[linux.git] / net / ipv4 / inet_lro.c

/*
 *  linux/net/ipv4/inet_lro.c
 *
 *  Large Receive Offload (ipv4 / tcp)
 *
 *  (C) Copyright IBM Corp. 2007
 *
 *  Authors:
 *       Jan-Bernd Themann <[email protected]>
 *       Christoph Raisch <[email protected]>
 *
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2, or (at your option)
 * any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */


#include <linux/module.h>
#include <linux/if_vlan.h>
#include <linux/inet_lro.h>
#include <net/checksum.h>

MODULE_LICENSE("GPL");
MODULE_AUTHOR("Jan-Bernd Themann <[email protected]>");
MODULE_DESCRIPTION("Large Receive Offload (ipv4 / tcp)");

#define TCP_HDR_LEN(tcph) (tcph->doff << 2)
#define IP_HDR_LEN(iph) (iph->ihl << 2)
#define TCP_PAYLOAD_LENGTH(iph, tcph) \
	(ntohs(iph->tot_len) - IP_HDR_LEN(iph) - TCP_HDR_LEN(tcph))

#define IPH_LEN_WO_OPTIONS 5
#define TCPH_LEN_WO_OPTIONS 5
#define TCPH_LEN_W_TIMESTAMP 8

#define LRO_MAX_PG_HLEN 64

#define LRO_INC_STATS(lro_mgr, attr) { lro_mgr->stats.attr++; }

/*
 * Basic tcp checks whether packet is suitable for LRO
 */

static int lro_tcp_ip_check(const struct iphdr *iph, const struct tcphdr *tcph,
			    int len, const struct net_lro_desc *lro_desc)
{
        /* check ip header: don't aggregate padded frames */
	if (ntohs(iph->tot_len) != len)
		return -1;

	if (TCP_PAYLOAD_LENGTH(iph, tcph) == 0)
		return -1;

	if (iph->ihl != IPH_LEN_WO_OPTIONS)
		return -1;

	if (tcph->cwr || tcph->ece || tcph->urg || !tcph->ack ||
	    tcph->rst || tcph->syn || tcph->fin)
		return -1;

	if (INET_ECN_is_ce(ipv4_get_dsfield(iph)))
		return -1;

	if (tcph->doff != TCPH_LEN_WO_OPTIONS &&
	    tcph->doff != TCPH_LEN_W_TIMESTAMP)
		return -1;

	/* check tcp options (only timestamp allowed) */
	if (tcph->doff == TCPH_LEN_W_TIMESTAMP) {
		__be32 *topt = (__be32 *)(tcph + 1);

		if (*topt != htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16)
				   | (TCPOPT_TIMESTAMP << 8)
				   | TCPOLEN_TIMESTAMP))
			return -1;

		/* timestamp should be in right order */
		topt++;
		if (lro_desc && after(ntohl(lro_desc->tcp_rcv_tsval),
				      ntohl(*topt)))
			return -1;

		/* timestamp reply should not be zero */
		topt++;
		if (*topt == 0)
			return -1;
	}

	return 0;
}

static void lro_update_tcp_ip_header(struct net_lro_desc *lro_desc)
{
	struct iphdr *iph = lro_desc->iph;
	struct tcphdr *tcph = lro_desc->tcph;
	__be32 *p;
	__wsum tcp_hdr_csum;

	tcph->ack_seq = lro_desc->tcp_ack;
	tcph->window = lro_desc->tcp_window;

	if (lro_desc->tcp_saw_tstamp) {
		p = (__be32 *)(tcph + 1);
		*(p+2) = lro_desc->tcp_rcv_tsecr;
	}

	csum_replace2(&iph->check, iph->tot_len, htons(lro_desc->ip_tot_len));
	iph->tot_len = htons(lro_desc->ip_tot_len);

	tcph->check = 0;
	tcp_hdr_csum = csum_partial(tcph, TCP_HDR_LEN(tcph), 0);
	lro_desc->data_csum = csum_add(lro_desc->data_csum, tcp_hdr_csum);
	tcph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
					lro_desc->ip_tot_len -
					IP_HDR_LEN(iph), IPPROTO_TCP,
					lro_desc->data_csum);
}

static __wsum lro_tcp_data_csum(struct iphdr *iph, struct tcphdr *tcph, int len)
{
	__wsum tcp_csum;
	__wsum tcp_hdr_csum;
	__wsum tcp_ps_hdr_csum;

	tcp_csum = ~csum_unfold(tcph->check);
	tcp_hdr_csum = csum_partial(tcph, TCP_HDR_LEN(tcph), tcp_csum);

	tcp_ps_hdr_csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
					     len + TCP_HDR_LEN(tcph),
					     IPPROTO_TCP, 0);

	return csum_sub(csum_sub(tcp_csum, tcp_hdr_csum),
			tcp_ps_hdr_csum);
}

static void lro_init_desc(struct net_lro_desc *lro_desc, struct sk_buff *skb,
			  struct iphdr *iph, struct tcphdr *tcph)
{
	int nr_frags;
	__be32 *ptr;
	u32 tcp_data_len = TCP_PAYLOAD_LENGTH(iph, tcph);

	nr_frags = skb_shinfo(skb)->nr_frags;
	lro_desc->parent = skb;
	lro_desc->next_frag = &(skb_shinfo(skb)->frags[nr_frags]);
	lro_desc->iph = iph;
	lro_desc->tcph = tcph;
	lro_desc->tcp_next_seq = ntohl(tcph->seq) + tcp_data_len;
	lro_desc->tcp_ack = tcph->ack_seq;
	lro_desc->tcp_window = tcph->window;

	lro_desc->pkt_aggr_cnt = 1;
	lro_desc->ip_tot_len = ntohs(iph->tot_len);

	if (tcph->doff == 8) {
		ptr = (__be32 *)(tcph+1);
		lro_desc->tcp_saw_tstamp = 1;
		lro_desc->tcp_rcv_tsval = *(ptr+1);
		lro_desc->tcp_rcv_tsecr = *(ptr+2);
	}

	lro_desc->mss = tcp_data_len;
	lro_desc->active = 1;

	lro_desc->data_csum = lro_tcp_data_csum(iph, tcph,
						tcp_data_len);
}

static inline void lro_clear_desc(struct net_lro_desc *lro_desc)
{
	memset(lro_desc, 0, sizeof(struct net_lro_desc));
}

static void lro_add_common(struct net_lro_desc *lro_desc, struct iphdr *iph,
			   struct tcphdr *tcph, int tcp_data_len)
{
	struct sk_buff *parent = lro_desc->parent;
	__be32 *topt;

	lro_desc->pkt_aggr_cnt++;
	lro_desc->ip_tot_len += tcp_data_len;
	lro_desc->tcp_next_seq += tcp_data_len;
	lro_desc->tcp_window = tcph->window;
	lro_desc->tcp_ack = tcph->ack_seq;

	/* don't update tcp_rcv_tsval, would not work with PAWS */
	if (lro_desc->tcp_saw_tstamp) {
		topt = (__be32 *) (tcph + 1);
		lro_desc->tcp_rcv_tsecr = *(topt + 2);
	}

	lro_desc->data_csum = csum_block_add(lro_desc->data_csum,
					     lro_tcp_data_csum(iph, tcph,
							       tcp_data_len),
					     parent->len);

	parent->len += tcp_data_len;
	parent->data_len += tcp_data_len;
	if (tcp_data_len > lro_desc->mss)
		lro_desc->mss = tcp_data_len;
}

static void lro_add_packet(struct net_lro_desc *lro_desc, struct sk_buff *skb,
			   struct iphdr *iph, struct tcphdr *tcph)
{
	struct sk_buff *parent = lro_desc->parent;
	int tcp_data_len = TCP_PAYLOAD_LENGTH(iph, tcph);

	lro_add_common(lro_desc, iph, tcph, tcp_data_len);

	skb_pull(skb, (skb->len - tcp_data_len));
	parent->truesize += skb->truesize;

	if (lro_desc->last_skb)
		lro_desc->last_skb->next = skb;
	else
		skb_shinfo(parent)->frag_list = skb;

	lro_desc->last_skb = skb;
}

static void lro_add_frags(struct net_lro_desc *lro_desc,
			  int len, int hlen, int truesize,
			  struct skb_frag_struct *skb_frags,
			  struct iphdr *iph, struct tcphdr *tcph)
{
	struct sk_buff *skb = lro_desc->parent;
	int tcp_data_len = TCP_PAYLOAD_LENGTH(iph, tcph);

	lro_add_common(lro_desc, iph, tcph, tcp_data_len);

	skb->truesize += truesize;

	skb_frags[0].page_offset += hlen;
	skb_frag_size_sub(&skb_frags[0], hlen);

	while (tcp_data_len > 0) {
		*(lro_desc->next_frag) = *skb_frags;
		tcp_data_len -= skb_frag_size(skb_frags);
		lro_desc->next_frag++;
		skb_frags++;
		skb_shinfo(skb)->nr_frags++;
	}
}

static int lro_check_tcp_conn(struct net_lro_desc *lro_desc,
			      struct iphdr *iph,
			      struct tcphdr *tcph)
{
	if ((lro_desc->iph->saddr != iph->saddr) ||
	    (lro_desc->iph->daddr != iph->daddr) ||
	    (lro_desc->tcph->source != tcph->source) ||
	    (lro_desc->tcph->dest != tcph->dest))
		return -1;
	return 0;
}

static struct net_lro_desc *lro_get_desc(struct net_lro_mgr *lro_mgr,
					 struct net_lro_desc *lro_arr,
					 struct iphdr *iph,
					 struct tcphdr *tcph)
{
	struct net_lro_desc *lro_desc = NULL;
	struct net_lro_desc *tmp;
	int max_desc = lro_mgr->max_desc;
	int i;

	for (i = 0; i < max_desc; i++) {
		tmp = &lro_arr[i];
		if (tmp->active)
			if (!lro_check_tcp_conn(tmp, iph, tcph)) {
				lro_desc = tmp;
				goto out;
			}
	}

	for (i = 0; i < max_desc; i++) {
		if (!lro_arr[i].active) {
			lro_desc = &lro_arr[i];
			goto out;
		}
	}

	LRO_INC_STATS(lro_mgr, no_desc);
out:
	return lro_desc;
}

static void lro_flush(struct net_lro_mgr *lro_mgr,
		      struct net_lro_desc *lro_desc)
{
	if (lro_desc->pkt_aggr_cnt > 1)
		lro_update_tcp_ip_header(lro_desc);

	skb_shinfo(lro_desc->parent)->gso_size = lro_desc->mss;

	if (lro_mgr->features & LRO_F_NAPI)
		netif_receive_skb(lro_desc->parent);
	else
		netif_rx(lro_desc->parent);

	LRO_INC_STATS(lro_mgr, flushed);
	lro_clear_desc(lro_desc);
}

static int __lro_proc_skb(struct net_lro_mgr *lro_mgr, struct sk_buff *skb,
			  void *priv)
{
	struct net_lro_desc *lro_desc;
	struct iphdr *iph;
	struct tcphdr *tcph;
	u64 flags;
	int vlan_hdr_len = 0;

	if (!lro_mgr->get_skb_header ||
	    lro_mgr->get_skb_header(skb, (void *)&iph, (void *)&tcph,
				    &flags, priv))
		goto out;

	if (!(flags & LRO_IPV4) || !(flags & LRO_TCP))
		goto out;

	lro_desc = lro_get_desc(lro_mgr, lro_mgr->lro_arr, iph, tcph);
	if (!lro_desc)
		goto out;

	if ((skb->protocol == htons(ETH_P_8021Q)) &&
	    !(lro_mgr->features & LRO_F_EXTRACT_VLAN_ID))
		vlan_hdr_len = VLAN_HLEN;

	if (!lro_desc->active) { /* start new lro session */
		if (lro_tcp_ip_check(iph, tcph, skb->len - vlan_hdr_len, NULL))
			goto out;

		skb->ip_summed = lro_mgr->ip_summed_aggr;
		lro_init_desc(lro_desc, skb, iph, tcph);
		LRO_INC_STATS(lro_mgr, aggregated);
		return 0;
	}

	if (lro_desc->tcp_next_seq != ntohl(tcph->seq))
		goto out2;

	if (lro_tcp_ip_check(iph, tcph, skb->len, lro_desc))
		goto out2;

	lro_add_packet(lro_desc, skb, iph, tcph);
	LRO_INC_STATS(lro_mgr, aggregated);

	if ((lro_desc->pkt_aggr_cnt >= lro_mgr->max_aggr) ||
	    lro_desc->parent->len > (0xFFFF - lro_mgr->dev->mtu))
		lro_flush(lro_mgr, lro_desc);

	return 0;

out2: /* send aggregated SKBs to stack */
	lro_flush(lro_mgr, lro_desc);

out:
	return 1;
}


static struct sk_buff *lro_gen_skb(struct net_lro_mgr *lro_mgr,
				   struct skb_frag_struct *frags,
				   int len, int true_size,
				   void *mac_hdr,
				   int hlen, __wsum sum,
				   u32 ip_summed)
{
	struct sk_buff *skb;
	struct skb_frag_struct *skb_frags;
	int data_len = len;
	int hdr_len = min(len, hlen);

	skb = netdev_alloc_skb(lro_mgr->dev, hlen + lro_mgr->frag_align_pad);
	if (!skb)
		return NULL;

	skb_reserve(skb, lro_mgr->frag_align_pad);
	skb->len = len;
	skb->data_len = len - hdr_len;
	skb->truesize += true_size;
	skb->tail += hdr_len;

	memcpy(skb->data, mac_hdr, hdr_len);

	skb_frags = skb_shinfo(skb)->frags;
	while (data_len > 0) {
		*skb_frags = *frags;
		data_len -= skb_frag_size(frags);
		skb_frags++;
		frags++;
		skb_shinfo(skb)->nr_frags++;
	}

	skb_shinfo(skb)->frags[0].page_offset += hdr_len;
	skb_frag_size_sub(&skb_shinfo(skb)->frags[0], hdr_len);

	skb->ip_summed = ip_summed;
	skb->csum = sum;
	skb->protocol = eth_type_trans(skb, lro_mgr->dev);
	return skb;
}

static struct sk_buff *__lro_proc_segment(struct net_lro_mgr *lro_mgr,
					  struct skb_frag_struct *frags,
					  int len, int true_size,
					  void *priv, __wsum sum)
{
	struct net_lro_desc *lro_desc;
	struct iphdr *iph;
	struct tcphdr *tcph;
	struct sk_buff *skb;
	u64 flags;
	void *mac_hdr;
	int mac_hdr_len;
	int hdr_len = LRO_MAX_PG_HLEN;
	int vlan_hdr_len = 0;

	if (!lro_mgr->get_frag_header ||
	    lro_mgr->get_frag_header(frags, (void *)&mac_hdr, (void *)&iph,
				     (void *)&tcph, &flags, priv)) {
		mac_hdr = skb_frag_address(frags);
		goto out1;
	}

	if (!(flags & LRO_IPV4) || !(flags & LRO_TCP))
		goto out1;

	hdr_len = (int)((void *)(tcph) + TCP_HDR_LEN(tcph) - mac_hdr);
	mac_hdr_len = (int)((void *)(iph) - mac_hdr);

	lro_desc = lro_get_desc(lro_mgr, lro_mgr->lro_arr, iph, tcph);
	if (!lro_desc)
		goto out1;

	if (!lro_desc->active) { /* start new lro session */
		if (lro_tcp_ip_check(iph, tcph, len - mac_hdr_len, NULL))
			goto out1;

		skb = lro_gen_skb(lro_mgr, frags, len, true_size, mac_hdr,
				  hdr_len, 0, lro_mgr->ip_summed_aggr);
		if (!skb)
			goto out;

		if ((skb->protocol == htons(ETH_P_8021Q)) &&
		    !(lro_mgr->features & LRO_F_EXTRACT_VLAN_ID))
			vlan_hdr_len = VLAN_HLEN;

		iph = (void *)(skb->data + vlan_hdr_len);
		tcph = (void *)((u8 *)skb->data + vlan_hdr_len
				+ IP_HDR_LEN(iph));

		lro_init_desc(lro_desc, skb, iph, tcph);
		LRO_INC_STATS(lro_mgr, aggregated);
		return NULL;
	}

	if (lro_desc->tcp_next_seq != ntohl(tcph->seq))
		goto out2;

	if (lro_tcp_ip_check(iph, tcph, len - mac_hdr_len, lro_desc))
		goto out2;

	lro_add_frags(lro_desc, len, hdr_len, true_size, frags, iph, tcph);
	LRO_INC_STATS(lro_mgr, aggregated);

	if ((skb_shinfo(lro_desc->parent)->nr_frags >= lro_mgr->max_aggr) ||
	    lro_desc->parent->len > (0xFFFF - lro_mgr->dev->mtu))
		lro_flush(lro_mgr, lro_desc);

	return NULL;

out2: /* send aggregated packets to the stack */
	lro_flush(lro_mgr, lro_desc);

out1:  /* Original packet has to be posted to the stack */
	skb = lro_gen_skb(lro_mgr, frags, len, true_size, mac_hdr,
			  hdr_len, sum, lro_mgr->ip_summed);
out:
	return skb;
}

void lro_receive_skb(struct net_lro_mgr *lro_mgr,
		     struct sk_buff *skb,
		     void *priv)
{
	if (__lro_proc_skb(lro_mgr, skb, priv)) {
		if (lro_mgr->features & LRO_F_NAPI)
			netif_receive_skb(skb);
		else
			netif_rx(skb);
	}
}
EXPORT_SYMBOL(lro_receive_skb);

void lro_receive_frags(struct net_lro_mgr *lro_mgr,
		       struct skb_frag_struct *frags,
		       int len, int true_size, void *priv, __wsum sum)
{
	struct sk_buff *skb;

	skb = __lro_proc_segment(lro_mgr, frags, len, true_size, priv, sum);
	if (!skb)
		return;

	if (lro_mgr->features & LRO_F_NAPI)
		netif_receive_skb(skb);
	else
		netif_rx(skb);
}
EXPORT_SYMBOL(lro_receive_frags);

void lro_flush_all(struct net_lro_mgr *lro_mgr)
{
	int i;
	struct net_lro_desc *lro_desc = lro_mgr->lro_arr;

	for (i = 0; i < lro_mgr->max_desc; i++) {
		if (lro_desc[i].active)
			lro_flush(lro_mgr, &lro_desc[i]);
	}
}
EXPORT_SYMBOL(lro_flush_all);

void lro_flush_pkt(struct net_lro_mgr *lro_mgr,
		  struct iphdr *iph, struct tcphdr *tcph)
{
	struct net_lro_desc *lro_desc;

	lro_desc = lro_get_desc(lro_mgr, lro_mgr->lro_arr, iph, tcph);
	if (lro_desc->active)
		lro_flush(lro_mgr, lro_desc);
}
EXPORT_SYMBOL(lro_flush_pkt);
Commit	Line	Data
71c87e0c JBT	1	/*
	2	* linux/net/ipv4/inet_lro.c
	3	*
	4	* Large Receive Offload (ipv4 / tcp)
	5	*
	6	* (C) Copyright IBM Corp. 2007
	7	*
	8	* Authors:
	9	* Jan-Bernd Themann <[email protected]>
	10	* Christoph Raisch <[email protected]>
	11	*
	12	*
	13	* This program is free software; you can redistribute it and/or modify
	14	* it under the terms of the GNU General Public License as published by
	15	* the Free Software Foundation; either version 2, or (at your option)
	16	* any later version.
	17	*
	18	* This program is distributed in the hope that it will be useful,
	19	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	20	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
	21	* GNU General Public License for more details.
	22	*
	23	* You should have received a copy of the GNU General Public License
	24	* along with this program; if not, write to the Free Software
	25	* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
	26	*/
	27
	28
	29	#include <linux/module.h>
	30	#include <linux/if_vlan.h>
	31	#include <linux/inet_lro.h>
35353c2b	32	#include <net/checksum.h>
71c87e0c JBT	33
	34	MODULE_LICENSE("GPL");
	35	MODULE_AUTHOR("Jan-Bernd Themann <[email protected]>");
	36	MODULE_DESCRIPTION("Large Receive Offload (ipv4 / tcp)");
	37
	38	#define TCP_HDR_LEN(tcph) (tcph->doff << 2)
	39	#define IP_HDR_LEN(iph) (iph->ihl << 2)
	40	#define TCP_PAYLOAD_LENGTH(iph, tcph) \
	41	(ntohs(iph->tot_len) - IP_HDR_LEN(iph) - TCP_HDR_LEN(tcph))
	42
	43	#define IPH_LEN_WO_OPTIONS 5
	44	#define TCPH_LEN_WO_OPTIONS 5
	45	#define TCPH_LEN_W_TIMESTAMP 8
	46
	47	#define LRO_MAX_PG_HLEN 64
	48
	49	#define LRO_INC_STATS(lro_mgr, attr) { lro_mgr->stats.attr++; }
	50
	51	/*
	52	* Basic tcp checks whether packet is suitable for LRO
	53	*/
	54
b71d1d42 ED	55	static int lro_tcp_ip_check(const struct iphdr iph, const struct tcphdr tcph,
b71d1d42 ED	56	int len, const struct net_lro_desc *lro_desc)
71c87e0c JBT	57	{
	58	/* check ip header: don't aggregate padded frames */
	59	if (ntohs(iph->tot_len) != len)
	60	return -1;
	61
	62	if (TCP_PAYLOAD_LENGTH(iph, tcph) == 0)
	63	return -1;
	64
	65	if (iph->ihl != IPH_LEN_WO_OPTIONS)
	66	return -1;
	67
9d4fb27d JP	68	if (tcph->cwr \|\| tcph->ece \|\| tcph->urg \|\| !tcph->ack \|\|
9d4fb27d JP	69	tcph->rst \|\| tcph->syn \|\| tcph->fin)
71c87e0c JBT	70	return -1;
	71
	72	if (INET_ECN_is_ce(ipv4_get_dsfield(iph)))
	73	return -1;
	74
9d4fb27d JP	75	if (tcph->doff != TCPH_LEN_WO_OPTIONS &&
9d4fb27d JP	76	tcph->doff != TCPH_LEN_W_TIMESTAMP)
71c87e0c JBT	77	return -1;
	78
	79	/* check tcp options (only timestamp allowed) */
	80	if (tcph->doff == TCPH_LEN_W_TIMESTAMP) {
9df7c98a	81	__be32 topt = (__be32 )(tcph + 1);
71c87e0c JBT	82
	83	if (*topt != htonl((TCPOPT_NOP << 24) \| (TCPOPT_NOP << 16)
	84	\| (TCPOPT_TIMESTAMP << 8)
	85	\| TCPOLEN_TIMESTAMP))
	86	return -1;
	87
	88	/* timestamp should be in right order */
	89	topt++;
	90	if (lro_desc && after(ntohl(lro_desc->tcp_rcv_tsval),
	91	ntohl(*topt)))
	92	return -1;
	93
	94	/* timestamp reply should not be zero */
	95	topt++;
	96	if (*topt == 0)
	97	return -1;
	98	}
	99
	100	return 0;
	101	}
	102
	103	static void lro_update_tcp_ip_header(struct net_lro_desc *lro_desc)
	104	{
	105	struct iphdr *iph = lro_desc->iph;
	106	struct tcphdr *tcph = lro_desc->tcph;
9df7c98a	107	__be32 *p;
71c87e0c JBT	108	__wsum tcp_hdr_csum;
	109
	110	tcph->ack_seq = lro_desc->tcp_ack;
	111	tcph->window = lro_desc->tcp_window;
	112
	113	if (lro_desc->tcp_saw_tstamp) {
9df7c98a	114	p = (__be32 *)(tcph + 1);
71c87e0c JBT	115	*(p+2) = lro_desc->tcp_rcv_tsecr;
	116	}
	117
35353c2b	118	csum_replace2(&iph->check, iph->tot_len, htons(lro_desc->ip_tot_len));
71c87e0c JBT	119	iph->tot_len = htons(lro_desc->ip_tot_len);
71c87e0c JBT	120
71c87e0c	121	tcph->check = 0;
07f0757a	122	tcp_hdr_csum = csum_partial(tcph, TCP_HDR_LEN(tcph), 0);
71c87e0c JBT	123	lro_desc->data_csum = csum_add(lro_desc->data_csum, tcp_hdr_csum);
	124	tcph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
	125	lro_desc->ip_tot_len -
	126	IP_HDR_LEN(iph), IPPROTO_TCP,
	127	lro_desc->data_csum);
	128	}
	129
	130	static __wsum lro_tcp_data_csum(struct iphdr iph, struct tcphdr tcph, int len)
	131	{
	132	__wsum tcp_csum;
	133	__wsum tcp_hdr_csum;
	134	__wsum tcp_ps_hdr_csum;
	135
	136	tcp_csum = ~csum_unfold(tcph->check);
07f0757a	137	tcp_hdr_csum = csum_partial(tcph, TCP_HDR_LEN(tcph), tcp_csum);
71c87e0c JBT	138
	139	tcp_ps_hdr_csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
	140	len + TCP_HDR_LEN(tcph),
	141	IPPROTO_TCP, 0);
	142
	143	return csum_sub(csum_sub(tcp_csum, tcp_hdr_csum),
	144	tcp_ps_hdr_csum);
	145	}
	146
	147	static void lro_init_desc(struct net_lro_desc lro_desc, struct sk_buff skb,
9fea0330	148	struct iphdr iph, struct tcphdr tcph)
71c87e0c JBT	149	{
71c87e0c JBT	150	int nr_frags;
9df7c98a	151	__be32 *ptr;
71c87e0c JBT	152	u32 tcp_data_len = TCP_PAYLOAD_LENGTH(iph, tcph);
	153
	154	nr_frags = skb_shinfo(skb)->nr_frags;
	155	lro_desc->parent = skb;
	156	lro_desc->next_frag = &(skb_shinfo(skb)->frags[nr_frags]);
	157	lro_desc->iph = iph;
	158	lro_desc->tcph = tcph;
	159	lro_desc->tcp_next_seq = ntohl(tcph->seq) + tcp_data_len;
f53f4137	160	lro_desc->tcp_ack = tcph->ack_seq;
71c87e0c JBT	161	lro_desc->tcp_window = tcph->window;
	162
	163	lro_desc->pkt_aggr_cnt = 1;
	164	lro_desc->ip_tot_len = ntohs(iph->tot_len);
	165
	166	if (tcph->doff == 8) {
9df7c98a	167	ptr = (__be32 *)(tcph+1);
71c87e0c JBT	168	lro_desc->tcp_saw_tstamp = 1;
	169	lro_desc->tcp_rcv_tsval = *(ptr+1);
	170	lro_desc->tcp_rcv_tsecr = *(ptr+2);
	171	}
	172
	173	lro_desc->mss = tcp_data_len;
71c87e0c JBT	174	lro_desc->active = 1;
	175
	176	lro_desc->data_csum = lro_tcp_data_csum(iph, tcph,
	177	tcp_data_len);
	178	}
	179
	180	static inline void lro_clear_desc(struct net_lro_desc *lro_desc)
	181	{
	182	memset(lro_desc, 0, sizeof(struct net_lro_desc));
	183	}
	184
	185	static void lro_add_common(struct net_lro_desc lro_desc, struct iphdr iph,
	186	struct tcphdr *tcph, int tcp_data_len)
	187	{
	188	struct sk_buff *parent = lro_desc->parent;
9df7c98a	189	__be32 *topt;
71c87e0c JBT	190
	191	lro_desc->pkt_aggr_cnt++;
	192	lro_desc->ip_tot_len += tcp_data_len;
	193	lro_desc->tcp_next_seq += tcp_data_len;
	194	lro_desc->tcp_window = tcph->window;
	195	lro_desc->tcp_ack = tcph->ack_seq;
	196
	197	/* don't update tcp_rcv_tsval, would not work with PAWS */
	198	if (lro_desc->tcp_saw_tstamp) {
9df7c98a	199	topt = (__be32 *) (tcph + 1);
71c87e0c JBT	200	lro_desc->tcp_rcv_tsecr = *(topt + 2);
	201	}
	202
	203	lro_desc->data_csum = csum_block_add(lro_desc->data_csum,
	204	lro_tcp_data_csum(iph, tcph,
	205	tcp_data_len),
	206	parent->len);
	207
	208	parent->len += tcp_data_len;
	209	parent->data_len += tcp_data_len;
	210	if (tcp_data_len > lro_desc->mss)
	211	lro_desc->mss = tcp_data_len;
	212	}
	213
	214	static void lro_add_packet(struct net_lro_desc lro_desc, struct sk_buff skb,
	215	struct iphdr iph, struct tcphdr tcph)
	216	{
	217	struct sk_buff *parent = lro_desc->parent;
	218	int tcp_data_len = TCP_PAYLOAD_LENGTH(iph, tcph);
	219
	220	lro_add_common(lro_desc, iph, tcph, tcp_data_len);
	221
	222	skb_pull(skb, (skb->len - tcp_data_len));
	223	parent->truesize += skb->truesize;
	224
	225	if (lro_desc->last_skb)
	226	lro_desc->last_skb->next = skb;
	227	else
	228	skb_shinfo(parent)->frag_list = skb;
	229
	230	lro_desc->last_skb = skb;
	231	}
	232
	233	static void lro_add_frags(struct net_lro_desc *lro_desc,
	234	int len, int hlen, int truesize,
	235	struct skb_frag_struct *skb_frags,
	236	struct iphdr iph, struct tcphdr tcph)
	237	{
	238	struct sk_buff *skb = lro_desc->parent;
	239	int tcp_data_len = TCP_PAYLOAD_LENGTH(iph, tcph);
	240
	241	lro_add_common(lro_desc, iph, tcph, tcp_data_len);
	242
	243	skb->truesize += truesize;
	244
	245	skb_frags[0].page_offset += hlen;
9e903e08	246	skb_frag_size_sub(&skb_frags[0], hlen);
71c87e0c JBT	247
	248	while (tcp_data_len > 0) {
	249	(lro_desc->next_frag) = skb_frags;
9e903e08	250	tcp_data_len -= skb_frag_size(skb_frags);
71c87e0c JBT	251	lro_desc->next_frag++;
	252	skb_frags++;
	253	skb_shinfo(skb)->nr_frags++;
	254	}
	255	}
	256
	257	static int lro_check_tcp_conn(struct net_lro_desc *lro_desc,
	258	struct iphdr *iph,
	259	struct tcphdr *tcph)
	260	{
9d4fb27d JP	261	if ((lro_desc->iph->saddr != iph->saddr) \|\|
	262	(lro_desc->iph->daddr != iph->daddr) \|\|
	263	(lro_desc->tcph->source != tcph->source) \|\|
	264	(lro_desc->tcph->dest != tcph->dest))
71c87e0c JBT	265	return -1;
	266	return 0;
	267	}
	268
	269	static struct net_lro_desc lro_get_desc(struct net_lro_mgr lro_mgr,
	270	struct net_lro_desc *lro_arr,
	271	struct iphdr *iph,
	272	struct tcphdr *tcph)
	273	{
	274	struct net_lro_desc *lro_desc = NULL;
	275	struct net_lro_desc *tmp;
	276	int max_desc = lro_mgr->max_desc;
	277	int i;
	278
	279	for (i = 0; i < max_desc; i++) {
	280	tmp = &lro_arr[i];
	281	if (tmp->active)
	282	if (!lro_check_tcp_conn(tmp, iph, tcph)) {
	283	lro_desc = tmp;
	284	goto out;
	285	}
	286	}
	287
	288	for (i = 0; i < max_desc; i++) {
	289	if (!lro_arr[i].active) {
	290	lro_desc = &lro_arr[i];
	291	goto out;
	292	}
	293	}
	294
	295	LRO_INC_STATS(lro_mgr, no_desc);
	296	out:
	297	return lro_desc;
	298	}
	299
	300	static void lro_flush(struct net_lro_mgr *lro_mgr,
	301	struct net_lro_desc *lro_desc)
	302	{
	303	if (lro_desc->pkt_aggr_cnt > 1)
	304	lro_update_tcp_ip_header(lro_desc);
	305
	306	skb_shinfo(lro_desc->parent)->gso_size = lro_desc->mss;
	307
9fea0330 JP	308	if (lro_mgr->features & LRO_F_NAPI)
	309	netif_receive_skb(lro_desc->parent);
	310	else
	311	netif_rx(lro_desc->parent);
71c87e0c JBT	312
	313	LRO_INC_STATS(lro_mgr, flushed);
	314	lro_clear_desc(lro_desc);
	315	}
	316
	317	static int __lro_proc_skb(struct net_lro_mgr lro_mgr, struct sk_buff skb,
9fea0330	318	void *priv)
71c87e0c JBT	319	{
	320	struct net_lro_desc *lro_desc;
	321	struct iphdr *iph;
	322	struct tcphdr *tcph;
	323	u64 flags;
	324	int vlan_hdr_len = 0;
	325
9d4fb27d JP	326	if (!lro_mgr->get_skb_header \|\|
	327	lro_mgr->get_skb_header(skb, (void )&iph, (void )&tcph,
	328	&flags, priv))
71c87e0c JBT	329	goto out;
	330
	331	if (!(flags & LRO_IPV4) \|\| !(flags & LRO_TCP))
	332	goto out;
	333
	334	lro_desc = lro_get_desc(lro_mgr, lro_mgr->lro_arr, iph, tcph);
	335	if (!lro_desc)
	336	goto out;
	337
9d4fb27d JP	338	if ((skb->protocol == htons(ETH_P_8021Q)) &&
9d4fb27d JP	339	!(lro_mgr->features & LRO_F_EXTRACT_VLAN_ID))
71c87e0c JBT	340	vlan_hdr_len = VLAN_HLEN;
	341
	342	if (!lro_desc->active) { /* start new lro session */
	343	if (lro_tcp_ip_check(iph, tcph, skb->len - vlan_hdr_len, NULL))
	344	goto out;
	345
	346	skb->ip_summed = lro_mgr->ip_summed_aggr;
9fea0330	347	lro_init_desc(lro_desc, skb, iph, tcph);
71c87e0c JBT	348	LRO_INC_STATS(lro_mgr, aggregated);
	349	return 0;
	350	}
	351
	352	if (lro_desc->tcp_next_seq != ntohl(tcph->seq))
	353	goto out2;
	354
	355	if (lro_tcp_ip_check(iph, tcph, skb->len, lro_desc))
	356	goto out2;
	357
	358	lro_add_packet(lro_desc, skb, iph, tcph);
	359	LRO_INC_STATS(lro_mgr, aggregated);
	360
	361	if ((lro_desc->pkt_aggr_cnt >= lro_mgr->max_aggr) \|\|
	362	lro_desc->parent->len > (0xFFFF - lro_mgr->dev->mtu))
	363	lro_flush(lro_mgr, lro_desc);
	364
	365	return 0;
	366
	367	out2: /* send aggregated SKBs to stack */
	368	lro_flush(lro_mgr, lro_desc);
	369
251a4b32	370	out:
71c87e0c JBT	371	return 1;
	372	}
	373
	374
	375	static struct sk_buff lro_gen_skb(struct net_lro_mgr lro_mgr,
	376	struct skb_frag_struct *frags,
	377	int len, int true_size,
	378	void *mac_hdr,
	379	int hlen, __wsum sum,
	380	u32 ip_summed)
	381	{
	382	struct sk_buff *skb;
	383	struct skb_frag_struct *skb_frags;
	384	int data_len = len;
	385	int hdr_len = min(len, hlen);
	386
621544eb	387	skb = netdev_alloc_skb(lro_mgr->dev, hlen + lro_mgr->frag_align_pad);
71c87e0c JBT	388	if (!skb)
	389	return NULL;
	390
621544eb	391	skb_reserve(skb, lro_mgr->frag_align_pad);
71c87e0c JBT	392	skb->len = len;
	393	skb->data_len = len - hdr_len;
	394	skb->truesize += true_size;
	395	skb->tail += hdr_len;
	396
	397	memcpy(skb->data, mac_hdr, hdr_len);
	398
	399	skb_frags = skb_shinfo(skb)->frags;
	400	while (data_len > 0) {
	401	skb_frags = frags;
9e903e08	402	data_len -= skb_frag_size(frags);
71c87e0c JBT	403	skb_frags++;
	404	frags++;
	405	skb_shinfo(skb)->nr_frags++;
	406	}
	407
	408	skb_shinfo(skb)->frags[0].page_offset += hdr_len;
9e903e08	409	skb_frag_size_sub(&skb_shinfo(skb)->frags[0], hdr_len);
71c87e0c JBT	410
	411	skb->ip_summed = ip_summed;
	412	skb->csum = sum;
	413	skb->protocol = eth_type_trans(skb, lro_mgr->dev);
	414	return skb;
	415	}
	416
	417	static struct sk_buff __lro_proc_segment(struct net_lro_mgr lro_mgr,
	418	struct skb_frag_struct *frags,
	419	int len, int true_size,
9fea0330	420	void *priv, __wsum sum)
71c87e0c JBT	421	{
	422	struct net_lro_desc *lro_desc;
	423	struct iphdr *iph;
	424	struct tcphdr *tcph;
	425	struct sk_buff *skb;
	426	u64 flags;
	427	void *mac_hdr;
	428	int mac_hdr_len;
	429	int hdr_len = LRO_MAX_PG_HLEN;
	430	int vlan_hdr_len = 0;
	431
9d4fb27d JP	432	if (!lro_mgr->get_frag_header \|\|
	433	lro_mgr->get_frag_header(frags, (void )&mac_hdr, (void )&iph,
	434	(void *)&tcph, &flags, priv)) {
aff65da0	435	mac_hdr = skb_frag_address(frags);
71c87e0c JBT	436	goto out1;
	437	}
	438
	439	if (!(flags & LRO_IPV4) \|\| !(flags & LRO_TCP))
	440	goto out1;
	441
	442	hdr_len = (int)((void *)(tcph) + TCP_HDR_LEN(tcph) - mac_hdr);
	443	mac_hdr_len = (int)((void *)(iph) - mac_hdr);
	444
	445	lro_desc = lro_get_desc(lro_mgr, lro_mgr->lro_arr, iph, tcph);
	446	if (!lro_desc)
	447	goto out1;
	448
	449	if (!lro_desc->active) { /* start new lro session */
	450	if (lro_tcp_ip_check(iph, tcph, len - mac_hdr_len, NULL))
	451	goto out1;
	452
	453	skb = lro_gen_skb(lro_mgr, frags, len, true_size, mac_hdr,
	454	hdr_len, 0, lro_mgr->ip_summed_aggr);
	455	if (!skb)
	456	goto out;
	457
9d4fb27d JP	458	if ((skb->protocol == htons(ETH_P_8021Q)) &&
9d4fb27d JP	459	!(lro_mgr->features & LRO_F_EXTRACT_VLAN_ID))
71c87e0c JBT	460	vlan_hdr_len = VLAN_HLEN;
	461
	462	iph = (void *)(skb->data + vlan_hdr_len);
	463	tcph = (void )((u8 )skb->data + vlan_hdr_len
	464	+ IP_HDR_LEN(iph));
	465
9fea0330	466	lro_init_desc(lro_desc, skb, iph, tcph);
71c87e0c	467	LRO_INC_STATS(lro_mgr, aggregated);
cfcabdcc	468	return NULL;
71c87e0c JBT	469	}
	470
	471	if (lro_desc->tcp_next_seq != ntohl(tcph->seq))
	472	goto out2;
	473
	474	if (lro_tcp_ip_check(iph, tcph, len - mac_hdr_len, lro_desc))
	475	goto out2;
	476
	477	lro_add_frags(lro_desc, len, hdr_len, true_size, frags, iph, tcph);
	478	LRO_INC_STATS(lro_mgr, aggregated);
	479
	480	if ((skb_shinfo(lro_desc->parent)->nr_frags >= lro_mgr->max_aggr) \|\|
	481	lro_desc->parent->len > (0xFFFF - lro_mgr->dev->mtu))
	482	lro_flush(lro_mgr, lro_desc);
	483
	484	return NULL;
	485
	486	out2: /* send aggregated packets to the stack */
	487	lro_flush(lro_mgr, lro_desc);
	488
	489	out1: /* Original packet has to be posted to the stack */
	490	skb = lro_gen_skb(lro_mgr, frags, len, true_size, mac_hdr,
	491	hdr_len, sum, lro_mgr->ip_summed);
	492	out:
	493	return skb;
	494	}
	495
	496	void lro_receive_skb(struct net_lro_mgr *lro_mgr,
	497	struct sk_buff *skb,
	498	void *priv)
	499	{
9fea0330	500	if (__lro_proc_skb(lro_mgr, skb, priv)) {
877364e6	501	if (lro_mgr->features & LRO_F_NAPI)
71c87e0c JBT	502	netif_receive_skb(skb);
	503	else
	504	netif_rx(skb);
	505	}
	506	}
	507	EXPORT_SYMBOL(lro_receive_skb);
	508
71c87e0c JBT	509	void lro_receive_frags(struct net_lro_mgr *lro_mgr,
	510	struct skb_frag_struct *frags,
	511	int len, int true_size, void *priv, __wsum sum)
	512	{
	513	struct sk_buff *skb;
	514
9fea0330	515	skb = __lro_proc_segment(lro_mgr, frags, len, true_size, priv, sum);
71c87e0c JBT	516	if (!skb)
	517	return;
	518
877364e6	519	if (lro_mgr->features & LRO_F_NAPI)
71c87e0c JBT	520	netif_receive_skb(skb);
	521	else
	522	netif_rx(skb);
	523	}
	524	EXPORT_SYMBOL(lro_receive_frags);
	525
71c87e0c JBT	526	void lro_flush_all(struct net_lro_mgr *lro_mgr)
	527	{
	528	int i;
	529	struct net_lro_desc *lro_desc = lro_mgr->lro_arr;
	530
	531	for (i = 0; i < lro_mgr->max_desc; i++) {
	532	if (lro_desc[i].active)
	533	lro_flush(lro_mgr, &lro_desc[i]);
	534	}
	535	}
	536	EXPORT_SYMBOL(lro_flush_all);
	537
	538	void lro_flush_pkt(struct net_lro_mgr *lro_mgr,
	539	struct iphdr iph, struct tcphdr tcph)
	540	{
	541	struct net_lro_desc *lro_desc;
	542
	543	lro_desc = lro_get_desc(lro_mgr, lro_mgr->lro_arr, iph, tcph);
	544	if (lro_desc->active)
	545	lro_flush(lro_mgr, lro_desc);
	546	}
	547	EXPORT_SYMBOL(lro_flush_pkt);