[CRIU] [PATCH v2] net: Create child with CLONE_VM in prepare_net_namespaces()

Andrei Vagin avagin at virtuozzo.com
Fri Mar 17 11:35:06 PDT 2017


On Fri, Mar 17, 2017 at 09:19:26PM +0300, Kirill Tkhai wrote:
> On 17.03.2017 20:52, Andrei Vagin wrote:
> > On Wed, Mar 15, 2017 at 07:56:31PM +0300, Kirill Tkhai wrote:
> >> Some functions in prepare_net_ns() use vmalloc(), and this
> >> memory should be visible to our children.
> > 
> > Why do our tests not fail? Can we fix them to not skip this issue?
> 
> I'm not in close relationship with network tests, but I suppose it's possible (by fixing, not via writing a new one).
> 
> But sadly at the moment I have 40+ patches for pid_ns and not in a time to dive into this one.
> We may keep in mind a test for this for the future. I suppose to do not implement partial cases,
> but just to do a feature (test option), which will unshare all namespaces in test_init().
> This should cover problems like this.

I don't like this sort of comments. All this logic is a part of
nested-userns support. And I think you need only half an our to make
this test. I know that nobody wants to spend time for tests, but we have
to do that.

> 
> ***
> There is one more problem in this function, I wrote. We do not restore net ns after unshare().
> Please, look at that one

Why do we need to restore netns after unshare()?

> 
> >>
> >> v2: No functional changes, just killed "continue".
> >>
> >> Signed-off-by: Kirill Tkhai <ktkhai at virtuozzo.com>
> >> ---
> >>  criu/net.c |   21 +++++++++++++++------
> >>  1 file changed, 15 insertions(+), 6 deletions(-)
> >>
> >> diff --git a/criu/net.c b/criu/net.c
> >> index 309e2bd66..d40ca7f11 100644
> >> --- a/criu/net.c
> >> +++ b/criu/net.c
> >> @@ -1756,20 +1756,27 @@ static int create_net_ns(void *arg)
> >>  
> >>  int prepare_net_namespaces()
> >>  {
> >> -	char stack[128] __stack_aligned__;
> >> +	int status, stack_size;
> >>  	struct ns_id *nsid;
> >> -	int status;
> >> +	char *stack;
> >>  	pid_t pid;
> >>  
> >>  	if (!(root_ns_mask & CLONE_NEWNET))
> >>  		return 0;
> >>  
> >> +	stack_size = 2 * 1024 * 1024;
> >> +	stack = mmap(NULL, stack_size, PROT_WRITE | PROT_READ, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
> >> +	if (stack == MAP_FAILED) {
> >> +		pr_perror("Can't allocate stack");
> >> +		return -1;
> >> +	}
> >> +
> >>  	for (nsid = ns_ids; nsid != NULL; nsid = nsid->next) {
> >>  		if (nsid->nd != &net_ns_desc)
> >>  			continue;
> >>  
> >>  		if (root_user_ns && nsid->user_ns != root_user_ns) {
> >> -			pid = clone(create_net_ns, stack + 128, SIGCHLD, nsid);
> >> +			pid = clone(create_net_ns, stack + stack_size, CLONE_VM | CLONE_FILES | SIGCHLD, nsid);
> >>  			if (pid < 0) {
> >>  				pr_perror("Can't clone");
> >>  				goto err;
> >> @@ -1778,13 +1785,15 @@ int prepare_net_namespaces()
> >>  				pr_perror("Child process waiting %d", status);
> >>  				goto err;
> >>  			}
> >> -			continue;
> >> +		} else {
> >> +			if (do_create_net_ns(nsid))
> >> +				goto err;
> >> +
> >>  		}
> >>  
> >> -		if (do_create_net_ns(nsid))
> >> -			goto err;
> >>  	}
> >>  
> >> +	munmap(stack, stack_size);
> >>  	close_service_fd(NS_FD_OFF);
> >>  
> >>  	return 0;
> >>


More information about the CRIU mailing list