Re: [PATCH 8/16] Masquerade the siginfo when sending a pid to a foreign namespace

[Pavel Emelianov <xemul@openvz.org>]

sukadev@us.ibm.com wrote:
> Pavel Emelianov [xemul@openvz.org] wrote:
> | When user send signal from (say) init namespace to any task in a sub
> | namespace the siginfo struct must not carry the sender's pid value, as
> | this value may refer to some task in the destination namespace and thus
> | may confuse the application.
> 
> Also, do you prevent signals to the child reaper of a container from within
> its container ? If so, can you show me where you handle it ? I can't
> seem to find it.
> 
> And I guess you do allow signals to the child-reaper of a container from
> its parent container.

See my comment below.

> | 
> | The consensus was to pretend in this case as if it is the kernel who
> | sends the signal.
> | 
> | The pid_ns_accessible() call is introduced to check this pid-to-ns
> | accessibility.
> | 
> | Signed-off-by: Pavel Emelianov <xemul@openvz.org>
> | 
> | ---
> | 
> |  include/linux/pid.h |   10 ++++++++++
> |  kernel/signal.c     |   34 ++++++++++++++++++++++++++++------
> |  2 files changed, 38 insertions(+), 6 deletions(-)
> | 
> | diff -upr linux-2.6.22-rc4-mm2.orig/include/linux/pid.h linux-2.6.22-rc4-mm2-2/include/linux/pid.h
> | --- linux-2.6.22-rc4-mm2.orig/include/linux/pid.h	2007-06-14 12:14:29.000000000 +0400
> | +++ linux-2.6.22-rc4-mm2-2/include/linux/pid.h	2007-07-04 19:00:38.000000000 +0400
> | @@ -83,6 +89,16 @@ extern void FASTCALL(detach_pid(struct t
> |  	return nr;
> |  }
> | 
> | +/*
> | + * checks whether the pid actually lives in the namespace ns, i.e. it was
> | + * created in this namespace or it was moved there.
> | + */
> | +
> | +static inline int pid_ns_accessible(struct pid_namespace *ns, struct pid *pid)
> | +{
> | +	return pid->numbers[pid->level].ns == ns;
> | +}
> | +
> |  #define do_each_pid_task(pid, type, task)				\
> |  	do {								\
> |  		struct hlist_node *pos___;				\
> | diff -upr linux-2.6.22-rc4-mm2.orig/kernel/signal.c linux-2.6.22-rc4-mm2-2/kernel/signal.c
> | --- linux-2.6.22-rc4-mm2.orig/kernel/signal.c	2007-07-04 19:00:38.000000000 +0400
> | +++ linux-2.6.22-rc4-mm2-2/kernel/signal.c	2007-07-04 19:00:38.000000000 +0400
> | @@ -1124,13 +1124,31 @@ EXPORT_SYMBOL_GPL(kill_pid_info_as_uid);
> |   * is probably wrong.  Should make it like BSD or SYSV.
> |   */
> | 
> | -static int kill_something_info(int sig, struct siginfo *info, int pid)
> | +static inline void masquerade_siginfo(struct pid_namespace *src_ns,
> | +		struct pid *tgt_pid, struct siginfo *info)
> | +{
> | +	if (tgt_pid != NULL && !pid_ns_accessible(src_ns, tgt_pid)) {
> | +		/*
> | +		 * current namespace is not seen from the taks we
> | +		 * want to send the signal to, so pretend as if it
> | +		 * is the kernel who does this to avoid pid messing
> | +		 * by the target
> | +		 */
> | +
> | +		info->si_pid = 0;
> | +		info->si_code = SI_KERNEL;
> | +	}
> | +}
> | +
> | +static int kill_something_info(int sig, struct siginfo *info, int pid_nr)
> |  {
> |  	int ret;
> | +	struct pid *pid;
> | +
> |  	rcu_read_lock();
> | -	if (!pid) {
> | +	if (!pid_nr) {
> |  		ret = kill_pgrp_info(sig, info, task_pgrp(current));
> | -	} else if (pid == -1) {
> | +	} else if (pid_nr == -1) {
> |  		int retval = 0, count = 0;
> |  		struct task_struct * p;
> 
> So what happens if we run "kill -s <sig> -1" from within a container ?
> Do you terminate all processes in the system or just the process in
> the container ?

That's the biggest problem in the whole set. I do not allow for
any signal to the namespaces init (and use "standart" init in my
experiences), since I have no ideas of how to make it look good. 

Checking for abilities in the sys_kill() is a solution, but why 
wasn't it such in the global init case? Why init checks for signals
in get_signal_to_deliver(). I have to think a bit more with this 
place. Maybe checking for permissions in sys_kill is a good solution.

On of the ideas I had is that the namespace's init has to accept 
all the signals with si_code == SI_KERNEL (this will include signals 
from parent namespaces as well), but the problem is that struct 
siginfo's do not reach the get_signal_to_deliver in 100% times. If 
we just could somehow push the siginfo to init, I would concern the 
problem to be solved.

Thanks,
Pavel

> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>