ALPHA-pl14r

[...]

From: torvalds@cc.helsinki.fi (Linus Torvalds)
Newsgroups: comp.os.linux.announce
Subject: Linux 1.0 codefreeze imminent
Date: 18 Jan 1994 23:40:37 +0200

This is a general announcement of the imminent code-freeze that will
hopefully make linux 1.0 a reality.  The plan has been discussed a bit
with various developers already, and is already late, but is still in
effect otherwise.

In short, the next version of linux (0.99.15) will be a "full-featured"
release, and only obvious bug-fixes to existing features will be applied
before calling it 1.0.  If this means that your favourite feature or
networking version won't make it, don't despair: there is life even
after beta (and it's probably not worth mailing me about it any more:
I've seen quite a few favourite features already ;-).

In fact, 1.0 has little "real meaning", as far as development goes, but
should be taken as an indication that it can be used for real work
(which has been true for some time, depending on your definition of
"real work").  Development won't stop or even slow down: some of it has
even been shelved pending a 1.0 already.

Calling it 1.0 will not necessarily make all bugs go away (quite the
opposite, judging by some other programs), but I hope it will be a
reasonably stable release.  In order to accomplish this, the code-freeze
after 0.99.15 will be about a month, and I hope people will test out
that kernel heavily, instead of waiting for "the real release" so that
any potential bugs can be found and fixed.

As to where we are now: as of this moment, the latest release is the 'r'
version of pl14 (aka "ALPHA-pl14r").  I've made ALPHA releases available
on ftp.funet.fi almost daily, and expect a final pl15 within a few more
days.  Testing out the ALPHA releases is not discouraged either if you
like recompiling kernels every day or two..

And finally: we also try to create a "credits" file that mentions the
developers of the kernel and essential linux utilities.  The credit file
compilator is jmartin@opus.starlab.csc.com (John A. Martin), and if you
feel you have cause to be mentioned in it, please contact him.

		Linus

7 files changed, 157 insertions, 81 deletions

diff --git a/Makefile b/Makefile
index 7028cdf..49dbefd 100644
--- a/Makefile
+++ b/Makefile
@@ -1,6 +1,6 @@
 VERSION = 0.99
 PATCHLEVEL = 14
-ALPHA = q
+ALPHA = r
 
 all:	Version zImage
 
diff --git a/drivers/char/serial.c b/drivers/char/serial.c
index 0330087..c25c17b 100644
--- a/drivers/char/serial.c
+++ b/drivers/char/serial.c
@@ -1574,7 +1574,7 @@ static int block_til_ready(struct tty_struct *tty, struct file * filp,
 	 */
 	if (info->flags & ASYNC_CLOSING) {
 		interruptible_sleep_on(&info->close_wait);
-		return -ERESTARTSYS;
+		return -EAGAIN;
 	}
 
 	/*
@@ -1632,7 +1632,7 @@ static int block_til_ready(struct tty_struct *tty, struct file * filp,
 		current->state = TASK_INTERRUPTIBLE;
 		if (tty_hung_up_p(filp) ||
 		    !(info->flags & ASYNC_INITIALIZED)) {
-			retval = -ERESTARTSYS;
+			retval = -EAGAIN;
 			break;
 		}
 		if (!(info->flags & ASYNC_CALLOUT_ACTIVE) &&
diff --git a/include/linux/mman.h b/include/linux/mman.h
index 18e2539..4b42f73 100644
--- a/include/linux/mman.h
+++ b/include/linux/mman.h
@@ -10,5 +10,6 @@
 #define MAP_PRIVATE      2         /* Changes are private */
 #define MAP_TYPE         0xf       /* Mask for type of mapping */
 #define MAP_FIXED        0x10      /* Interpret addr exactly */
+#define MAP_ANONYMOUS    0x20      /* don't use a file */
 
 #endif /* _LINUX_MMAN_H */
diff --git a/mm/mmap.c b/mm/mmap.c
index 10e24df..5fea243 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -137,17 +137,20 @@ int do_mmap(struct file * file, unsigned long addr, unsigned long len,
 asmlinkage int sys_mmap(unsigned long *buffer)
 {
 	int error;
-	unsigned long fd;
-	struct file * file;
+	unsigned long flags;
+	struct file * file = NULL;
 
 	error = verify_area(VERIFY_READ, buffer, 6*4);
 	if (error)
 		return error;
-	fd = get_fs_long(buffer+4);
-	if (fd >= NR_OPEN || !(file = current->filp[fd]))
-		return -EBADF;
+	flags = get_fs_long(buffer+3);
+	if (!(flags & MAP_ANONYMOUS)) {
+		unsigned long fd = get_fs_long(buffer+4);
+		if (fd >= NR_OPEN || !(file = current->filp[fd]))
+			return -EBADF;
+	}
 	return do_mmap(file, get_fs_long(buffer), get_fs_long(buffer+1),
-		get_fs_long(buffer+2), get_fs_long(buffer+3), get_fs_long(buffer+5));
+		get_fs_long(buffer+2), flags, get_fs_long(buffer+5));
 }
 
 /*
diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index 6cbaa15..99845dd 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -158,7 +158,7 @@ void * vmalloc(unsigned long size)
 	area->size = size + PAGE_SIZE;
 	area->next = NULL;
 	for (p = &vmlist; (tmp = *p) ; p = &tmp->next) {
-		if (size + (unsigned long) addr <= (unsigned long) tmp->addr)
+		if (size + (unsigned long) addr < (unsigned long) tmp->addr)
 			break;
 		addr = (void *) (tmp->size + (unsigned long) tmp->addr);
 	}
diff --git a/net/inet/ip.c b/net/inet/ip.c
index 761f5f1..b31ee4c 100644
--- a/net/inet/ip.c
+++ b/net/inet/ip.c
@@ -800,6 +800,7 @@ static struct sk_buff *ip_glue(struct ipq *qp)
    	iph = skb->h.iph;
    	iph->frag_off = 0;
    	iph->tot_len = htons((iph->ihl * sizeof(unsigned long)) + count);
+   	skb->ip_hdr = iph;
    	return(skb);
 }
  
diff --git a/net/inet/tcp.c b/net/inet/tcp.c
index 15d2d13..f24226a 100644
--- a/net/inet/tcp.c
+++ b/net/inet/tcp.c
@@ -208,8 +208,15 @@ tcp_retransmit(struct sock *sk, int all)
 	return;
   }
 
-  if (sk->cong_window > 4)
-       sk->cong_window = sk->cong_window / 2;
+/*
+ *  If we had the full V-J mechanism, this might be right.  But
+ *  for the moment we want simple slow start after error.
+ *
+ *  if (sk->cong_window > 4)
+ *       sk->cong_window = sk->cong_window / 2;
+ */
+ 
+  sk->cong_window = 1;
   sk->exp_growth = 0;
 
   /* Do the actuall retransmit. */
@@ -613,6 +620,7 @@ tcp_send_partial(struct sock *sk)
   
   skb->h.seq = sk->send_seq;
   if (after(sk->send_seq , sk->window_seq) ||
+      sk->retransmits ||
       sk->packets_out >= sk->cong_window) {
 	DPRINTF((DBG_TCP, "sk->cong_window = %d, sk->packets_out = %d\n",
 					sk->cong_window, sk->packets_out));
@@ -833,67 +841,75 @@ tcp_write(struct sock *sk, unsigned char *from,
 
 	/* Now we need to check if we have a half built packet. */
 	if (sk->send_tmp != NULL) {
-		/* If sk->mss has been changed this could cause problems. */
+	        int hdrlen;
 
-		/* Add more stuff to the end of skb->len */
 		skb = sk->send_tmp;
+
+		         /* IP header + TCP header */
+		hdrlen = ((unsigned long)skb->h.th - (unsigned long)(skb+1))
+		         + sizeof(struct tcphdr);
+
+		/* If sk->mtu has been changed this could cause problems. */
+
+		/* Add more stuff to the end of skb->len */
 		if (!(flags & MSG_OOB)) {
-			copy = min(sk->mss - skb->len + 128 +
-				   prot->max_header, len);
-	      
-		/* FIXME: this is really a bug. */
-		if (copy <= 0) {
-			printk("TCP: **bug**: \"copy\" <= 0!!\n");
-			copy = 0;
-		}
+			copy = min(sk->mtu - (skb->len - hdrlen), len);
+			/* FIXME: this is really a bug. */
+			if (copy <= 0) {
+			  printk("TCP: **bug**: \"copy\" <= 0!!\n");
+			  copy = 0;
+			}
 	  
-		memcpy_fromfs((unsigned char *)(skb+1) + skb->len, from, copy);
-		skb->len += copy;
-		from += copy;
-		copied += copy;
-		len -= copy;
-		sk->send_seq += copy;
-	}
-
-	if (skb->len -(unsigned long)skb->h.th +
-	   (unsigned long)(skb+1) >= sk->mss ||(flags & MSG_OOB)) {
-		tcp_send_partial(sk);
+			memcpy_fromfs((unsigned char *)(skb+1) + skb->len, 
+				      from, copy);
+			skb->len += copy;
+			from += copy;
+			copied += copy;
+			len -= copy;
+			sk->send_seq += copy;
+		      }
+
+		if ((skb->len - hdrlen) > sk->mtu || (flags & MSG_OOB)) {
+		  tcp_send_partial(sk);
+		}
+		continue;
 	}
-	continue;
-  }
 
-  /*
-   * We also need to worry about the window.
-   * The smallest we will send is about 200 bytes.
-   * This is a bit sad for TCP/AMPR people running
-   * 196 byte windows! - FIXME
-   */
-  copy = min(sk->mtu, diff(sk->window_seq, sk->send_seq));
+	/*
+	 * We also need to worry about the window.
+	 * If window < 1/4 offered window, don't use it.  That's
+	 *   silly window prevention.  What we actually do is 
+	 *   use the whole MTU.  Since the results in the right
+	 *   edge of the packet being outside the window, it will
+	 *   be queued for later rather than sent.
+	 */
 
-  /* FIXME: redundent check here. */
-  if (copy < 200 || copy > sk->mtu) copy = sk->mtu;
-  copy = min(copy, len);
+	copy = diff(sk->window_seq, sk->send_seq);
+	if (copy < (diff(sk->window_seq, sk->rcv_ack_seq) >> 2))
+	  copy = sk->mtu;
+	copy = min(copy, sk->mtu);
+	copy = min(copy, len);
 
   /* We should really check the window here also. */
-  if (sk->packets_out && copy < sk->mss && !(flags & MSG_OOB)) {
+	if (sk->packets_out && copy < sk->mtu && !(flags & MSG_OOB)) {
 	/* We will release the socket incase we sleep here. */
-	release_sock(sk);
-	skb = (struct sk_buff *) prot->wmalloc(sk,
-			sk->mss + 128 + prot->max_header +
+	  release_sock(sk);
+	  skb = (struct sk_buff *) prot->wmalloc(sk,
+			sk->mtu + 128 + prot->max_header +
 			sizeof(*skb), 0, GFP_KERNEL);
-	sk->inuse = 1;
-	sk->send_tmp = skb;
-	if (skb != NULL)
-		skb->mem_len = sk->mss + 128 + prot->max_header + sizeof(*skb);
+	  sk->inuse = 1;
+	  sk->send_tmp = skb;
+	  if (skb != NULL)
+		skb->mem_len = sk->mtu + 128 + prot->max_header + sizeof(*skb);
 	} else {
 		/* We will release the socket incase we sleep here. */
-		release_sock(sk);
-		skb = (struct sk_buff *) prot->wmalloc(sk,
-				copy + prot->max_header +
-				sizeof(*skb), 0, GFP_KERNEL);
-		sk->inuse = 1;
-		if (skb != NULL)
-			skb->mem_len = copy+prot->max_header + sizeof(*skb);
+	  release_sock(sk);
+	  skb = (struct sk_buff *) prot->wmalloc(sk,
+						 copy + prot->max_header +
+						 sizeof(*skb), 0, GFP_KERNEL);
+	  sk->inuse = 1;
+	  if (skb != NULL)
+	    skb->mem_len = copy+prot->max_header + sizeof(*skb);
 	}
 
 	/* If we didn't get any memory, we need to sleep. */
@@ -980,6 +996,7 @@ tcp_write(struct sock *sk, unsigned char *from,
 
 	skb->h.seq = sk->send_seq;
 	if (after(sk->send_seq , sk->window_seq) ||
+	          sk->retransmits ||
 		  sk->packets_out >= sk->cong_window) {
 		DPRINTF((DBG_TCP, "sk->cong_window = %d, sk->packets_out = %d\n",
 					sk->cong_window, sk->packets_out));
@@ -998,8 +1015,23 @@ tcp_write(struct sock *sk, unsigned char *from,
 	}
   }
   sk->err = 0;
+
+/*
+ * In its original form, the following code implements
+ * Nagle's rule.  Everybody recommends it, and probably
+ * production code should use it.  But it slows down response enough
+ * that I dont' like it.  So as long as we're not bumping into 
+ * the window, I go ahead and send it.  If you want Nagle, just
+ * define USE_NAGLE  --C. Hedrick
+ */
+
   /* Avoid possible race on send_tmp - c/o Johannes Stille */
-  if(sk->send_tmp && !sk->packets_out)
+  if(sk->send_tmp && 
+     ((!sk->packets_out) 
+#ifndef USE_NAGLE
+      || before(sk->send_seq , sk->window_seq)
+#endif
+      ))
   	tcp_send_partial(sk);
   /* -- */
   release_sock(sk);
@@ -1696,7 +1728,6 @@ tcp_options(struct sock *sk, struct tcphdr *th)
 {
   unsigned char *ptr;
   int length=(th->doff*4)-sizeof(struct tcphdr);
-  int mtuset=0;
     
   ptr = (unsigned char *)(th + 1);
   
@@ -1721,7 +1752,6 @@ tcp_options(struct sock *sk, struct tcphdr *th)
   					if(opsize==4)
   					{
   						sk->mtu=min(sk->mtu,ntohs(*(unsigned short *)ptr));
-  						mtuset=1;
   					}
   					break;
   				/* Add other options here as people feel the urge to implement stuff like large windows */
@@ -1730,12 +1760,7 @@ tcp_options(struct sock *sk, struct tcphdr *th)
   			length-=opsize;
   	}
   }
-  					
-  if (!mtuset) 
-  {
-	sk->mtu = min(sk->mtu, 576 - HEADER_SIZE);
-	return;
-  }
+
 }
 
 /*
@@ -1857,6 +1882,16 @@ tcp_conn_request(struct sock *sk, struct sk_buff *skb,
   newsk->ip_ttl=sk->ip_ttl;
   newsk->ip_tos=skb->ip_hdr->tos;
 
+/* use 512 or whatever user asked for */
+/* note use of sk->mss, since user has no direct access to newsk */
+  if (sk->mss)
+    newsk->mtu = sk->mss;
+  else
+    newsk->mtu = 576 - HEADER_SIZE;
+/* but not bigger than device MTU */
+  newsk->mtu = min(newsk->mtu, dev->mtu - HEADER_SIZE);
+
+/* this will min with what arrived in the packet */
   tcp_options(newsk,skb->h.th);
 
   buff = (struct sk_buff *) newsk->prot->wmalloc(newsk, MAX_SYN_SIZE, 1, GFP_ATOMIC);
@@ -1916,8 +1951,8 @@ tcp_conn_request(struct sock *sk, struct sk_buff *skb,
   ptr =(unsigned char *)(t1+1);
   ptr[0] = 2;
   ptr[1] = 4;
-  ptr[2] =((dev->mtu - HEADER_SIZE) >> 8) & 0xff;
-  ptr[3] =(dev->mtu - HEADER_SIZE) & 0xff;
+  ptr[2] = ((newsk->mtu) >> 8) & 0xff;
+  ptr[3] =(newsk->mtu) & 0xff;
 
   tcp_send_check(t1, daddr, saddr, sizeof(*t1)+4, newsk);
   newsk->prot->queue_xmit(newsk, dev, buff, 0);
@@ -2097,7 +2132,8 @@ tcp_write_xmit(struct sock *sk)
 
   while(sk->wfront != NULL &&
         before(sk->wfront->h.seq, sk->window_seq) &&
-        sk->packets_out < sk->cong_window) {
+	(sk->retransmits == 0 || before(sk->wfront->h.seq, sk->rcv_ack_seq +1))
+        && sk->packets_out < sk->cong_window) {
 		skb = sk->wfront;
 		IS_SKB(skb);
 		sk->wfront =(struct sk_buff *)skb->next;
@@ -2270,8 +2306,25 @@ tcp_ack(struct sock *sk, struct tcphdr *th, unsigned long saddr, int len)
 	if (before(sk->send_head->h.seq, ack+1)) {
 		struct sk_buff *oskb;
 
-		sk->retransmits = 0;
+		if (sk->retransmits) {
+
+		  /* if we're retransmitting, don't start any new
+		   * packets until after everything in retransmit queue
+		   * is acked.  That's as close as I can come at the
+		   * moment to slow start the way this code is organized
+		   */
+		  if (sk->send_head->link3)
+		    sk->retransmits = 1;
+		  else
+		    sk->retransmits = 0;
+		}
 
+		/*
+		 * need to restart backoff whenever we get a response,
+		 * or things get impossible if we lose a window-full of
+		 * data with very small MSS
+		 */
+		sk->backoff = 0;
 		/* We have one less packet out there. */
 		if (sk->packets_out > 0) sk->packets_out --;
 		DPRINTF((DBG_TCP, "skb=%X skb->h.seq = %d acked ack=%d\n",
@@ -2282,13 +2335,22 @@ tcp_ack(struct sock *sk, struct tcphdr *th, unsigned long saddr, int len)
 
 		oskb = sk->send_head;
 
-		/* Estimate the RTT. Ignore the ones right after a retransmit. */
-		if (sk->retransmits == 0 && !(flag&2)) {
+		/* 
+		 * In theory we're supposed to ignore rtt's when there's
+		 * retransmission in process.  Unfortunately this means
+		 * that if there's a sharp increase in RTT, we may 
+		 * never get out of retransmission.  For the moment
+		 * ignore the test.
+		 */
+
+		if (/* sk->retransmits == 0 && */ !(flag&2)) {
 		  long abserr, rtt = jiffies - oskb->when;
 
-		  if (sk->state == TCP_SYN_SENT || sk->state == TCP_SYN_RECV)
+		  if (sk->state == TCP_SYN_SENT || sk->state == TCP_SYN_RECV) {
 		    /* first ack, so nothing else to average with */
 		    sk->rtt = rtt;
+		    sk->mdev = rtt; /* overcautious initial estimate */
+		  }
 		  else {
 		    abserr = (rtt > sk->rtt) ? rtt - sk->rtt : sk->rtt - rtt;
 		    sk->rtt = (7 * sk->rtt + rtt) >> 3;
@@ -2328,7 +2390,9 @@ tcp_ack(struct sock *sk, struct tcphdr *th, unsigned long saddr, int len)
    */
   if (sk->wfront != NULL) {
 	if (after (sk->window_seq, sk->wfront->h.seq) &&
-		sk->packets_out < sk->cong_window) {
+	        (sk->retransmits == 0 || 
+		 before(sk->wfront->h.seq, sk->rcv_ack_seq +1))
+		&& sk->packets_out < sk->cong_window) {
 		flag |= 1;
 		tcp_write_xmit(sk);
 	}
@@ -2390,7 +2454,7 @@ tcp_ack(struct sock *sk, struct tcphdr *th, unsigned long saddr, int len)
       (sk->send_head->when + backoff(sk->backoff) * (2 * sk->mdev + sk->rtt)
        < jiffies)) {
 	sk->exp_growth = 0;
-	ip_retransmit(sk, 0);
+	ip_retransmit(sk, 1);
   }
 
   DPRINTF((DBG_TCP, "leaving tcp_ack\n"));
@@ -2855,13 +2919,20 @@ tcp_connect(struct sock *sk, struct sockaddr_in *usin, int addr_len)
   t1->urg_ptr = 0;
   t1->doff = 6;
 
+/* use 512 or whatever user asked for */
+  if (sk->mss)
+    sk->mtu = sk->mss;
+  else
+    sk->mtu = 576 - HEADER_SIZE;
+/* but not bigger than device MTU */
+  sk->mtu = min(sk->mtu, dev->mtu - HEADER_SIZE);
+
   /* Put in the TCP options to say MTU. */
   ptr = (unsigned char *)(t1+1);
   ptr[0] = 2;
   ptr[1] = 4;
-  ptr[2] = (dev->mtu- HEADER_SIZE) >> 8;
-  ptr[3] = (dev->mtu- HEADER_SIZE) & 0xff;
-  sk->mtu = dev->mtu - HEADER_SIZE;
+  ptr[2] = (sk->mtu) >> 8;
+  ptr[3] = (sk->mtu) & 0xff;
   tcp_send_check(t1, sk->saddr, sk->daddr,
 		  sizeof(struct tcphdr) + 4, sk);