From 73b40395c71c510b82fd6104416ecde1e5ac6418 Mon Sep 17 00:00:00 2001 From: Kir Kolyshkin Date: Tue, 6 Oct 2026 23:23:56 -0700 Subject: [PATCH] mount: add OPEN_TREE_DROP_MNTNS_MOUNTS A recursive open_tree(OPEN_TREE_CLONE) copies the nsfs mounts of any mount namespaces pinned below the source. When the clone is attached inside a mount namespace younger than a pinned one, move_mount(2) fails with ELOOP from check_for_nsfs_mounts(), per the rule from commit 8823c079ba71 ("vfs: Add setns support for the mount namespace") that prevents mount namespace reference loops. This breaks container runtimes that use OPEN_TREE_NAMESPACE, such as crun [1]. The new namespace is younger than everything else, so bind mounting e.g. the host root fails on any host that pins a mount namespace (snapd does, under /run/snapd/ns). By the time it fails, setns() has already run and userspace cannot recover: the host tree is out of reach, and the offending mounts cannot be unmounted from the detached copy. copy_mnt_ns() and create_new_namespace() already leave these mounts behind; only get_detached_copy() copies them. Add an open_tree() flag to drop them from the clone, as suggested by Aleksa [2]. Locked nsfs mounts are dropped the same way copy_mnt_ns() does it, so nothing new is exposed. Keep it opt-in: open_tree(OPEN_TREE_CLONE) plus move_mount(2) is how mount --rbind works through a file descriptor, and in the caller's own or an older namespace such mounts remain usable. The flag requires AT_RECURSIVE, since a non-recursive clone has no submounts to drop, and OPEN_TREE_CLONE or OPEN_TREE_NAMESPACE. With the latter it is a no-op, so a runtime can pass it with any recursive open_tree() call. Link: https://github.com/containers/crun/issues/2262 [1] Link: https://lore.kernel.org/all/2026-01-07-oldest-grim-captions-spills-ywC2O3@cyphar.com/ [2] Signed-off-by: Kir Kolyshkin Link: https://patch.msgid.link/20261006-nsfs-prune-rfc-v3-1-adb97ea52868@gmail.com Signed-off-by: Christian Brauner (Amutable) --- fs/namespace.c | 16 ++++++++++++++-- include/uapi/linux/mount.h | 1 + 2 files changed, 15 insertions(+), 2 deletions(-) diff --git a/fs/namespace.c b/fs/namespace.c index 3ed3766f3a18..15c3c68b7f6b 100644 --- a/fs/namespace.c +++ b/fs/namespace.c @@ -3165,7 +3165,8 @@ static struct mnt_namespace *get_detached_copy(const struct path *path, unsigned ns->seq_origin = src_mnt_ns->ns.ns_id; } - mnt = __do_loopback(path, (flags & AT_RECURSIVE), CL_COPY_MNT_NS_FILE); + mnt = __do_loopback(path, (flags & AT_RECURSIVE), + (flags & OPEN_TREE_DROP_MNTNS_MOUNTS) ? 0 : CL_COPY_MNT_NS_FILE); if (IS_ERR(mnt)) { emptied_ns = ns; return ERR_CAST(mnt); @@ -3317,7 +3318,18 @@ static struct file *vfs_open_tree(int dfd, const char __user *filename, unsigned if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | - OPEN_TREE_CLOEXEC | OPEN_TREE_NAMESPACE)) + OPEN_TREE_CLOEXEC | OPEN_TREE_NAMESPACE | + OPEN_TREE_DROP_MNTNS_MOUNTS)) + return ERR_PTR(-EINVAL); + + /* + * Only meaningful when a tree is copied recursively: a non-recursive + * clone has no submounts to drop. OPEN_TREE_NAMESPACE never copies + * pinned mount namespaces, so there the flag is a no-op. + */ + if ((flags & OPEN_TREE_DROP_MNTNS_MOUNTS) && + (!(flags & AT_RECURSIVE) || + !(flags & (OPEN_TREE_CLONE | OPEN_TREE_NAMESPACE)))) return ERR_PTR(-EINVAL); if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE | OPEN_TREE_NAMESPACE)) == diff --git a/include/uapi/linux/mount.h b/include/uapi/linux/mount.h index 2204708dbf7a..ac865edac517 100644 --- a/include/uapi/linux/mount.h +++ b/include/uapi/linux/mount.h @@ -63,6 +63,7 @@ */ #define OPEN_TREE_CLONE (1 << 0) /* Clone the target tree and attach the clone */ #define OPEN_TREE_NAMESPACE (1 << 1) /* Clone the target tree into a new mount namespace */ +#define OPEN_TREE_DROP_MNTNS_MOUNTS (1 << 2) /* Drop mntns mounts from the clone */ #define OPEN_TREE_CLOEXEC O_CLOEXEC /* Close the file on execve() */ /*