{"id":"CVE-2026-97902","published":"2026-09-25T11:17:17.000","lastModified":"2026-09-25T11:17:17.000","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nfs: don't return -EINVAL for successful nested thaw\n\nCommit 7366f8b6fc6a (\"fs: handle freezing from multiple devices\")\nreplaced the freeze_holders bitmask with per-holder counters to allow\nnested freezes. In the bitmask version, a thaw that released a shared\nhold while another holder remained returned 0. Since the rework,\nthaw_super_locked() drops the freeze reference via freeze_dec() but\nthen returns -EINVAL when other freezers remain, misinforming the\ncaller: the thaw did succeed, the superblock just stays frozen for the\nremaining holders.\n\nThis breaks bdev-initiated freezing. When a filesystem is frozen with\nFIFREEZE and additionally frozen via bdev_freeze() -- which nests by\ndesign, see fs_bdev_freeze() -- the subsequent bdev_thaw() receives\n-EINVAL from the holder op although its freeze reference was dropped,\nand therefore keeps bd_fsfreeze_count elevated. Then device-mapper's\nunlock_fs() ignores bdev_thaw()'s return value, so nothing rebalances\nthe count. After the user's FITHAW and umount, the block device can\nnever be mounted again:\n\n    dm-1: Can't mount, blockdev is frozen\n\nThere is no way for userspace to drop the leaked count; only\ndestroying the block device (or a reboot) recovers the device.\n\nReproducer (any kernel since v6.8):\n\n    dmsetup create dut --table \"0 $(blockdev --getsz \"$DEV\") linear $DEV 0\"\n    mkfs.ext4 /dev/mapper/dut\n    mount /dev/mapper/dut /mnt\n    fsfreeze --freeze /mnt      # freeze_ucount == 1\n    dmsetup suspend dut         # bd_fsfreeze_count == 1, ucount == 2\n    dmsetup resume dut          # ucount 2 -> 1, but thaw_super()\n                                # returns -EINVAL, so bdev_thaw()\n                                # keeps bd_fsfreeze_count at 1\n    fsfreeze --unfreeze /mnt    # filesystem thaws fine\n    umount /mnt\n    mount /dev/mapper/dut /mnt  # EBUSY, forever\n\nThe same happens with fsfreeze held across an LVM snapshot of the\norigin volume.\n\nfs_bdev_thaw()'s documentation already describes the intended\nsemantics: \"If this function returns zero it doesn't mean that the\nfilesystem is unfrozen as it may have been frozen multiple times\".\nRestore them by returning 0 when a nested thaw drops its hold while\nother freezers remain. Thawing without holding a freeze still fails\nwith -EINVAL as may_unfreeze() rejects that case before the reference\ncount is touched.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/111339816138f042a59778322f941edc26903484","tags":[]},{"url":"https://git.kernel.org/stable/c/76e478499913dc1f368aebe85b28b7d1265985ff","tags":[]},{"url":"https://git.kernel.org/stable/c/fe967191e5851ea79818c5fe4e781c3882139218","tags":[]},{"url":"https://git.kernel.org/stable/c/ff2a694f6613d54ebd77ba72583d4d94446f55bd","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}