CVE-2026-93197
UnknownSummary
A bug in the Linux kernel's memcg subsystem causes LRU size accounting to be copied to the parent instead of moved when a memory cgroup is offlined, leaving stale counters on the child. This makes the LRU (or MGLRU) scanner read phantom sizes and grind through empty lists on dying cgroups, over-budgeting limits and wasting CPU. On one host, counters described 476 GiB of pages—1.89x the machine's RAM.
Risk Assessment
Organizations may experience excessive CPU consumption from futile LRU scanning on dead cgroups and incorrect memory management decisions due to inflated counters. This can degrade performance and, in extreme cases, disrupt memory reclaim behavior.
Recommendation
Update the Linux kernel to a version containing the fix for CVE-2026-93197, noting it depends on prior commit bf4ade7dbd76 ("memcg: keep folio's objcg same as its node") and must not be backported ahead of it. After updating, verify system stability and CPU usage on hosts with heavy memory cgroup usage.
Other vulnerabilities in Linux kernel
See all- CVE-2026-93204Unknown
In the Linux kernel, the batman-adv module's dat handling updated MAC addresses using a simple copy, which could be partially observed by parallel readers. A reader might transport a half-updated MAC address over the network or use it in ARP responses, poisoning the ARP cache. The fix uses atomic64_t to store the 48-bit MAC address, ensuring readers see either the old or new address, never a mixture.
- CVE-2026-93202Unknown
A bug in the Linux kernel's I3C subsystem causes recursive locking during registration of new I3C devices. The function i3c_master_register_new_i3c_devs() registers devices while holding i3c_bus_normaluse_lock(), and device_register() can immediately trigger a probe callback that tries to acquire the same lock again, leading to a deadlock. The fix separates device creation from registration and uses the maintenance lock.
- CVE-2026-93201Unknown
In the Linux kernel, the dm-pcache module did not validate cache segment IDs from persistent memory before indexing into cache->segments[]. With only CRC protection using a public seed, an attacker with CAP_SYS_ADMIN supplying the cache device could cause out-of-bounds reads and writes via crafted metadata. The fix adds validation and rejects invalid segment IDs with -EIO.
- CVE-2026-93200Unknown
The Linux kernel's i3c master subsystem had a use-after-free vulnerability in the master->this pointer. Sysfs attribute callbacks dereferenced master->this, which was freed in i3c_master_detach_free_devs() before the master device was released. The fix keeps master->this alive until i3c_masterdev_release() and resets it on error paths.
- CVE-2026-93199Unknown
In the Linux kernel, the i3c_master_search_i3c_dev_duplicate() function that searches for duplicate I3C devices can incorrectly match the master device itself (master->this), returning the controller as a duplicate. Since the controller is not a target device, it cannot be a duplicate of one. The fix excludes master->this from matching.
- CVE-2026-93198Unknown
In the Linux kernel, the dm-pcache module does not validate the persisted dirty_tail chain at load. A crafted image, whose on-media fields are authenticated only by a crc32c with a fixed seed, can aim dirty_tail at a chain of last ksets that never terminates, causing cache_writeback_fn() to re-arm itself forever. The fix adds validation of the dirty_tail chain at load with a hop cap.
- CVE-2026-93196Unknown
A use-after-free vulnerability was found in the Linux kernel's virtio_pmem driver for NVDIMM persistent memory. The issue occurs when a request has already been freed by the submitter but is still reachable via the virtqueue, leading to waking up freed memory during interrupt handling.
- CVE-2026-93195Unknown
In the Linux kernel, the drm/bridge synopsys dw-dp driver initializes and registers the DisplayPort AUX channel during dw_dp_bind(), but never unregisters it. This may lead to resource leaks and/or use-after-free. The fix adds the missing dw_dp_unbind() function to allow cleanup.
- CVE-2026-93194Unknown
In the Linux kernel, the drm/rockchip dw_dp driver initializes and registers core resources such as the DisplayPort AUX channel during dw_dp_bind(), but never unregisters them. This leads to memory leaks and/or use-after-free, as shown in a KASAN report. The issue is fixed by using the exported dw_dp_unbind() function in the component's unbind() callback and in its bind() error path.
- CVE-2026-93193Unknown
In the Linux kernel, the drm/rockchip analogix_dp driver has an OF node reference leak in rockchip_dp_drm_encoder_enable(), because of_get_child_by_name() does not call of_node_put() symmetrically. The fix uses __free(device_node) to automatically manage reference release.
Original NVD description (English source)
In the Linux kernel, the following vulnerability has been resolved: memcg: move LRU size accounting on reparenting instead of copying it When a memory cgroup is offlined its LRU folios are reparented to the parent. lruvec_reparent_lru() splices the child's lists into the parent's and credits the parent with the child's per-zone lru_zone_size[], but never clears the child's copy, so the size is copied rather than moved. lru_gen_reparent_memcg() does the same for MGLRU. The parent is left correct, credited with exactly the folios it took over. The stale value sits on the child and nothing will correct it: folio->memcg_data now resolves to the parent, so every later update_lru_size() for those folios goes there. Dying cgroups are not freed immediately and mem_cgroup_iter() still walks them, so shrink_lruvec() keeps being called on them. get_scan_count() reads the phantom counter through lruvec_lru_size() and the scan loop then grinds through nr[] in SWAP_CLUSTER_MAX steps against an empty list, for as long as the dead cgroup lives. Under MGLRU the MGLRU scanner runs instead, but count_shadow_nodes() sums all of NR_LRU_LISTS through lruvec_lru_size() and over-budgets the shadow node limit just the same. On one 251 GiB host a sweep of every mz->lru_zone_size[] found 380 counters describing folios on no list at all: 124777314 pages, 476 GiB, 1.89x the machine's RAM, across 57 cgroups. All were on memcgs with CSS_DYING set and CSS_ONLINE clear, and parent/child pairs reported byte-identical sizes. LRU_UNEVICTABLE needs its size moved too. Its list is deliberately not spliced because lruvec_init() poisons the head - the unevictable LRU is imaginary and folios are never threaded on it - but the size is kept by lruvec_add_folio()/lruvec_del_folio() and those folios account to the parent from here on. This depends on commit bf4ade7dbd76 ("memcg: keep folio's objcg same as its node") and must not be backported ahead of it. Without that invariant a folio's objcg can belong to another node, so a folio already spliced onto the parent's list can still resolve to the child's lruvec until the objcg's node is reparented in a later iteration of memcg_reparent_objcgs(); clearing the child's counter early then lets lruvec_del_folio() underflow it and trip the WARN_ONCE()/VM_BUG_ON() in mem_cgroup_update_lru_size().