*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*/
#define DEBUG_SUBSYSTEM S_LNET
#include <linux/if.h>
#include <linux/in.h>
#include <linux/net.h>
+#include <net/addrconf.h>
+#include <net/ipv6.h>
#include <linux/file.h>
#include <linux/pagemap.h>
/* For sys_open & sys_close */
#include <linux/syscalls.h>
#include <net/sock.h>
+#include <linux/inetdevice.h>
+#include <libcfs/linux/linux-time.h>
+#include <libcfs/linux/linux-net.h>
#include <libcfs/libcfs.h>
#include <lnet/lib-lnet.h>
-/*
- * kernel 5.1: commit 7f1bc6e95d7840d4305595b3e4025cddda88cee5
- * Y2038 64-bit time.
- * SO_TIMESTAMP, SO_TIMESTAMPNS and SO_TIMESTAMPING options, the
- * way they are currently defined, are not y2038 safe.
- * Subsequent patches in the series add new y2038 safe versions
- * of these options which provide 64 bit timestamps on all
- * architectures uniformly.
- * Hence, rename existing options with OLD tag suffixes.
- *
- * NOTE: When updating to timespec64 change change these to '_NEW'.
- *
- */
-#ifndef SO_SNDTIMEO
-#define SO_SNDTIMEO SO_SNDTIMEO_OLD
-#endif
-
-#ifndef SO_RCVTIMEO
-#define SO_RCVTIMEO SO_RCVTIMEO_OLD
-#endif
-
int
lnet_sock_write(struct socket *sock, void *buffer, int nob, int timeout)
{
- int rc;
- long jiffies_left = cfs_time_seconds(timeout);
- unsigned long then;
- struct timeval tv;
+ int rc;
+ long jiffies_left = cfs_time_seconds(timeout);
+ unsigned long then;
LASSERT(nob > 0);
/* Caller may pass a zero timeout if she thinks the socket buffer is
};
if (timeout != 0) {
+ struct sock *sk = sock->sk;
+
/* Set send timeout to remaining time */
- jiffies_to_timeval(jiffies_left, &tv);
- rc = kernel_setsockopt(sock, SOL_SOCKET, SO_SNDTIMEO,
- (char *)&tv, sizeof(tv));
- if (rc != 0) {
- CERROR("Can't set socket send timeout "
- "%ld.%06d: %d\n",
- (long)tv.tv_sec, (int)tv.tv_usec, rc);
- return rc;
- }
+ lock_sock(sk);
+ sk->sk_sndtimeo = jiffies_left;
+ release_sock(sk);
}
then = jiffies;
int
lnet_sock_read(struct socket *sock, void *buffer, int nob, int timeout)
{
- int rc;
- long jiffies_left = cfs_time_seconds(timeout);
- unsigned long then;
- struct timeval tv;
+ int rc;
+ long jiffies_left = cfs_time_seconds(timeout);
+ unsigned long then;
LASSERT(nob > 0);
LASSERT(jiffies_left > 0);
struct msghdr msg = {
.msg_flags = 0
};
+ struct sock *sk = sock->sk;
/* Set receive timeout to remaining time */
- jiffies_to_timeval(jiffies_left, &tv);
- rc = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVTIMEO,
- (char *)&tv, sizeof(tv));
- if (rc != 0) {
- CERROR("Can't set socket recv timeout %ld.%06d: %d\n",
- (long)tv.tv_sec, (int)tv.tv_usec, rc);
- return rc;
- }
+ lock_sock(sk);
+ sk->sk_rcvtimeo = jiffies_left;
+ release_sock(sk);
then = jiffies;
rc = kernel_recvmsg(sock, &msg, &iov, 1, nob, 0);
}
EXPORT_SYMBOL(lnet_sock_read);
-static int
-lnet_sock_create(struct socket **sockp, int *fatal,
- __u32 local_ip, int local_port, struct net *ns)
+int choose_ipv4_src(__u32 *ret, int interface, __u32 dst_ipaddr, struct net *ns)
{
- struct sockaddr_in locaddr;
- struct socket *sock;
- int rc;
- int option;
+ struct net_device *dev;
+ struct in_device *in_dev;
+ int err;
+ DECLARE_CONST_IN_IFADDR(ifa);
+
+ rcu_read_lock();
+ dev = dev_get_by_index_rcu(ns, interface);
+ err = -EINVAL;
+ if (!dev || !(dev->flags & IFF_UP))
+ goto out;
+ in_dev = __in_dev_get_rcu(dev);
+ if (!in_dev)
+ goto out;
+ err = -ENOENT;
+ in_dev_for_each_ifa_rcu(ifa, in_dev) {
+ if (err ||
+ ((dst_ipaddr ^ ntohl(ifa->ifa_local))
+ & ntohl(ifa->ifa_mask)) == 0) {
+ /* This address at least as good as what we
+ * already have
+ */
+ *ret = ntohl(ifa->ifa_local);
+ err = 0;
+ }
+ }
+ endfor_ifa(in_dev);
+out:
+ rcu_read_unlock();
+ return err;
+}
+EXPORT_SYMBOL(choose_ipv4_src);
- /* All errors are fatal except bind failure if the port is in use */
- *fatal = 1;
+static struct socket *
+lnet_sock_create(int interface, struct sockaddr *remaddr,
+ int local_port, struct net *ns)
+{
+ struct socket *sock;
+ int rc;
+ int family;
+ family = AF_INET6;
+ if (remaddr)
+ family = remaddr->sa_family;
+retry:
#ifdef HAVE_SOCK_CREATE_KERN_USE_NET
- rc = sock_create_kern(ns, PF_INET, SOCK_STREAM, 0, &sock);
+ rc = sock_create_kern(ns, family, SOCK_STREAM, 0, &sock);
#else
- rc = sock_create_kern(PF_INET, SOCK_STREAM, 0, &sock);
+ rc = sock_create_kern(family, SOCK_STREAM, 0, &sock);
#endif
- *sockp = sock;
- if (rc != 0) {
- CERROR("Can't create socket: %d\n", rc);
- return rc;
+ if (rc == -EAFNOSUPPORT && family == AF_INET6 && !remaddr) {
+ family = AF_INET;
+ goto retry;
}
- option = 1;
- rc = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
- (char *)&option, sizeof(option));
- if (rc != 0) {
- CERROR("Can't set SO_REUSEADDR for socket: %d\n", rc);
- goto failed;
+ if (rc) {
+ CERROR("Can't create socket: %d\n", rc);
+ return ERR_PTR(rc);
}
- if (local_ip != 0 || local_port != 0) {
- memset(&locaddr, 0, sizeof(locaddr));
- locaddr.sin_family = AF_INET;
- locaddr.sin_port = htons(local_port);
- locaddr.sin_addr.s_addr = (local_ip == 0) ?
- INADDR_ANY : htonl(local_ip);
+ sock->sk->sk_reuseport = 1;
+
+ if (interface >= 0 || local_port != 0) {
+ struct sockaddr_storage locaddr = {};
+
+ switch (family) {
+ case AF_INET: {
+ struct sockaddr_in *sin = (void *)&locaddr;
+
+ sin->sin_family = AF_INET;
+ sin->sin_addr.s_addr = INADDR_ANY;
+ if (interface >= 0 && remaddr) {
+ struct sockaddr_in *rem = (void *)remaddr;
+ __u32 ip;
+ rc = choose_ipv4_src(&ip,
+ interface,
+ ntohl(rem->sin_addr.s_addr),
+ ns);
+ if (rc)
+ goto failed;
+ sin->sin_addr.s_addr = htonl(ip);
+ }
+ sin->sin_port = htons(local_port);
+ break;
+ }
+#if IS_ENABLED(CONFIG_IPV6)
+ case AF_INET6: {
+ struct sockaddr_in6 *sin6 = (void *)&locaddr;
+ int val = 0;
+
+ sin6->sin6_family = AF_INET6;
+ sin6->sin6_addr = in6addr_any;
+
+ /* Make sure we get both IPv4 and IPv6 connections.
+ * This is the default, but it can be overridden so we
+ * force it back.
+ */
+#ifdef HAVE_KERNEL_SETSOCKOPT
+ kernel_setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY,
+ (char *) &val, sizeof(val));
+#elif defined(_LINUX_SOCKPTR_H)
+ /* sockptr_t was introduced around
+ * v5.8-rc4-1952-ga7b75c5a8c41 and allows a
+ * kernel address to be passed to ->setsockopt
+ */
+ if (ipv6_only_sock(sock->sk)) {
+ sockptr_t optval = KERNEL_SOCKPTR(&val);
+
+ sock->ops->setsockopt(sock,
+ IPPROTO_IPV6, IPV6_V6ONLY,
+ optval, sizeof(val));
+ }
+#else
+ /* From v5.7-rc6-2614-g5a892ff2facb when
+ * kernel_setsockopt() was removed until
+ * sockptr_t (above) there is no clean way to
+ * pass kernel address to setsockopt. We could
+ * use get_fs()/set_fs(), but in this particular
+ * situation there is an easier way. It depends
+ * on the fact that at least for these few
+ * kernels a NULL address to ipv6_setsockopt()
+ * is treated like the address of a zero.
+ */
+ if (ipv6_only_sock(sock->sk) && !val) {
+ void *optval = NULL;
+
+ sock->ops->setsockopt(sock,
+ IPPROTO_IPV6, IPV6_V6ONLY,
+ optval, sizeof(val));
+ }
+#endif /* HAVE_KERNEL_SETSOCKOPT */
+
+ if (interface >= 0 && remaddr) {
+ struct sockaddr_in6 *rem = (void *)remaddr;
+
+ ipv6_dev_get_saddr(ns,
+ dev_get_by_index(ns,
+ interface),
+ &rem->sin6_addr, 0,
+ &sin6->sin6_addr);
+ }
+ sin6->sin6_port = htons(local_port);
+ break;
+ }
+#endif /* IS_ENABLED(CONFIG_IPV6) */
+ }
rc = kernel_bind(sock, (struct sockaddr *)&locaddr,
sizeof(locaddr));
if (rc == -EADDRINUSE) {
CDEBUG(D_NET, "Port %d already in use\n", local_port);
- *fatal = 0;
goto failed;
}
if (rc != 0) {
goto failed;
}
}
- return 0;
+ return sock;
failed:
sock_release(sock);
- return rc;
+ return ERR_PTR(rc);
}
-int
+void
lnet_sock_setbuf(struct socket *sock, int txbufsize, int rxbufsize)
{
- int option;
- int rc;
+ struct sock *sk = sock->sk;
if (txbufsize != 0) {
- option = txbufsize;
- rc = kernel_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
- (char *)&option, sizeof(option));
- if (rc != 0) {
- CERROR("Can't set send buffer %d: %d\n",
- option, rc);
- return rc;
- }
+ sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
+ sk->sk_sndbuf = txbufsize;
+ sk->sk_write_space(sk);
}
if (rxbufsize != 0) {
- option = rxbufsize;
- rc = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
- (char *)&option, sizeof(option));
- if (rc != 0) {
- CERROR("Can't set receive buffer %d: %d\n",
- option, rc);
- return rc;
- }
+ sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
+ sk->sk_sndbuf = rxbufsize;
}
- return 0;
}
EXPORT_SYMBOL(lnet_sock_setbuf);
int
-lnet_sock_getaddr(struct socket *sock, bool remote, __u32 *ip, int *port)
+lnet_sock_getaddr(struct socket *sock, bool remote,
+ struct sockaddr_storage *peer)
{
- struct sockaddr_in sin;
int rc;
#ifndef HAVE_KERN_SOCK_GETNAME_2ARGS
- int len = sizeof(sin);
+ int len = sizeof(*peer);
#endif
if (remote)
rc = lnet_kernel_getpeername(sock,
- (struct sockaddr *)&sin, &len);
+ (struct sockaddr *)peer, &len);
else
rc = lnet_kernel_getsockname(sock,
- (struct sockaddr *)&sin, &len);
+ (struct sockaddr *)peer, &len);
if (rc < 0) {
CERROR("Error %d getting sock %s IP/port\n",
rc, remote ? "peer" : "local");
return rc;
}
-
- if (ip != NULL)
- *ip = ntohl(sin.sin_addr.s_addr);
-
- if (port != NULL)
- *port = ntohs(sin.sin_port);
-
+ if (peer->ss_family == AF_INET6) {
+ struct sockaddr_in6 *in6 = (void *)peer;
+ struct sockaddr_in *in = (void *)peer;
+ short port = in6->sin6_port;
+
+ if (ipv6_addr_v4mapped(&in6->sin6_addr)) {
+ /* Pretend it is a v4 socket */
+ memset(in, 0, sizeof(*in));
+ in->sin_family = AF_INET;
+ in->sin_port = port;
+ memcpy(&in->sin_addr, &in6->sin6_addr.s6_addr32[3], 4);
+ }
+ }
return 0;
}
EXPORT_SYMBOL(lnet_sock_getaddr);
-int
-lnet_sock_getbuf(struct socket *sock, int *txbufsize, int *rxbufsize)
+void lnet_sock_getbuf(struct socket *sock, int *txbufsize, int *rxbufsize)
{
if (txbufsize != NULL)
*txbufsize = sock->sk->sk_sndbuf;
if (rxbufsize != NULL)
*rxbufsize = sock->sk->sk_rcvbuf;
-
- return 0;
}
EXPORT_SYMBOL(lnet_sock_getbuf);
-int
-lnet_sock_listen(struct socket **sockp,
- __u32 local_ip, int local_port, int backlog, struct net *ns)
+struct socket *
+lnet_sock_listen(int local_port, int backlog, struct net *ns)
{
- int fatal;
- int rc;
+ struct socket *sock;
+ int rc;
- rc = lnet_sock_create(sockp, &fatal, local_ip, local_port, ns);
- if (rc != 0) {
- if (!fatal)
+ sock = lnet_sock_create(-1, NULL, local_port, ns);
+ if (IS_ERR(sock)) {
+ rc = PTR_ERR(sock);
+ if (rc == -EADDRINUSE)
CERROR("Can't create socket: port %d already in use\n",
local_port);
- return rc;
+ return ERR_PTR(rc);
}
- rc = kernel_listen(*sockp, backlog);
+ rc = kernel_listen(sock, backlog);
if (rc == 0)
- return 0;
+ return sock;
CERROR("Can't set listen backlog %d: %d\n", backlog, rc);
- sock_release(*sockp);
- return rc;
-}
-
-int
-lnet_sock_accept(struct socket **newsockp, struct socket *sock)
-{
- wait_queue_entry_t wait;
- struct socket *newsock;
- int rc;
-
- /* XXX this should add a ref to sock->ops->owner, if
- * TCP could be a module */
- rc = sock_create_lite(PF_PACKET, sock->type, IPPROTO_TCP, &newsock);
- if (rc) {
- CERROR("Can't allocate socket\n");
- return rc;
- }
-
- newsock->ops = sock->ops;
-
-#ifdef HAVE_KERN_SOCK_ACCEPT_FLAG_ARG
- rc = sock->ops->accept(sock, newsock, O_NONBLOCK, false);
-#else
- rc = sock->ops->accept(sock, newsock, O_NONBLOCK);
-#endif
- if (rc == -EAGAIN) {
- /* Nothing ready, so wait for activity */
- init_waitqueue_entry(&wait, current);
- add_wait_queue(sk_sleep(sock->sk), &wait);
- set_current_state(TASK_INTERRUPTIBLE);
- schedule();
- remove_wait_queue(sk_sleep(sock->sk), &wait);
-#ifdef HAVE_KERN_SOCK_ACCEPT_FLAG_ARG
- rc = sock->ops->accept(sock, newsock, O_NONBLOCK, false);
-#else
- rc = sock->ops->accept(sock, newsock, O_NONBLOCK);
-#endif
- }
-
- if (rc != 0)
- goto failed;
-
- *newsockp = newsock;
- return 0;
-
-failed:
- sock_release(newsock);
- return rc;
+ sock_release(sock);
+ return ERR_PTR(rc);
}
-int
-lnet_sock_connect(struct socket **sockp, int *fatal,
- __u32 local_ip, int local_port,
- __u32 peer_ip, int peer_port,
+struct socket *
+lnet_sock_connect(int interface, int local_port,
+ struct sockaddr *peeraddr,
struct net *ns)
{
- struct sockaddr_in srvaddr;
- int rc;
-
- rc = lnet_sock_create(sockp, fatal, local_ip, local_port, ns);
- if (rc != 0)
- return rc;
+ struct socket *sock;
+ int rc;
- memset(&srvaddr, 0, sizeof(srvaddr));
- srvaddr.sin_family = AF_INET;
- srvaddr.sin_port = htons(peer_port);
- srvaddr.sin_addr.s_addr = htonl(peer_ip);
+ sock = lnet_sock_create(interface, peeraddr, local_port, ns);
+ if (IS_ERR(sock))
+ return sock;
- rc = kernel_connect(*sockp, (struct sockaddr *)&srvaddr,
- sizeof(srvaddr), 0);
+ rc = kernel_connect(sock, peeraddr, sizeof(struct sockaddr_in6), 0);
if (rc == 0)
- return 0;
+ return sock;
/* EADDRNOTAVAIL probably means we're already connected to the same
* peer/port on the same local port on a differently typed
* connection. Let our caller retry with a different local
* port... */
- *fatal = !(rc == -EADDRNOTAVAIL);
- CDEBUG_LIMIT(*fatal ? D_NETERROR : D_NET,
- "Error %d connecting %pI4h/%d -> %pI4h/%d\n", rc,
- &local_ip, local_port, &peer_ip, peer_port);
+ CDEBUG_LIMIT(rc == -EADDRNOTAVAIL ? D_NET : D_NETERROR,
+ "Error %d connecting %d -> %pISp\n", rc,
+ local_port, peeraddr);
- sock_release(*sockp);
- return rc;
+ sock_release(sock);
+ return ERR_PTR(rc);
}