vfs: fix dumpable race in create_userns_hierarchy()
PATCH 1/1 vfs: fix dumpable race in create_userns_hierarchy()
All processes in the userns hierarchy share the same mm_struct due to CLONE_VM. When a child calls setresuid() in switch_ids(), the kernel clears the dumpable flag on the shared mm. The child immediately re-sets it via prctl(PR_SET_DUMPABLE, 1), but there is a window between the two where the parent may be opening /proc/PID/ns/user for the previous level's child. That open hits ptrace_may_access() which checks get_dumpable(mm) and, finding it zero, falls back to ptrace_has_cap(mm->user_ns, mode). Since mm->user_ns is init_user_ns (the mm was created by real root) and the parent is only ns-root, the capability check fails and the open returns -EACCES.
Fix this by extending the existing two-way socketpair handshake into a three-way handshake: the child signals the parent after becoming dumpable, then blocks until the parent has opened /proc/PID/ns/user and signals back. Only then does the child proceed to create_userns_hierarchy() for the next level. This guarantees no concurrent setresuid() can clear dumpable while the parent's open() is in flight.
--- src/vfs/utils.c | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+)
src/vfs/utils.c +20 -0
@@ -312,6 +312,18 @@ static int userns_fd_cb(void *data)if (ret <0 )return syserror ("failure: write to socketpair" ); +/* +* Wait for the parent to open our /proc/PID/ns/user before +* proceeding to create the next userns level. Without this, +* setresuid() in the next-level child clears the dumpable flag +* on the shared mm (CLONE_VM) and the parent's open() fails the +* ptrace_may_access() check with -EACCES because mm->user_ns is +* init_user_ns where userns processes lack CAP_SYS_PTRACE. +*/ + ret =read_nointr (h->fd_event, &c,1 ); +if (ret <0 ) +return syserror ("failure: read from socketpair" ); + ret =create_userns_hierarchy (++h);if (ret <0 )return syserror ("failure: userns level %d" , h->level);
@@ -377,6 +389,14 @@ int create_userns_hierarchy(struct userns_hierarchy *h) goto out_wait; } +/* Tell the child it can now proceed to create the next level. */ + bytes =write_nointr (fd_socket[0 ],"1" ,1 ); +if (bytes <0 ) { +kill (pid,SIGKILL ); +syserror ("failure: write to socketpair" ); + goto out_wait; + } + fret =0 ; out_wait:
---
base-commit: 3ded3e13c008326d197d11ac975049ed1f8ec922
change-id: 20260323-fix-vfs-bugs-b4f8a492d6d7
Quoted 19 lines · Christian Brauner
All processes in the userns hierarchy share the same
mm_structdue toCLONE_VM. When a child callssetresuid()inswitch_ids(), the kernel clears the dumpable flag on the shared mm. The child immediately re-sets it viaprctl(PR_SET_DUMPABLE, 1), but there is a window between the two where the parent may be opening/proc/PID/ns/userfor the previous level's child. That open hitsptrace_may_access()which checksget_dumpable(mm)and, finding it zero, falls back toptrace_has_cap(mm->user_ns, mode). Sincemm->user_nsisinit_user_ns(the mm was created by real root) and the parent is only ns-root, the capability check fails and the open returns-EACCES.Fix this by extending the existing two-way socketpair handshake into a three-way handshake: the child signals the parent after becoming dumpable, then blocks until the parent has opened
/proc/PID/ns/userand signals back. Only then does the child proceed tocreate_userns_hierarchy()for the next level. This guarantees no concurrentsetresuid()can clear dumpable while the parent'sopen()is in flight.
Quoted 2 lines
Signed-off-by Christian Brauner <[email protected]>---
Sorry, a few unforeseen things came up on my end and I lost track of this patch.
Quoted 47 lines
src/vfs/utils.c | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+)src/vfs/utils.c +20
-1@@ -312,6 +312,18 @@ static int userns_fd_cb(void *data)if (ret <0 )return syserror ("failure: write to socketpair" ); +/* +* Wait for the parent to open our /proc/PID/ns/user before +* proceeding to create the next userns level. Without this, +* setresuid() in the next-level child clears the dumpable flag +* on the shared mm (CLONE_VM) and the parent's open() fails the +* ptrace_may_access() check with -EACCES because mm->user_ns is +* init_user_ns where userns processes lack CAP_SYS_PTRACE. +*/ + ret =read_nointr (h->fd_event, &c,1 ); +if (ret <0 ) +return syserror ("failure: read from socketpair" ); + ret =create_userns_hierarchy (++h);if (ret <0 )return syserror ("failure: userns level %d" , h->level);@@ -377,6 +389,14 @@ int create_userns_hierarchy(struct userns_hierarchy *h) goto out_wait; } +/* Tell the child it can now proceed to create the next level. */ + bytes =write_nointr (fd_socket[0 ],"1" ,1 ); +if (bytes <0 ) { +kill (pid,SIGKILL ); +syserror ("failure: write to socketpair" ); + goto out_wait; + } + fret =0 ; out_wait: ---base-commit:
3ded3e13c008326d197d11ac975049ed1f8ec922
change-id: 20260323-fix-vfs-bugs-b4f8a492d6d7