IPv6: reassembly: replace magic number with macro definitions
[safe/jmp/linux-2.6] / net / ipv6 / netfilter / nf_conntrack_reasm.c
index 862d089..744ea49 100644 (file)
@@ -27,7 +27,6 @@
 #include <linux/ipv6.h>
 #include <linux/icmpv6.h>
 #include <linux/random.h>
-#include <linux/jhash.h>
 
 #include <net/sock.h>
 #include <net/snmp.h>
 #include <net/rawv6.h>
 #include <net/ndisc.h>
 #include <net/addrconf.h>
+#include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
 #include <linux/sysctl.h>
 #include <linux/netfilter.h>
 #include <linux/netfilter_ipv6.h>
 #include <linux/kernel.h>
 #include <linux/module.h>
 
-#define NF_CT_FRAG6_HIGH_THRESH 262144 /* == 256*1024 */
-#define NF_CT_FRAG6_LOW_THRESH 196608  /* == 192*1024 */
-#define NF_CT_FRAG6_TIMEOUT IPV6_FRAG_TIMEOUT
 
 struct nf_ct_frag6_skb_cb
 {
@@ -70,48 +67,42 @@ struct nf_ct_frag6_queue
        __u16                   nhoffset;
 };
 
-struct inet_frags_ctl nf_frags_ctl __read_mostly = {
-       .high_thresh     = 256 * 1024,
-       .low_thresh      = 192 * 1024,
-       .timeout         = IPV6_FRAG_TIMEOUT,
-       .secret_interval = 10 * 60 * HZ,
-};
-
 static struct inet_frags nf_frags;
-
-static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
-                              struct in6_addr *daddr)
-{
-       u32 a, b, c;
-
-       a = (__force u32)saddr->s6_addr32[0];
-       b = (__force u32)saddr->s6_addr32[1];
-       c = (__force u32)saddr->s6_addr32[2];
-
-       a += JHASH_GOLDEN_RATIO;
-       b += JHASH_GOLDEN_RATIO;
-       c += nf_frags.rnd;
-       __jhash_mix(a, b, c);
-
-       a += (__force u32)saddr->s6_addr32[3];
-       b += (__force u32)daddr->s6_addr32[0];
-       c += (__force u32)daddr->s6_addr32[1];
-       __jhash_mix(a, b, c);
-
-       a += (__force u32)daddr->s6_addr32[2];
-       b += (__force u32)daddr->s6_addr32[3];
-       c += (__force u32)id;
-       __jhash_mix(a, b, c);
-
-       return c & (INETFRAGS_HASHSZ - 1);
-}
+static struct netns_frags nf_init_frags;
+
+#ifdef CONFIG_SYSCTL
+struct ctl_table nf_ct_ipv6_sysctl_table[] = {
+       {
+               .procname       = "nf_conntrack_frag6_timeout",
+               .data           = &nf_init_frags.timeout,
+               .maxlen         = sizeof(unsigned int),
+               .mode           = 0644,
+               .proc_handler   = proc_dointvec_jiffies,
+       },
+       {
+               .procname       = "nf_conntrack_frag6_low_thresh",
+               .data           = &nf_init_frags.low_thresh,
+               .maxlen         = sizeof(unsigned int),
+               .mode           = 0644,
+               .proc_handler   = proc_dointvec,
+       },
+       {
+               .procname       = "nf_conntrack_frag6_high_thresh",
+               .data           = &nf_init_frags.high_thresh,
+               .maxlen         = sizeof(unsigned int),
+               .mode           = 0644,
+               .proc_handler   = proc_dointvec,
+       },
+       { }
+};
+#endif
 
 static unsigned int nf_hashfn(struct inet_frag_queue *q)
 {
-       struct nf_ct_frag6_queue *nq;
+       const struct nf_ct_frag6_queue *nq;
 
        nq = container_of(q, struct nf_ct_frag6_queue, q);
-       return ip6qhashfn(nq->id, &nq->saddr, &nq->daddr);
+       return inet6_hash_frag(nq->id, &nq->saddr, &nq->daddr, nf_frags.rnd);
 }
 
 static void nf_skb_free(struct sk_buff *skb)
@@ -125,32 +116,16 @@ static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
 {
        if (work)
                *work -= skb->truesize;
-       atomic_sub(skb->truesize, &nf_frags.mem);
+       atomic_sub(skb->truesize, &nf_init_frags.mem);
        nf_skb_free(skb);
        kfree_skb(skb);
 }
 
-static void nf_frag_free(struct inet_frag_queue *q)
-{
-       kfree(container_of(q, struct nf_ct_frag6_queue, q));
-}
-
-static inline struct nf_ct_frag6_queue *frag_alloc_queue(void)
-{
-       struct nf_ct_frag6_queue *fq = kmalloc(sizeof(struct nf_ct_frag6_queue), GFP_ATOMIC);
-
-       if (!fq)
-               return NULL;
-       atomic_add(sizeof(struct nf_ct_frag6_queue), &nf_frags.mem);
-       return fq;
-}
-
 /* Destruction primitives. */
 
 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
 {
-       if (atomic_dec_and_test(&fq->q.refcnt))
-               inet_frag_destroy(&fq->q, &nf_frags, NULL);
+       inet_frag_put(&fq->q, &nf_frags);
 }
 
 /* Kill fq entry. It is not destroyed immediately,
@@ -163,16 +138,21 @@ static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
 
 static void nf_ct_frag6_evictor(void)
 {
-       inet_frag_evictor(&nf_frags);
+       local_bh_disable();
+       inet_frag_evictor(&nf_init_frags, &nf_frags);
+       local_bh_enable();
 }
 
 static void nf_ct_frag6_expire(unsigned long data)
 {
-       struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data;
+       struct nf_ct_frag6_queue *fq;
+
+       fq = container_of((struct inet_frag_queue *)data,
+                       struct nf_ct_frag6_queue, q);
 
        spin_lock(&fq->q.lock);
 
-       if (fq->q.last_in & COMPLETE)
+       if (fq->q.last_in & INET_FRAG_COMPLETE)
                goto out;
 
        fq_kill(fq);
@@ -184,99 +164,41 @@ out:
 
 /* Creation primitives. */
 
-static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash,
-                                         struct nf_ct_frag6_queue *fq_in)
+static __inline__ struct nf_ct_frag6_queue *
+fq_find(__be32 id, u32 user, struct in6_addr *src, struct in6_addr *dst)
 {
-       struct nf_ct_frag6_queue *fq;
-#ifdef CONFIG_SMP
-       struct hlist_node *n;
-#endif
-
-       write_lock(&nf_frags.lock);
-#ifdef CONFIG_SMP
-       hlist_for_each_entry(fq, n, &nf_frags.hash[hash], q.list) {
-               if (fq->id == fq_in->id &&
-                   ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
-                   ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
-                       atomic_inc(&fq->q.refcnt);
-                       write_unlock(&nf_frags.lock);
-                       fq_in->q.last_in |= COMPLETE;
-                       fq_put(fq_in);
-                       return fq;
-               }
-       }
-#endif
-       fq = fq_in;
-
-       if (!mod_timer(&fq->q.timer, jiffies + nf_frags_ctl.timeout))
-               atomic_inc(&fq->q.refcnt);
-
-       atomic_inc(&fq->q.refcnt);
-       hlist_add_head(&fq->q.list, &nf_frags.hash[hash]);
-       INIT_LIST_HEAD(&fq->q.lru_list);
-       list_add_tail(&fq->q.lru_list, &nf_frags.lru_list);
-       nf_frags.nqueues++;
-       write_unlock(&nf_frags.lock);
-       return fq;
-}
+       struct inet_frag_queue *q;
+       struct ip6_create_arg arg;
+       unsigned int hash;
 
+       arg.id = id;
+       arg.user = user;
+       arg.src = src;
+       arg.dst = dst;
 
-static struct nf_ct_frag6_queue *
-nf_ct_frag6_create(unsigned int hash, __be32 id, struct in6_addr *src,                            struct in6_addr *dst)
-{
-       struct nf_ct_frag6_queue *fq;
+       read_lock_bh(&nf_frags.lock);
+       hash = inet6_hash_frag(id, src, dst, nf_frags.rnd);
 
-       if ((fq = frag_alloc_queue()) == NULL) {
-               pr_debug("Can't alloc new queue\n");
+       q = inet_frag_find(&nf_init_frags, &nf_frags, &arg, hash);
+       local_bh_enable();
+       if (q == NULL)
                goto oom;
-       }
-
-       memset(fq, 0, sizeof(struct nf_ct_frag6_queue));
-
-       fq->id = id;
-       ipv6_addr_copy(&fq->saddr, src);
-       ipv6_addr_copy(&fq->daddr, dst);
 
-       setup_timer(&fq->q.timer, nf_ct_frag6_expire, (unsigned long)fq);
-       spin_lock_init(&fq->q.lock);
-       atomic_set(&fq->q.refcnt, 1);
-
-       return nf_ct_frag6_intern(hash, fq);
+       return container_of(q, struct nf_ct_frag6_queue, q);
 
 oom:
+       pr_debug("Can't alloc new queue\n");
        return NULL;
 }
 
-static __inline__ struct nf_ct_frag6_queue *
-fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst)
-{
-       struct nf_ct_frag6_queue *fq;
-       struct hlist_node *n;
-       unsigned int hash = ip6qhashfn(id, src, dst);
-
-       read_lock(&nf_frags.lock);
-       hlist_for_each_entry(fq, n, &nf_frags.hash[hash], q.list) {
-               if (fq->id == id &&
-                   ipv6_addr_equal(src, &fq->saddr) &&
-                   ipv6_addr_equal(dst, &fq->daddr)) {
-                       atomic_inc(&fq->q.refcnt);
-                       read_unlock(&nf_frags.lock);
-                       return fq;
-               }
-       }
-       read_unlock(&nf_frags.lock);
-
-       return nf_ct_frag6_create(hash, id, src, dst);
-}
-
 
 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
-                            struct frag_hdr *fhdr, int nhoff)
+                            const struct frag_hdr *fhdr, int nhoff)
 {
        struct sk_buff *prev, *next;
        int offset, end;
 
-       if (fq->q.last_in & COMPLETE) {
+       if (fq->q.last_in & INET_FRAG_COMPLETE) {
                pr_debug("Allready completed\n");
                goto err;
        }
@@ -303,11 +225,11 @@ static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
                 * or have different end, the segment is corrupted.
                 */
                if (end < fq->q.len ||
-                   ((fq->q.last_in & LAST_IN) && end != fq->q.len)) {
+                   ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len)) {
                        pr_debug("already received last fragment\n");
                        goto err;
                }
-               fq->q.last_in |= LAST_IN;
+               fq->q.last_in |= INET_FRAG_LAST_IN;
                fq->q.len = end;
        } else {
                /* Check if the fragment is rounded to 8 bytes.
@@ -322,7 +244,7 @@ static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
                }
                if (end > fq->q.len) {
                        /* Some bits beyond end -> corruption. */
-                       if (fq->q.last_in & LAST_IN) {
+                       if (fq->q.last_in & INET_FRAG_LAST_IN) {
                                pr_debug("last packet already reached.\n");
                                goto err;
                        }
@@ -427,17 +349,17 @@ static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
        skb->dev = NULL;
        fq->q.stamp = skb->tstamp;
        fq->q.meat += skb->len;
-       atomic_add(skb->truesize, &nf_frags.mem);
+       atomic_add(skb->truesize, &nf_init_frags.mem);
 
        /* The first fragment.
         * nhoffset is obtained from the first fragment, of course.
         */
        if (offset == 0) {
                fq->nhoffset = nhoff;
-               fq->q.last_in |= FIRST_IN;
+               fq->q.last_in |= INET_FRAG_FIRST_IN;
        }
        write_lock(&nf_frags.lock);
-       list_move_tail(&fq->q.lru_list, &nf_frags.lru_list);
+       list_move_tail(&fq->q.lru_list, &nf_init_frags.lru_list);
        write_unlock(&nf_frags.lock);
        return 0;
 
@@ -462,8 +384,8 @@ nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
 
        fq_kill(fq);
 
-       BUG_TRAP(head != NULL);
-       BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0);
+       WARN_ON(head == NULL);
+       WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
 
        /* Unfragmented part is taken from the first segment. */
        payload_len = ((head->data - skb_network_header(head)) -
@@ -483,7 +405,7 @@ nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
        /* 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_shinfo(head)->frag_list) {
+       if (skb_has_frags(head)) {
                struct sk_buff *clone;
                int i, plen = 0;
 
@@ -494,7 +416,7 @@ nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
                clone->next = head->next;
                head->next = clone;
                skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
-               skb_shinfo(head)->frag_list = NULL;
+               skb_frag_list_init(head);
                for (i=0; i<skb_shinfo(head)->nr_frags; i++)
                        plen += skb_shinfo(head)->frags[i].size;
                clone->len = clone->data_len = head->data_len - plen;
@@ -504,7 +426,7 @@ nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
                clone->ip_summed = head->ip_summed;
 
                NFCT_FRAG6_CB(clone)->orig = NULL;
-               atomic_add(clone->truesize, &nf_frags.mem);
+               atomic_add(clone->truesize, &nf_init_frags.mem);
        }
 
        /* We have to remove fragment header from datagram and to relocate
@@ -518,7 +440,7 @@ nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
        skb_shinfo(head)->frag_list = head->next;
        skb_reset_transport_header(head);
        skb_push(head, head->data - skb_network_header(head));
-       atomic_sub(head->truesize, &nf_frags.mem);
+       atomic_sub(head->truesize, &nf_init_frags.mem);
 
        for (fp=head->next; fp; fp = fp->next) {
                head->data_len += fp->len;
@@ -528,7 +450,7 @@ nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
                else if (head->ip_summed == CHECKSUM_COMPLETE)
                        head->csum = csum_add(head->csum, fp->csum);
                head->truesize += fp->truesize;
-               atomic_sub(fp->truesize, &nf_frags.mem);
+               atomic_sub(fp->truesize, &nf_init_frags.mem);
        }
 
        head->next = NULL;
@@ -602,14 +524,14 @@ find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
                if (!ipv6_ext_hdr(nexthdr)) {
                        return -1;
                }
-               if (len < (int)sizeof(struct ipv6_opt_hdr)) {
-                       pr_debug("too short\n");
-                       return -1;
-               }
                if (nexthdr == NEXTHDR_NONE) {
                        pr_debug("next header is none\n");
                        return -1;
                }
+               if (len < (int)sizeof(struct ipv6_opt_hdr)) {
+                       pr_debug("too short\n");
+                       return -1;
+               }
                if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
                        BUG();
                if (nexthdr == NEXTHDR_AUTH)
@@ -635,7 +557,7 @@ find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
        return 0;
 }
 
-struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
+struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
 {
        struct sk_buff *clone;
        struct net_device *dev = skb->dev;
@@ -678,30 +600,31 @@ struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
                goto ret_orig;
        }
 
-       if (atomic_read(&nf_frags.mem) > nf_frags_ctl.high_thresh)
+       if (atomic_read(&nf_init_frags.mem) > nf_init_frags.high_thresh)
                nf_ct_frag6_evictor();
 
-       fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr);
+       fq = fq_find(fhdr->identification, user, &hdr->saddr, &hdr->daddr);
        if (fq == NULL) {
                pr_debug("Can't find and can't create new queue\n");
                goto ret_orig;
        }
 
-       spin_lock(&fq->q.lock);
+       spin_lock_bh(&fq->q.lock);
 
        if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
-               spin_unlock(&fq->q.lock);
+               spin_unlock_bh(&fq->q.lock);
                pr_debug("Can't insert skb to queue\n");
                fq_put(fq);
                goto ret_orig;
        }
 
-       if (fq->q.last_in == (FIRST_IN|LAST_IN) && fq->q.meat == fq->q.len) {
+       if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
+           fq->q.meat == fq->q.len) {
                ret_skb = nf_ct_frag6_reasm(fq, dev);
                if (ret_skb == NULL)
                        pr_debug("Can't reassemble fragmented packets\n");
        }
-       spin_unlock(&fq->q.lock);
+       spin_unlock_bh(&fq->q.lock);
 
        fq_put(fq);
        return ret_skb;
@@ -732,28 +655,20 @@ void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
        nf_conntrack_put_reasm(skb);
 }
 
-int nf_ct_frag6_kfree_frags(struct sk_buff *skb)
-{
-       struct sk_buff *s, *s2;
-
-       for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) {
-
-               s2 = s->next;
-               kfree_skb(s);
-       }
-
-       kfree_skb(skb);
-
-       return 0;
-}
-
 int nf_ct_frag6_init(void)
 {
-       nf_frags.ctl = &nf_frags_ctl;
        nf_frags.hashfn = nf_hashfn;
-       nf_frags.destructor = nf_frag_free;
+       nf_frags.constructor = ip6_frag_init;
+       nf_frags.destructor = NULL;
        nf_frags.skb_free = nf_skb_free;
        nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
+       nf_frags.match = ip6_frag_match;
+       nf_frags.frag_expire = nf_ct_frag6_expire;
+       nf_frags.secret_interval = 10 * 60 * HZ;
+       nf_init_frags.timeout = IPV6_FRAG_TIMEOUT;
+       nf_init_frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
+       nf_init_frags.low_thresh = IPV6_FRAG_LOW_THRESH;
+       inet_frags_init_net(&nf_init_frags);
        inet_frags_init(&nf_frags);
 
        return 0;
@@ -763,6 +678,6 @@ void nf_ct_frag6_cleanup(void)
 {
        inet_frags_fini(&nf_frags);
 
-       nf_frags_ctl.low_thresh = 0;
+       nf_init_frags.low_thresh = 0;
        nf_ct_frag6_evictor();
 }