{"id":"CVE-2026-98996","title":"fs: don't silently unmount busy mounts","summary":"In the Linux kernel, the following vulnerability has been resolved:\n\nfs: don't silently unmount busy mounts\n\nThe propagate_mount_busy() helper exists to decide whether a synchronous\ncan suceed. It does that by looking at the copies of th…","severity":"none","vendor":"Linux","product":"Linux","affected":["Linux >= 99b19d16471e9c3faa85cad38abc9cbbe04c6d55 < bf2c8dec26fb792876b5e9bde7030c45d8be3908","Linux >= 99b19d16471e9c3faa85cad38abc9cbbe04c6d55 < a210ac218c307513d43e98a08f23c2dfbeeebc2f","Linux >= 99b19d16471e9c3faa85cad38abc9cbbe04c6d55 < 6857e209015f95be7a8f19a9cd5dac345b9b9859","Linux dc5ce95b0252d143497283f7f659840d4f8a79d0","Linux fdb8f10499924d96b8eb60cc1354b278f6fcf0e9","Linux bb4fbf094b440a9209ed88c8681960d4b26eec0f","Linux b43f81ef0babcc07db1eaaece37c989204e4ff00","Linux 81311d1698c0c87c972ba71d3ce44a617b52ea21","Linux >= 4.1.43 < 4.2","Linux >= 4.4.78 < 4.5","Linux >= 4.9.39 < 4.10","Linux >= 4.11.12 < 4.12","Linux >= 4.12.3 < 4.13","Linux 4.13"],"published":"2026-10-23","updated":"2026-10-23","sourceUpdated":"2026-10-23T12:20:15.188Z","source":"CVEORG","sourceUrl":"https://www.cve.org/CVERecord?id=CVE-2026-98996","references":[{"url":"https://git.kernel.org/stable/c/bf2c8dec26fb792876b5e9bde7030c45d8be3908"},{"url":"https://git.kernel.org/stable/c/a210ac218c307513d43e98a08f23c2dfbeeebc2f"},{"url":"https://git.kernel.org/stable/c/6857e209015f95be7a8f19a9cd5dac345b9b9859"}],"tags":["cve.org"],"ingestedAt":"2026-10-23T13:24:55.543Z","slug":"CVE-2026-98996","body":"## Overview\n\nIn the Linux kernel, the following vulnerability has been resolved:\n\nfs: don't silently unmount busy mounts\n\nThe propagate_mount_busy() helper exists to decide whether a synchronous\ncan suceed. It does that by looking at the copies of the victims in the\nmounts its parent propagates to. One of those functions from propagation\nhell.\n\nIt skips a copy that has child mounts except for the case of a single\nchild mount covering the copy's root. That algorithm used to be exact.\nIt isn't anymore.\n\nOriginally only childless copies were unmounted during propagation.\nAnd 1064f874abc0 (\"mnt: Tuck mounts under others instead of creating\nshadow/side mounts.\") ended up adding covered mounts to both sides in\none go.\n\nAlong came 99b19d16471e (\"mnt: In propgate_umount handle visiting mounts\nin any order\") and that started to unmount more than before:\n\n[1]: A copy of a mount is unmounted when each of its children is either\n     its overmount or a copy of the victim itself. In the former case\n     the overmount gets reparented. In the latter case the copy of the\n     victim will end up being unmounted.\n\nSo propagate_umount() ended up being rewritten without adjusting\npropagate_mount_busy(). And currently propagate_mount_busy() doesn't\nlook at copies of the victim that have two or more children nor a single\nchild that is not the overmount.\n\nSo any such copy in [1] is not checked for\nreferences. The result is that a synchronous unmount succeeds, pulling\nout a mount that is still in use.\n\nA container can run into this with its own mounts. The host shares a\ntree with the container and the host has a mount in that tree:\n\n  mount -t tmpfs none /mnt\n  mount --make-shared /mnt\n  mount -t tmpfs none /mnt/a\n\nThe container's /mnt is a slave copy of the host mount. The container\nnow takes a detached copy of the tree and mounts that copy beneath its\npropagated copy of /mnt/a and keeps the file descriptor to the tree\nopen:\n\n  unshare -m --propagation unchanged\n  mount --make-rslave /\n  fd = open_tree(AT_FDCWD, \"/mnt\", OPEN_TREE_CLONE | AT_RECURSIVE)\n  move_mount(fd, \"\", AT_FDCWD, \"/mnt/a\", MOVE_MOUNT_BENEATH)\n\n  grep /mnt /proc/self/mountinfo\n  775 355 ... /mnt ...     master:646\n  777 775 ... /mnt/a ...   master:646\n  776 777 ... /mnt/a ...   master:647\n  778 777 ... /mnt/a/a ... master:647\n\nThe now attached tree (777) is now mounted beneath the container's copy\nof /mnt/a. That's also where the host mount sits (as seen from the host\nmount namespace). So the propagated copy (776) of the host's mount is on\ntop of its root and the copy of that (778) is inside it. Now the host\nruns:\n\n  umount /mnt/a\n\nand it succeeds. The moved tree tree and the copy inside of it are gone\nfrom the container. Now only the propagated copy is left and it got\nreparented to where it was before:\n\n  grep /mnt /proc/self/mountinfo\n  775 355 ... /mnt ... master:646\n  776 775 ... /mnt/a ...\n\nBut the file descriptor still refers to the moved tree. So a synchronous\numount detached a mount that is still in use.\n\nTeach propagate_mount_busy() the same checks that propagate_umount()\ndoes. It now checks for a single victim mount without children\nfor which it can assume that MNT_LOCKED is cleared on every copy like\npropagate_umount() does. The copies form chains. IOW, when a copy itself\nreceives propagation from the victim's parent then the mount at the same\nmountpoint inside that copy is a copy as well. Whether a mount receives\npropagation from the victim's can simply be checked by walking its\nmasters. Walk each chain once from its top so we are sure that nested\ncopies stay linear. Check every reference once of every copy that gets\nunmounted.\n\n## Affected\n\n- `Linux >= 99b19d16471e9c3faa85cad38abc9cbbe04c6d55 < bf2c8dec26fb792876b5e9bde7030c45d8be3908`\n- `Linux >= 99b19d16471e9c3faa85cad38abc9cbbe04c6d55 < a210ac218c307513d43e98a08f23c2dfbeeebc2f`\n- `Linux >= 99b19d16471e9c3faa85cad38abc9cbbe04c6d55 < 6857e209015f95be7a8f19a9cd5dac345b9b9859`\n- `Linux dc5ce95b0252d143497283f7f659840d4f8a79d0`\n- `Linux fdb8f10499924d96b8eb60cc1354b278f6fcf0e9`\n- `Linux bb4fbf094b440a9209ed88c8681960d4b26eec0f`\n- `Linux b43f81ef0babcc07db1eaaece37c989204e4ff00`\n- `Linux 81311d1698c0c87c972ba71d3ce44a617b52ea21`\n- `Linux >= 4.1.43 < 4.2`\n- `Linux >= 4.4.78 < 4.5`\n- `Linux >= 4.9.39 < 4.10`\n- `Linux >= 4.11.12 < 4.12`\n- `Linux >= 4.12.3 < 4.13`\n- `Linux 4.13`\n\n## Remediation\n\nRefer to the linked advisories for vendor-supplied fixes and affected version ranges.","depth":"sunlit","depthScore":3,"depthScoreParts":{"impact":2.8,"likelihood":0,"exploitation":0,"ransomware":0},"changes":[]}