[v4,3/3] prctl: Allow ptrace capable processes to change /proc/self/exe
diff mbox series

Message ID 20200701064906.323185-4-areber@redhat.com
State New
Headers show
Series
  • capabilities: Introduce CAP_CHECKPOINT_RESTORE
Related show

Commit Message

Adrian Reber July 1, 2020, 6:49 a.m. UTC
From: Nicolas Viennot <Nicolas.Viennot@twosigma.com>

Previously, the current process could only change the /proc/self/exe
link with local CAP_SYS_ADMIN.
This commit relaxes this restriction by permitting such change with
CAP_CHECKPOINT_RESTORE, and the ability to use ptrace.

With access to ptrace facilities, a process can do the following: fork a
child, execve() the target executable, and have the child use ptrace()
to replace the memory content of the current process. This technique
makes it possible to masquerade an arbitrary program as any executable,
even setuid ones.

Signed-off-by: Nicolas Viennot <Nicolas.Viennot@twosigma.com>
Signed-off-by: Adrian Reber <areber@redhat.com>
---
 include/linux/lsm_hook_defs.h |  1 +
 include/linux/security.h      |  6 ++++++
 kernel/sys.c                  | 12 ++++--------
 security/commoncap.c          | 26 ++++++++++++++++++++++++++
 security/security.c           |  5 +++++
 security/selinux/hooks.c      | 14 ++++++++++++++
 6 files changed, 56 insertions(+), 8 deletions(-)

Comments

Christian Brauner July 1, 2020, 8:55 a.m. UTC | #1
On Wed, Jul 01, 2020 at 08:49:06AM +0200, Adrian Reber wrote:
> From: Nicolas Viennot <Nicolas.Viennot@twosigma.com>
> 
> Previously, the current process could only change the /proc/self/exe
> link with local CAP_SYS_ADMIN.
> This commit relaxes this restriction by permitting such change with
> CAP_CHECKPOINT_RESTORE, and the ability to use ptrace.
> 
> With access to ptrace facilities, a process can do the following: fork a
> child, execve() the target executable, and have the child use ptrace()
> to replace the memory content of the current process. This technique
> makes it possible to masquerade an arbitrary program as any executable,
> even setuid ones.
> 
> Signed-off-by: Nicolas Viennot <Nicolas.Viennot@twosigma.com>
> Signed-off-by: Adrian Reber <areber@redhat.com>
> ---
>  include/linux/lsm_hook_defs.h |  1 +
>  include/linux/security.h      |  6 ++++++
>  kernel/sys.c                  | 12 ++++--------
>  security/commoncap.c          | 26 ++++++++++++++++++++++++++
>  security/security.c           |  5 +++++
>  security/selinux/hooks.c      | 14 ++++++++++++++
>  6 files changed, 56 insertions(+), 8 deletions(-)
> 
> diff --git a/include/linux/lsm_hook_defs.h b/include/linux/lsm_hook_defs.h
> index 0098852bb56a..90e51d5e093b 100644
> --- a/include/linux/lsm_hook_defs.h
> +++ b/include/linux/lsm_hook_defs.h
> @@ -211,6 +211,7 @@ LSM_HOOK(int, 0, task_kill, struct task_struct *p, struct kernel_siginfo *info,
>  	 int sig, const struct cred *cred)
>  LSM_HOOK(int, -ENOSYS, task_prctl, int option, unsigned long arg2,
>  	 unsigned long arg3, unsigned long arg4, unsigned long arg5)
> +LSM_HOOK(int, 0, prctl_set_mm_exe_file, struct file *exe_file)
>  LSM_HOOK(void, LSM_RET_VOID, task_to_inode, struct task_struct *p,
>  	 struct inode *inode)
>  LSM_HOOK(int, 0, ipc_permission, struct kern_ipc_perm *ipcp, short flag)
> diff --git a/include/linux/security.h b/include/linux/security.h
> index 2797e7f6418e..0f594eb7e766 100644
> --- a/include/linux/security.h
> +++ b/include/linux/security.h
> @@ -412,6 +412,7 @@ int security_task_kill(struct task_struct *p, struct kernel_siginfo *info,
>  			int sig, const struct cred *cred);
>  int security_task_prctl(int option, unsigned long arg2, unsigned long arg3,
>  			unsigned long arg4, unsigned long arg5);
> +int security_prctl_set_mm_exe_file(struct file *exe_file);
>  void security_task_to_inode(struct task_struct *p, struct inode *inode);
>  int security_ipc_permission(struct kern_ipc_perm *ipcp, short flag);
>  void security_ipc_getsecid(struct kern_ipc_perm *ipcp, u32 *secid);
> @@ -1124,6 +1125,11 @@ static inline int security_task_prctl(int option, unsigned long arg2,
>  	return cap_task_prctl(option, arg2, arg3, arg4, arg5);
>  }
>  
> +static inline int security_prctl_set_mm_exe_file(struct file *exe_file)
> +{
> +	return cap_prctl_set_mm_exe_file(exe_file);
> +}
> +
>  static inline void security_task_to_inode(struct task_struct *p, struct inode *inode)
>  { }
>  
> diff --git a/kernel/sys.c b/kernel/sys.c
> index 00a96746e28a..bb53e8408c63 100644
> --- a/kernel/sys.c
> +++ b/kernel/sys.c
> @@ -1851,6 +1851,10 @@ static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
>  	if (err)
>  		goto exit;
>  
> +	err = security_prctl_set_mm_exe_file(exe.file);
> +	if (err)
> +		goto exit;
> +
>  	/*
>  	 * Forbid mm->exe_file change if old file still mapped.
>  	 */
> @@ -2006,14 +2010,6 @@ static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data
>  	}
>  
>  	if (prctl_map.exe_fd != (u32)-1) {
> -		/*
> -		 * Make sure the caller has the rights to
> -		 * change /proc/pid/exe link: only local sys admin should
> -		 * be allowed to.
> -		 */
> -		if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
> -			return -EINVAL;
> -
>  		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
>  		if (error)
>  			return error;
> diff --git a/security/commoncap.c b/security/commoncap.c
> index 59bf3c1674c8..663d00fe2ecc 100644
> --- a/security/commoncap.c
> +++ b/security/commoncap.c
> @@ -1291,6 +1291,31 @@ int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
>  	}
>  }
>  
> +/**
> + * cap_prctl_set_mm_exe_file - Determine whether /proc/self/exe can be changed
> + * by the current process.
> + * @exe_file: The new exe file
> + * Returns 0 if permission is granted, -ve if denied.
> + *
> + * The current process is permitted to change its /proc/self/exe link via two policies:
> + * 1) The current user can do checkpoint/restore. At the time of this writing,
> + *    this means CAP_SYS_ADMIN or CAP_CHECKPOINT_RESTORE capable.
> + * 2) The current user can use ptrace.
> + *
> + * With access to ptrace facilities, a process can do the following:
> + * fork a child, execve() the target executable, and have the child use
> + * ptrace() to replace the memory content of the current process.
> + * This technique makes it possible to masquerade an arbitrary program as the
> + * target executable, even if it is setuid.
> + */
> +int cap_prctl_set_mm_exe_file(struct file *exe_file)
> +{
> +	if (checkpoint_restore_ns_capable(current_user_ns()))
> +		return 0;
> +
> +	return security_ptrace_access_check(current, PTRACE_MODE_ATTACH_REALCREDS);
> +}

What is the reason for having this be a new security hook? Doesn't look
like it needs to be unless I'm missing something. This just seems more
complex than it needs to be.

I might be wrong here but if you look at the callsites for
security_ptrace_access_check() right now, you'll see that it's only
called from kernel/ptrace.c in __ptrace_may_access() and that function
checks right at the top:

	/* Don't let security modules deny introspection */
	if (same_thread_group(task, current))
		return 0;

since you're passing in same_thread_group(current, current) you're
passing this check and never even hitting
security_ptrace_access_check(). So the contract seems to be (as is
obvious from the comment) that a task can't be denied ptrace
introspection. But if you're using security_ptrace_access_check(current)
here and _if_ there would be any lsm that would deny ptrace
introspection to current you'd suddenly introduce a callsite where
ptrace introspection is denied. That seems wrong. So either you meant to
do something else here or you really just want:

checkpoint_restore_ns_capable(current_user_ns())

and none of the rest. But I might be missing the big picture in this
patch.

> +	if (checkpoint_restore_ns_capable(current_user_ns()))
> +		return 0;
> +
>  /**
>   * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
>   * @mm: The VM space in which the new mapping is to be made
> @@ -1356,6 +1381,7 @@ static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
>  	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
>  	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
>  	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
> +	LSM_HOOK_INIT(prctl_set_mm_exe_file, cap_prctl_set_mm_exe_file),
>  	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
>  	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
>  	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
> diff --git a/security/security.c b/security/security.c
> index 2bb912496232..13a1ed32f9e3 100644
> --- a/security/security.c
> +++ b/security/security.c
> @@ -1790,6 +1790,11 @@ int security_task_prctl(int option, unsigned long arg2, unsigned long arg3,
>  	return rc;
>  }
>  
> +int security_prctl_set_mm_exe_file(struct file *exe_file)
> +{
> +	return call_int_hook(prctl_set_mm_exe_file, 0, exe_file);
> +}
> +
>  void security_task_to_inode(struct task_struct *p, struct inode *inode)
>  {
>  	call_void_hook(task_to_inode, p, inode);
> diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
> index ca901025802a..fca5581392b8 100644
> --- a/security/selinux/hooks.c
> +++ b/security/selinux/hooks.c
> @@ -4156,6 +4156,19 @@ static int selinux_task_kill(struct task_struct *p, struct kernel_siginfo *info,
>  			    secid, task_sid(p), SECCLASS_PROCESS, perm, NULL);
>  }
>  
> +static int selinux_prctl_set_mm_exe_file(struct file *exe_file)
> +{
> +	u32 sid = current_sid();
> +
> +	struct common_audit_data ad = {
> +		.type = LSM_AUDIT_DATA_FILE,
> +		.u.file = exe_file,
> +	};
> +
> +	return avc_has_perm(&selinux_state, sid, sid,
> +			    SECCLASS_FILE, FILE__EXECUTE_NO_TRANS, &ad);
> +}
> +
>  static void selinux_task_to_inode(struct task_struct *p,
>  				  struct inode *inode)
>  {
> @@ -7057,6 +7070,7 @@ static struct security_hook_list selinux_hooks[] __lsm_ro_after_init = {
>  	LSM_HOOK_INIT(task_getscheduler, selinux_task_getscheduler),
>  	LSM_HOOK_INIT(task_movememory, selinux_task_movememory),
>  	LSM_HOOK_INIT(task_kill, selinux_task_kill),
> +	LSM_HOOK_INIT(prctl_set_mm_exe_file, selinux_prctl_set_mm_exe_file),
>  	LSM_HOOK_INIT(task_to_inode, selinux_task_to_inode),
>  
>  	LSM_HOOK_INIT(ipc_permission, selinux_ipc_permission),
> -- 
> 2.26.2
>
Serge E. Hallyn July 2, 2020, 9:16 p.m. UTC | #2
On Wed, Jul 01, 2020 at 08:49:06AM +0200, Adrian Reber wrote:
> From: Nicolas Viennot <Nicolas.Viennot@twosigma.com>
> 
> Previously, the current process could only change the /proc/self/exe
> link with local CAP_SYS_ADMIN.
> This commit relaxes this restriction by permitting such change with
> CAP_CHECKPOINT_RESTORE, and the ability to use ptrace.
> 
> With access to ptrace facilities, a process can do the following: fork a
> child, execve() the target executable, and have the child use ptrace()
> to replace the memory content of the current process. This technique
> makes it possible to masquerade an arbitrary program as any executable,
> even setuid ones.
> 
> Signed-off-by: Nicolas Viennot <Nicolas.Viennot@twosigma.com>
> Signed-off-by: Adrian Reber <areber@redhat.com>

This is scary.  But I believe it is safe.

Reviewed-by: Serge Hallyn <serge@hallyn.com>

I am a bit curious about the implications of the selinux patch.
IIUC you are using the permission of the tracing process to
execute the file without transition, so this is a way to work
around the policy which might prevent the tracee from doing so.
Given that SELinux wants to be MAC, I'm not *quite* sure that's
considered kosher.  You also are skipping the PROCESS__PTRACE
to SECCLASS_PROCESS check which selinux_bprm_set_creds does later
on.  Again I'm just not quite sure what's considered normal there
these days.

Paul, do you have input there?

> ---
>  include/linux/lsm_hook_defs.h |  1 +
>  include/linux/security.h      |  6 ++++++
>  kernel/sys.c                  | 12 ++++--------
>  security/commoncap.c          | 26 ++++++++++++++++++++++++++
>  security/security.c           |  5 +++++
>  security/selinux/hooks.c      | 14 ++++++++++++++
>  6 files changed, 56 insertions(+), 8 deletions(-)
> 
> diff --git a/include/linux/lsm_hook_defs.h b/include/linux/lsm_hook_defs.h
> index 0098852bb56a..90e51d5e093b 100644
> --- a/include/linux/lsm_hook_defs.h
> +++ b/include/linux/lsm_hook_defs.h
> @@ -211,6 +211,7 @@ LSM_HOOK(int, 0, task_kill, struct task_struct *p, struct kernel_siginfo *info,
>  	 int sig, const struct cred *cred)
>  LSM_HOOK(int, -ENOSYS, task_prctl, int option, unsigned long arg2,
>  	 unsigned long arg3, unsigned long arg4, unsigned long arg5)
> +LSM_HOOK(int, 0, prctl_set_mm_exe_file, struct file *exe_file)
>  LSM_HOOK(void, LSM_RET_VOID, task_to_inode, struct task_struct *p,
>  	 struct inode *inode)
>  LSM_HOOK(int, 0, ipc_permission, struct kern_ipc_perm *ipcp, short flag)
> diff --git a/include/linux/security.h b/include/linux/security.h
> index 2797e7f6418e..0f594eb7e766 100644
> --- a/include/linux/security.h
> +++ b/include/linux/security.h
> @@ -412,6 +412,7 @@ int security_task_kill(struct task_struct *p, struct kernel_siginfo *info,
>  			int sig, const struct cred *cred);
>  int security_task_prctl(int option, unsigned long arg2, unsigned long arg3,
>  			unsigned long arg4, unsigned long arg5);
> +int security_prctl_set_mm_exe_file(struct file *exe_file);
>  void security_task_to_inode(struct task_struct *p, struct inode *inode);
>  int security_ipc_permission(struct kern_ipc_perm *ipcp, short flag);
>  void security_ipc_getsecid(struct kern_ipc_perm *ipcp, u32 *secid);
> @@ -1124,6 +1125,11 @@ static inline int security_task_prctl(int option, unsigned long arg2,
>  	return cap_task_prctl(option, arg2, arg3, arg4, arg5);
>  }
>  
> +static inline int security_prctl_set_mm_exe_file(struct file *exe_file)
> +{
> +	return cap_prctl_set_mm_exe_file(exe_file);
> +}
> +
>  static inline void security_task_to_inode(struct task_struct *p, struct inode *inode)
>  { }
>  
> diff --git a/kernel/sys.c b/kernel/sys.c
> index 00a96746e28a..bb53e8408c63 100644
> --- a/kernel/sys.c
> +++ b/kernel/sys.c
> @@ -1851,6 +1851,10 @@ static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
>  	if (err)
>  		goto exit;
>  
> +	err = security_prctl_set_mm_exe_file(exe.file);
> +	if (err)
> +		goto exit;
> +
>  	/*
>  	 * Forbid mm->exe_file change if old file still mapped.
>  	 */
> @@ -2006,14 +2010,6 @@ static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data
>  	}
>  
>  	if (prctl_map.exe_fd != (u32)-1) {
> -		/*
> -		 * Make sure the caller has the rights to
> -		 * change /proc/pid/exe link: only local sys admin should
> -		 * be allowed to.
> -		 */
> -		if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
> -			return -EINVAL;
> -
>  		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
>  		if (error)
>  			return error;
> diff --git a/security/commoncap.c b/security/commoncap.c
> index 59bf3c1674c8..663d00fe2ecc 100644
> --- a/security/commoncap.c
> +++ b/security/commoncap.c
> @@ -1291,6 +1291,31 @@ int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
>  	}
>  }
>  
> +/**
> + * cap_prctl_set_mm_exe_file - Determine whether /proc/self/exe can be changed
> + * by the current process.
> + * @exe_file: The new exe file
> + * Returns 0 if permission is granted, -ve if denied.
> + *
> + * The current process is permitted to change its /proc/self/exe link via two policies:
> + * 1) The current user can do checkpoint/restore. At the time of this writing,
> + *    this means CAP_SYS_ADMIN or CAP_CHECKPOINT_RESTORE capable.
> + * 2) The current user can use ptrace.
> + *
> + * With access to ptrace facilities, a process can do the following:
> + * fork a child, execve() the target executable, and have the child use
> + * ptrace() to replace the memory content of the current process.
> + * This technique makes it possible to masquerade an arbitrary program as the
> + * target executable, even if it is setuid.
> + */
> +int cap_prctl_set_mm_exe_file(struct file *exe_file)
> +{
> +	if (checkpoint_restore_ns_capable(current_user_ns()))
> +		return 0;
> +
> +	return security_ptrace_access_check(current, PTRACE_MODE_ATTACH_REALCREDS);
> +}
> +
>  /**
>   * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
>   * @mm: The VM space in which the new mapping is to be made
> @@ -1356,6 +1381,7 @@ static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
>  	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
>  	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
>  	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
> +	LSM_HOOK_INIT(prctl_set_mm_exe_file, cap_prctl_set_mm_exe_file),
>  	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
>  	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
>  	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
> diff --git a/security/security.c b/security/security.c
> index 2bb912496232..13a1ed32f9e3 100644
> --- a/security/security.c
> +++ b/security/security.c
> @@ -1790,6 +1790,11 @@ int security_task_prctl(int option, unsigned long arg2, unsigned long arg3,
>  	return rc;
>  }
>  
> +int security_prctl_set_mm_exe_file(struct file *exe_file)
> +{
> +	return call_int_hook(prctl_set_mm_exe_file, 0, exe_file);
> +}
> +
>  void security_task_to_inode(struct task_struct *p, struct inode *inode)
>  {
>  	call_void_hook(task_to_inode, p, inode);
> diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
> index ca901025802a..fca5581392b8 100644
> --- a/security/selinux/hooks.c
> +++ b/security/selinux/hooks.c
> @@ -4156,6 +4156,19 @@ static int selinux_task_kill(struct task_struct *p, struct kernel_siginfo *info,
>  			    secid, task_sid(p), SECCLASS_PROCESS, perm, NULL);
>  }
>  
> +static int selinux_prctl_set_mm_exe_file(struct file *exe_file)
> +{
> +	u32 sid = current_sid();
> +
> +	struct common_audit_data ad = {
> +		.type = LSM_AUDIT_DATA_FILE,
> +		.u.file = exe_file,
> +	};
> +
> +	return avc_has_perm(&selinux_state, sid, sid,
> +			    SECCLASS_FILE, FILE__EXECUTE_NO_TRANS, &ad);
> +}
> +
>  static void selinux_task_to_inode(struct task_struct *p,
>  				  struct inode *inode)
>  {
> @@ -7057,6 +7070,7 @@ static struct security_hook_list selinux_hooks[] __lsm_ro_after_init = {
>  	LSM_HOOK_INIT(task_getscheduler, selinux_task_getscheduler),
>  	LSM_HOOK_INIT(task_movememory, selinux_task_movememory),
>  	LSM_HOOK_INIT(task_kill, selinux_task_kill),
> +	LSM_HOOK_INIT(prctl_set_mm_exe_file, selinux_prctl_set_mm_exe_file),
>  	LSM_HOOK_INIT(task_to_inode, selinux_task_to_inode),
>  
>  	LSM_HOOK_INIT(ipc_permission, selinux_ipc_permission),
> -- 
> 2.26.2
Serge E. Hallyn July 2, 2020, 9:58 p.m. UTC | #3
On Wed, Jul 01, 2020 at 10:55:37AM +0200, Christian Brauner wrote:
> On Wed, Jul 01, 2020 at 08:49:06AM +0200, Adrian Reber wrote:
> > From: Nicolas Viennot <Nicolas.Viennot@twosigma.com>
> > 
> > Previously, the current process could only change the /proc/self/exe
> > link with local CAP_SYS_ADMIN.
> > This commit relaxes this restriction by permitting such change with
> > CAP_CHECKPOINT_RESTORE, and the ability to use ptrace.
> > 
> > With access to ptrace facilities, a process can do the following: fork a
> > child, execve() the target executable, and have the child use ptrace()
> > to replace the memory content of the current process. This technique
> > makes it possible to masquerade an arbitrary program as any executable,
> > even setuid ones.
> > 
> > Signed-off-by: Nicolas Viennot <Nicolas.Viennot@twosigma.com>
> > Signed-off-by: Adrian Reber <areber@redhat.com>
> > ---
> >  include/linux/lsm_hook_defs.h |  1 +
> >  include/linux/security.h      |  6 ++++++
> >  kernel/sys.c                  | 12 ++++--------
> >  security/commoncap.c          | 26 ++++++++++++++++++++++++++
> >  security/security.c           |  5 +++++
> >  security/selinux/hooks.c      | 14 ++++++++++++++
> >  6 files changed, 56 insertions(+), 8 deletions(-)
> > 
> > diff --git a/include/linux/lsm_hook_defs.h b/include/linux/lsm_hook_defs.h
> > index 0098852bb56a..90e51d5e093b 100644
> > --- a/include/linux/lsm_hook_defs.h
> > +++ b/include/linux/lsm_hook_defs.h
> > @@ -211,6 +211,7 @@ LSM_HOOK(int, 0, task_kill, struct task_struct *p, struct kernel_siginfo *info,
> >  	 int sig, const struct cred *cred)
> >  LSM_HOOK(int, -ENOSYS, task_prctl, int option, unsigned long arg2,
> >  	 unsigned long arg3, unsigned long arg4, unsigned long arg5)
> > +LSM_HOOK(int, 0, prctl_set_mm_exe_file, struct file *exe_file)
> >  LSM_HOOK(void, LSM_RET_VOID, task_to_inode, struct task_struct *p,
> >  	 struct inode *inode)
> >  LSM_HOOK(int, 0, ipc_permission, struct kern_ipc_perm *ipcp, short flag)
> > diff --git a/include/linux/security.h b/include/linux/security.h
> > index 2797e7f6418e..0f594eb7e766 100644
> > --- a/include/linux/security.h
> > +++ b/include/linux/security.h
> > @@ -412,6 +412,7 @@ int security_task_kill(struct task_struct *p, struct kernel_siginfo *info,
> >  			int sig, const struct cred *cred);
> >  int security_task_prctl(int option, unsigned long arg2, unsigned long arg3,
> >  			unsigned long arg4, unsigned long arg5);
> > +int security_prctl_set_mm_exe_file(struct file *exe_file);
> >  void security_task_to_inode(struct task_struct *p, struct inode *inode);
> >  int security_ipc_permission(struct kern_ipc_perm *ipcp, short flag);
> >  void security_ipc_getsecid(struct kern_ipc_perm *ipcp, u32 *secid);
> > @@ -1124,6 +1125,11 @@ static inline int security_task_prctl(int option, unsigned long arg2,
> >  	return cap_task_prctl(option, arg2, arg3, arg4, arg5);
> >  }
> >  
> > +static inline int security_prctl_set_mm_exe_file(struct file *exe_file)
> > +{
> > +	return cap_prctl_set_mm_exe_file(exe_file);
> > +}
> > +
> >  static inline void security_task_to_inode(struct task_struct *p, struct inode *inode)
> >  { }
> >  
> > diff --git a/kernel/sys.c b/kernel/sys.c
> > index 00a96746e28a..bb53e8408c63 100644
> > --- a/kernel/sys.c
> > +++ b/kernel/sys.c
> > @@ -1851,6 +1851,10 @@ static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
> >  	if (err)
> >  		goto exit;
> >  
> > +	err = security_prctl_set_mm_exe_file(exe.file);
> > +	if (err)
> > +		goto exit;
> > +
> >  	/*
> >  	 * Forbid mm->exe_file change if old file still mapped.
> >  	 */
> > @@ -2006,14 +2010,6 @@ static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data
> >  	}
> >  
> >  	if (prctl_map.exe_fd != (u32)-1) {
> > -		/*
> > -		 * Make sure the caller has the rights to
> > -		 * change /proc/pid/exe link: only local sys admin should
> > -		 * be allowed to.
> > -		 */
> > -		if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
> > -			return -EINVAL;
> > -
> >  		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
> >  		if (error)
> >  			return error;
> > diff --git a/security/commoncap.c b/security/commoncap.c
> > index 59bf3c1674c8..663d00fe2ecc 100644
> > --- a/security/commoncap.c
> > +++ b/security/commoncap.c
> > @@ -1291,6 +1291,31 @@ int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
> >  	}
> >  }
> >  
> > +/**
> > + * cap_prctl_set_mm_exe_file - Determine whether /proc/self/exe can be changed
> > + * by the current process.
> > + * @exe_file: The new exe file
> > + * Returns 0 if permission is granted, -ve if denied.
> > + *
> > + * The current process is permitted to change its /proc/self/exe link via two policies:
> > + * 1) The current user can do checkpoint/restore. At the time of this writing,
> > + *    this means CAP_SYS_ADMIN or CAP_CHECKPOINT_RESTORE capable.
> > + * 2) The current user can use ptrace.
> > + *
> > + * With access to ptrace facilities, a process can do the following:
> > + * fork a child, execve() the target executable, and have the child use
> > + * ptrace() to replace the memory content of the current process.
> > + * This technique makes it possible to masquerade an arbitrary program as the
> > + * target executable, even if it is setuid.
> > + */
> > +int cap_prctl_set_mm_exe_file(struct file *exe_file)
> > +{
> > +	if (checkpoint_restore_ns_capable(current_user_ns()))
> > +		return 0;
> > +
> > +	return security_ptrace_access_check(current, PTRACE_MODE_ATTACH_REALCREDS);
> > +}
> 
> What is the reason for having this be a new security hook? Doesn't look
> like it needs to be unless I'm missing something. This just seems more
> complex than it needs to be.

Yeah, agreed, and that in turn risks actually making it less safe.  Is
there a good reason not to do what Christian suggests?

> I might be wrong here but if you look at the callsites for
> security_ptrace_access_check() right now, you'll see that it's only
> called from kernel/ptrace.c in __ptrace_may_access() and that function
> checks right at the top:
> 
> 	/* Don't let security modules deny introspection */
> 	if (same_thread_group(task, current))
> 		return 0;
> 
> since you're passing in same_thread_group(current, current) you're
> passing this check and never even hitting
> security_ptrace_access_check(). So the contract seems to be (as is
> obvious from the comment) that a task can't be denied ptrace
> introspection. But if you're using security_ptrace_access_check(current)
> here and _if_ there would be any lsm that would deny ptrace
> introspection to current you'd suddenly introduce a callsite where
> ptrace introspection is denied. That seems wrong. So either you meant to
> do something else here or you really just want:
> 
> checkpoint_restore_ns_capable(current_user_ns())
> 
> and none of the rest. But I might be missing the big picture in this
> patch.
> 
> > +	if (checkpoint_restore_ns_capable(current_user_ns()))
> > +		return 0;
> > +
> >  /**
> >   * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
> >   * @mm: The VM space in which the new mapping is to be made
> > @@ -1356,6 +1381,7 @@ static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
> >  	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
> >  	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
> >  	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
> > +	LSM_HOOK_INIT(prctl_set_mm_exe_file, cap_prctl_set_mm_exe_file),
> >  	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
> >  	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
> >  	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
> > diff --git a/security/security.c b/security/security.c
> > index 2bb912496232..13a1ed32f9e3 100644
> > --- a/security/security.c
> > +++ b/security/security.c
> > @@ -1790,6 +1790,11 @@ int security_task_prctl(int option, unsigned long arg2, unsigned long arg3,
> >  	return rc;
> >  }
> >  
> > +int security_prctl_set_mm_exe_file(struct file *exe_file)
> > +{
> > +	return call_int_hook(prctl_set_mm_exe_file, 0, exe_file);
> > +}
> > +
> >  void security_task_to_inode(struct task_struct *p, struct inode *inode)
> >  {
> >  	call_void_hook(task_to_inode, p, inode);
> > diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
> > index ca901025802a..fca5581392b8 100644
> > --- a/security/selinux/hooks.c
> > +++ b/security/selinux/hooks.c
> > @@ -4156,6 +4156,19 @@ static int selinux_task_kill(struct task_struct *p, struct kernel_siginfo *info,
> >  			    secid, task_sid(p), SECCLASS_PROCESS, perm, NULL);
> >  }
> >  
> > +static int selinux_prctl_set_mm_exe_file(struct file *exe_file)
> > +{
> > +	u32 sid = current_sid();
> > +
> > +	struct common_audit_data ad = {
> > +		.type = LSM_AUDIT_DATA_FILE,
> > +		.u.file = exe_file,
> > +	};
> > +
> > +	return avc_has_perm(&selinux_state, sid, sid,
> > +			    SECCLASS_FILE, FILE__EXECUTE_NO_TRANS, &ad);
> > +}
> > +
> >  static void selinux_task_to_inode(struct task_struct *p,
> >  				  struct inode *inode)
> >  {
> > @@ -7057,6 +7070,7 @@ static struct security_hook_list selinux_hooks[] __lsm_ro_after_init = {
> >  	LSM_HOOK_INIT(task_getscheduler, selinux_task_getscheduler),
> >  	LSM_HOOK_INIT(task_movememory, selinux_task_movememory),
> >  	LSM_HOOK_INIT(task_kill, selinux_task_kill),
> > +	LSM_HOOK_INIT(prctl_set_mm_exe_file, selinux_prctl_set_mm_exe_file),
> >  	LSM_HOOK_INIT(task_to_inode, selinux_task_to_inode),
> >  
> >  	LSM_HOOK_INIT(ipc_permission, selinux_ipc_permission),
> > -- 
> > 2.26.2
> >
Paul Moore July 2, 2020, 10 p.m. UTC | #4
On Thu, Jul 2, 2020 at 5:16 PM Serge E. Hallyn <serge@hallyn.com> wrote:
> On Wed, Jul 01, 2020 at 08:49:06AM +0200, Adrian Reber wrote:
> > From: Nicolas Viennot <Nicolas.Viennot@twosigma.com>
> >
> > Previously, the current process could only change the /proc/self/exe
> > link with local CAP_SYS_ADMIN.
> > This commit relaxes this restriction by permitting such change with
> > CAP_CHECKPOINT_RESTORE, and the ability to use ptrace.
> >
> > With access to ptrace facilities, a process can do the following: fork a
> > child, execve() the target executable, and have the child use ptrace()
> > to replace the memory content of the current process. This technique
> > makes it possible to masquerade an arbitrary program as any executable,
> > even setuid ones.
> >
> > Signed-off-by: Nicolas Viennot <Nicolas.Viennot@twosigma.com>
> > Signed-off-by: Adrian Reber <areber@redhat.com>
>
> This is scary.  But I believe it is safe.
>
> Reviewed-by: Serge Hallyn <serge@hallyn.com>
>
> I am a bit curious about the implications of the selinux patch.
> IIUC you are using the permission of the tracing process to
> execute the file without transition, so this is a way to work
> around the policy which might prevent the tracee from doing so.
> Given that SELinux wants to be MAC, I'm not *quite* sure that's
> considered kosher.  You also are skipping the PROCESS__PTRACE
> to SECCLASS_PROCESS check which selinux_bprm_set_creds does later
> on.  Again I'm just not quite sure what's considered normal there
> these days.
>
> Paul, do you have input there?

I agree, the SELinux hook looks wrong.  Building on what Christian
said, this looks more like a ptrace operation than an exec operation.
Nicolas Viennot July 6, 2020, 5:13 p.m. UTC | #5
> > This is scary.  But I believe it is safe.
> >
> > Reviewed-by: Serge Hallyn <serge@hallyn.com>
> >
> > I am a bit curious about the implications of the selinux patch.
> > IIUC you are using the permission of the tracing process to execute
> > the file without transition, so this is a way to work around the
> > policy which might prevent the tracee from doing so.
> > Given that SELinux wants to be MAC, I'm not *quite* sure that's
> > considered kosher.  You also are skipping the PROCESS__PTRACE to
> > SECCLASS_PROCESS check which selinux_bprm_set_creds does later on.
> > Again I'm just not quite sure what's considered normal there these
> > days.
> >
> > Paul, do you have input there?
>
> I agree, the SELinux hook looks wrong.  Building on what Christian said, this looks more like a ptrace operation than an exec operation.

Serge, Paul, Christian,

I made a PoC to demonstrate the change of /proc/self/exe without CAP_SYS_ADMIN using only ptrace and execve.
You may find it here: https://github.com/nviennot/run_as_exe

What do you recommend to relax the security checks in the kernel when it comes to changing the exe link?

    Nico
Christian Brauner July 6, 2020, 5:44 p.m. UTC | #6
On Mon, Jul 06, 2020 at 05:13:35PM +0000, Nicolas Viennot wrote:
> > > This is scary.  But I believe it is safe.
> > >
> > > Reviewed-by: Serge Hallyn <serge@hallyn.com>
> > >
> > > I am a bit curious about the implications of the selinux patch.
> > > IIUC you are using the permission of the tracing process to execute
> > > the file without transition, so this is a way to work around the
> > > policy which might prevent the tracee from doing so.
> > > Given that SELinux wants to be MAC, I'm not *quite* sure that's
> > > considered kosher.  You also are skipping the PROCESS__PTRACE to
> > > SECCLASS_PROCESS check which selinux_bprm_set_creds does later on.
> > > Again I'm just not quite sure what's considered normal there these
> > > days.
> > >
> > > Paul, do you have input there?
> >
> > I agree, the SELinux hook looks wrong.  Building on what Christian said, this looks more like a ptrace operation than an exec operation.
> 
> Serge, Paul, Christian,
> 
> I made a PoC to demonstrate the change of /proc/self/exe without CAP_SYS_ADMIN using only ptrace and execve.
> You may find it here: https://github.com/nviennot/run_as_exe
> 
> What do you recommend to relax the security checks in the kernel when it comes to changing the exe link?

Looks fun! Yeah, so that this is possible is known afaict. But you're
not really circumventing the kernel check but are mucking with the EFL
by changing the auxv, right?

Originally, you needed to be userns root, i.e. only uid 0 could
change the /proc/self/exe link (cf. [1]). This was changed to
ns_capable(CAP_SYS_ADMIN) in [2].

The original reasoning in [1] is interesting as it basically already
points to your poc:

"Still note that updating exe-file link now doesn't require sys-resource
 capability anymore, after all there is no much profit in preventing
 setup own file link (there are a number of ways to execute own code --
 ptrace, ld-preload, so that the only reliable way to find which exactly
 code is executed is to inspect running program memory).  Still we
 require the caller to be at least user-namespace root user."

There were arguments being made that /proc/<pid>/exe needs to be sm that
userspace can have a decent amount of trust in but I believe that that's
not a great argument.

But let me dig a little into the original discussion and see what the
thread-model was.
At this point I'm starting to believe that it was people being cautios
but better be sure.

[1]: f606b77f1a9e ("prctl: PR_SET_MM -- introduce PR_SET_MM_MAP operation")
[2]: 4d28df6152aa ("prctl: Allow local CAP_SYS_ADMIN changing exe_file")
[3]: https://lore.kernel.org/patchwork/patch/697304/

Christian
Christian Brauner July 7, 2020, 3:45 p.m. UTC | #7
On Mon, Jul 06, 2020 at 07:44:38PM +0200, Christian Brauner wrote:
> On Mon, Jul 06, 2020 at 05:13:35PM +0000, Nicolas Viennot wrote:
> > > > This is scary.  But I believe it is safe.
> > > >
> > > > Reviewed-by: Serge Hallyn <serge@hallyn.com>
> > > >
> > > > I am a bit curious about the implications of the selinux patch.
> > > > IIUC you are using the permission of the tracing process to execute
> > > > the file without transition, so this is a way to work around the
> > > > policy which might prevent the tracee from doing so.
> > > > Given that SELinux wants to be MAC, I'm not *quite* sure that's
> > > > considered kosher.  You also are skipping the PROCESS__PTRACE to
> > > > SECCLASS_PROCESS check which selinux_bprm_set_creds does later on.
> > > > Again I'm just not quite sure what's considered normal there these
> > > > days.
> > > >
> > > > Paul, do you have input there?
> > >
> > > I agree, the SELinux hook looks wrong.  Building on what Christian said, this looks more like a ptrace operation than an exec operation.
> > 
> > Serge, Paul, Christian,
> > 
> > I made a PoC to demonstrate the change of /proc/self/exe without CAP_SYS_ADMIN using only ptrace and execve.
> > You may find it here: https://github.com/nviennot/run_as_exe
> > 
> > What do you recommend to relax the security checks in the kernel when it comes to changing the exe link?
> 
> Looks fun! Yeah, so that this is possible is known afaict. But you're
> not really circumventing the kernel check but are mucking with the EFL
> by changing the auxv, right?
> 
> Originally, you needed to be userns root, i.e. only uid 0 could
> change the /proc/self/exe link (cf. [1]). This was changed to
> ns_capable(CAP_SYS_ADMIN) in [2].
> 
> The original reasoning in [1] is interesting as it basically already
> points to your poc:
> 
> "Still note that updating exe-file link now doesn't require sys-resource
>  capability anymore, after all there is no much profit in preventing
>  setup own file link (there are a number of ways to execute own code --
>  ptrace, ld-preload, so that the only reliable way to find which exactly
>  code is executed is to inspect running program memory).  Still we
>  require the caller to be at least user-namespace root user."
> 
> There were arguments being made that /proc/<pid>/exe needs to be sm that
> userspace can have a decent amount of trust in but I believe that that's
> not a great argument.
> 
> But let me dig a little into the original discussion and see what the
> thread-model was.
> At this point I'm starting to believe that it was people being cautios
> but better be sure.

Ok, so the original patch proposal was presented in [4] in 2014. The
final version of that patch added the PR_SET_MM_MAP we know today. The
initial version presented in [4] did not require _any_ privilege.

So the reasoning for only placing the /proc/<pid>/exe link under
ns_capable(CAP_SYS_ADMIN) is very thin. to quote from [5]:

"Controlling exe_fd without privileges may turn out to be dangerous. At
 least things like tomoyo examine it for making policy decisions (see
 tomoyo_manager())."

So yes, tomoyo_get_exe() is what this was retained for apparently:

const char *tomoyo_get_exe(void)
{
	struct file *exe_file;
	const char *cp;
	struct mm_struct *mm = current->mm;

	if (!mm)
		return NULL;
	exe_file = get_mm_exe_file(mm);
	if (!exe_file)
		return NULL;

	cp = tomoyo_realpath_from_path(&exe_file->f_path);
	fput(exe_file);
	return cp;
}

The exe path is literally used in tomoyo_manager() to verify that you
are allowed to change policy. That seems like a bad idea to me but then
again, I don't know enough about Tomoyo. In any case, I think that means
we can't remove CAP_SYS_ADMIN because that would make things worse than
they are right now for Tomoyo but I also don't see why placing this
under ns_capable(CAP_SYS_ADMIN) || ns_capable(CAP_CHECKPOINT_RESTORE)
would make this any worse.

And Cyrill (and later in that thread Andrei) already mentioned it in [6]:
"@exe_fd is just a hint and as I mentioned if we have ptrace/preload
 rights there damn a lot of ways to inject own code into any program so
 that a user won't even notice ;)"

Another place where the exe file is relevant is for the coredump with
the -E option. But it only uses the path when generating the coredump
pattern and if that's a security issue than your poc shows that this can
already be achieved today.

Christian

> 
> [1]: f606b77f1a9e ("prctl: PR_SET_MM -- introduce PR_SET_MM_MAP operation")
> [2]: 4d28df6152aa ("prctl: Allow local CAP_SYS_ADMIN changing exe_file")
> [3]: https://lore.kernel.org/patchwork/patch/697304/

[4]: https://lore.kernel.org/lkml/20140703151102.842945837@openvz.org/ 
[5]: https://lore.kernel.org/lkml/CAGXu5jL3exT4j+8rjMv1O54uJWQ5UHL69Z-24b61rhXROqZamQ@mail.gmail.com/
[6]: https://lore.kernel.org/lkml/20140722203614.GF838@moon/
Cyrill Gorcunov July 7, 2020, 8:27 p.m. UTC | #8
On Tue, Jul 07, 2020 at 05:45:04PM +0200, Christian Brauner wrote:
...
> 
> Ok, so the original patch proposal was presented in [4] in 2014. The
> final version of that patch added the PR_SET_MM_MAP we know today. The
> initial version presented in [4] did not require _any_ privilege.
> 

True. I still think that relyng on /proc/<pid>/exe being immutable (or
guarded by caps) in a sake of security is a bit misleading, this link
only a hint without any guarantees of what code is being executed once
we pass cs:rip to userspace right after exec is completed. Nowadays I rather
think we might need to call audit_log() here or something similar to point
that exe link is changed (by criu or someone else) and simply notify
node's administrator, that's all. But as you pointed tomoyo may be
affected if we simply drops all caps from here. Thus I agree that
the new cap won't make situation worse.

Still I'm not in touch with kernel code for a couple of years already
and might be missing something obvious here.

Patch
diff mbox series

diff --git a/include/linux/lsm_hook_defs.h b/include/linux/lsm_hook_defs.h
index 0098852bb56a..90e51d5e093b 100644
--- a/include/linux/lsm_hook_defs.h
+++ b/include/linux/lsm_hook_defs.h
@@ -211,6 +211,7 @@  LSM_HOOK(int, 0, task_kill, struct task_struct *p, struct kernel_siginfo *info,
 	 int sig, const struct cred *cred)
 LSM_HOOK(int, -ENOSYS, task_prctl, int option, unsigned long arg2,
 	 unsigned long arg3, unsigned long arg4, unsigned long arg5)
+LSM_HOOK(int, 0, prctl_set_mm_exe_file, struct file *exe_file)
 LSM_HOOK(void, LSM_RET_VOID, task_to_inode, struct task_struct *p,
 	 struct inode *inode)
 LSM_HOOK(int, 0, ipc_permission, struct kern_ipc_perm *ipcp, short flag)
diff --git a/include/linux/security.h b/include/linux/security.h
index 2797e7f6418e..0f594eb7e766 100644
--- a/include/linux/security.h
+++ b/include/linux/security.h
@@ -412,6 +412,7 @@  int security_task_kill(struct task_struct *p, struct kernel_siginfo *info,
 			int sig, const struct cred *cred);
 int security_task_prctl(int option, unsigned long arg2, unsigned long arg3,
 			unsigned long arg4, unsigned long arg5);
+int security_prctl_set_mm_exe_file(struct file *exe_file);
 void security_task_to_inode(struct task_struct *p, struct inode *inode);
 int security_ipc_permission(struct kern_ipc_perm *ipcp, short flag);
 void security_ipc_getsecid(struct kern_ipc_perm *ipcp, u32 *secid);
@@ -1124,6 +1125,11 @@  static inline int security_task_prctl(int option, unsigned long arg2,
 	return cap_task_prctl(option, arg2, arg3, arg4, arg5);
 }
 
+static inline int security_prctl_set_mm_exe_file(struct file *exe_file)
+{
+	return cap_prctl_set_mm_exe_file(exe_file);
+}
+
 static inline void security_task_to_inode(struct task_struct *p, struct inode *inode)
 { }
 
diff --git a/kernel/sys.c b/kernel/sys.c
index 00a96746e28a..bb53e8408c63 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -1851,6 +1851,10 @@  static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
 	if (err)
 		goto exit;
 
+	err = security_prctl_set_mm_exe_file(exe.file);
+	if (err)
+		goto exit;
+
 	/*
 	 * Forbid mm->exe_file change if old file still mapped.
 	 */
@@ -2006,14 +2010,6 @@  static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data
 	}
 
 	if (prctl_map.exe_fd != (u32)-1) {
-		/*
-		 * Make sure the caller has the rights to
-		 * change /proc/pid/exe link: only local sys admin should
-		 * be allowed to.
-		 */
-		if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
-			return -EINVAL;
-
 		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
 		if (error)
 			return error;
diff --git a/security/commoncap.c b/security/commoncap.c
index 59bf3c1674c8..663d00fe2ecc 100644
--- a/security/commoncap.c
+++ b/security/commoncap.c
@@ -1291,6 +1291,31 @@  int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
 	}
 }
 
+/**
+ * cap_prctl_set_mm_exe_file - Determine whether /proc/self/exe can be changed
+ * by the current process.
+ * @exe_file: The new exe file
+ * Returns 0 if permission is granted, -ve if denied.
+ *
+ * The current process is permitted to change its /proc/self/exe link via two policies:
+ * 1) The current user can do checkpoint/restore. At the time of this writing,
+ *    this means CAP_SYS_ADMIN or CAP_CHECKPOINT_RESTORE capable.
+ * 2) The current user can use ptrace.
+ *
+ * With access to ptrace facilities, a process can do the following:
+ * fork a child, execve() the target executable, and have the child use
+ * ptrace() to replace the memory content of the current process.
+ * This technique makes it possible to masquerade an arbitrary program as the
+ * target executable, even if it is setuid.
+ */
+int cap_prctl_set_mm_exe_file(struct file *exe_file)
+{
+	if (checkpoint_restore_ns_capable(current_user_ns()))
+		return 0;
+
+	return security_ptrace_access_check(current, PTRACE_MODE_ATTACH_REALCREDS);
+}
+
 /**
  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
  * @mm: The VM space in which the new mapping is to be made
@@ -1356,6 +1381,7 @@  static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
+	LSM_HOOK_INIT(prctl_set_mm_exe_file, cap_prctl_set_mm_exe_file),
 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
diff --git a/security/security.c b/security/security.c
index 2bb912496232..13a1ed32f9e3 100644
--- a/security/security.c
+++ b/security/security.c
@@ -1790,6 +1790,11 @@  int security_task_prctl(int option, unsigned long arg2, unsigned long arg3,
 	return rc;
 }
 
+int security_prctl_set_mm_exe_file(struct file *exe_file)
+{
+	return call_int_hook(prctl_set_mm_exe_file, 0, exe_file);
+}
+
 void security_task_to_inode(struct task_struct *p, struct inode *inode)
 {
 	call_void_hook(task_to_inode, p, inode);
diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
index ca901025802a..fca5581392b8 100644
--- a/security/selinux/hooks.c
+++ b/security/selinux/hooks.c
@@ -4156,6 +4156,19 @@  static int selinux_task_kill(struct task_struct *p, struct kernel_siginfo *info,
 			    secid, task_sid(p), SECCLASS_PROCESS, perm, NULL);
 }
 
+static int selinux_prctl_set_mm_exe_file(struct file *exe_file)
+{
+	u32 sid = current_sid();
+
+	struct common_audit_data ad = {
+		.type = LSM_AUDIT_DATA_FILE,
+		.u.file = exe_file,
+	};
+
+	return avc_has_perm(&selinux_state, sid, sid,
+			    SECCLASS_FILE, FILE__EXECUTE_NO_TRANS, &ad);
+}
+
 static void selinux_task_to_inode(struct task_struct *p,
 				  struct inode *inode)
 {
@@ -7057,6 +7070,7 @@  static struct security_hook_list selinux_hooks[] __lsm_ro_after_init = {
 	LSM_HOOK_INIT(task_getscheduler, selinux_task_getscheduler),
 	LSM_HOOK_INIT(task_movememory, selinux_task_movememory),
 	LSM_HOOK_INIT(task_kill, selinux_task_kill),
+	LSM_HOOK_INIT(prctl_set_mm_exe_file, selinux_prctl_set_mm_exe_file),
 	LSM_HOOK_INIT(task_to_inode, selinux_task_to_inode),
 
 	LSM_HOOK_INIT(ipc_permission, selinux_ipc_permission),