[Devel] Re: [PATCH] cryo: Improve socket support: add UDP and IPv6

[Dave Hansen]

On Mon, 2008-06-16 at 17:38 +0200, Benjamin Thery wrote:
> Today, I played a bit with inet sockets and cryo. With few modifications
> I managed to "checkpoint-restart" programs that use UDP sockets to listen
> and send data.
> 
> This patch improves socket checkpointing in cryo:
> * Save/restore socket family: we can checkpoint IPv6 sockets now.
> * Save/restore socket type: we can checkpoint UDP sockets too.
> 
> Simple test programs for UDP sockets available upon request :)
> 
> Regards,
> Benjamin
> 
> Signed-off-by: Benjamin Thery <benjamin.thery at bull.net>
> ---
>  cr.c |   63 ++++++++++++++++++++++++++++++++++++++++++++++++++-------------
>  1 file changed, 50 insertions(+), 13 deletions(-)
> 
> Index: cryodev/cr.c
> ===================================================================
> --- cryodev.orig/cr.c
> +++ cryodev/cr.c
> @@ -67,11 +67,15 @@ int pagesize = 0;
> 
>  typedef struct isockinfo_t {
>  	int fdnum;
> -	int type;			/* socket type: UDP, TCP */
> +	int type;			/* socket type: SOCK_STREAM, SOCK_DGRAM */
> +	int family;			/* socket family: AF_INET, AF_INET6 */
>  	int mode;			/* bind, listen, connect */
>  	int opt;			/* various sockopt */
>  	int backlog;			/* for listen(2) */
> -	struct sockaddr_in locaddr;	/* for bind(2) */
> +	union {
> +		struct sockaddr_in  saddr4;
> +		struct sockaddr_in6 saddr6;
> +	} locaddr;			/* for bind(2) */
>  	char tcpstate[TPI_LEN];
>  } isockinfo_t;
> 
> @@ -179,7 +183,7 @@ int getmaps(pid_t pid, memseg_t **p)
> 
>  int getsockinfo(pid_t pid, pinfo_t *pi, int num)
>  {
> -	struct sockaddr name;
> +	struct sockaddr_storage name;
>  	int namelen = (int)sizeof(name), ret = 0;
>  	int tpi_len = TPI_LEN;
>  	int flag, flaglen;
> @@ -188,9 +192,11 @@ int getsockinfo(pid_t pid, pinfo_t *pi, 
>  	/* get socketname and see if it is an unix or an inet socket */
>  	memset(&name, 0, sizeof(name));
>  	PT_GETSOCKNAME(pid, num, &name, &namelen);
> -	t_d(name.sa_family);
> +	t_d(name.ss_family);
> 
> -	if (name.sa_family != (sa_family_t)AF_INET) return ret;
> +	if (name.ss_family != (sa_family_t)AF_INET &&
> +	    name.ss_family != (sa_family_t)AF_INET6)
> +		return ret;
>  		
>  	if (! (pi->si = (isockinfo_t *)realloc(pi->si, sizeof(isockinfo_t) * (pi->ns+1)))) return -1;
>  	psi = &pi->si[pi->ns];
> @@ -198,11 +204,25 @@ int getsockinfo(pid_t pid, pinfo_t *pi, 
> 
>  	//pi->si[pi->ns].fdnum = num;
>  	psi->fdnum = num;
> +	psi->family = name.ss_family;
> 
>  	psi->mode = ISOCK_UNKNOWN;
> -	memcpy((void *)&psi->locaddr, (void *)&name, sizeof(name));
> -	if (psi->locaddr.sin_addr.s_addr != 0 || psi->locaddr.sin_port != 0) psi->mode |= ISOCK_BIND;
> -	
> +	memcpy((void *)&psi->locaddr, (void *)&name, namelen);
> +
> +	if ((psi->family == AF_INET &&
> +	     (psi->locaddr.saddr4.sin_addr.s_addr != INADDR_ANY ||
> +	      psi->locaddr.saddr4.sin_port != 0)) ||
> +	    (psi->family == AF_INET6 &&
> +	     (memcmp(&psi->locaddr.saddr6.sin6_addr, &in6addr_any, 16) != 0 ||
> +	      psi->locaddr.saddr6.sin6_port != 0)))
> +		psi->mode |= ISOCK_BIND;

Could you put this into a little helper?  It is getting a little
convoluted.

> +	flag = 0;
> +	flaglen = (int)sizeof(flag);
> +	PT_GETSOCKOPT(pid, num, SOL_SOCKET, SO_TYPE, &flag, &flaglen);
> +	if (flag)
> +		psi->type = flag;
> +
>  	flag = 0;
>  	flaglen = (int)sizeof(flag);

Why are you casting this?

>  	PT_GETSOCKOPT(pid, num, SOL_SOCKET, SO_REUSEADDR, &flag, &flaglen);
> @@ -866,6 +886,7 @@ int restore_sock(int fd, pid_t pid)
>  	int sock;
>  	size_t bufsz;
>  	isockinfo_t *isockinfo = NULL;
> +	int len;
> 
>  	for (;;) {
>  		read_item(fd, item, sizeof(item), &buf, &bufsz);
> @@ -873,20 +894,36 @@ int restore_sock(int fd, pid_t pid)
>  		else ITEM_SET(isockinfo, isockinfo_t);
>  		else break;	// unknown item
> 
> -		t_d(sock = PT_SOCKET(pid, AF_INET, SOCK_STREAM, 0));
> +		t_d(sock = PT_SOCKET(pid, isockinfo->family, isockinfo->type, 0));
>  		if (sock < 0) {
>  			ERROR("PT_SOCKET(%d, %d, %d, %d) errno=%d: %s",
> -				pid, AF_INET, SOCK_STREAM, 0, sock, strerror(-sock)); 
> +			      pid, isockinfo->family, isockinfo->type, 0,
> +			      sock, strerror(-sock));
>  		}
>  		t_d(PT_DUP2(pid, sock, isockinfo->fdnum));
>  		if (sock != isockinfo->fdnum) t_d(PT_CLOSE(pid, sock));
>  		
>  		if (isockinfo->mode & ISOCK_BIND) {
> +			struct sockaddr *sa;
>  			struct sockaddr_in sin;
> +			struct sockaddr_in6 sin6;
> 
> -			memcpy((void *)&sin, (void *)&isockinfo->locaddr, sizeof(sin));
> -			if (isockinfo->mode & ISOCK_CONNECT) sin.sin_port = htons(0);
> -			t_d(PT_BIND(pid, isockinfo->fdnum, &sin, sizeof(sin)));
> +			if (isockinfo->family == AF_INET) {
> +				len = sizeof(sin);
> +				memcpy((void *)&sin,
> +				       (void *)&isockinfo->locaddr.saddr4, len);

I see that these casts were there already, but could we kill them since
you're touching it now?

> +				if (isockinfo->mode & ISOCK_CONNECT)
> +					sin6.sin6_port = htons(0);
> +				sa = (struct sockaddr *) &sin;
> +			} else {

How about an explicit check for AF_INET6 here? 

> +				len = sizeof(sin6);
> +				memcpy((void *)&sin6,
> +				       (void *)&isockinfo->locaddr.saddr6, len);
> +				if (isockinfo->mode & ISOCK_CONNECT)
> +					sin.sin_port = htons(0);
> +				sa = (struct sockaddr *) &sin6;
> +			}
> +			t_d(PT_BIND(pid, isockinfo->fdnum, sa, len));

-- Dave

_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers