.
Embedded network storage devices that use ksmbd should also read CVE-2026-99309 in the cloud section. It applies to 32-bit kernels only. Products with Bluetooth should read CVE-2026-99793 and CVE-2026-99794 in the mobile section.
Cloud and datacenter Linux kernel CVEs
Most of this week’s high-severity records apply to servers. All 3 CRITICAL records and 9 of the 14 HIGH records are in this section. Network file systems come first, then virtualization, then the IP stack.
CVE-2026-99304: CephFS client reads past a snapshot message
The CephFS client adds two 32-bit counts from the network before it checks the message length. The sum can wrap to zero. The check then passes and the client reads past the end of the message buffer. The record says a malicious or compromised metadata server (MDS) can send this message with no action on the client. The code is in the CephFS client. The gate is CONFIG_CEPH_FS. The CNA score is 9.8 (CRITICAL). The first affected version is 2.6.34. The record lists fixes in 6.12.117, 6.18.63, 7.3.4 and mainline 7.4-rc1. It affects 5.10, 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99304 on cve.org.
CVE-2026-99306: CephFS client accepts an MDS map with too many ranks
A Ceph monitor can announce more MDS ranks than the client supports. The client then reads past the end of a bitmap. The record says the map arrives on the monitor connection, and that this connection is not authenticated in the legacy messenger mode. The CNA rationale says the legacy mode is the default. The code is in the CephFS client. The gate is CONFIG_CEPH_FS. The CNA score is 9.8 (CRITICAL). The first affected version is 5.15. The record lists fixes in 6.12.117, 6.18.63, 7.3.4 and mainline 7.4-rc1. It affects 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99306 on cve.org.
CVE-2026-99296: libceph reads four bytes that the monitor never sent
The Ceph library checks a monitor reply against the size of the reply buffer, not against the number of bytes received. A short reply passes the check, and the client reads four bytes that were never received. The value reaches user space. The record shows a kernel panic in the RBD block driver, which passes the value on without a check. The CNA rationale says this panic needs an administrator to add an RBD device at that time. The gate is CONFIG_CEPH_LIB. The CNA score is 7.1 (HIGH). The first affected version is 4.9. The record lists fixes in 6.12.117, 6.18.61, 7.3.2 and mainline 7.4-rc1. It affects 5.10, 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99296 on cve.org.
CVE-2026-99310: NFS server use-after-free during EXCHANGE_ID
The NFSv4 server walks the list of open owners of a client without taking the lock that protects the list. Another thread can free an open owner at the same time. The record says an EXCHANGE_ID request with different credentials can then use the freed open owner. The CNA rationale says the sender needs two connections to the server and only AUTH_SYS credentials, which the client states itself. The gate is CONFIG_NFSD. The CNA score is 9.8 (CRITICAL). The first affected version is 4.4. The record lists fixes in 6.12.117, 6.18.61, 7.3.2 and mainline 7.4-rc1. It affects 5.10, 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99310 on cve.org.
CVE-2026-99289: ksmbd uses a freed session when one open replaces another
ksmbd is the SMB server inside the kernel. An open with the same AppInstanceId can replace an earlier open. The server reaches the session of the earlier open through a pointer that has no reference count. If that session was already freed, the server takes a lock in freed memory. The CNA rationale says the sender must already be a user of the share. The record also affects 7.2, which is end of life and has no fix. The gate is CONFIG_SMB_SERVER. The CNA score is 8.8 (HIGH). The first affected version is 7.2. The record lists fixes in 7.3.2 and mainline 7.4-rc1. See CVE-2026-99289 on cve.org.
CVE-2026-99290: ksmbd use-after-free after a failed durable reconnect
Two reconnect requests for the same durable handle can run at the same time. If the first one fails, the handle is freed but stays in the file table of the first session. Closing that session then writes to freed memory. The CNA rationale says the sender must be the authenticated owner of the handle. It also says the sender cannot force the failure that starts the bug, so the attack complexity is high. The gate is CONFIG_SMB_SERVER. The CNA score is 7.5 (HIGH). The first affected version is 6.9. The change that caused the bug was backported, so the record also lists 6.6.32 as affected. Later releases of that branch are affected too. The record lists fixes in 6.12.117, 6.18.61, 7.3.2 and mainline 7.4-rc1. It affects 6.6 and lists no fix for that branch. See CVE-2026-99290 on cve.org.
CVE-2026-99291: ksmbd writes past the SMB2 CREATE response buffer
ksmbd uses a 448-byte buffer for a single SMB2 CREATE response. A request with AAPL response contexts can need 456 bytes. In that case the server writes 12 bytes past the end of the buffer. The CNA rationale says the share must have the Time Machine flag and the sender must be an authenticated share user. Kernels older than 7.2.6 are not affected. The gate is CONFIG_SMB_SERVER. The CNA score is 8.8 (HIGH). The first affected version is 7.3. The change that caused the bug was backported, so the record also lists 7.2.6 as affected. Later releases of that branch are affected too. The record lists fixes in 7.3.2 and mainline 7.4-rc1. See CVE-2026-99291 on cve.org.
CVE-2026-99309: ksmbd security descriptor check overflows on 32-bit kernels
This record applies to 32-bit builds only. A large DACL offset makes an addition overflow, and a bounds check passes. The record describes a read of up to 8 bytes before the start of a buffer. The CNA rationale also describes a write past a heap buffer. It says the share must use the acl-xattr flag and the sender must be an authenticated share user. This is relevant to 32-bit network storage devices that use ksmbd. The gate is CONFIG_SMB_SERVER on a 32-bit kernel. The CNA score is 8.8 (HIGH). The first affected version is 6.0. The change that caused the bug was backported, so the record also lists 5.15.62 as affected. Later releases of that branch are affected too. The record lists fixes in 6.12.117, 6.18.61, 7.3.2 and mainline 7.4-rc1. It affects 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99309 on cve.org.
CVE-2026-99292: SMB client writes past a buffer during chmod
The SMB client builds a new security descriptor when a program calls chmod() on a file on an SMB mount. If the existing descriptor has compact or overlapping owner and group SIDs, or no DACL offset, the client allocates too little memory. It then writes past the end of the buffer. The CNA rationale says the descriptor comes from the server. It also says the mount must use the modefromsid or cifsacl option, and a user of the mount must change a file mode. The gate is CONFIG_CIFS. The CNA score is 8.8 (HIGH). The first affected version is 3.8. The record lists fixes in 6.12.117, 6.18.61, 7.3.2 and mainline 7.4-rc1. It affects 5.10, 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99292 on cve.org.
CVE-2026-99295: SMB client uses a freed read request after a receive error
A receive error during a read can start a reconnect. The reconnect ends the pending requests and can free the read request. The receive code then continues and uses the freed request. The CNA rationale says the SMB server can cause this when it closes the connection during a read response. The CNA vector lists an impact on availability only. A system that only mounts an SMB share is affected. The gate is CONFIG_CIFS. The CNA score is 7.5 (HIGH). The first affected version is 6.3. The record lists fixes in 6.12.117, 6.18.61, 7.3.2 and mainline 7.4-rc1. It affects 6.6 and lists no fix for that branch. See CVE-2026-99295 on cve.org.
CVE-2026-99312: KVM runs a nested guest with a stale MMU context after INIT
This record applies to nested virtualization on AMD processors (nested SVM). If KVM emulates an INIT while the nested guest (L2) runs with paging disabled, KVM later runs the first-level guest (L1) with a stale MMU context. The CNA vector has a changed scope (S:C). The CNA rationale says a non-root user who can run a virtual machine can start the bug. It says the processor then uses a freed host page as a page table root, and that nested support is on by default in kvm_amd. The gate is CONFIG_KVM_AMD with nested virtualization enabled. The CNA score is 8.8 (HIGH). The first affected version is 6.1. The change that caused the bug was backported, so the record also lists 5.15.81 as affected. Later releases of that branch are affected too. The record lists fixes in 6.12.118, 6.18.64, 7.3.5 and mainline 7.4-rc2. It affects 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99312 on cve.org.
CVE-2026-99591: AMD IOMMU device table is forced to its maximum size
The AMD IOMMU can read the device table entry that follows the last entry of the table. The record says that if this entry is on a page owned by a guest, the guest can corrupt hypervisor pages or stop the system. The fix forces the table to its maximum size of 2 MB on AMD processor families 0x17 to 0x1a. The record calls it a mitigation for CVE-2024-21963. The record also says to update the production kernel and the kdump crash kernel together, because a mixed pair can make the crash kernel panic on SNP-capable systems. The gate is CONFIG_AMD_IOMMU on an AMD host that runs guests. The record has no CNA score. The record names no first affected version. It lists all earlier releases of the fixed branches as affected. The record lists fixes in 6.12.118, 6.18.64, 7.3.5 and mainline 7.4-rc3. See CVE-2026-99591 on cve.org.
CVE-2026-99709: IP fragment reassembly size counter can wrap
The kernel adds the memory size of all fragments in a reassembly queue in a 32-bit signed value. If the sum reaches 2^32 it wraps, and the kernel can free a socket that is still in use. The record says this can be started from an unprivileged user and network namespace with padded TUN packets. This is relevant to hosts that run containers for other users. There is no specific configuration gate. The code is part of the IP stack. The record has no CNA score. The first affected version is 6.9. The change that caused the bug was backported, so the record also lists these stable releases as affected: 5.10.227, 5.15.168, 6.1.85 and 6.6.26. Later releases of those branches are affected too. The record lists fixes in 7.3.5 and mainline 7.4-rc2. It affects 5.10, 5.15, 6.1, 6.6, 6.12 and 6.18 and lists no fix for those branches. See CVE-2026-99709 on cve.org.
CVE-2026-99320: IPv6 raw socket use-after-free with VRF
The IPv6 raw socket send path keeps a pointer to the route. The VRF output path can replace the route and release the old one. The send path then uses the old pointer. The CNA vector is local with high attack complexity. The CNA rationale says the sender needs CAP_NET_RAW and CAP_NET_ADMIN, and that an unprivileged user can get both inside a user and network namespace. The gate is IPv6 with a VRF device. The CNA score is 7.0 (HIGH). The first affected version is 4.9. The record lists fixes in 6.12.118, 6.18.64, 7.3.5 and mainline 7.4-rc2. It affects 5.10, 5.15, 6.1 and 6.6 and lists no fix for those branches. See CVE-2026-99320 on cve.org.
The scored records are a small part of the network file system work this week. In total there are 29 ksmbd records, 17 SMB client records, 19 NFS server records, 8 SUNRPC records and 6 Ceph records. There are also 23 KVM records and 62 memory management records. One update per branch covers all of them.
Medical devices
Connected medical devices matter for this list when they use Bluetooth, TCP/IP, USB or network file systems. All four apply this week. Wearables and patient monitors that use Bluetooth are affected by CVE-2026-99793 and CVE-2026-99794. A device that stores or reads clinical data on an SMB share is affected by CVE-2026-99295, even if it only mounts the share. It is affected by CVE-2026-99292 if the mount uses the modefromsid or cifsacl option. A clinical data store that exports NFSv4 from a Linux server is affected by the CRITICAL record CVE-2026-99310. Bedside devices with USB ports are affected by CVE-2026-99308. Devices that pair with Logitech HID++ input devices over Bluetooth are affected by CVE-2026-99313. Devices that exchange DICOM images or HL7 and FHIR messages use the IP stack. CVE-2026-99709 is in the IP stack, but the record describes a start from an unprivileged namespace on the device itself. It matters where untrusted local code can run. Medical-device patching is gated by validation and regulatory revalidation (US FDA postmarket cybersecurity guidance, EU MDR and IEC 62304), so the action is to plan the stable update through the manufacturer’s change-control process.
How to check if a CVE applies to you
Ask three questions for each record.
The first question is the version. Run uname -r and compare the result with the table above. If your version is at or above the value for your branch, the records with a listed fix for that branch are cleared.
The second question is the configuration. A record applies only if the code is built into your kernel or loaded as a module. Check the gate that is named for the record.
raghu@techveda.org:~$ zcat /proc/config.gz | grep -E "CONFIG_(BT|CAN|NFSD|SMB_SERVER|CIFS|CEPH_FS|KVM_AMD)="
CONFIG_BT=m
CONFIG_CAN=m
CONFIG_CIFS=mThe output above is an example. On that system the Bluetooth, CAN and SMB client records apply. The NFS server, ksmbd, CephFS and KVM records do not apply, because those options are not set.
The third question is reachability. It sets the order of work. Records that a remote or over-the-air peer can reach are highest priority. This week that means the NFS server, SMB, Ceph and Bluetooth records. Records that a local unprivileged user can reach come next. Records that a guest can use against a host come next on shared hosts. Records that need privilege or a rare configuration come last.
Key takeaways
- 516 Linux kernel CVEs were published this week. 3 are CRITICAL, 14 are HIGH and 499 have no CNA score.
- Update to at least 6.12.118, 6.18.64 or 7.3.5. These are the newest releases, so no record this week can name a higher version.
- If you export NFS, serve SMB with ksmbd, or mount CephFS or SMB shares, treat this update as highest priority. 88 records are in that code, including all 3 CRITICAL records.
- On 5.10, 5.15, 6.1 and 6.6, most of the records that affect the branch list no fix for it. Check the next stable release, and plan the move away from 5.10 and 5.15 before December 2026.
- Bluetooth and CAN products should take the update for
CVE-2026-99793,CVE-2026-99794,CVE-2026-99672andCVE-2026-99674. - KVM hosts with AMD processors should take the update and should update the kdump crash kernel at the same time.
- No record says a bug is being exploited. The action is a planned stable update, not an emergency change.
Frequently asked questions
How many Linux kernel CVEs were published from 15 to 21 November 2026?
NVD lists 516 records for the kernel CNA in that window, and none was rejected. The lore.kernel.org announcement list shows the same 516 announcements. All 516 records were read for this post.
Are any of this week’s CVEs CRITICAL?
Yes. 3 records have a CNA score of 9.0 or higher: CVE-2026-99304, CVE-2026-99306 and CVE-2026-99310. Two are in the CephFS client and one is in the kernel NFS server. 14 more records are HIGH. The other 499 records have no CNA score.
Which kernel version do I need?
Update to at least 6.12.118, 6.18.64 or 7.3.5 for your branch. 7.2 is end of life, so move to 7.3.5. On 5.10, 5.15, 6.1 and 6.6, stay on the newest release and check the next stable release for more fixes.
Are any of these CVEs being exploited?
No record says a bug is being exploited, and no record contains exploit code. Some records include a crash report from a test. Every record names a fix in a kernel release that is already available. The action is to schedule the stable update.
I use 5.10, 5.15, 6.1 or 6.6. Am I affected?
Probably yes for some records. Between 242 and 349 records affect each of these branches and list no fix for it. This includes the CRITICAL NFS server and Ceph records. If your product uses that code, ask your kernel vendor about a backport or plan a move to 6.12 or 6.18. kernel.org lists December 2026 as the projected end of life for 5.10 and 5.15.
Does a device that only mounts a network share need the update?
Yes, if it mounts an SMB or NFS share or a CephFS file system and the record matches its configuration. 17 records are in the SMB client and 6 are in the NFS client. The two CRITICAL Ceph records are also in client code. For several of these records, the data that starts the bug comes from the server.




