{"id":"CVE-2026-98992","kind":"vulnerability","aliases":["CVE-2026-98992"],"related":[],"title":"Linux kernel: recursive bind mount of unmounted nsfs or pidfd tree","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nnamespace: don't copy an unmounted mount tree\n\nmay_copy_tree() accepts a mount whose root is an nsfs or pidfs file\nbefore it checks whether that mount is mounted. This exception exists\nfor nsfs and pidfs mounts. They are part of no mount namespace so\nneither check_mnt() nor is_mounted() would let them through and\nbind-mounting a nsfsfd or a pidfd wouldn't work.\n\nProblem is that exception also admits a bind mount of such a file when\nit was lazily unmounted and is only kept alive by a file descriptor:\n\n  mount --bind /proc/self/ns/net /tmp/x\n  exec 3< /tmp/x\n  umount -l /tmp/x\n  mount --rbind /proc/self/fd/3 /tmp/y\n\nWith MS_REC or AT_RECURSIVE copy_tree() doesn't stop at the unmounted\nmount. It copies every mount that stayed attached to it as well. So if\nyou have a bunch of overmounts over the nsfsfd or the pidfd the whole\nunmounted stack is copied. That's a bug.\n\nSo refuse to walk an unmounted tree. Copying its root alone stays\npossible. So a bind mount of a namespace that is gone can still be\nrecreated from a file descriptor on the old one. The file descriptor\npins that mount and its root and __has_locked_children() only reads the\ndirect children which don't leave the parent's list without\nnamespace_sem. The internal mounts carry MNT_NS_INTERNAL instead of NULL\nand are still walked. Copying an unmounted tree recursively fails with\nEINVAL.","summary":"The Linux kernel let a recursive bind mount copy a lazily unmounted mount tree rooted at an nsfs or pidfd file, which it should not walk. The fix refuses recursive copies of unmounted trees, so the kernel no longer duplicates mounts that stayed attached to the unmounted one.","severity":"unknown","cvss":null,"epss":null,"exploited":false,"exploited_at":null,"products":[],"ecosystems":[],"tags":["infrastructure","linux-kernel","self-hosted"],"not_affected_if":[],"how_to_check":"- unknown","fix":"- Update the Linux kernel to a release that includes the upstream commit resolving this issue; the sources do not name a version.","status":"open","published_at":"2026-10-23T13:17:21+00:00","first_seen_at":"2026-10-23T21:57:33+00:00","triaged_at":"2026-10-25T04:56:05+00:00","updated_at":"2026-10-25T04:56:05+00:00","category":"infrastructure","confidence":0.4,"triage":{"at":"2026-10-25T04:56:05Z","model":"claude-haiku-5-5","output":{"fix":"- Update the Linux kernel to a release that includes the upstream commit resolving this issue; the sources do not name a version.","reach":"local","title":"Linux kernel: recursive bind mount of unmounted nsfs or pidfd tree","impact":"low","summary":"The Linux kernel let a recursive bind mount copy a lazily unmounted mount tree rooted at an nsfs or pidfd file, which it should not walk. The fix refuses recursive copies of unmounted trees, so the kernel no longer duplicates mounts that stayed attached to the unmounted one.","category":"infrastructure","nicknames":[],"who_cares":["linux-kernel","self-hosted"],"confidence":0.4,"needs_auth":"unknown","worst_case":"unknown","analyst_note":"Sources give only the NVD description; check the kernel stable commit and distro advisories for affected versions, the fixed version, and the real security impact.","how_to_check":"- unknown","needs_analyst":true,"not_affected_if":[],"who_is_affected":"Systems running a Linux kernel without the upstream fix that handle nsfs or pidfd bind mounts, typically hosts where unprivileged or container workloads can create mounts from file descriptors. The sources do not say which kernel versions are affected or which configurations are exposed, so teams should check their distribution's advisory.","user_interaction":"unknown","needs_special_setup":"unknown","severity_if_unknown":"unknown"},"cost_usd":0.000628,"input_tokens":3772,"output_tokens":502,"severity_before":"unknown"},"analyst_status":null,"analyst_note":null,"search":"'-2026':13A,16A '-98992':14A,17A '/proc/self/fd/3':190C '/proc/self/ns/net':180C '/tmp/x':181C,184C,187C '/tmp/y':191C '3':183C 'accept':94C 'admit':156C 'aliv':173C 'alon':252C 'also':155C 'attach':68B,213C 'bind':4A,31B,141C,158C,179C,257C 'bind-mount':140C 'bug':241C 'bunch':223C 'carri':309C 'check':107C,130C 'children':289C,294C 'copi':33B,55B,86C,92C,198C,208C,237C,249C,320C 'cve':12A,15A 'descriptor':177C,272C,279C 'direct':293C 'doesn':200C 'duplic':64B 'einval':327C 'everi':209C 'except':114C,154C 'exec':182C 'exist':115C 'fail':325C 'file':45B,104C,163C,176C,271C,278C 'fix':52B 'follow':78C 'gone':264C 'host':24B 'infrastructur':18B 'instead':313C 'intern':307C,312C 'kept':172C 'kernel':2A,21B,27B,61B,76C 'l':186C 'lazili':35B,167C 'leav':298C 'let':28B,136C 'linux':1A,20B,26B,75C 'linux-kernel':19B 'list':302C 'lock':288C 'longer':63B 'may':91C 'mnt':131C,310C 'mount':5A,32B,37B,65B,89C,96C,110C,112C,120C,126C,134C,142C,159C,178C,188C,206C,210C,258C,282C,308C 'ms':193C 'namespac':83C,127C,261C,304C 'neither':129C 'ns':311C 'nsfs':8A,42B,101C,117C 'nsfsfd':144C,228C 'null':315C 'old':275C 'one':72B,276C 'overmount':225C 'parent':300C 'part':123C 'pidf':103C,119C 'pidfd':10A,44B,147C,231C 'pin':280C 'possibl':254C 'problem':151C 'rbind':189C 'read':291C 'rec':194C 'recreat':268C 'recurs':3A,30B,54B,197C,324C 'refus':53B,243C 'resolv':82C 'root':39B,98C,251C,285C 'self':23B 'self-host':22B 'sem':305C 'stack':235C 'stay':67B,212C,253C 'still':266C,318C 'stop':202C 'tree':11A,38B,58B,90C,93C,199C,248C,323C 'umount':185C 'unmount':7A,36B,57B,71B,88C,168C,205C,234C,247C,322C 'vulner':79C 'walk':50B,245C,319C 'well':217C 'whether':108C 'whole':233C 'whose':97C 'without':303C 'work':150C 'would':135C 'wouldn':148C","nicknames":[],"mitigations":[],"who_affected":"Systems running a Linux kernel without the upstream fix that handle nsfs or pidfd bind mounts, typically hosts where unprivileged or container workloads can create mounts from file descriptors. The sources do not say which kernel versions are affected or which configurations are exposed, so teams should check their distribution's advisory.","tier":"C","verified":{},"analyst_revisit_at":null,"first_public_at":"2026-10-23T13:17:21+00:00","kev":null,"cvss_vector":null,"cvss_by":null,"poc_count":0,"sources":[{"source":"nvd","external_id":"CVE-2026-98992","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-98992","severity":null,"published_at":"2026-10-23T13:17:21+00:00","modified_at":"2026-10-23T13:17:21+00:00","first_seen_at":"2026-10-23T21:57:33+00:00","fetched_at":"2026-10-23T22:29:03+00:00","note":null}],"documents":[],"notice":"Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai"}