xen/ipxe-ipv4-fragment.patch

Index: xen-4.1.2-testing/tools/firmware/etherboot/patches/ipxe-git-13186b64b6c3
===================================================================
--- /dev/null	2010-05-08 03:31:08.000000000 -0600
+++ xen-4.1.2-testing/tools/firmware/etherboot/patches/ipxe-git-13186b64b6c3	2011-12-19 15:05:32.000000000 -0700
@@ -0,0 +1,354 @@
+commit 13186b64b6c3d5cbe9ed13bda1532e79b1afe81d
+Author: Michael Brown <mcb30@ipxe.org>
+Date:   Sat Jul 16 01:15:53 2011 +0100
+
+    [ipv4] Fix fragment reassembly
+    
+    Signed-off-by: Michael Brown <mcb30@ipxe.org>
+    Signed-off-by: Michal Kubecek <mkubecek@suse.cz>
+
+diff -up a/src/include/gpxe/ip.h.orig-frag b/src/include/gpxe/ip.h
+--- a/src/include/gpxe/ip.h.orig-frag	2010-02-02 17:12:44.000000000 +0100
++++ b/src/include/gpxe/ip.h	2011-11-18 15:49:17.202660163 +0100
+@@ -32,9 +32,6 @@ struct net_protocol;
+ #define IP_TOS		0
+ #define IP_TTL		64
+ 
+-#define IP_FRAG_IOB_SIZE	1500
+-#define IP_FRAG_TIMEOUT		50
+-
+ /** An IPv4 packet header */
+ struct iphdr {
+ 	uint8_t  verhdrlen;
+@@ -74,20 +71,16 @@ struct ipv4_miniroute {
+ 	struct in_addr gateway;
+ };
+ 
+-/* Fragment reassembly buffer */
+-struct frag_buffer {
+-	/* Identification number */
+-	uint16_t ident;
+-	/* Source network address */
+-	struct in_addr src;
+-	/* Destination network address */
+-	struct in_addr dest;
+-	/* Reassembled I/O buffer */
+-	struct io_buffer *frag_iob;
+-	/* Reassembly timer */
+-	struct retry_timer frag_timer;
++/* IPv4 fragment reassembly buffer */
++struct ipv4_fragment {
+ 	/* List of fragment reassembly buffers */
+ 	struct list_head list;
++	/** Reassembled packet */
++	struct io_buffer *iobuf;
++	/** Current offset */
++	size_t offset;
++	/** Reassembly timer */
++	struct retry_timer timer;
+ };
+ 
+ extern struct list_head ipv4_miniroutes;
+diff -up a/src/include/gpxe/retry.h.orig-frag b/src/include/gpxe/retry.h
+--- a/src/include/gpxe/retry.h.orig-frag	2010-02-02 17:12:44.000000000 +0100
++++ b/src/include/gpxe/retry.h	2011-11-18 15:59:25.258837891 +0100
+@@ -51,6 +51,19 @@ struct retry_timer {
+ 	void ( * expired ) ( struct retry_timer *timer, int over );
+ };
+ 
++/**
++ * Initialise a timer
++ *
++ * @v timer		Retry timer
++ * @v expired		Timer expired callback
++ */
++static inline __attribute__ (( always_inline )) void
++timer_init ( struct retry_timer *timer,
++	     void ( * expired ) ( struct retry_timer *timer, int over ) )
++{
++	timer->expired = expired;
++}
++
+ extern void start_timer ( struct retry_timer *timer );
+ extern void start_timer_fixed ( struct retry_timer *timer,
+ 				unsigned long timeout );
+diff -up a/src/net/ipv4.c.orig-frag b/src/net/ipv4.c
+--- a/src/net/ipv4.c.orig-frag	2010-02-02 17:12:44.000000000 +0100
++++ b/src/net/ipv4.c	2011-11-18 15:49:17.203660142 +0100
+@@ -14,6 +14,7 @@
+ #include <gpxe/tcpip.h>
+ #include <gpxe/dhcp.h>
+ #include <gpxe/settings.h>
++#include <gpxe/timer.h>
+ 
+ /** @file
+  *
+@@ -32,7 +33,10 @@ struct net_protocol ipv4_protocol;
+ struct list_head ipv4_miniroutes = LIST_HEAD_INIT ( ipv4_miniroutes );
+ 
+ /** List of fragment reassembly buffers */
+-static LIST_HEAD ( frag_buffers );
++static LIST_HEAD ( ipv4_fragments );
++
++/** Fragment reassembly timeout */
++#define IP_FRAG_TIMEOUT ( TICKS_PER_SEC / 2 )
+ 
+ /**
+  * Add IPv4 minirouting table entry
+@@ -134,104 +138,126 @@ static struct ipv4_miniroute * ipv4_rout
+ }
+ 
+ /**
+- * Fragment reassembly counter timeout
++ * Expire fragment reassembly buffer
+  *
+- * @v timer	Retry timer
+- * @v over	If asserted, the timer is greater than @c MAX_TIMEOUT 
++ * @v timer		Retry timer
++ * @v fail		Failure indicator
+  */
+-static void ipv4_frag_expired ( struct retry_timer *timer __unused,
+-				int over ) {
+-	if ( over ) {
+-		DBG ( "Fragment reassembly timeout" );
+-		/* Free the fragment buffer */
+-	}
++static void ipv4_fragment_expired ( struct retry_timer *timer,
++				    int fail __unused ) {
++	struct ipv4_fragment *frag =
++		container_of ( timer, struct ipv4_fragment, timer );
++	struct iphdr *iphdr = frag->iobuf->data;
++
++	DBG ( "IPv4 fragment %04x expired\n", ntohs ( iphdr->ident ) );
++	free_iob ( frag->iobuf );
++	list_del ( &frag->list );
++	free ( frag );
+ }
+ 
+ /**
+- * Free fragment buffer
++ * Find matching fragment reassembly buffer
+  *
+- * @v fragbug	Fragment buffer
++ * @v iphdr		IPv4 header
++ * @ret frag		Fragment reassembly buffer, or NULL
+  */
+-static void free_fragbuf ( struct frag_buffer *fragbuf ) {
+-	free ( fragbuf );
++static struct ipv4_fragment * ipv4_fragment ( struct iphdr *iphdr ) {
++	struct ipv4_fragment *frag;
++	struct iphdr *frag_iphdr;
++
++	list_for_each_entry ( frag, &ipv4_fragments, list ) {
++		frag_iphdr = frag->iobuf->data;
++
++		if ( ( iphdr->src.s_addr == frag_iphdr->src.s_addr ) &&
++		     ( iphdr->ident == frag_iphdr->ident ) ) {
++			return frag;
++		}
++	}
++
++	return NULL;
+ }
+ 
+ /**
+  * Fragment reassembler
+  *
+- * @v iobuf		I/O buffer, fragment of the datagram
+- * @ret frag_iob	Reassembled packet, or NULL
++ * @v iobuf		I/O buffer
++ * @ret iobuf		Reassembled packet, or NULL
+  */
+-static struct io_buffer * ipv4_reassemble ( struct io_buffer * iobuf ) {
++static struct io_buffer * ipv4_reassemble ( struct io_buffer *iobuf ) {
+ 	struct iphdr *iphdr = iobuf->data;
+-	struct frag_buffer *fragbuf;
+-	
+-	/**
+-	 * Check if the fragment belongs to any fragment series
+-	 */
+-	list_for_each_entry ( fragbuf, &frag_buffers, list ) {
+-		if ( fragbuf->ident == iphdr->ident &&
+-		     fragbuf->src.s_addr == iphdr->src.s_addr ) {
+-			/**
+-			 * Check if the packet is the expected fragment
+-			 * 
+-			 * The offset of the new packet must be equal to the
+-			 * length of the data accumulated so far (the length of
+-			 * the reassembled I/O buffer
+-			 */
+-			if ( iob_len ( fragbuf->frag_iob ) == 
+-			      ( iphdr->frags & IP_MASK_OFFSET ) ) {
+-				/**
+-				 * Append the contents of the fragment to the
+-				 * reassembled I/O buffer
+-				 */
+-				iob_pull ( iobuf, sizeof ( *iphdr ) );
+-				memcpy ( iob_put ( fragbuf->frag_iob,
+-							iob_len ( iobuf ) ),
+-					 iobuf->data, iob_len ( iobuf ) );
+-				free_iob ( iobuf );
+-
+-				/** Check if the fragment series is over */
+-				if ( ! ( iphdr->frags & IP_MASK_MOREFRAGS ) ) {
+-					iobuf = fragbuf->frag_iob;
+-					free_fragbuf ( fragbuf );
+-					return iobuf;
+-				}
+-
+-			} else {
+-				/* Discard the fragment series */
+-				free_fragbuf ( fragbuf );
+-				free_iob ( iobuf );
+-			}
+-			return NULL;
++	size_t offset = ( ( ntohs ( iphdr->frags ) & IP_MASK_OFFSET ) << 3 );
++	unsigned int more_frags = ( iphdr->frags & htons ( IP_MASK_MOREFRAGS ));
++	size_t hdrlen = ( ( iphdr->verhdrlen & IP_MASK_HLEN ) * 4 );
++	struct ipv4_fragment *frag;
++	size_t expected_offset;
++	struct io_buffer *new_iobuf;
++
++	/* Find matching fragment reassembly buffer, if any */
++	frag = ipv4_fragment ( iphdr );
++
++	/* Drop out-of-order fragments */
++	expected_offset = ( frag ? frag->offset : 0 );
++	if ( offset != expected_offset ) {
++		DBG ( "IPv4 dropping out-of-sequence fragment %04x (%zd+%zd, "
++		      "expected %zd)\n", ntohs ( iphdr->ident ), offset,
++		      ( iob_len ( iobuf ) - hdrlen ), expected_offset );
++		goto drop;
++	}
++
++	/* Create or extend fragment reassembly buffer as applicable */
++	if ( frag == NULL ) {
++
++		/* Create new fragment reassembly buffer */
++		frag = zalloc ( sizeof ( *frag ) );
++		if ( ! frag )
++			goto drop;
++		list_add ( &frag->list, &ipv4_fragments );
++		frag->iobuf = iobuf;
++		frag->offset = ( iob_len ( iobuf ) - hdrlen );
++		timer_init ( &frag->timer, ipv4_fragment_expired );
++
++	} else {
++
++		/* Extend reassembly buffer */
++		iob_pull ( iobuf, hdrlen );
++		new_iobuf = alloc_iob ( iob_len ( frag->iobuf ) +
++					iob_len ( iobuf ) );
++		if ( ! new_iobuf ) {
++			DBG ( "IPv4 could not extend reassembly buffer to "
++			      "%zd bytes\n",
++			      ( iob_len ( frag->iobuf ) + iob_len ( iobuf ) ) );
++			goto drop;
+ 		}
+-	}
+-	
+-	/** Check if the fragment is the first in the fragment series */
+-	if ( iphdr->frags & IP_MASK_MOREFRAGS &&
+-			( ( iphdr->frags & IP_MASK_OFFSET ) == 0 ) ) {
+-	
+-		/** Create a new fragment buffer */
+-		fragbuf = ( struct frag_buffer* ) malloc ( sizeof( *fragbuf ) );
+-		fragbuf->ident = iphdr->ident;
+-		fragbuf->src = iphdr->src;
+-
+-		/* Set up the reassembly I/O buffer */
+-		fragbuf->frag_iob = alloc_iob ( IP_FRAG_IOB_SIZE );
+-		iob_pull ( iobuf, sizeof ( *iphdr ) );
+-		memcpy ( iob_put ( fragbuf->frag_iob, iob_len ( iobuf ) ),
++		memcpy ( iob_put ( new_iobuf, iob_len ( frag->iobuf ) ),
++			 frag->iobuf->data, iob_len ( frag->iobuf ) );
++		memcpy ( iob_put ( new_iobuf, iob_len ( iobuf ) ),
+ 			 iobuf->data, iob_len ( iobuf ) );
++		free_iob ( frag->iobuf );
++		frag->iobuf = new_iobuf;
++		frag->offset += iob_len ( iobuf );
+ 		free_iob ( iobuf );
++		iphdr = frag->iobuf->data;
++		iphdr->len = ntohs ( iob_len ( frag->iobuf ) );
+ 
+-		/* Set the reassembly timer */
+-		fragbuf->frag_timer.timeout = IP_FRAG_TIMEOUT;
+-		fragbuf->frag_timer.expired = ipv4_frag_expired;
+-		start_timer ( &fragbuf->frag_timer );
++		/* Stop fragment reassembly timer */
++		stop_timer ( &frag->timer );
+ 
+-		/* Add the fragment buffer to the list of fragment buffers */
+-		list_add ( &fragbuf->list, &frag_buffers );
++		/* If this is the final fragment, return it */
++		if ( ! more_frags ) {
++			iobuf = frag->iobuf;
++			list_del ( &frag->list );
++			free ( frag );
++			return iobuf;
++		}
+ 	}
+-	
++
++	/* (Re)start fragment reassembly timer */
++	start_timer_fixed ( &frag->timer, IP_FRAG_TIMEOUT );
++
++	return NULL;
++
++ drop:
++	free_iob ( iobuf );
+ 	return NULL;
+ }
+ 
+@@ -432,37 +458,38 @@ static int ipv4_rx ( struct io_buffer *i
+ 		goto err;
+ 	}
+ 
++	/* Truncate packet to correct length */
++	iob_unput ( iobuf, ( iob_len ( iobuf ) - len ) );
++
+ 	/* Print IPv4 header for debugging */
+ 	DBG ( "IPv4 RX %s<-", inet_ntoa ( iphdr->dest ) );
+ 	DBG ( "%s len %d proto %d id %04x csum %04x\n",
+ 	      inet_ntoa ( iphdr->src ), ntohs ( iphdr->len ), iphdr->protocol,
+ 	      ntohs ( iphdr->ident ), ntohs ( iphdr->chksum ) );
+ 
+-	/* Truncate packet to correct length, calculate pseudo-header
+-	 * checksum and then strip off the IPv4 header.
+-	 */
+-	iob_unput ( iobuf, ( iob_len ( iobuf ) - len ) );
+-	pshdr_csum = ipv4_pshdr_chksum ( iobuf, TCPIP_EMPTY_CSUM );
+-	iob_pull ( iobuf, hdrlen );
+-
+-	/* Fragment reassembly */
+-	if ( ( iphdr->frags & htons ( IP_MASK_MOREFRAGS ) ) || 
+-	     ( ( iphdr->frags & htons ( IP_MASK_OFFSET ) ) != 0 ) ) {
+-		/* Pass the fragment to ipv4_reassemble() which either
+-		 * returns a fully reassembled I/O buffer or NULL.
++	/* Perform fragment reassembly if applicable */
++	if ( iphdr->frags & htons ( IP_MASK_OFFSET | IP_MASK_MOREFRAGS ) ) {
++		/* Pass the fragment to ipv4_reassemble() which returns
++		 * either a fully reassembled I/O buffer or NULL.
+ 		 */
+ 		iobuf = ipv4_reassemble ( iobuf );
+ 		if ( ! iobuf )
+ 			return 0;
++		iphdr = iobuf->data;
++		hdrlen = ( ( iphdr->verhdrlen & IP_MASK_HLEN ) * 4 );
+ 	}
+ 
+-	/* Construct socket addresses and hand off to transport layer */
++	/* Construct socket addresses, calculate pseudo-header
++	 * checksum, and hand off to transport layer
++	 */
+ 	memset ( &src, 0, sizeof ( src ) );
+ 	src.sin.sin_family = AF_INET;
+ 	src.sin.sin_addr = iphdr->src;
+ 	memset ( &dest, 0, sizeof ( dest ) );
+ 	dest.sin.sin_family = AF_INET;
+ 	dest.sin.sin_addr = iphdr->dest;
++	pshdr_csum = ipv4_pshdr_chksum ( iobuf, TCPIP_EMPTY_CSUM );
++	iob_pull ( iobuf, hdrlen );
+ 	if ( ( rc = tcpip_rx ( iobuf, iphdr->protocol, &src.st,
+ 			       &dest.st, pshdr_csum ) ) != 0 ) {
+ 		DBG ( "IPv4 received packet rejected by stack: %s\n",
Index: xen-4.1.2-testing/tools/firmware/etherboot/patches/series
===================================================================
--- xen-4.1.2-testing.orig/tools/firmware/etherboot/patches/series
+++ xen-4.1.2-testing/tools/firmware/etherboot/patches/series
@@ -2,3 +2,4 @@ boot_prompt_option.patch
 gpxe-git-0edf2405b457
 gpxe-git-a803ef3dfeac
 ipxe-git-f7c5918b179b
+ipxe-git-13186b64b6c3