RE: [Patch v4 13/18] security: Update security level of a process when modifying its dumpability

[Schaufler, Casey]

> -----Original Message-----
> From: Tim Chen [mailto:tim.c.chen@xxxxxxxxxxxxxxx]
> Sent: Tuesday, October 30, 2018 11:49 AM
> To: Jiri Kosina <jikos@xxxxxxxxxx>; Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> Cc: Tim Chen <tim.c.chen@xxxxxxxxxxxxxxx>; Tom Lendacky
> <thomas.lendacky@xxxxxxx>; Ingo Molnar <mingo@xxxxxxxxxx>; Peter
> Zijlstra <peterz@xxxxxxxxxxxxx>; Josh Poimboeuf <jpoimboe@xxxxxxxxxx>;
> Andrea Arcangeli <aarcange@xxxxxxxxxx>; David Woodhouse
> <dwmw@xxxxxxxxxxxx>; Andi Kleen <ak@xxxxxxxxxxxxxxx>; Hansen, Dave
> <dave.hansen@xxxxxxxxx>; Schaufler, Casey <casey.schaufler@xxxxxxxxx>;
> Mallick, Asit K <asit.k.mallick@xxxxxxxxx>; Arjan van de Ven
> <arjan@xxxxxxxxxxxxxxx>; Jon Masters <jcm@xxxxxxxxxx>; Waiman Long
> <longman9394@xxxxxxxxx>; linux-kernel@xxxxxxxxxxxxxxx; x86@xxxxxxxxxx

Added LSM mail list to the CC:

> Subject: [Patch v4 13/18] security: Update security level of a process when
> modifying its dumpability

"Level" isn't a good description of security characteristics
as it has been overloaded in too many ways already. More
on that later.

> When a process is made non-dumpable, the action implies a higher level
> of security implicitly as its memory is imposed with access restriction.

How is this "higher level" manifest?

> A call to update_process_security() is added to update security defenses
> according to a process's dumpability and its implied security level.

Can you describe the set of security attributes that are implied by the
process' dumpability? I would think that the dumpability would be an
artifact of these attributes, not the other way around.

> Architecture specific defenses is erected for threads in the process

nit: s/is/are/

> by calling arch_set_security(task, SECURITY_LEVEL_HIGH) or the defenses
> relaxed via arch_set_security(task, SECURITY_LEVEL_NORMAL).  Such defenses
> may incur extra overhead and is reserved for tasks needing high security.

nit: s/is/are/

> Signed-off-by: Tim Chen <tim.c.chen@xxxxxxxxxxxxxxx>
> ---
>  fs/exec.c                |  2 ++
>  include/linux/security.h |  6 ++++++
>  kernel/cred.c            |  5 ++++-
>  kernel/sys.c             |  1 +
>  security/security.c      | 31 +++++++++++++++++++++++++++++++
>  5 files changed, 44 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/exec.c b/fs/exec.c
> index 1ebf6e5..e70c8a7 100644
> --- a/fs/exec.c
> +++ b/fs/exec.c
> @@ -1366,6 +1366,8 @@ void setup_new_exec(struct linux_binprm * bprm)
>  	else
>  		set_dumpable(current->mm, SUID_DUMP_USER);
> 
> +	update_process_security(current);
> +
>  	arch_setup_new_exec();
>  	perf_event_exec();
>  	__set_task_comm(current, kbasename(bprm->filename), true);
> diff --git a/include/linux/security.h b/include/linux/security.h
> index 75f4156..469d05f 100644
> --- a/include/linux/security.h
> +++ b/include/linux/security.h
> @@ -61,6 +61,12 @@ struct mm_struct;
>  /* LSM Agnostic defines for sb_set_mnt_opts */
>  #define SECURITY_LSM_NATIVE_LABELS	1
> 
> +/* Security level */
> +#define SECURITY_NORMAL	0
> +#define SECURITY_HIGH	1

NAK: "NORMAL" and "HIGH" are meaningless in this context.
If you intend to differentiate between "dumpable" and "undumpable"
you could use those, although I would recommend something that
describes the reason the task is dumpable.

> +
> +extern int update_process_security(struct task_struct *task);
> +
>  struct ctl_table;
>  struct audit_krule;
>  struct user_namespace;
> diff --git a/kernel/cred.c b/kernel/cred.c
> index ecf0365..0806a74 100644
> --- a/kernel/cred.c
> +++ b/kernel/cred.c
> @@ -19,6 +19,7 @@
>  #include <linux/security.h>
>  #include <linux/binfmts.h>
>  #include <linux/cn_proc.h>
> +#include <linux/security.h>
> 
>  #if 0
>  #define kdebug(FMT, ...)						\
> @@ -445,8 +446,10 @@ int commit_creds(struct cred *new)
>  	    !uid_eq(old->fsuid, new->fsuid) ||
>  	    !gid_eq(old->fsgid, new->fsgid) ||
>  	    !cred_cap_issubset(old, new)) {
> -		if (task->mm)
> +		if (task->mm) {
>  			set_dumpable(task->mm, suid_dumpable);
> +			update_process_security(task);
> +		}
>  		task->pdeath_signal = 0;
>  		smp_wmb();
>  	}
> diff --git a/kernel/sys.c b/kernel/sys.c
> index cf5c675..c6f179a 100644
> --- a/kernel/sys.c
> +++ b/kernel/sys.c
> @@ -2293,6 +2293,7 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long,
> arg2, unsigned long, arg3,
>  			break;
>  		}
>  		set_dumpable(me->mm, arg2);
> +		update_process_security(me);
>  		break;
> 
>  	case PR_SET_UNALIGN:
> diff --git a/security/security.c b/security/security.c
> index 736e78d..12460f2 100644
> --- a/security/security.c
> +++ b/security/security.c
> @@ -28,6 +28,8 @@
>  #include <linux/personality.h>
>  #include <linux/backing-dev.h>
>  #include <linux/string.h>
> +#include <linux/coredump.h>
> +#include <linux/sched/signal.h>
>  #include <net/flow.h>
> 
>  #include <trace/events/initcall.h>
> @@ -1353,6 +1355,35 @@ int security_inode_getsecctx(struct inode *inode,
> void **ctx, u32 *ctxlen)
>  }
>  EXPORT_SYMBOL(security_inode_getsecctx);
> 
> +void __weak arch_set_security(struct task_struct *task,
> +			      unsigned int security_level)
> +{
> +}

This isn't an LSM hook and hence does not belong in this file.
arch_set_security() isn't descriptive, and is in fact a bad choice
as task_struct has a field "security". This function has nothing
to do with the task->security field, which is what I would expect
based on the name.

> +
> +int update_process_security(struct task_struct *task)

Again, this isn't an LSM hook and does not belong in this file.
Also again, "security" isn't descriptive in the name.

> +{
> +	unsigned long flags;
> +	struct task_struct *t;
> +	int security_level;
> +
> +	if (!task->mm)
> +		return -EINVAL;
> +
> +	if (!lock_task_sighand(task, &flags))
> +		return -ESRCH;
> +
> +	if (get_dumpable(task->mm) != SUID_DUMP_USER)
> +		security_level = SECURITY_HIGH;
> +	else
> +		security_level = SECURITY_NORMAL;
> +
> +	for_each_thread(task, t)
> +		arch_set_security(task, security_level);
> +
> +	unlock_task_sighand(task, &flags);
> +	return 0;
> +}
> +
>  #ifdef CONFIG_SECURITY_NETWORK
> 
>  int security_unix_stream_connect(struct sock *sock, struct sock *other, struct
> sock *newsk)
> --
> 2.9.4