Skip to contentSkip to conversation

cifs: Allow binding to local IP address.

cifs Ben Greear 2 participants 6 messages Aug 31, 2010 4 messages not in the archive

Subsystems (4)
lore
Share & tools
lore

Ben Greear

author
#0 lore raw

When using multi-homed machines, it's nice to be able to specify the local IP to use for outbound connections. This patch gives cifs the ability to bind to a particular IP address.

Usage: mount -t cifs -o srcaddr=192.168.1.50,user=foo, ...
Usage: mount -t cifs -o srcaddr=2002::100:1,user=foo, ...

---
:100644 100644 b7431af... 4510f4d... M fs/cifs/cifsfs.c
:100644 100644 d82f5fb... d6f0a7d... M fs/cifs/cifsfs.h
:100644 100644 c9d0cfc... 784fd4a... M fs/cifs/cifsglob.h
:100644 100644 ec0ea4a... 9a3175b... M fs/cifs/connect.c
:100644 100644 ab70a3f... 735989a... M net/ipv6/addrconf.c

 fs/cifs/cifsfs.c    |   32 ++++++++++++++++++
 fs/cifs/cifsfs.h    |    3 ++
 fs/cifs/cifsglob.h  |    1 +
 fs/cifs/connect.c   |   91 +++++++++++++++++++++++++++++++++++++++++++++++++-
 net/ipv6/addrconf.c |    1 +
 5 files changed, 126 insertions(+), 2 deletions(-)
fs/cifs/cifsfs.c +32 -0
index b7431af..4510f4d 100644
@@ -357,6 +357,23 @@ cifs_show_address(struct seq_file *s, struct TCP_Server_Info *server)
 	}
 }
 
+bool
+cifs_addr_is_specified(struct sockaddr *srcaddr) {
+	struct sockaddr_in *saddr4 = (struct sockaddr_in *)srcaddr;
+	struct sockaddr_in6 *saddr6 = (struct sockaddr_in6 *)srcaddr;
+	switch (srcaddr->sa_family) {
+	case AF_INET:
+		return saddr4->sin_addr.s_addr != 0;
+#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
+	case AF_INET6:
+		/* in6addr_any isn't available if IPv6 isn't compiled in */
+		return (memcmp(&in6addr_any, &saddr6->sin6_addr,
+			       sizeof(in6addr_any)) != 0);
+#endif
+	}
+	return false;
+}
+
 /*
  * cifs_show_options() is for displaying mount options in /proc/mounts.
  * Not all settable options are displayed but most of the important
@@ -367,6 +384,8 @@ cifs_show_options(struct seq_file *s, struct vfsmount *m)
 {
 	struct cifs_sb_info *cifs_sb = CIFS_SB(m->mnt_sb);
 	struct cifsTconInfo *tcon = cifs_sb->tcon;
+	struct sockaddr *srcaddr;
+	srcaddr = (struct sockaddr *)(&tcon->ses->server->srcaddr);
 
 	seq_printf(s, ",unc=%s", tcon->treeName);
 	if (tcon->ses->userName)
@@ -374,6 +393,19 @@ cifs_show_options(struct seq_file *s, struct vfsmount *m)
 	if (tcon->ses->domainName)
 		seq_printf(s, ",domain=%s", tcon->ses->domainName);
 
+	if (cifs_addr_is_specified(srcaddr)) {
+		struct sockaddr_in *saddr4;
+		struct sockaddr_in6 *saddr6;
+		saddr4 = (struct sockaddr_in *)srcaddr;
+		saddr6 = (struct sockaddr_in6 *)srcaddr;
+		if (saddr6->sin6_family == AF_INET6)
+			seq_printf(s, ",srcaddr=%pI6c",
+				   &saddr6->sin6_addr);
+		else
+			seq_printf(s, ",srcaddr=%pI4",
+				   &saddr4->sin_addr.s_addr);
+	}
+
 	seq_printf(s, ",uid=%d", cifs_sb->mnt_uid);
 	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_OVERR_UID)
 		seq_printf(s, ",forceuid");
fs/cifs/cifsfs.h +3 -0
index d82f5fb..d6f0a7d 100644
@@ -110,6 +110,9 @@ extern ssize_t	cifs_getxattr(struct dentry *, const char *, void *, size_t);
 extern ssize_t	cifs_listxattr(struct dentry *, char *, size_t);
 extern long cifs_ioctl(struct file *filep, unsigned int cmd, unsigned long arg);
 
+struct sockaddr;
+extern bool cifs_addr_is_specified(struct sockaddr *srcaddr);
+
 #ifdef CONFIG_CIFS_EXPERIMENTAL
 extern const struct export_operations cifs_export_ops;
 #endif /* EXPERIMENTAL */
fs/cifs/cifsglob.h +1 -0
index c9d0cfc..784fd4a 100644
@@ -157,6 +157,7 @@ struct TCP_Server_Info {
 		struct sockaddr_in sockAddr;
 		struct sockaddr_in6 sockAddr6;
 	} addr;
+	struct sockaddr_storage srcaddr; /* locally bind to this IP */
 	wait_queue_head_t response_q;
 	wait_queue_head_t request_q; /* if more than maxmpx to srvr must block*/
 	struct list_head pending_mid_q;
fs/cifs/connect.c +89 -2

158 lines not on this page · raw message

net/ipv6/addrconf.c +1 -0
index ab70a3f..735989a 100644
@@ -230,6 +230,7 @@ static struct ipv6_devconf ipv6_devconf_dflt __read_mostly = {
 
 /* IPv6 Wildcard Address and Loopback Address defined by RFC2553 */
 const struct in6_addr in6addr_any = IN6ADDR_ANY_INIT;
+EXPORT_SYMBOL(in6addr_any);
 const struct in6_addr in6addr_loopback = IN6ADDR_LOOPBACK_INIT;
 const struct in6_addr in6addr_linklocal_allnodes = IN6ADDR_LINKLOCAL_ALLNODES_INIT;
 const struct in6_addr in6addr_linklocal_allrouters = IN6ADDR_LINKLOCAL_ALLROUTERS_INIT;
Signature · 1 line
1.6.2.5

A message that is not in the archive (sent off-list or not yet received): 1283233222-8702-1-git-send-email-greearb-my8/[email protected]

author
#2 lore raw
Quoted 6 lines · Ben Greear

When using multi-homed machines, it's nice to be able to specify the local IP to use for outbound connections. This patch gives cifs the ability to bind to a particular IP address.

   Usage:  mount -t cifs -o srcaddr=192.168.1.50,user=foo, ...
   Usage:  mount -t cifs -o srcaddr=2002::100:1,user=foo, ...

I tested IPv6 and IPv4 and it seems to work as designed.

checkpatch warns of a few long lines, but I saw no way to break them up w/out adding even worse (IMO) cruft. But, I can change that if it's important.

Please let me know if any other changes are needed.

Thanks,
Ben

Signature · 2 lines
Ben Greear <greearb-my8/[email protected]>
Candela Technologies Inc  http://www.candelatech.com
#3 lore raw
In reply to a message not in the archive
Quoted 26 lines · On Mon, 30 Aug 2010 22:40:22 -0700 Ben Greear…

When using multi-homed machines, it's nice to be able to specify the local IP to use for outbound connections. This patch gives cifs the ability to bind to a particular IP address.

Usage: mount -t cifs -o srcaddr=192.168.1.50,user=foo, ...
Usage: mount -t cifs -o srcaddr=2002::100:1,user=foo, ...

---
:100644 100644 b7431af... 4510f4d... M fs/cifs/cifsfs.c
:100644 100644 d82f5fb... d6f0a7d... M fs/cifs/cifsfs.h
:100644 100644 c9d0cfc... 784fd4a... M fs/cifs/cifsglob.h
:100644 100644 ec0ea4a... 9a3175b... M fs/cifs/connect.c
:100644 100644 ab70a3f... 735989a... M net/ipv6/addrconf.c

 fs/cifs/cifsfs.c    |   32 ++++++++++++++++++
 fs/cifs/cifsfs.h    |    3 ++
 fs/cifs/cifsglob.h  |    1 +
 fs/cifs/connect.c   |   91 +++++++++++++++++++++++++++++++++++++++++++++++++-
 net/ipv6/addrconf.c |    1 +
 5 files changed, 126 insertions(+), 2 deletions(-)
fs/cifs/cifsfs.c +0 -0
index b7431af..4510f4d 100644

fs/cifs/cifsfs.c cifs_show_address(struct seq_file *s, struct TCP_Server_Info *server) · line 366

Quoted 10 lines
fs/cifs/cifsfs.c +7 -0
@@ -357,6 +357,23 @@ cifs_show_address(struct seq_file *s, struct TCP_Server_Info *server)
 	}
 }
 
+bool
+cifs_addr_is_specified(struct sockaddr *srcaddr) {
+	struct sockaddr_in *saddr4 = (struct sockaddr_in *)srcaddr;
+	struct sockaddr_in6 *saddr6 = (struct sockaddr_in6 *)srcaddr;
+	switch (srcaddr->sa_family) {
+	case AF_INET:
+		return saddr4->sin_addr.s_addr != 0;
						^^^^
					nit: should probably be htonl(INADDR_ANY)
Quoted 89 lines
fs/cifs/cifsfs.c +25 -0
+#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
+	case AF_INET6:
+		/* in6addr_any isn't available if IPv6 isn't compiled in */
+		return (memcmp(&in6addr_any, &saddr6->sin6_addr,
+			       sizeof(in6addr_any)) != 0);
+#endif
+	}
+	return false;
+}
+
 /*
  * cifs_show_options() is for displaying mount options in /proc/mounts.
  * Not all settable options are displayed but most of the important
@@ -367,6 +384,8 @@ cifs_show_options(struct seq_file *s, struct vfsmount *m)
 {
 	struct cifs_sb_info *cifs_sb = CIFS_SB(m->mnt_sb);
 	struct cifsTconInfo *tcon = cifs_sb->tcon;
+	struct sockaddr *srcaddr;
+	srcaddr = (struct sockaddr *)(&tcon->ses->server->srcaddr);
 
 	seq_printf(s, ",unc=%s", tcon->treeName);
 	if (tcon->ses->userName)
@@ -374,6 +393,19 @@ cifs_show_options(struct seq_file *s, struct vfsmount *m)
 	if (tcon->ses->domainName)
 		seq_printf(s, ",domain=%s", tcon->ses->domainName);
 
+	if (cifs_addr_is_specified(srcaddr)) {
+		struct sockaddr_in *saddr4;
+		struct sockaddr_in6 *saddr6;
+		saddr4 = (struct sockaddr_in *)srcaddr;
+		saddr6 = (struct sockaddr_in6 *)srcaddr;
+		if (saddr6->sin6_family == AF_INET6)
+			seq_printf(s, ",srcaddr=%pI6c",
+				   &saddr6->sin6_addr);
+		else
+			seq_printf(s, ",srcaddr=%pI4",
+				   &saddr4->sin_addr.s_addr);
+	}
+
 	seq_printf(s, ",uid=%d", cifs_sb->mnt_uid);
 	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_OVERR_UID)
 		seq_printf(s, ",forceuid");
fs/cifs/cifsfs.h +3 -0
index d82f5fb..d6f0a7d 100644
@@ -110,6 +110,9 @@ extern ssize_t	cifs_getxattr(struct dentry *, const char *, void *, size_t);
 extern ssize_t	cifs_listxattr(struct dentry *, char *, size_t);
 extern long cifs_ioctl(struct file *filep, unsigned int cmd, unsigned long arg);
 
+struct sockaddr;
+extern bool cifs_addr_is_specified(struct sockaddr *srcaddr);
+
 #ifdef CONFIG_CIFS_EXPERIMENTAL
 extern const struct export_operations cifs_export_ops;
 #endif /* EXPERIMENTAL */
fs/cifs/cifsglob.h +1 -0
index c9d0cfc..784fd4a 100644
@@ -157,6 +157,7 @@ struct TCP_Server_Info {
 		struct sockaddr_in sockAddr;
 		struct sockaddr_in6 sockAddr6;
 	} addr;
+	struct sockaddr_storage srcaddr; /* locally bind to this IP */
 	wait_queue_head_t response_q;
 	wait_queue_head_t request_q; /* if more than maxmpx to srvr must block*/
 	struct list_head pending_mid_q;
fs/cifs/connect.c +2 -0
index ec0ea4a..9a3175b 100644
@@ -49,6 +49,7 @@
 #include "rfc1002pdu.h"
 #include "cn_cifs.h"
 #include "fscache.h"
+#include "cifsfs.h"
 
 #define CIFS_PORT 445
 #define RFC1001_PORT 139
@@ -105,6 +106,7 @@ struct smb_vol {
 	bool sockopt_tcp_nodelay:1;
 	unsigned short int port;
 	char *prepath;
+	struct sockaddr_storage srcaddr; /* allow binding to a local IP */
 	struct nls_table *local_nls;
 };
 

fs/cifs/connect.c cifs_parse_mount_options(char *options, const char *devname, · line 1070

Quoted 5 lines
fs/cifs/connect.c +2 -0
@@ -1064,6 +1066,22 @@ cifs_parse_mount_options(char *options, const char *devname,
 						    "long\n");
 				return 1;
 			}
+		} else if (strnicmp(data, "srcaddr", 8) == 0) {
+			memset(&vol->srcaddr, 0, sizeof(vol->srcaddr));

^^^^^^^^^
Is this necessary? vol is kzalloc'ed so it should be zeroed out already. Or are you worried about people specifying srcaddr= more than once?

Quoted 19 lines
fs/cifs/connect.c +14 -0
+
+			if (!value || !*value) {
+				printk(KERN_WARNING "CIFS: srcaddr value"
+				       " not specified.\n");
+				return 1;	/* needs_arg; */
+			}
+			i = cifs_convert_address((struct sockaddr *)&vol->srcaddr,
+						 value, strlen(value));
+			if (i < 0) {
+				printk(KERN_WARNING "CIFS:  Could not parse"
+				       " srcaddr: %s\n",
+				       value);
+				return 1;
+			}
 		} else if (strnicmp(data, "prefixpath", 10) == 0) {
 			if (!value || !*value) {
 				printk(KERN_WARNING
@@ -1392,8 +1410,37 @@ cifs_parse_mount_options(char *options, const char *devname,
 	return 0;

fs/cifs/connect.c cifs_parse_mount_options(char *options, const char *devname, · line 1419

Quoted 10 lines
fs/cifs/connect.c +6 -1
 }
 
+/** Returns true if srcaddr isn't specified or if it matches
+ * the IP address of the rhs argument.
+ */
 static bool
-match_address(struct TCP_Server_Info *server, struct sockaddr *addr)
+srcip_matches(struct sockaddr *srcaddr, struct sockaddr *rhs)
+{
+	if (cifs_addr_is_specified(srcaddr)) {

Just wondering what the expected behavior would be here...

Suppose I have a clustered IP address set up, and I want to ensure that one of my CIFS mounts use that as the srcaddr. Now, suppose I make another mount to the same server, but don't specify the srcaddr there. That mount uses the same socket as the first one (that is bound to the srcaddr). Now, suppose I unmount the first mount and then fail over the IP address. What happens to the second one? (Nothing good, I'd wager...)

Perhaps we should consider using a different socket when the srcaddr isn't specified?

fs/cifs/connect.c cifs_parse_mount_options(char *options, const char *devname, · line 1432

Quoted 13 lines
fs/cifs/connect.c +13 -0
+		struct sockaddr_in *saddr4 = (struct sockaddr_in *)srcaddr;
+		struct sockaddr_in6 *saddr6 = (struct sockaddr_in6 *)srcaddr;
+		struct sockaddr_in *vaddr4 = (struct sockaddr_in *)rhs;
+		struct sockaddr_in6 *vaddr6 = (struct sockaddr_in6 *)&rhs;
+
+		switch (srcaddr->sa_family) {
+		case AF_INET:
+			if (saddr4->sin_addr.s_addr != vaddr4->sin_addr.s_addr)
+				return false;
+			break;
+		case AF_INET6:
+			if (memcmp(&saddr6->sin6_addr, &vaddr6->sin6_addr,
+				   sizeof(saddr6->sin6_addr)) != 0)

^^^ should use ipv6_addr_equal, unless there's some reason not to do so...

Quoted 44 lines
fs/cifs/connect.c +18 -1
+				return false;
+			break;
+		}
+	}
+	return true;
+}
+
+
+static bool
+match_address(struct TCP_Server_Info *server, struct sockaddr *addr,
+	      struct sockaddr *srcaddr)
 {
 	struct sockaddr_in *addr4 = (struct sockaddr_in *)addr;
 	struct sockaddr_in6 *addr6 = (struct sockaddr_in6 *)addr;
@@ -1420,6 +1467,9 @@ match_address(struct TCP_Server_Info *server, struct sockaddr *addr)
 		break;
 	}
 
+	if (!srcip_matches(srcaddr, (struct sockaddr *)&server->srcaddr))
+		return false;
+
 	return true;
 }
 
@@ -1487,7 +1537,8 @@ cifs_find_tcp_session(struct sockaddr *addr, struct smb_vol *vol)
 		if (server->tcpStatus == CifsNew)
 			continue;
 
-		if (!match_address(server, addr))
+		if (!match_address(server, addr,
+				   (struct sockaddr *)&vol->srcaddr))
 			continue;
 
 		if (!match_security(server, vol))
@@ -1602,6 +1653,8 @@ cifs_get_tcp_session(struct smb_vol *volume_info)
 	 * no need to spinlock this init of tcpStatus or srv_count
 	 */
 	tcp_ses->tcpStatus = CifsNew;
+	memcpy(&tcp_ses->srcaddr, &volume_info->srcaddr,
+	       sizeof(tcp_ses->srcaddr));
 	++tcp_ses->srv_count;
 
 	if (addr.ss_family == AF_INET6) {

fs/cifs/connect.c cifs_find_smb_ses(struct TCP_Server_Info *server, struct smb_vol *vol) · line 1736

Quoted 6 lines
fs/cifs/connect.c +3 -0
@@ -1678,6 +1731,9 @@ cifs_find_smb_ses(struct TCP_Server_Info *server, struct smb_vol *vol)
 				    vol->password ? vol->password : "",
 				    MAX_PASSWORD_SIZE))
 				continue;
+			if (!srcip_matches((struct sockaddr *)&server->srcaddr,
+					   (struct sockaddr *)&ses->server->srcaddr))
+				continue;

^^^ there should be no need for this. By the time you get here, you've already determined whether you can use the same socket as another one (via the check in match_address).

Quoted 60 lines
fs/cifs/connect.c +31 -0
 		}
 		++ses->ses_count;
 		write_unlock(&cifs_tcp_ses_lock);
@@ -2026,6 +2082,33 @@ static void rfc1002mangle(char *target, char *source, unsigned int length)
 
 }
 
+static int
+bind_socket(struct TCP_Server_Info *server)
+{
+	int rc = 0;
+	if (cifs_addr_is_specified((struct sockaddr *)&server->srcaddr)) {
+		/* Bind to the local IP address if specified */
+		struct socket *socket = server->ssocket;
+		rc = socket->ops->bind(socket,
+				       (struct sockaddr *) &server->srcaddr,
+				       sizeof(server->srcaddr));
+		if (rc < 0) {
+			struct sockaddr_in *saddr4;
+			struct sockaddr_in6 *saddr6;
+			saddr4 = (struct sockaddr_in *)&server->srcaddr;
+			saddr6 = (struct sockaddr_in6 *)&server->srcaddr;
+			if (saddr6->sin6_family == AF_INET6)
+				printk(KERN_WARNING "cifs: "
+				       "Failed to bind to: %pI6c, error: %d\n",
+				       &saddr6->sin6_addr, rc);
+			else
+				printk(KERN_WARNING "cifs: "
+				       "Failed to bind to: %pI4, error: %d\n",
+				       &saddr4->sin_addr.s_addr, rc);
+		}
+	}
+	return rc;
+}
 
 static int
 ipv4_connect(struct TCP_Server_Info *server)
@@ -2051,6 +2134,8 @@ ipv4_connect(struct TCP_Server_Info *server)
 		cifs_reclassify_socket4(socket);
 	}
 
+	bind_socket(server);
+
 	/* user overrode default port */
 	if (server->addr.sockAddr.sin_port) {
 		rc = socket->ops->connect(socket, (struct sockaddr *)
@@ -2213,6 +2298,8 @@ ipv6_connect(struct TCP_Server_Info *server)
 		cifs_reclassify_socket6(socket);
 	}
 
+	bind_socket(server);
+
 	/* user overrode default port */
 	if (server->addr.sockAddr6.sin6_port) {
 		rc = socket->ops->connect(socket,
net/ipv6/addrconf.c +0 -0
index ab70a3f..735989a 100644

net/ipv6/addrconf.c static struct ipv6_devconf ipv6_devconf_dflt __read_mostly = { · line 233

Quoted 4 lines
net/ipv6/addrconf.c +1 -0
@@ -230,6 +230,7 @@ static struct ipv6_devconf ipv6_devconf_dflt __read_mostly = {
 
 /* IPv6 Wildcard Address and Loopback Address defined by RFC2553 */
 const struct in6_addr in6addr_any = IN6ADDR_ANY_INIT;
+EXPORT_SYMBOL(in6addr_any);

There should be no need at all to touch stuff outside of cifs.ko, at least, not in this patch. If you want to do this, you need to do it as a separate patch and float it in linux-netdev or someplace like that.

In the meantime, I'd suggest declaring a private in6_addr variable and initializing it to IN6ADDR_ANY_INIT (similar to what the sunrpc code does).

Quoted 3 lines
net/ipv6/addrconf.c +0 -0
 const struct in6_addr in6addr_loopback = IN6ADDR_LOOPBACK_INIT;
 const struct in6_addr in6addr_linklocal_allnodes = IN6ADDR_LINKLOCAL_ALLNODES_INIT;
 const struct in6_addr in6addr_linklocal_allrouters = IN6ADDR_LINKLOCAL_ALLROUTERS_INIT;
Signature · 1 line
Jeff Layton <[email protected]>

A message that is not in the archive (sent off-list or not yet received): 20100831100649.3c9121df-4QP7MXygkU+dMjc06nkz3ljfA9RmPOcC@public.gmane.org

author
#5 lore raw
Quoted 10 lines · Jeff Layton
Quoted 6 lines · On Mon, 30 Aug 2010 22:40:22 -0700 Ben Greear<greearb-my8/4N5VtI7c+919tysfdA@pu…

When using multi-homed machines, it's nice to be able to specify the local IP to use for outbound connections. This patch gives cifs the ability to bind to a particular IP address.

   Usage:  mount -t cifs -o srcaddr=192.168.1.50,user=foo, ...
   Usage:  mount -t cifs -o srcaddr=2002::100:1,user=foo, ...
Quoted 9 lines

fs/cifs/cifsfs.c cifs_show_address(struct seq_file *s, struct TCP_Server_Info *server) · line 366

Quoted 7 lines
fs/cifs/cifsfs.c +7 -0
+bool
+cifs_addr_is_specified(struct sockaddr *srcaddr) {
+	struct sockaddr_in *saddr4 = (struct sockaddr_in *)srcaddr;
+	struct sockaddr_in6 *saddr6 = (struct sockaddr_in6 *)srcaddr;
+	switch (srcaddr->sa_family) {
+	case AF_INET:
+		return saddr4->sin_addr.s_addr != 0;
						^^^^
					nit: should probably be htonl(INADDR_ANY)

We don't initialize it to INADDR_ANY, so I'd rather not check against it, even though now and forever it's going to be zero. But, will change if that's the consensus.

Quoted 11 lines

fs/cifs/connect.c cifs_parse_mount_options(char *options, const char *devname, · line 1070

Quoted 6 lines
fs/cifs/connect.c +2 -0
@@ -1064,6 +1066,22 @@ cifs_parse_mount_options(char *options, const char *devname,
  						    "long\n");
  				return 1;
  			}
+		} else if (strnicmp(data, "srcaddr", 8) == 0) {
+			memset(&vol->srcaddr, 0, sizeof(vol->srcaddr));

^^^^^^^^^
Is this necessary? vol is kzalloc'ed so it should be zeroed out already. Or are you worried about people specifying srcaddr= more than once?

It just seems cleaner in general to set it to known value in case errors later in that method keep us from setting it to a proper new value. And, certainly useful if it can be set twice somehow..but not sure if that's currently the case or not.

Quoted 41 lines
Quoted 20 lines
fs/cifs/connect.c +14 -0
+
+			if (!value || !*value) {
+				printk(KERN_WARNING "CIFS: srcaddr value"
+				       " not specified.\n");
+				return 1;	/* needs_arg; */
+			}
+			i = cifs_convert_address((struct sockaddr *)&vol->srcaddr,
+						 value, strlen(value));
+			if (i<  0) {
+				printk(KERN_WARNING "CIFS:  Could not parse"
+				       " srcaddr: %s\n",
+				       value);
+				return 1;
+			}
  		} else if (strnicmp(data, "prefixpath", 10) == 0) {
  			if (!value || !*value) {
  				printk(KERN_WARNING
@@ -1392,8 +1410,37 @@ cifs_parse_mount_options(char *options, const char *devname,
  	return 0;

fs/cifs/connect.c cifs_parse_mount_options(char *options, const char *devname, · line 1419

Quoted 10 lines
fs/cifs/connect.c +6 -1
  }
 
+/** Returns true if srcaddr isn't specified or if it matches
+ * the IP address of the rhs argument.
+ */
  static bool
-match_address(struct TCP_Server_Info *server, struct sockaddr *addr)
+srcip_matches(struct sockaddr *srcaddr, struct sockaddr *rhs)
+{
+	if (cifs_addr_is_specified(srcaddr)) {

Just wondering what the expected behavior would be here...

Suppose I have a clustered IP address set up, and I want to ensure that one of my CIFS mounts use that as the srcaddr. Now, suppose I make another mount to the same server, but don't specify the srcaddr there. That mount uses the same socket as the first one (that is bound to the srcaddr). Now, suppose I unmount the first mount and then fail over the IP address. What happens to the second one? (Nothing good, I'd wager...)

Quoted 2 lines

Perhaps we should consider using a different socket when the srcaddr isn't specified?

The idea is that each mount with unique source addr gets it's own cifs session. They will not be shared even if the server is the same. If srcaddr isn't specified, then you get today's behaviour (shared so long as server is the same).

In your example, the bound and the un-bound shouldn't share anything, but it's possible my code doesn't fully do that. If so, I would consider that a bug.

Quoted 5 lines

fs/cifs/connect.c cifs_parse_mount_options(char *options, const char *devname, · line 1432

Quoted 3 lines
fs/cifs/connect.c +3 -0
+		case AF_INET6:
+			if (memcmp(&saddr6->sin6_addr,&vaddr6->sin6_addr,
+				   sizeof(saddr6->sin6_addr)) != 0)

^^^ should use ipv6_addr_equal, unless there's some reason not to do so...

addr_equal should be fine, I will change that.

Quoted 43 lines
Quoted 43 lines
fs/cifs/connect.c +18 -1
+				return false;
+			break;
+		}
+	}
+	return true;
+}
+
+
+static bool
+match_address(struct TCP_Server_Info *server, struct sockaddr *addr,
+	      struct sockaddr *srcaddr)
  {
  	struct sockaddr_in *addr4 = (struct sockaddr_in *)addr;
  	struct sockaddr_in6 *addr6 = (struct sockaddr_in6 *)addr;
@@ -1420,6 +1467,9 @@ match_address(struct TCP_Server_Info *server, struct sockaddr *addr)
  		break;
  	}
 
+	if (!srcip_matches(srcaddr, (struct sockaddr *)&server->srcaddr))
+		return false;
+
  	return true;
  }
 
@@ -1487,7 +1537,8 @@ cifs_find_tcp_session(struct sockaddr *addr, struct smb_vol *vol)
  		if (server->tcpStatus == CifsNew)
  			continue;
 
-		if (!match_address(server, addr))
+		if (!match_address(server, addr,
+				   (struct sockaddr *)&vol->srcaddr))
  			continue;
 
  		if (!match_security(server, vol))
@@ -1602,6 +1653,8 @@ cifs_get_tcp_session(struct smb_vol *volume_info)
  	 * no need to spinlock this init of tcpStatus or srv_count
  	 */
  	tcp_ses->tcpStatus = CifsNew;
+	memcpy(&tcp_ses->srcaddr,&volume_info->srcaddr,
+	       sizeof(tcp_ses->srcaddr));
  	++tcp_ses->srv_count;
 
Quoted 13 lines

fs/cifs/connect.c cifs_find_smb_ses(struct TCP_Server_Info *server, struct smb_vol *vol) · line 1736

Quoted 8 lines
fs/cifs/connect.c +3 -0
  	if (addr.ss_family == AF_INET6) {
@@ -1678,6 +1731,9 @@ cifs_find_smb_ses(struct TCP_Server_Info *server, struct smb_vol *vol)
  				    vol->password ? vol->password : "",
  				    MAX_PASSWORD_SIZE))
  				continue;
+			if (!srcip_matches((struct sockaddr *)&server->srcaddr,
+					   (struct sockaddr *)&ses->server->srcaddr))
+				continue;

^^^ there should be no need for this. By the time you get here, you've already determined whether you can use the same socket as another one (via the check in match_address).

Ok, I'll remove that.

Quoted 13 lines

net/ipv6/addrconf.c static struct ipv6_devconf ipv6_devconf_dflt __read_mostly = { · line 233

Quoted 8 lines

index ab70a3f..735989a 100644

net/ipv6/addrconf.c +1 -0
@@ -230,6 +230,7 @@ static struct ipv6_devconf ipv6_devconf_dflt __read_mostly = {
 
  /* IPv6 Wildcard Address and Loopback Address defined by RFC2553 */
  const struct in6_addr in6addr_any = IN6ADDR_ANY_INIT;
+EXPORT_SYMBOL(in6addr_any);

There should be no need at all to touch stuff outside of cifs.ko, at least, not in this patch. If you want to do this, you need to do it as a separate patch and float it in linux-netdev or someplace like that.

Quoted 3 lines

In the meantime, I'd suggest declaring a private in6_addr variable and initializing it to IN6ADDR_ANY_INIT (similar to what the sunrpc code does).

Sounds good. I'll see if davem will accept such a patch, and will use a private copy in cifs until such time as the ipv6 patch goes in.

Thanks for the review!

Thanks,
Ben

Signature · 2 lines
Ben Greear <greearb-my8/[email protected]>
Candela Technologies Inc  http://www.candelatech.com

A message that is not in the archive (sent off-list or not yet received): 4C7D3177.7070906-my8/[email protected]

#7 lore raw
Quoted 21 lines · On Tue, 31 Aug 2010 09:44:39 -0700 Ben Greear…
Quoted 10 lines · Jeff Layton
Quoted 6 lines · On Mon, 30 Aug 2010 22:40:22 -0700 Ben Greear<greearb-my8/4N5VtI7c+919tysfdA@pu…

When using multi-homed machines, it's nice to be able to specify the local IP to use for outbound connections. This patch gives cifs the ability to bind to a particular IP address.

   Usage:  mount -t cifs -o srcaddr=192.168.1.50,user=foo, ...
   Usage:  mount -t cifs -o srcaddr=2002::100:1,user=foo, ...
Quoted 9 lines

fs/cifs/cifsfs.c cifs_show_address(struct seq_file *s, struct TCP_Server_Info *server) · line 366

Quoted 7 lines
fs/cifs/cifsfs.c +7 -0
+bool
+cifs_addr_is_specified(struct sockaddr *srcaddr) {
+	struct sockaddr_in *saddr4 = (struct sockaddr_in *)srcaddr;
+	struct sockaddr_in6 *saddr6 = (struct sockaddr_in6 *)srcaddr;
+	switch (srcaddr->sa_family) {
+	case AF_INET:
+		return saddr4->sin_addr.s_addr != 0;
						^^^^
					nit: should probably be htonl(INADDR_ANY)
Quoted 4 lines

We don't initialize it to INADDR_ANY, so I'd rather not check against it, even though now and forever it's going to be zero. But, will change if that's the consensus.

What may be best instead is to just set sa_family to AF_UNSPEC when it's not specified. An sa_family that was AF_INET or AF_INET6 would imply that it had been specified. That way someone could reasonably force a ->bind to occur to INADDR_ANY (as silly as that sounds).

Quoted 12 lines
Quoted 11 lines

fs/cifs/connect.c cifs_parse_mount_options(char *options, const char *devname, · line 1070

Quoted 6 lines
fs/cifs/connect.c +2 -0
@@ -1064,6 +1066,22 @@ cifs_parse_mount_options(char *options, const char *devname,
  						    "long\n");
  				return 1;
  			}
+		} else if (strnicmp(data, "srcaddr", 8) == 0) {
+			memset(&vol->srcaddr, 0, sizeof(vol->srcaddr));

^^^^^^^^^
Is this necessary? vol is kzalloc'ed so it should be zeroed out already. Or are you worried about people specifying srcaddr= more than once?

Quoted 5 lines

It just seems cleaner in general to set it to known value in case errors later in that method keep us from setting it to a proper new value. And, certainly useful if it can be set twice somehow..but not sure if that's currently the case or not.

Fair enough. There's probably no need to zero this out entirely though. Setting the family to AF_UNSPEC should be sufficient.

Quoted 44 lines
Quoted 41 lines
Quoted 20 lines
fs/cifs/connect.c +14 -0
+
+			if (!value || !*value) {
+				printk(KERN_WARNING "CIFS: srcaddr value"
+				       " not specified.\n");
+				return 1;	/* needs_arg; */
+			}
+			i = cifs_convert_address((struct sockaddr *)&vol->srcaddr,
+						 value, strlen(value));
+			if (i<  0) {
+				printk(KERN_WARNING "CIFS:  Could not parse"
+				       " srcaddr: %s\n",
+				       value);
+				return 1;
+			}
  		} else if (strnicmp(data, "prefixpath", 10) == 0) {
  			if (!value || !*value) {
  				printk(KERN_WARNING
@@ -1392,8 +1410,37 @@ cifs_parse_mount_options(char *options, const char *devname,
  	return 0;

fs/cifs/connect.c cifs_parse_mount_options(char *options, const char *devname, · line 1419

Quoted 10 lines
fs/cifs/connect.c +6 -1
  }
 
+/** Returns true if srcaddr isn't specified or if it matches
+ * the IP address of the rhs argument.
+ */
  static bool
-match_address(struct TCP_Server_Info *server, struct sockaddr *addr)
+srcip_matches(struct sockaddr *srcaddr, struct sockaddr *rhs)
+{
+	if (cifs_addr_is_specified(srcaddr)) {

Just wondering what the expected behavior would be here...

Suppose I have a clustered IP address set up, and I want to ensure that one of my CIFS mounts use that as the srcaddr. Now, suppose I make another mount to the same server, but don't specify the srcaddr there. That mount uses the same socket as the first one (that is bound to the srcaddr). Now, suppose I unmount the first mount and then fail over the IP address. What happens to the second one? (Nothing good, I'd wager...)

Quoted 2 lines

Perhaps we should consider using a different socket when the srcaddr isn't specified?

Quoted 9 lines

The idea is that each mount with unique source addr gets it's own cifs session. They will not be shared even if the server is the same. If srcaddr isn't specified, then you get today's behaviour (shared so long as server is the same).

In your example, the bound and the un-bound shouldn't share anything, but it's possible my code doesn't fully do that. If so, I would consider that a bug.

Ok, I think that is a bug here. srcip_matches returns true when cifs_addr_is_specified returns false. So if you don't specify a srcaddr= option, you'll match any session, right?

Quoted 80 lines
Quoted 5 lines

fs/cifs/connect.c cifs_parse_mount_options(char *options, const char *devname, · line 1432

Quoted 3 lines
fs/cifs/connect.c +3 -0
+		case AF_INET6:
+			if (memcmp(&saddr6->sin6_addr,&vaddr6->sin6_addr,
+				   sizeof(saddr6->sin6_addr)) != 0)

^^^ should use ipv6_addr_equal, unless there's some reason not to do so...

addr_equal should be fine, I will change that.

Quoted 43 lines
Quoted 43 lines
fs/cifs/connect.c +18 -1
+				return false;
+			break;
+		}
+	}
+	return true;
+}
+
+
+static bool
+match_address(struct TCP_Server_Info *server, struct sockaddr *addr,
+	      struct sockaddr *srcaddr)
  {
  	struct sockaddr_in *addr4 = (struct sockaddr_in *)addr;
  	struct sockaddr_in6 *addr6 = (struct sockaddr_in6 *)addr;
@@ -1420,6 +1467,9 @@ match_address(struct TCP_Server_Info *server, struct sockaddr *addr)
  		break;
  	}
 
+	if (!srcip_matches(srcaddr, (struct sockaddr *)&server->srcaddr))
+		return false;
+
  	return true;
  }
 
@@ -1487,7 +1537,8 @@ cifs_find_tcp_session(struct sockaddr *addr, struct smb_vol *vol)
  		if (server->tcpStatus == CifsNew)
  			continue;
 
-		if (!match_address(server, addr))
+		if (!match_address(server, addr,
+				   (struct sockaddr *)&vol->srcaddr))
  			continue;
 
  		if (!match_security(server, vol))
@@ -1602,6 +1653,8 @@ cifs_get_tcp_session(struct smb_vol *volume_info)
  	 * no need to spinlock this init of tcpStatus or srv_count
  	 */
  	tcp_ses->tcpStatus = CifsNew;
+	memcpy(&tcp_ses->srcaddr,&volume_info->srcaddr,
+	       sizeof(tcp_ses->srcaddr));
  	++tcp_ses->srv_count;
 
Quoted 13 lines

fs/cifs/connect.c cifs_find_smb_ses(struct TCP_Server_Info *server, struct smb_vol *vol) · line 1736

Quoted 8 lines
fs/cifs/connect.c +3 -0
  	if (addr.ss_family == AF_INET6) {
@@ -1678,6 +1731,9 @@ cifs_find_smb_ses(struct TCP_Server_Info *server, struct smb_vol *vol)
  				    vol->password ? vol->password : "",
  				    MAX_PASSWORD_SIZE))
  				continue;
+			if (!srcip_matches((struct sockaddr *)&server->srcaddr,
+					   (struct sockaddr *)&ses->server->srcaddr))
+				continue;

^^^ there should be no need for this. By the time you get here, you've already determined whether you can use the same socket as another one (via the check in match_address).

Ok, I'll remove that.

Quoted 13 lines

net/ipv6/addrconf.c static struct ipv6_devconf ipv6_devconf_dflt __read_mostly = { · line 233

Quoted 8 lines

index ab70a3f..735989a 100644

net/ipv6/addrconf.c +1 -0
@@ -230,6 +230,7 @@ static struct ipv6_devconf ipv6_devconf_dflt __read_mostly = {
 
  /* IPv6 Wildcard Address and Loopback Address defined by RFC2553 */
  const struct in6_addr in6addr_any = IN6ADDR_ANY_INIT;
+EXPORT_SYMBOL(in6addr_any);

There should be no need at all to touch stuff outside of cifs.ko, at least, not in this patch. If you want to do this, you need to do it as a separate patch and float it in linux-netdev or someplace like that.

Quoted 9 lines
Quoted 3 lines

In the meantime, I'd suggest declaring a private in6_addr variable and initializing it to IN6ADDR_ANY_INIT (similar to what the sunrpc code does).

Sounds good. I'll see if davem will accept such a patch, and will use a private copy in cifs until such time as the ipv6 patch goes in.

Thanks for the review!

No problem.

Signature · 1 line
Jeff Layton <[email protected]>

A message that is not in the archive (sent off-list or not yet received): 20100831145023.09253a26-9yPaYZwiELC+kQycOl6kW4xkIHaj4LzF@public.gmane.org

author
#9 lore raw
Quoted 6 lines · Jeff Layton

On Tue, 31 Aug 2010 09:44:39 -0700

What may be best instead is to just set sa_family to AF_UNSPEC when it's not specified. An sa_family that was AF_INET or AF_INET6 would imply that it had been specified. That way someone could reasonably force a ->bind to occur to INADDR_ANY (as silly as that sounds).

Ok, I'll do that.

Quoted 8 lines
Quoted 5 lines

It just seems cleaner in general to set it to known value in case errors later in that method keep us from setting it to a proper new value. And, certainly useful if it can be set twice somehow..but not sure if that's currently the case or not.

Fair enough. There's probably no need to zero this out entirely though. Setting the family to AF_UNSPEC should be sufficient.

Will do that in next patch.

Quoted 23 lines
Quoted 23 lines
Quoted 15 lines

Review

Quoted 4 lines
(quoted diff) +3 -1
-match_address(struct TCP_Server_Info *server, struct sockaddr *addr)
+srcip_matches(struct sockaddr *srcaddr, struct sockaddr *rhs)
+{
+	if (cifs_addr_is_specified(srcaddr)) {

Just wondering what the expected behavior would be here...

Suppose I have a clustered IP address set up, and I want to ensure that one of my CIFS mounts use that as the srcaddr. Now, suppose I make another mount to the same server, but don't specify the srcaddr there. That mount uses the same socket as the first one (that is bound to the srcaddr). Now, suppose I unmount the first mount and then fail over the IP address. What happens to the second one? (Nothing good, I'd wager...)

Quoted 2 lines

Perhaps we should consider using a different socket when the srcaddr isn't specified?

The idea is that each mount with unique source addr gets it's own cifs session. They will not be shared even if the server is the same. If srcaddr isn't specified, then you get today's behaviour (shared so long as server is the same).

Quoted 8 lines
Quoted 4 lines

In your example, the bound and the un-bound shouldn't share anything, but it's possible my code doesn't fully do that. If so, I would consider that a bug.

Ok, I think that is a bug here. srcip_matches returns true when cifs_addr_is_specified returns false. So if you don't specify a srcaddr= option, you'll match any session, right?

Ok, I think you're right, I'll work on a fix and post a new patch shortly.

Thanks,
Ben

Signature · 2 lines
Ben Greear <greearb-my8/[email protected]>
Candela Technologies Inc  http://www.candelatech.com