iproute2/Patches/af_unix.dif

diff -ur ../vger3-011229/linux/net/unix/af_unix.c linux/net/unix/af_unix.c
--- ../vger3-011229/linux/net/unix/af_unix.c	Mon Dec  3 20:24:03 2001
+++ linux/net/unix/af_unix.c	Sat Jan  5 04:30:19 2002
@@ -112,6 +112,7 @@
 #include <asm/checksum.h>
 
 int sysctl_unix_max_dgram_qlen = 10;
+int sysctl_unix_stream_pages = MAX_SKB_FRAGS;
 
 unix_socket *unix_socket_table[UNIX_HASH_SIZE+1];
 rwlock_t unix_table_lock = RW_LOCK_UNLOCKED;
@@ -1123,9 +1124,6 @@
 	struct scm_cookie scm;
 	memset(&scm, 0, sizeof(scm));
 	unix_detach_fds(&scm, skb);
-
-	/* Alas, it calls VFS */
-	/* So fscking what? fput() had been SMP-safe since the last Summer */
 	scm_destroy(&scm);
 	sock_wfree(skb);
 }
@@ -1140,6 +1138,67 @@
 	scm->fp = NULL;
 }
 
+int datagram_copy_fromiovec(struct iovec *iov, struct sk_buff *skb, int size)
+{
+	struct sock *sk;
+	struct sk_buff **tail, *skb1;
+	int copy = min_t(int, size, skb_tailroom(skb));
+
+	if (memcpy_fromiovec(skb_put(skb, copy), iov, copy))
+		goto do_fault;
+
+	if ((size -= copy) == 0)
+		return 0;
+
+	sk = skb->sk;
+	skb1 = skb;
+	tail = &skb_shinfo(skb)->frag_list;
+
+	do {
+		struct page *page;
+		int i = skb_shinfo(skb1)->nr_frags;
+
+		if (i == MAX_SKB_FRAGS) {
+			skb1 = alloc_skb(0, sk->allocation);
+			if (skb1 == NULL)
+				goto do_oom;
+			*tail = skb1;
+			tail = &skb1->next;
+			i = 0;
+			skb->truesize += skb1->truesize;
+			atomic_add(skb1->truesize, &sk->wmem_alloc);
+		}
+
+		page = alloc_pages(sk->allocation, 0);
+		if (page == NULL)
+			goto do_oom;
+
+		copy = min_t(int, size, PAGE_SIZE);
+		skb_shinfo(skb1)->nr_frags=i+1;
+		skb_shinfo(skb1)->frags[i].page = page;
+		skb_shinfo(skb1)->frags[i].page_offset = 0;
+		skb_shinfo(skb1)->frags[i].size = copy;
+
+		skb1->len += copy;
+		skb1->data_len += copy;
+		if (skb != skb1) {
+			skb->len += copy;
+			skb->data_len += copy;
+		}
+		skb->truesize += PAGE_SIZE;
+		atomic_add(PAGE_SIZE, &sk->wmem_alloc);
+		if (memcpy_fromiovec(page_address(page), iov, copy))
+			goto do_fault;
+	} while ((size -= copy) > 0);
+	return 0;
+
+do_oom:
+	return -ENOMEM;
+
+do_fault:
+	return -EFAULT;
+}
+
 /*
  *	Send AF_UNIX data.
  */
@@ -1155,6 +1214,7 @@
 	unsigned hash;
 	struct sk_buff *skb;
 	long timeo;
+	int alloc;
 
 	err = -EOPNOTSUPP;
 	if (msg->msg_flags&MSG_OOB)
@@ -1178,10 +1238,14 @@
 		goto out;
 
 	err = -EMSGSIZE;
-	if ((unsigned)len > sk->sndbuf - 32)
+	if ((unsigned)len > sk->sndbuf)
 		goto out;
 
-	skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
+ 	alloc = len;
+ 	if (alloc > SKB_MAX_HEAD(0))
+ 		alloc = SKB_MAX_HEAD(0);
+
+	skb = sock_alloc_send_skb(sk, alloc, msg->msg_flags&MSG_DONTWAIT, &err);
 	if (skb==NULL)
 		goto out;
 
@@ -1190,7 +1254,7 @@
 		unix_attach_fds(scm, skb);
 
 	skb->h.raw = skb->data;
-	err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
+	err = datagram_copy_fromiovec(msg->msg_iov, skb, len);
 	if (err)
 		goto out_free;
 
@@ -1275,74 +1339,57 @@
 	return err;
 }
 
-		
 static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg, int len,
 			       struct scm_cookie *scm)
 {
 	struct sock *sk = sock->sk;
 	unix_socket *other = NULL;
-	struct sockaddr_un *sunaddr=msg->msg_name;
-	int err,size;
 	struct sk_buff *skb;
+	int err;
 	int sent=0;
 
 	err = -EOPNOTSUPP;
 	if (msg->msg_flags&MSG_OOB)
 		goto out_err;
 
-	if (msg->msg_namelen) {
-		err = (sk->state==TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP);
+	err = -ENOTCONN;
+	other = unix_peer_get(sk);
+	if (!other)
 		goto out_err;
-	} else {
-		sunaddr = NULL;
-		err = -ENOTCONN;
-		other = unix_peer_get(sk);
-		if (!other)
-			goto out_err;
-	}
 
 	if (sk->shutdown&SEND_SHUTDOWN)
 		goto pipe_err;
 
-	while(sent < len)
-	{
-		/*
-		 *	Optimisation for the fact that under 0.01% of X messages typically
-		 *	need breaking up.
-		 */
+	while(sent < len) {
+		int size, alloc;
 
-		size=len-sent;
+		size = len-sent;
 
 		/* Keep two messages in the pipe so it schedules better */
-		if (size > sk->sndbuf/2 - 64)
-			size = sk->sndbuf/2 - 64;
+		if (size > sk->sndbuf/2)
+			size = sk->sndbuf/2;
 
-		if (size > SKB_MAX_ALLOC)
-			size = SKB_MAX_ALLOC;
-			
 		/*
 		 *	Grab a buffer
 		 */
-		 
-		skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
+		alloc = size;
+
+		if (size > SKB_MAX_HEAD(0)) {
+			alloc = SKB_MAX_HEAD(0);
+			if (size > alloc + sysctl_unix_stream_pages*PAGE_SIZE)
+				size = alloc + sysctl_unix_stream_pages*PAGE_SIZE;
+		}
+
+ 		skb=sock_alloc_send_skb(sk,alloc,msg->msg_flags&MSG_DONTWAIT, &err);
 
 		if (skb==NULL)
 			goto out_err;
 
-		/*
-		 *	If you pass two values to the sock_alloc_send_skb
-		 *	it tries to grab the large buffer with GFP_NOFS
-		 *	(which can fail easily), and if it fails grab the
-		 *	fallback size buffer which is under a page and will
-		 *	succeed. [Alan]
-		 */
-		size = min_t(int, size, skb_tailroom(skb));
-
 		memcpy(UNIXCREDS(skb), &scm->creds, sizeof(struct ucred));
 		if (scm->fp)
 			unix_attach_fds(scm, skb);
 
-		if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
+ 		if ((err = datagram_copy_fromiovec(msg->msg_iov, skb, size)) != 0) {
 			kfree_skb(skb);
 			goto out_err;
 		}
@@ -1418,13 +1465,10 @@
 
 	scm->creds = *UNIXCREDS(skb);
 
-	if (!(flags & MSG_PEEK))
-	{
+	if (!(flags & MSG_PEEK)) {
 		if (UNIXCB(skb).fp)
 			unix_detach_fds(scm, skb);
-	}
-	else 
-	{
+	} else {
 		/* It is questionable: on PEEK we could:
 		   - do not return fds - good, but too simple 8)
 		   - return fds, and do not return them on read (old strategy,
@@ -1483,13 +1527,10 @@
 	return timeo;
 }
 
-
-
 static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg, int size,
 			       int flags, struct scm_cookie *scm)
 {
 	struct sock *sk = sock->sk;
-	struct sockaddr_un *sunaddr=msg->msg_name;
 	int copied = 0;
 	int check_creds = 0;
 	int target;
@@ -1515,21 +1556,18 @@
 
 	down(&sk->protinfo.af_unix.readsem);
 
-	do
-	{
+	do {
 		int chunk;
 		struct sk_buff *skb;
 
 		skb=skb_dequeue(&sk->receive_queue);
-		if (skb==NULL)
-		{
+		if (skb==NULL) {
 			if (copied >= target)
 				break;
 
 			/*
 			 *	POSIX 1003.1g mandates this order.
 			 */
-			 
 			if ((err = sock_error(sk)) != 0)
 				break;
 			if (sk->shutdown & RCV_SHUTDOWN)
@@ -1551,60 +1589,44 @@
 
 		if (check_creds) {
 			/* Never glue messages from different writers */
-			if (memcmp(UNIXCREDS(skb), &scm->creds, sizeof(scm->creds)) != 0) {
-				skb_queue_head(&sk->receive_queue, skb);
-				break;
-			}
+			if (memcmp(UNIXCREDS(skb), &scm->creds, sizeof(scm->creds)) != 0)
+				goto out_put_back;
 		} else {
 			/* Copy credentials */
 			scm->creds = *UNIXCREDS(skb);
 			check_creds = 1;
 		}
 
-		/* Copy address just once */
-		if (sunaddr)
-		{
-			unix_copy_addr(msg, skb->sk);
-			sunaddr = NULL;
-		}
+		chunk = min_t(int, skb->len - sk->protinfo.af_unix.copied, size);
+		err = skb_copy_datagram_iovec(skb, sk->protinfo.af_unix.copied, msg->msg_iov, chunk);
+		if (err)
+			goto out_put_back;
 
-		chunk = min_t(unsigned int, skb->len, size);
-		if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
-			skb_queue_head(&sk->receive_queue, skb);
-			if (copied == 0)
-				copied = -EFAULT;
-			break;
-		}
 		copied += chunk;
 		size -= chunk;
 
 		/* Mark read part of skb as used */
-		if (!(flags & MSG_PEEK))
-		{
-			skb_pull(skb, chunk);
-
+		if (!(flags & MSG_PEEK)) {
 			if (UNIXCB(skb).fp)
 				unix_detach_fds(scm, skb);
 
 			/* put the skb back if we didn't use it up.. */
-			if (skb->len)
-			{
-				skb_queue_head(&sk->receive_queue, skb);
-				break;
-			}
+			if ((sk->protinfo.af_unix.copied += chunk) < skb->len)
+				goto out_put_back;
+
+			sk->protinfo.af_unix.copied = 0;
 
 			kfree_skb(skb);
 
 			if (scm->fp)
 				break;
-		}
-		else
-		{
+		} else {
 			/* It is questionable, see note in unix_dgram_recvmsg.
 			 */
 			if (UNIXCB(skb).fp)
 				scm->fp = scm_fp_dup(UNIXCB(skb).fp);
 
+out_put_back:
 			/* put message back and return */
 			skb_queue_head(&sk->receive_queue, skb);
 			break;
@@ -1676,10 +1698,12 @@
 				break;
 			}
 
+			down(&sk->protinfo.af_unix.readsem);
 			spin_lock(&sk->receive_queue.lock);
 			if((skb=skb_peek(&sk->receive_queue))!=NULL)
-				amount=skb->len;
+				amount=skb->len - sk->protinfo.af_unix.copied;
 			spin_unlock(&sk->receive_queue.lock);
+			up(&sk->protinfo.af_unix.readsem);
 			err = put_user(amount, (int *)arg);
 			break;
 		}
@@ -1734,7 +1758,7 @@
 	int i;
 	unix_socket *s;
 	
-	len+= sprintf(buffer,"Num       RefCount Protocol Flags    Type St "
+	len+= sprintf(buffer,"Peer      RcvQueue WMem     Flags    Type St "
 	    "Inode Path\n");
 
 	read_lock(&unix_table_lock);
@@ -1742,10 +1766,10 @@
 	{
 		unix_state_rlock(s);
 
-		len+=sprintf(buffer+len,"%p: %08X %08X %08X %04X %02X %5ld",
-			s,
-			atomic_read(&s->refcnt),
-			0,
+		len+=sprintf(buffer+len,"%08lX: %08X %08X %08X %04X %02X %5ld",
+			unix_peer(s) ? sock_i_ino(unix_peer(s)) : 0,
+			skb_queue_len(&s->receive_queue),
+			atomic_read(&s->wmem_alloc),
 			s->state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
 			s->type,
 			s->socket ?
diff -ur ../vger3-011229/linux/net/unix/sysctl_net_unix.c linux/net/unix/sysctl_net_unix.c
--- ../vger3-011229/linux/net/unix/sysctl_net_unix.c	Tue Jan 30 21:20:16 2001
+++ linux/net/unix/sysctl_net_unix.c	Sat Jan  5 04:10:58 2002
@@ -13,10 +13,14 @@
 #include <linux/sysctl.h>
 
 extern int sysctl_unix_max_dgram_qlen;
+extern int sysctl_unix_stream_pages;
 
 ctl_table unix_table[] = {
 	{NET_UNIX_MAX_DGRAM_QLEN, "max_dgram_qlen",
 	&sysctl_unix_max_dgram_qlen, sizeof(int), 0600, NULL, 
+	 &proc_dointvec },
+	{NET_UNIX_STREAM_PAGES, "stream_pages",
+	&sysctl_unix_stream_pages, sizeof(int), 0600, NULL, 
 	 &proc_dointvec },
 	{0}
 };