tcp: skip cwnd moderation in TCP_CA_Open in tcp_try_to_open
[linux-2.6.git] / net / ipv4 / ip_fragment.c
index b59430b..fdaabf2 100644 (file)
@@ -32,6 +32,7 @@
 #include <linux/netdevice.h>
 #include <linux/jhash.h>
 #include <linux/random.h>
+#include <linux/slab.h>
 #include <net/route.h>
 #include <net/dst.h>
 #include <net/sock.h>
@@ -44,6 +45,7 @@
 #include <linux/udp.h>
 #include <linux/inet.h>
 #include <linux/netfilter_ipv4.h>
+#include <net/inet_ecn.h>
 
 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
  * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
@@ -69,11 +71,46 @@ struct ipq {
        __be32          daddr;
        __be16          id;
        u8              protocol;
+       u8              ecn; /* RFC3168 support */
        int             iif;
        unsigned int    rid;
        struct inet_peer *peer;
 };
 
+/* RFC 3168 support :
+ * We want to check ECN values of all fragments, do detect invalid combinations.
+ * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
+ */
+#define        IPFRAG_ECN_NOT_ECT      0x01 /* one frag had ECN_NOT_ECT */
+#define        IPFRAG_ECN_ECT_1        0x02 /* one frag had ECN_ECT_1 */
+#define        IPFRAG_ECN_ECT_0        0x04 /* one frag had ECN_ECT_0 */
+#define        IPFRAG_ECN_CE           0x08 /* one frag had ECN_CE */
+
+static inline u8 ip4_frag_ecn(u8 tos)
+{
+       return 1 << (tos & INET_ECN_MASK);
+}
+
+/* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
+ * Value : 0xff if frame should be dropped.
+ *         0 or INET_ECN_CE value, to be ORed in to final iph->tos field
+ */
+static const u8 ip4_frag_ecn_table[16] = {
+       /* at least one fragment had CE, and others ECT_0 or ECT_1 */
+       [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0]                      = INET_ECN_CE,
+       [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1]                      = INET_ECN_CE,
+       [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1]   = INET_ECN_CE,
+
+       /* invalid combinations : drop frame */
+       [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
+       [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
+       [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
+       [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
+       [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
+       [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
+       [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
+};
+
 static struct inet_frags ip4_frags;
 
 int ip_frag_nqueues(struct net *net)
@@ -115,19 +152,16 @@ static int ip4_frag_match(struct inet_frag_queue *q, void *a)
        struct ip4_create_arg *arg = a;
 
        qp = container_of(q, struct ipq, q);
-       return (qp->id == arg->iph->id &&
+       return  qp->id == arg->iph->id &&
                        qp->saddr == arg->iph->saddr &&
                        qp->daddr == arg->iph->daddr &&
                        qp->protocol == arg->iph->protocol &&
-                       qp->user == arg->user);
+                       qp->user == arg->user;
 }
 
 /* Memory Tracking Functions. */
-static __inline__ void frag_kfree_skb(struct netns_frags *nf,
-               struct sk_buff *skb, int *work)
+static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
 {
-       if (work)
-               *work -= skb->truesize;
        atomic_sub(skb->truesize, &nf->mem);
        kfree_skb(skb);
 }
@@ -139,11 +173,12 @@ static void ip4_frag_init(struct inet_frag_queue *q, void *a)
 
        qp->protocol = arg->iph->protocol;
        qp->id = arg->iph->id;
+       qp->ecn = ip4_frag_ecn(arg->iph->tos);
        qp->saddr = arg->iph->saddr;
        qp->daddr = arg->iph->daddr;
        qp->user = arg->user;
        qp->peer = sysctl_ipfrag_max_dist ?
-               inet_getpeer(arg->iph->saddr, 1) : NULL;
+               inet_getpeer_v4(arg->iph->saddr, 1) : NULL;
 }
 
 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
@@ -206,31 +241,31 @@ static void ip_expire(unsigned long arg)
 
        if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
                struct sk_buff *head = qp->q.fragments;
+               const struct iphdr *iph;
+               int err;
 
                rcu_read_lock();
                head->dev = dev_get_by_index_rcu(net, qp->iif);
                if (!head->dev)
                        goto out_rcu_unlock;
 
+               /* skb dst is stale, drop it, and perform route lookup again */
+               skb_dst_drop(head);
+               iph = ip_hdr(head);
+               err = ip_route_input_noref(head, iph->daddr, iph->saddr,
+                                          iph->tos, head->dev);
+               if (err)
+                       goto out_rcu_unlock;
+
                /*
-                * Only search router table for the head fragment,
-                * when defraging timeout at PRE_ROUTING HOOK.
+                * Only an end host needs to send an ICMP
+                * "Fragment Reassembly Timeout" message, per RFC792.
                 */
-               if (qp->user == IP_DEFRAG_CONNTRACK_IN && !skb_dst(head)) {
-                       const struct iphdr *iph = ip_hdr(head);
-                       int err = ip_route_input(head, iph->daddr, iph->saddr,
-                                                iph->tos, head->dev);
-                       if (unlikely(err))
-                               goto out_rcu_unlock;
-
-                       /*
-                        * Only an end host needs to send an ICMP
-                        * "Fragment Reassembly Timeout" message, per RFC792.
-                        */
-                       if (skb_rtable(head)->rt_type != RTN_LOCAL)
-                               goto out_rcu_unlock;
+               if (qp->user == IP_DEFRAG_AF_PACKET ||
+                   (qp->user == IP_DEFRAG_CONNTRACK_IN &&
+                    skb_rtable(head)->rt_type != RTN_LOCAL))
+                       goto out_rcu_unlock;
 
-               }
 
                /* Send an ICMP "Fragment Reassembly Timeout" message. */
                icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
@@ -308,7 +343,7 @@ static int ip_frag_reinit(struct ipq *qp)
        fp = qp->q.fragments;
        do {
                struct sk_buff *xp = fp->next;
-               frag_kfree_skb(qp->q.net, fp, NULL);
+               frag_kfree_skb(qp->q.net, fp);
                fp = xp;
        } while (fp);
 
@@ -316,7 +351,9 @@ static int ip_frag_reinit(struct ipq *qp)
        qp->q.len = 0;
        qp->q.meat = 0;
        qp->q.fragments = NULL;
+       qp->q.fragments_tail = NULL;
        qp->iif = 0;
+       qp->ecn = 0;
 
        return 0;
 }
@@ -329,6 +366,7 @@ static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
        int flags, offset;
        int ihl, end;
        int err = -ENOENT;
+       u8 ecn;
 
        if (qp->q.last_in & INET_FRAG_COMPLETE)
                goto err;
@@ -340,6 +378,7 @@ static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
                goto err;
        }
 
+       ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
        offset = ntohs(ip_hdr(skb)->frag_off);
        flags = offset & ~IP_OFFSET;
        offset &= IP_OFFSET;
@@ -388,6 +427,11 @@ static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
         * in the chain of fragments so far.  We must know where to put
         * this fragment, right?
         */
+       prev = qp->q.fragments_tail;
+       if (!prev || FRAG_CB(prev)->offset < offset) {
+               next = NULL;
+               goto found;
+       }
        prev = NULL;
        for (next = qp->q.fragments; next != NULL; next = next->next) {
                if (FRAG_CB(next)->offset >= offset)
@@ -395,6 +439,7 @@ static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
                prev = next;
        }
 
+found:
        /* We found where to put this one.  Check for overlap with
         * preceding fragment, and, if needed, align things so that
         * any overlaps are eliminated.
@@ -445,7 +490,7 @@ static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
                                qp->q.fragments = next;
 
                        qp->q.meat -= free_it->len;
-                       frag_kfree_skb(qp->q.net, free_it, NULL);
+                       frag_kfree_skb(qp->q.net, free_it);
                }
        }
 
@@ -453,6 +498,8 @@ static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
 
        /* Insert this fragment in the chain of fragments. */
        skb->next = next;
+       if (!next)
+               qp->q.fragments_tail = skb;
        if (prev)
                prev->next = skb;
        else
@@ -465,6 +512,7 @@ static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
        }
        qp->q.stamp = skb->tstamp;
        qp->q.meat += skb->len;
+       qp->ecn |= ecn;
        atomic_add(skb->truesize, &qp->q.net->mem);
        if (offset == 0)
                qp->q.last_in |= INET_FRAG_FIRST_IN;
@@ -495,9 +543,15 @@ static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
        int len;
        int ihlen;
        int err;
+       u8 ecn;
 
        ipq_kill(qp);
 
+       ecn = ip4_frag_ecn_table[qp->ecn];
+       if (unlikely(ecn == 0xff)) {
+               err = -EINVAL;
+               goto out_fail;
+       }
        /* Make the one we just received the head. */
        if (prev) {
                head = prev->next;
@@ -506,6 +560,8 @@ static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
                        goto out_nomem;
 
                fp->next = head->next;
+               if (!fp->next)
+                       qp->q.fragments_tail = fp;
                prev->next = fp;
 
                skb_morph(head, qp->q.fragments);
@@ -533,7 +589,7 @@ static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
        /* If the first fragment is fragmented itself, we split
         * it to two chunks: the first with data and paged part
         * and the second, holding only fragments. */
-       if (skb_has_frags(head)) {
+       if (skb_has_frag_list(head)) {
                struct sk_buff *clone;
                int i, plen = 0;
 
@@ -543,8 +599,8 @@ static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
                head->next = clone;
                skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
                skb_frag_list_init(head);
-               for (i=0; i<skb_shinfo(head)->nr_frags; i++)
-                       plen += skb_shinfo(head)->frags[i].size;
+               for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
+                       plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
                clone->len = clone->data_len = head->data_len - plen;
                head->data_len -= clone->len;
                head->len -= clone->len;
@@ -555,7 +611,6 @@ static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
 
        skb_shinfo(head)->frag_list = head->next;
        skb_push(head, head->data - skb_network_header(head));
-       atomic_sub(head->truesize, &qp->q.net->mem);
 
        for (fp=head->next; fp; fp = fp->next) {
                head->data_len += fp->len;
@@ -565,8 +620,8 @@ static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
                else if (head->ip_summed == CHECKSUM_COMPLETE)
                        head->csum = csum_add(head->csum, fp->csum);
                head->truesize += fp->truesize;
-               atomic_sub(fp->truesize, &qp->q.net->mem);
        }
+       atomic_sub(head->truesize, &qp->q.net->mem);
 
        head->next = NULL;
        head->dev = dev;
@@ -575,8 +630,10 @@ static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
        iph = ip_hdr(head);
        iph->frag_off = 0;
        iph->tot_len = htons(len);
+       iph->tos |= ecn;
        IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
        qp->q.fragments = NULL;
+       qp->q.fragments_tail = NULL;
        return 0;
 
 out_nomem:
@@ -623,6 +680,43 @@ int ip_defrag(struct sk_buff *skb, u32 user)
        kfree_skb(skb);
        return -ENOMEM;
 }
+EXPORT_SYMBOL(ip_defrag);
+
+struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
+{
+       const struct iphdr *iph;
+       u32 len;
+
+       if (skb->protocol != htons(ETH_P_IP))
+               return skb;
+
+       if (!pskb_may_pull(skb, sizeof(struct iphdr)))
+               return skb;
+
+       iph = ip_hdr(skb);
+       if (iph->ihl < 5 || iph->version != 4)
+               return skb;
+       if (!pskb_may_pull(skb, iph->ihl*4))
+               return skb;
+       iph = ip_hdr(skb);
+       len = ntohs(iph->tot_len);
+       if (skb->len < len || len < (iph->ihl * 4))
+               return skb;
+
+       if (ip_is_fragment(ip_hdr(skb))) {
+               skb = skb_share_check(skb, GFP_ATOMIC);
+               if (skb) {
+                       if (pskb_trim_rcsum(skb, len))
+                               return skb;
+                       memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
+                       if (ip_defrag(skb, user))
+                               return NULL;
+                       skb->rxhash = 0;
+               }
+       }
+       return skb;
+}
+EXPORT_SYMBOL(ip_check_defrag);
 
 #ifdef CONFIG_SYSCTL
 static int zero;
@@ -776,5 +870,3 @@ void __init ipfrag_init(void)
        ip4_frags.secret_interval = 10 * 60 * HZ;
        inet_frags_init(&ip4_frags);
 }
-
-EXPORT_SYMBOL(ip_defrag);