X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lnet%2Flnet%2Flib-socket.c;h=b801b4727344894aee8bf59e05b8c708c7b03e16;hb=d720fbaadad9343a481863ae3b04ba580541328d;hp=57678b3d8a0be72cb325d2f922098b934121b58a;hpb=15045c9067fd021baa0ec925bcc245949945d01e;p=fs%2Flustre-release.git

diff --git a/lnet/lnet/lib-socket.c b/lnet/lnet/lib-socket.c
index 57678b3..b801b47 100644
--- a/lnet/lnet/lib-socket.c
+++ b/lnet/lnet/lib-socket.c
@@ -23,7 +23,7 @@
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2015, 2016, Intel Corporation.
+ * Copyright (c) 2015, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -39,252 +39,37 @@
 /* For sys_open & sys_close */
 #include <linux/syscalls.h>
 #include <net/sock.h>
+#include <linux/inetdevice.h>
 
 #include <libcfs/libcfs.h>
 #include <lnet/lib-lnet.h>
 
-static int
-kernel_sock_unlocked_ioctl(struct file *filp, int cmd, unsigned long arg)
-{
-	mm_segment_t oldfs = get_fs();
-	int err;
-
-	set_fs(KERNEL_DS);
-	err = filp->f_op->unlocked_ioctl(filp, cmd, arg);
-	set_fs(oldfs);
-
-	return err;
-}
-
-static int
-lnet_sock_ioctl(int cmd, unsigned long arg)
-{
-	struct file    *sock_filp;
-	struct socket  *sock;
-	int		fd = -1;
-	int		rc;
-
-#ifdef HAVE_SOCK_CREATE_KERN_USE_NET
-	rc = sock_create_kern(&init_net, PF_INET, SOCK_STREAM, 0, &sock);
-#else
-	rc = sock_create_kern(PF_INET, SOCK_STREAM, 0, &sock);
+/*
+ * 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
-	if (rc != 0) {
-		CERROR("Can't create socket: %d\n", rc);
-		return rc;
-	}
 
-#if !defined(HAVE_SOCK_ALLOC_FILE) && !defined(HAVE_SOCK_ALLOC_FILE_3ARGS)
-	fd = sock_map_fd(sock, 0);
-	if (fd < 0) {
-		rc = fd;
-		sock_release(sock);
-		goto out;
-	}
-	sock_filp = fget(fd);
-#else
-# ifdef HAVE_SOCK_ALLOC_FILE_3ARGS
-	sock_filp = sock_alloc_file(sock, 0, NULL);
-# else
-	sock_filp = sock_alloc_file(sock, 0);
-# endif
+#ifndef SO_RCVTIMEO
+#define SO_RCVTIMEO SO_RCVTIMEO_OLD
 #endif
-	if (IS_ERR(sock_filp)) {
-		rc = PTR_ERR(sock_filp);
-		sock_release(sock);
-		goto out;
-	}
-
-	rc = kernel_sock_unlocked_ioctl(sock_filp, cmd, arg);
-
-	fput(sock_filp);
-out:
-	if (fd >= 0)
-		sys_close(fd);
-	return rc;
-}
-
-int
-lnet_ipif_query(char *name, int *up, __u32 *ip, __u32 *mask)
-{
-	struct ifreq	ifr;
-	int		nob;
-	int		rc;
-	__u32		val;
-
-	nob = strnlen(name, IFNAMSIZ);
-	if (nob == IFNAMSIZ) {
-		CERROR("Interface name %s too long\n", name);
-		return -EINVAL;
-	}
-
-	CLASSERT(sizeof(ifr.ifr_name) >= IFNAMSIZ);
-
-	if (strlen(name) > sizeof(ifr.ifr_name)-1)
-		return -E2BIG;
-	strncpy(ifr.ifr_name, name, sizeof(ifr.ifr_name));
-
-	rc = lnet_sock_ioctl(SIOCGIFFLAGS, (unsigned long)&ifr);
-	if (rc != 0) {
-		CERROR("Can't get flags for interface %s\n", name);
-		return rc;
-	}
-
-	if ((ifr.ifr_flags & IFF_UP) == 0) {
-		CDEBUG(D_NET, "Interface %s down\n", name);
-		*up = 0;
-		*ip = *mask = 0;
-		return 0;
-	}
-	*up = 1;
-
-	if (strlen(name) > sizeof(ifr.ifr_name)-1)
-		return -E2BIG;
-	strncpy(ifr.ifr_name, name, sizeof(ifr.ifr_name));
-
-	ifr.ifr_addr.sa_family = AF_INET;
-	rc = lnet_sock_ioctl(SIOCGIFADDR, (unsigned long)&ifr);
-
-	if (rc != 0) {
-		CERROR("Can't get IP address for interface %s\n", name);
-		return rc;
-	}
-
-	val = ((struct sockaddr_in *)&ifr.ifr_addr)->sin_addr.s_addr;
-	*ip = ntohl(val);
-
-	if (strlen(name) > sizeof(ifr.ifr_name)-1)
-		return -E2BIG;
-	strncpy(ifr.ifr_name, name, sizeof(ifr.ifr_name));
-
-	ifr.ifr_addr.sa_family = AF_INET;
-	rc = lnet_sock_ioctl(SIOCGIFNETMASK, (unsigned long)&ifr);
-	if (rc != 0) {
-		CERROR("Can't get netmask for interface %s\n", name);
-		return rc;
-	}
-
-	val = ((struct sockaddr_in *)&ifr.ifr_netmask)->sin_addr.s_addr;
-	*mask = ntohl(val);
-
-	return 0;
-}
-EXPORT_SYMBOL(lnet_ipif_query);
-
-void
-lnet_ipif_free_enumeration(char **names, int n)
-{
-	int	i;
-
-	LASSERT(n > 0);
-
-	for (i = 0; i < n && names[i] != NULL; i++)
-		LIBCFS_FREE(names[i], IFNAMSIZ);
-
-	LIBCFS_FREE(names, n * sizeof(*names));
-}
-EXPORT_SYMBOL(lnet_ipif_free_enumeration);
-
-int
-lnet_ipif_enumerate(char ***namesp)
-{
-	/* Allocate and fill in 'names', returning # interfaces/error */
-	char	      **names;
-	int		toobig;
-	int		nalloc;
-	int		nfound;
-	struct ifreq   *ifr;
-	struct ifconf	ifc;
-	int		rc;
-	int		nob;
-	int		i;
-
-	nalloc = 16;	/* first guess at max interfaces */
-	toobig = 0;
-	for (;;) {
-		if (nalloc * sizeof(*ifr) > PAGE_SIZE) {
-			toobig = 1;
-			nalloc = PAGE_SIZE / sizeof(*ifr);
-			CWARN("Too many interfaces: only enumerating "
-			      "first %d\n", nalloc);
-		}
-
-		LIBCFS_ALLOC(ifr, nalloc * sizeof(*ifr));
-		if (ifr == NULL) {
-			CERROR("ENOMEM enumerating up to %d interfaces\n",
-			       nalloc);
-			rc = -ENOMEM;
-			goto out0;
-		}
-
-		ifc.ifc_buf = (char *)ifr;
-		ifc.ifc_len = nalloc * sizeof(*ifr);
-
-		rc = lnet_sock_ioctl(SIOCGIFCONF, (unsigned long)&ifc);
-		if (rc < 0) {
-			CERROR("Error %d enumerating interfaces\n", rc);
-			goto out1;
-		}
-
-		LASSERT(rc == 0);
-
-		nfound = ifc.ifc_len/sizeof(*ifr);
-		LASSERT(nfound <= nalloc);
-
-		if (nfound < nalloc || toobig)
-			break;
-
-		LIBCFS_FREE(ifr, nalloc * sizeof(*ifr));
-		nalloc *= 2;
-	}
-
-	if (nfound == 0)
-		goto out1;
-
-	LIBCFS_ALLOC(names, nfound * sizeof(*names));
-	if (names == NULL) {
-		rc = -ENOMEM;
-		goto out1;
-	}
-
-	for (i = 0; i < nfound; i++) {
-		nob = strnlen(ifr[i].ifr_name, IFNAMSIZ);
-		if (nob == IFNAMSIZ) {
-			/* no space for terminating NULL */
-			CERROR("interface name %.*s too long (%d max)\n",
-			       nob, ifr[i].ifr_name, IFNAMSIZ);
-			rc = -ENAMETOOLONG;
-			goto out2;
-		}
-
-		LIBCFS_ALLOC(names[i], IFNAMSIZ);
-		if (names[i] == NULL) {
-			rc = -ENOMEM;
-			goto out2;
-		}
-
-		memcpy(names[i], ifr[i].ifr_name, nob);
-		names[i][nob] = 0;
-	}
-
-	*namesp = names;
-	rc = nfound;
-
- out2:
-	if (rc < 0)
-		lnet_ipif_free_enumeration(names, nfound);
- out1:
-	LIBCFS_FREE(ifr, nalloc * sizeof(*ifr));
- out0:
-	return rc;
-}
-EXPORT_SYMBOL(lnet_ipif_enumerate);
 
 int
 lnet_sock_write(struct socket *sock, void *buffer, int nob, int timeout)
 {
 	int		rc;
-	long		jiffies_left = timeout * msecs_to_jiffies(MSEC_PER_SEC);
+	long		jiffies_left = cfs_time_seconds(timeout);
 	unsigned long	then;
 	struct timeval	tv;
 
@@ -303,15 +88,7 @@ lnet_sock_write(struct socket *sock, void *buffer, int nob, int timeout)
 
 		if (timeout != 0) {
 			/* Set send timeout to remaining time */
-			tv = (struct timeval) {
-				.tv_sec = jiffies_left /
-					  msecs_to_jiffies(MSEC_PER_SEC),
-				.tv_usec = ((jiffies_left %
-					     msecs_to_jiffies(MSEC_PER_SEC)) *
-					     USEC_PER_SEC) /
-					     msecs_to_jiffies(MSEC_PER_SEC)
-			};
-
+			jiffies_to_timeval(jiffies_left, &tv);
 			rc = kernel_setsockopt(sock, SOL_SOCKET, SO_SNDTIMEO,
 					       (char *)&tv, sizeof(tv));
 			if (rc != 0) {
@@ -351,7 +128,7 @@ int
 lnet_sock_read(struct socket *sock, void *buffer, int nob, int timeout)
 {
 	int		rc;
-	long		jiffies_left = timeout * msecs_to_jiffies(MSEC_PER_SEC);
+	long		jiffies_left = cfs_time_seconds(timeout);
 	unsigned long	then;
 	struct timeval	tv;
 
@@ -368,13 +145,7 @@ lnet_sock_read(struct socket *sock, void *buffer, int nob, int timeout)
 		};
 
 		/* Set receive timeout to remaining time */
-		tv = (struct timeval) {
-			.tv_sec = jiffies_left / msecs_to_jiffies(MSEC_PER_SEC),
-			.tv_usec = ((jiffies_left %
-					msecs_to_jiffies(MSEC_PER_SEC)) *
-					USEC_PER_SEC) /
-					msecs_to_jiffies(MSEC_PER_SEC)
-		};
+		jiffies_to_timeval(jiffies_left, &tv);
 		rc = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVTIMEO,
 				       (char *)&tv, sizeof(tv));
 		if (rc != 0) {
@@ -405,49 +176,90 @@ lnet_sock_read(struct socket *sock, void *buffer, int nob, int timeout)
 }
 EXPORT_SYMBOL(lnet_sock_read);
 
-static int
-lnet_sock_create(struct socket **sockp, int *fatal,
-		 __u32 local_ip, int local_port)
+int choose_ipv4_src(__u32 *ret, int interface, __u32 dst_ipaddr, struct net *ns)
+{
+	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 (*ret == 0 ||
+		    ((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);
+
+static struct socket *
+lnet_sock_create(int interface, struct sockaddr *remaddr,
+		 int local_port, struct net *ns)
 {
-	struct sockaddr_in  locaddr;
 	struct socket	   *sock;
 	int		    rc;
 	int		    option;
 
-	/* All errors are fatal except bind failure if the port is in use */
-	*fatal = 1;
-
 #ifdef HAVE_SOCK_CREATE_KERN_USE_NET
-	rc = sock_create_kern(&init_net, PF_INET, SOCK_STREAM, 0, &sock);
+	rc = sock_create_kern(ns, PF_INET, SOCK_STREAM, 0, &sock);
 #else
 	rc = sock_create_kern(PF_INET, SOCK_STREAM, 0, &sock);
 #endif
-	*sockp = sock;
-	if (rc != 0) {
+	if (rc) {
 		CERROR("Can't create socket: %d\n", rc);
-		return rc;
+		return ERR_PTR(rc);
 	}
 
 	option = 1;
 	rc = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
 			       (char *)&option, sizeof(option));
-	if (rc != 0) {
+	if (rc) {
 		CERROR("Can't set SO_REUSEADDR for socket: %d\n", rc);
 		goto failed;
 	}
 
-	if (local_ip != 0 || local_port != 0) {
-		memset(&locaddr, 0, sizeof(locaddr));
+	if (interface >= 0 || local_port != 0) {
+		struct sockaddr_in locaddr = {};
+
 		locaddr.sin_family = AF_INET;
+		locaddr.sin_addr.s_addr = INADDR_ANY;
+		if (interface >= 0) {
+			struct sockaddr_in *sin = (void *)remaddr;
+			__u32 ip;
+
+			rc = choose_ipv4_src(&ip,
+					     interface,
+					     ntohl(sin->sin_addr.s_addr),
+					     ns);
+			if (rc)
+				goto failed;
+			locaddr.sin_addr.s_addr = htonl(ip);
+		}
+
 		locaddr.sin_port = htons(local_port);
-		locaddr.sin_addr.s_addr = (local_ip == 0) ?
-					  INADDR_ANY : htonl(local_ip);
 
 		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) {
@@ -456,11 +268,11 @@ lnet_sock_create(struct socket **sockp, int *fatal,
 			goto failed;
 		}
 	}
-	return 0;
+	return sock;
 
 failed:
 	sock_release(sock);
-	return rc;
+	return ERR_PTR(rc);
 }
 
 int
@@ -498,14 +310,18 @@ int
 lnet_sock_getaddr(struct socket *sock, bool remote, __u32 *ip, int *port)
 {
 	struct sockaddr_in sin;
-	int		   len = sizeof(sin);
-	int		   rc;
+	int rc;
+#ifndef HAVE_KERN_SOCK_GETNAME_2ARGS
+	int len = sizeof(sin);
+#endif
 
 	if (remote)
-		rc = kernel_getpeername(sock, (struct sockaddr *)&sin, &len);
+		rc = lnet_kernel_getpeername(sock,
+					     (struct sockaddr *)&sin, &len);
 	else
-		rc = kernel_getsockname(sock, (struct sockaddr *)&sin, &len);
-	if (rc != 0) {
+		rc = lnet_kernel_getsockname(sock,
+					     (struct sockaddr *)&sin, &len);
+	if (rc < 0) {
 		CERROR("Error %d getting sock %s IP/port\n",
 			rc, remote ? "peer" : "local");
 		return rc;
@@ -534,116 +350,63 @@ lnet_sock_getbuf(struct socket *sock, int *txbufsize, int *rxbufsize)
 }
 EXPORT_SYMBOL(lnet_sock_getbuf);
 
-int
-lnet_sock_listen(struct socket **sockp,
-		   __u32 local_ip, int local_port, int backlog)
+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);
-	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;
-}
-
-#ifndef HAVE_SK_SLEEP
-static inline wait_queue_head_t *sk_sleep(struct sock *sk)
-{
-	return sk->sk_sleep;
-}
-#endif
-
-int
-lnet_sock_accept(struct socket **newsockp, struct socket *sock)
-{
-	wait_queue_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,
+		  __u32 peer_ip, int peer_port,
+		  struct net *ns)
 {
-	struct sockaddr_in  srvaddr;
-	int		    rc;
-
-	rc = lnet_sock_create(sockp, fatal, local_ip, local_port);
-	if (rc != 0)
-		return rc;
+	struct socket *sock;
+	struct sockaddr_in srvaddr;
+	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);
 
-	rc = kernel_connect(*sockp, (struct sockaddr *)&srvaddr,
+	sock = lnet_sock_create(interface, (struct sockaddr *)&srvaddr,
+				local_port, ns);
+	if (IS_ERR(sock))
+		return sock;
+
+	rc = kernel_connect(sock, (struct sockaddr *)&srvaddr,
 			    sizeof(srvaddr), 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 -> %pI4h/%d\n", rc,
+		     local_port, &peer_ip, peer_port);
 
-	sock_release(*sockp);
-	return rc;
+	sock_release(sock);
+	return ERR_PTR(rc);
 }