2 * linux/net/ipv4/inet_lro.c
4 * Large Receive Offload (ipv4 / tcp)
6 * (C) Copyright IBM Corp. 2007
9 * Jan-Bernd Themann <themann@de.ibm.com>
10 * Christoph Raisch <raisch@de.ibm.com>
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)
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.
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.
29 #include <linux/module.h>
30 #include <linux/if_vlan.h>
31 #include <linux/inet_lro.h>
33 MODULE_LICENSE("GPL");
34 MODULE_AUTHOR("Jan-Bernd Themann <themann@de.ibm.com>");
35 MODULE_DESCRIPTION("Large Receive Offload (ipv4 / tcp)");
37 #define TCP_HDR_LEN(tcph) (tcph->doff << 2)
38 #define IP_HDR_LEN(iph) (iph->ihl << 2)
39 #define TCP_PAYLOAD_LENGTH(iph, tcph) \
40 (ntohs(iph->tot_len) - IP_HDR_LEN(iph) - TCP_HDR_LEN(tcph))
42 #define IPH_LEN_WO_OPTIONS 5
43 #define TCPH_LEN_WO_OPTIONS 5
44 #define TCPH_LEN_W_TIMESTAMP 8
46 #define LRO_MAX_PG_HLEN 64
48 #define LRO_INC_STATS(lro_mgr, attr) { lro_mgr->stats.attr++; }
51 * Basic tcp checks whether packet is suitable for LRO
54 static int lro_tcp_ip_check(struct iphdr *iph, struct tcphdr *tcph,
55 int len, struct net_lro_desc *lro_desc)
57 /* check ip header: don't aggregate padded frames */
58 if (ntohs(iph->tot_len) != len)
61 if (TCP_PAYLOAD_LENGTH(iph, tcph) == 0)
64 if (iph->ihl != IPH_LEN_WO_OPTIONS)
67 if (tcph->cwr || tcph->ece || tcph->urg || !tcph->ack
68 || tcph->rst || tcph->syn || tcph->fin)
71 if (INET_ECN_is_ce(ipv4_get_dsfield(iph)))
74 if (tcph->doff != TCPH_LEN_WO_OPTIONS
75 && tcph->doff != TCPH_LEN_W_TIMESTAMP)
78 /* check tcp options (only timestamp allowed) */
79 if (tcph->doff == TCPH_LEN_W_TIMESTAMP) {
80 __be32 *topt = (__be32 *)(tcph + 1);
82 if (*topt != htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16)
83 | (TCPOPT_TIMESTAMP << 8)
87 /* timestamp should be in right order */
89 if (lro_desc && after(ntohl(lro_desc->tcp_rcv_tsval),
93 /* timestamp reply should not be zero */
102 static void lro_update_tcp_ip_header(struct net_lro_desc *lro_desc)
104 struct iphdr *iph = lro_desc->iph;
105 struct tcphdr *tcph = lro_desc->tcph;
109 tcph->ack_seq = lro_desc->tcp_ack;
110 tcph->window = lro_desc->tcp_window;
112 if (lro_desc->tcp_saw_tstamp) {
113 p = (__be32 *)(tcph + 1);
114 *(p+2) = lro_desc->tcp_rcv_tsecr;
117 iph->tot_len = htons(lro_desc->ip_tot_len);
120 iph->check = ip_fast_csum((u8 *)lro_desc->iph, iph->ihl);
123 tcp_hdr_csum = csum_partial((u8 *)tcph, TCP_HDR_LEN(tcph), 0);
124 lro_desc->data_csum = csum_add(lro_desc->data_csum, tcp_hdr_csum);
125 tcph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
126 lro_desc->ip_tot_len -
127 IP_HDR_LEN(iph), IPPROTO_TCP,
128 lro_desc->data_csum);
131 static __wsum lro_tcp_data_csum(struct iphdr *iph, struct tcphdr *tcph, int len)
135 __wsum tcp_ps_hdr_csum;
137 tcp_csum = ~csum_unfold(tcph->check);
138 tcp_hdr_csum = csum_partial((u8 *)tcph, TCP_HDR_LEN(tcph), tcp_csum);
140 tcp_ps_hdr_csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
141 len + TCP_HDR_LEN(tcph),
144 return csum_sub(csum_sub(tcp_csum, tcp_hdr_csum),
148 static void lro_init_desc(struct net_lro_desc *lro_desc, struct sk_buff *skb,
149 struct iphdr *iph, struct tcphdr *tcph,
150 u16 vlan_tag, struct vlan_group *vgrp)
154 u32 tcp_data_len = TCP_PAYLOAD_LENGTH(iph, tcph);
156 nr_frags = skb_shinfo(skb)->nr_frags;
157 lro_desc->parent = skb;
158 lro_desc->next_frag = &(skb_shinfo(skb)->frags[nr_frags]);
160 lro_desc->tcph = tcph;
161 lro_desc->tcp_next_seq = ntohl(tcph->seq) + tcp_data_len;
162 lro_desc->tcp_ack = tcph->ack_seq;
163 lro_desc->tcp_window = tcph->window;
165 lro_desc->pkt_aggr_cnt = 1;
166 lro_desc->ip_tot_len = ntohs(iph->tot_len);
168 if (tcph->doff == 8) {
169 ptr = (__be32 *)(tcph+1);
170 lro_desc->tcp_saw_tstamp = 1;
171 lro_desc->tcp_rcv_tsval = *(ptr+1);
172 lro_desc->tcp_rcv_tsecr = *(ptr+2);
175 lro_desc->mss = tcp_data_len;
176 lro_desc->vgrp = vgrp;
177 lro_desc->vlan_tag = vlan_tag;
178 lro_desc->active = 1;
180 lro_desc->data_csum = lro_tcp_data_csum(iph, tcph,
184 static inline void lro_clear_desc(struct net_lro_desc *lro_desc)
186 memset(lro_desc, 0, sizeof(struct net_lro_desc));
189 static void lro_add_common(struct net_lro_desc *lro_desc, struct iphdr *iph,
190 struct tcphdr *tcph, int tcp_data_len)
192 struct sk_buff *parent = lro_desc->parent;
195 lro_desc->pkt_aggr_cnt++;
196 lro_desc->ip_tot_len += tcp_data_len;
197 lro_desc->tcp_next_seq += tcp_data_len;
198 lro_desc->tcp_window = tcph->window;
199 lro_desc->tcp_ack = tcph->ack_seq;
201 /* don't update tcp_rcv_tsval, would not work with PAWS */
202 if (lro_desc->tcp_saw_tstamp) {
203 topt = (__be32 *) (tcph + 1);
204 lro_desc->tcp_rcv_tsecr = *(topt + 2);
207 lro_desc->data_csum = csum_block_add(lro_desc->data_csum,
208 lro_tcp_data_csum(iph, tcph,
212 parent->len += tcp_data_len;
213 parent->data_len += tcp_data_len;
214 if (tcp_data_len > lro_desc->mss)
215 lro_desc->mss = tcp_data_len;
218 static void lro_add_packet(struct net_lro_desc *lro_desc, struct sk_buff *skb,
219 struct iphdr *iph, struct tcphdr *tcph)
221 struct sk_buff *parent = lro_desc->parent;
222 int tcp_data_len = TCP_PAYLOAD_LENGTH(iph, tcph);
224 lro_add_common(lro_desc, iph, tcph, tcp_data_len);
226 skb_pull(skb, (skb->len - tcp_data_len));
227 parent->truesize += skb->truesize;
229 if (lro_desc->last_skb)
230 lro_desc->last_skb->next = skb;
232 skb_shinfo(parent)->frag_list = skb;
234 lro_desc->last_skb = skb;
237 static void lro_add_frags(struct net_lro_desc *lro_desc,
238 int len, int hlen, int truesize,
239 struct skb_frag_struct *skb_frags,
240 struct iphdr *iph, struct tcphdr *tcph)
242 struct sk_buff *skb = lro_desc->parent;
243 int tcp_data_len = TCP_PAYLOAD_LENGTH(iph, tcph);
245 lro_add_common(lro_desc, iph, tcph, tcp_data_len);
247 skb->truesize += truesize;
249 skb_frags[0].page_offset += hlen;
250 skb_frags[0].size -= hlen;
252 while (tcp_data_len > 0) {
253 *(lro_desc->next_frag) = *skb_frags;
254 tcp_data_len -= skb_frags->size;
255 lro_desc->next_frag++;
257 skb_shinfo(skb)->nr_frags++;
261 static int lro_check_tcp_conn(struct net_lro_desc *lro_desc,
265 if ((lro_desc->iph->saddr != iph->saddr)
266 || (lro_desc->iph->daddr != iph->daddr)
267 || (lro_desc->tcph->source != tcph->source)
268 || (lro_desc->tcph->dest != tcph->dest))
273 static struct net_lro_desc *lro_get_desc(struct net_lro_mgr *lro_mgr,
274 struct net_lro_desc *lro_arr,
278 struct net_lro_desc *lro_desc = NULL;
279 struct net_lro_desc *tmp;
280 int max_desc = lro_mgr->max_desc;
283 for (i = 0; i < max_desc; i++) {
286 if (!lro_check_tcp_conn(tmp, iph, tcph)) {
292 for (i = 0; i < max_desc; i++) {
293 if (!lro_arr[i].active) {
294 lro_desc = &lro_arr[i];
299 LRO_INC_STATS(lro_mgr, no_desc);
304 static void lro_flush(struct net_lro_mgr *lro_mgr,
305 struct net_lro_desc *lro_desc)
307 if (lro_desc->pkt_aggr_cnt > 1)
308 lro_update_tcp_ip_header(lro_desc);
310 skb_shinfo(lro_desc->parent)->gso_size = lro_desc->mss;
312 if (lro_desc->vgrp) {
313 if (test_bit(LRO_F_NAPI, &lro_mgr->features))
314 vlan_hwaccel_receive_skb(lro_desc->parent,
318 vlan_hwaccel_rx(lro_desc->parent,
323 if (test_bit(LRO_F_NAPI, &lro_mgr->features))
324 netif_receive_skb(lro_desc->parent);
326 netif_rx(lro_desc->parent);
329 LRO_INC_STATS(lro_mgr, flushed);
330 lro_clear_desc(lro_desc);
333 static int __lro_proc_skb(struct net_lro_mgr *lro_mgr, struct sk_buff *skb,
334 struct vlan_group *vgrp, u16 vlan_tag, void *priv)
336 struct net_lro_desc *lro_desc;
340 int vlan_hdr_len = 0;
342 if (!lro_mgr->get_skb_header
343 || lro_mgr->get_skb_header(skb, (void *)&iph, (void *)&tcph,
347 if (!(flags & LRO_IPV4) || !(flags & LRO_TCP))
350 lro_desc = lro_get_desc(lro_mgr, lro_mgr->lro_arr, iph, tcph);
354 if ((skb->protocol == htons(ETH_P_8021Q))
355 && !test_bit(LRO_F_EXTRACT_VLAN_ID, &lro_mgr->features))
356 vlan_hdr_len = VLAN_HLEN;
358 if (!lro_desc->active) { /* start new lro session */
359 if (lro_tcp_ip_check(iph, tcph, skb->len - vlan_hdr_len, NULL))
362 skb->ip_summed = lro_mgr->ip_summed_aggr;
363 lro_init_desc(lro_desc, skb, iph, tcph, vlan_tag, vgrp);
364 LRO_INC_STATS(lro_mgr, aggregated);
368 if (lro_desc->tcp_next_seq != ntohl(tcph->seq))
371 if (lro_tcp_ip_check(iph, tcph, skb->len, lro_desc))
374 lro_add_packet(lro_desc, skb, iph, tcph);
375 LRO_INC_STATS(lro_mgr, aggregated);
377 if ((lro_desc->pkt_aggr_cnt >= lro_mgr->max_aggr) ||
378 lro_desc->parent->len > (0xFFFF - lro_mgr->dev->mtu))
379 lro_flush(lro_mgr, lro_desc);
383 out2: /* send aggregated SKBs to stack */
384 lro_flush(lro_mgr, lro_desc);
386 out: /* Original SKB has to be posted to stack */
387 skb->ip_summed = lro_mgr->ip_summed;
392 static struct sk_buff *lro_gen_skb(struct net_lro_mgr *lro_mgr,
393 struct skb_frag_struct *frags,
394 int len, int true_size,
396 int hlen, __wsum sum,
400 struct skb_frag_struct *skb_frags;
402 int hdr_len = min(len, hlen);
404 skb = netdev_alloc_skb(lro_mgr->dev, hlen);
409 skb->data_len = len - hdr_len;
410 skb->truesize += true_size;
411 skb->tail += hdr_len;
413 memcpy(skb->data, mac_hdr, hdr_len);
415 skb_frags = skb_shinfo(skb)->frags;
416 while (data_len > 0) {
418 data_len -= frags->size;
421 skb_shinfo(skb)->nr_frags++;
424 skb_shinfo(skb)->frags[0].page_offset += hdr_len;
425 skb_shinfo(skb)->frags[0].size -= hdr_len;
427 skb->ip_summed = ip_summed;
429 skb->protocol = eth_type_trans(skb, lro_mgr->dev);
433 static struct sk_buff *__lro_proc_segment(struct net_lro_mgr *lro_mgr,
434 struct skb_frag_struct *frags,
435 int len, int true_size,
436 struct vlan_group *vgrp,
437 u16 vlan_tag, void *priv, __wsum sum)
439 struct net_lro_desc *lro_desc;
446 int hdr_len = LRO_MAX_PG_HLEN;
447 int vlan_hdr_len = 0;
449 if (!lro_mgr->get_frag_header
450 || lro_mgr->get_frag_header(frags, (void *)&mac_hdr, (void *)&iph,
451 (void *)&tcph, &flags, priv)) {
452 mac_hdr = page_address(frags->page) + frags->page_offset;
456 if (!(flags & LRO_IPV4) || !(flags & LRO_TCP))
459 hdr_len = (int)((void *)(tcph) + TCP_HDR_LEN(tcph) - mac_hdr);
460 mac_hdr_len = (int)((void *)(iph) - mac_hdr);
462 lro_desc = lro_get_desc(lro_mgr, lro_mgr->lro_arr, iph, tcph);
466 if (!lro_desc->active) { /* start new lro session */
467 if (lro_tcp_ip_check(iph, tcph, len - mac_hdr_len, NULL))
470 skb = lro_gen_skb(lro_mgr, frags, len, true_size, mac_hdr,
471 hdr_len, 0, lro_mgr->ip_summed_aggr);
475 if ((skb->protocol == htons(ETH_P_8021Q))
476 && !test_bit(LRO_F_EXTRACT_VLAN_ID, &lro_mgr->features))
477 vlan_hdr_len = VLAN_HLEN;
479 iph = (void *)(skb->data + vlan_hdr_len);
480 tcph = (void *)((u8 *)skb->data + vlan_hdr_len
483 lro_init_desc(lro_desc, skb, iph, tcph, 0, NULL);
484 LRO_INC_STATS(lro_mgr, aggregated);
488 if (lro_desc->tcp_next_seq != ntohl(tcph->seq))
491 if (lro_tcp_ip_check(iph, tcph, len - mac_hdr_len, lro_desc))
494 lro_add_frags(lro_desc, len, hdr_len, true_size, frags, iph, tcph);
495 LRO_INC_STATS(lro_mgr, aggregated);
497 if ((skb_shinfo(lro_desc->parent)->nr_frags >= lro_mgr->max_aggr) ||
498 lro_desc->parent->len > (0xFFFF - lro_mgr->dev->mtu))
499 lro_flush(lro_mgr, lro_desc);
503 out2: /* send aggregated packets to the stack */
504 lro_flush(lro_mgr, lro_desc);
506 out1: /* Original packet has to be posted to the stack */
507 skb = lro_gen_skb(lro_mgr, frags, len, true_size, mac_hdr,
508 hdr_len, sum, lro_mgr->ip_summed);
513 void lro_receive_skb(struct net_lro_mgr *lro_mgr,
517 if (__lro_proc_skb(lro_mgr, skb, NULL, 0, priv)) {
518 if (test_bit(LRO_F_NAPI, &lro_mgr->features))
519 netif_receive_skb(skb);
524 EXPORT_SYMBOL(lro_receive_skb);
526 void lro_vlan_hwaccel_receive_skb(struct net_lro_mgr *lro_mgr,
528 struct vlan_group *vgrp,
532 if (__lro_proc_skb(lro_mgr, skb, vgrp, vlan_tag, priv)) {
533 if (test_bit(LRO_F_NAPI, &lro_mgr->features))
534 vlan_hwaccel_receive_skb(skb, vgrp, vlan_tag);
536 vlan_hwaccel_rx(skb, vgrp, vlan_tag);
539 EXPORT_SYMBOL(lro_vlan_hwaccel_receive_skb);
541 void lro_receive_frags(struct net_lro_mgr *lro_mgr,
542 struct skb_frag_struct *frags,
543 int len, int true_size, void *priv, __wsum sum)
547 skb = __lro_proc_segment(lro_mgr, frags, len, true_size, NULL, 0,
552 if (test_bit(LRO_F_NAPI, &lro_mgr->features))
553 netif_receive_skb(skb);
557 EXPORT_SYMBOL(lro_receive_frags);
559 void lro_vlan_hwaccel_receive_frags(struct net_lro_mgr *lro_mgr,
560 struct skb_frag_struct *frags,
561 int len, int true_size,
562 struct vlan_group *vgrp,
563 u16 vlan_tag, void *priv, __wsum sum)
567 skb = __lro_proc_segment(lro_mgr, frags, len, true_size, vgrp,
568 vlan_tag, priv, sum);
572 if (test_bit(LRO_F_NAPI, &lro_mgr->features))
573 vlan_hwaccel_receive_skb(skb, vgrp, vlan_tag);
575 vlan_hwaccel_rx(skb, vgrp, vlan_tag);
577 EXPORT_SYMBOL(lro_vlan_hwaccel_receive_frags);
579 void lro_flush_all(struct net_lro_mgr *lro_mgr)
582 struct net_lro_desc *lro_desc = lro_mgr->lro_arr;
584 for (i = 0; i < lro_mgr->max_desc; i++) {
585 if (lro_desc[i].active)
586 lro_flush(lro_mgr, &lro_desc[i]);
589 EXPORT_SYMBOL(lro_flush_all);
591 void lro_flush_pkt(struct net_lro_mgr *lro_mgr,
592 struct iphdr *iph, struct tcphdr *tcph)
594 struct net_lro_desc *lro_desc;
596 lro_desc = lro_get_desc(lro_mgr, lro_mgr->lro_arr, iph, tcph);
597 if (lro_desc->active)
598 lro_flush(lro_mgr, lro_desc);
600 EXPORT_SYMBOL(lro_flush_pkt);