vserver 1.9.5.x5
[linux-2.6.git] / net / ipv4 / ipcomp.c
index 2a5a7df..89e320d 100644 (file)
 #include <linux/config.h>
 #include <linux/module.h>
 #include <asm/scatterlist.h>
+#include <asm/semaphore.h>
 #include <linux/crypto.h>
 #include <linux/pfkeyv2.h>
-#include <net/inet_ecn.h>
+#include <linux/percpu.h>
+#include <linux/smp.h>
+#include <linux/list.h>
+#include <linux/vmalloc.h>
+#include <linux/rtnetlink.h>
 #include <net/ip.h>
 #include <net/xfrm.h>
 #include <net/icmp.h>
 #include <net/ipcomp.h>
 
+struct ipcomp_tfms {
+       struct list_head list;
+       struct crypto_tfm **tfms;
+       int users;
+};
+
+static DECLARE_MUTEX(ipcomp_resource_sem);
+static void **ipcomp_scratches;
+static int ipcomp_scratch_users;
+static LIST_HEAD(ipcomp_tfms_list);
+
 static int ipcomp_decompress(struct xfrm_state *x, struct sk_buff *skb)
 {
        int err, plen, dlen;
        struct iphdr *iph;
        struct ipcomp_data *ipcd = x->data;
-       u8 *start, *scratch = ipcd->scratch;
+       u8 *start, *scratch;
+       struct crypto_tfm *tfm;
+       int cpu;
        
        plen = skb->len;
        dlen = IPCOMP_SCRATCH_SIZE;
        start = skb->data;
 
-       err = crypto_comp_decompress(ipcd->tfm, start, plen, scratch, &dlen);
+       cpu = get_cpu();
+       scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
+       tfm = *per_cpu_ptr(ipcd->tfms, cpu);
+
+       err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
        if (err)
                goto out;
 
@@ -53,6 +75,7 @@ static int ipcomp_decompress(struct xfrm_state *x, struct sk_buff *skb)
        iph = skb->nh.iph;
        iph->tot_len = htons(dlen + iph->ihl * 4);
 out:   
+       put_cpu();
        return err;
 }
 
@@ -98,14 +121,20 @@ static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
        int err, plen, dlen, ihlen;
        struct iphdr *iph = skb->nh.iph;
        struct ipcomp_data *ipcd = x->data;
-       u8 *start, *scratch = ipcd->scratch;
+       u8 *start, *scratch;
+       struct crypto_tfm *tfm;
+       int cpu;
        
        ihlen = iph->ihl * 4;
        plen = skb->len - ihlen;
        dlen = IPCOMP_SCRATCH_SIZE;
        start = skb->data + ihlen;
 
-       err = crypto_comp_compress(ipcd->tfm, start, plen, scratch, &dlen);
+       cpu = get_cpu();
+       scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
+       tfm = *per_cpu_ptr(ipcd->tfms, cpu);
+
+       err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
        if (err)
                goto out;
 
@@ -114,77 +143,38 @@ static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
                goto out;
        }
        
-       memcpy(start, scratch, dlen);
-       pskb_trim(skb, ihlen + dlen);
+       memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
+       put_cpu();
+
+       pskb_trim(skb, ihlen + dlen + sizeof(struct ip_comp_hdr));
+       return 0;
        
 out:   
+       put_cpu();
        return err;
 }
 
-static void ipcomp_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
-{
-       struct dst_entry *dst = skb->dst;
-       struct iphdr *iph, *top_iph;
-
-       iph = skb->nh.iph;
-       top_iph = (struct iphdr *)skb_push(skb, sizeof(struct iphdr));
-       top_iph->ihl = 5;
-       top_iph->version = 4;
-       top_iph->tos = iph->tos;
-       top_iph->tot_len = htons(skb->len);
-       if (!(iph->frag_off&htons(IP_DF)))
-               __ip_select_ident(top_iph, dst, 0);
-       top_iph->ttl = iph->ttl;
-       top_iph->check = 0;
-       top_iph->saddr = x->props.saddr.a4;
-       top_iph->daddr = x->id.daddr.a4;
-       top_iph->frag_off = iph->frag_off&~htons(IP_MF|IP_OFFSET);
-       memset(&(IPCB(skb)->opt), 0, sizeof(struct ip_options));
-       skb->nh.raw = skb->data;
-}
-
 static int ipcomp_output(struct sk_buff *skb)
 {
        int err;
        struct dst_entry *dst = skb->dst;
        struct xfrm_state *x = dst->xfrm;
-       struct iphdr *iph, *top_iph;
+       struct iphdr *iph;
        struct ip_comp_hdr *ipch;
        struct ipcomp_data *ipcd = x->data;
-       union {
-               struct iphdr    iph;
-               char            buf[60];
-       } tmp_iph;
        int hdr_len = 0;
 
-       if (skb->ip_summed == CHECKSUM_HW && skb_checksum_help(skb) == NULL) {
-               err = -EINVAL;
-               goto error_nolock;
-       }
-
-       spin_lock_bh(&x->lock);
-       err = xfrm_check_output(x, skb, AF_INET);
-       if (err)
-               goto error;
-
-       /* Don't bother compressing */
-       if (!x->props.mode) {
-               iph = skb->nh.iph;
-               hdr_len = iph->ihl * 4;
-       }
+       iph = skb->nh.iph;
+       iph->tot_len = htons(skb->len);
+       hdr_len = iph->ihl * 4;
        if ((skb->len - hdr_len) < ipcd->threshold) {
+               /* Don't bother compressing */
                if (x->props.mode) {
-                       ipcomp_tunnel_encap(x, skb);
-                       iph = skb->nh.iph;
-                       iph->protocol = IPPROTO_IPIP;
                        ip_send_check(iph);
                }
                goto out_ok;
        }
 
-       if (x->props.mode) 
-               ipcomp_tunnel_encap(x, skb);
-
        if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
            skb_linearize(skb, GFP_ATOMIC) != 0) {
                err = -ENOMEM;
@@ -196,7 +186,6 @@ static int ipcomp_output(struct sk_buff *skb)
                if (err == -EMSGSIZE) {
                        if (x->props.mode) {
                                iph = skb->nh.iph;
-                               iph->protocol = IPPROTO_IPIP;
                                ip_send_check(iph);
                        }
                        goto out_ok;
@@ -206,40 +195,19 @@ static int ipcomp_output(struct sk_buff *skb)
 
        /* Install ipcomp header, convert into ipcomp datagram. */
        iph = skb->nh.iph;
-       memcpy(&tmp_iph, iph, iph->ihl * 4);
-       top_iph = (struct iphdr *)skb_push(skb, sizeof(struct ip_comp_hdr));
-       memcpy(top_iph, &tmp_iph, iph->ihl * 4);
-       iph = top_iph;
-       if (x->props.mode && (x->props.flags & XFRM_STATE_NOECN))
-               IP_ECN_clear(iph);
        iph->tot_len = htons(skb->len);
-       iph->protocol = IPPROTO_COMP;
-       iph->check = 0;
        ipch = (struct ip_comp_hdr *)((char *)iph + iph->ihl * 4);
-       ipch->nexthdr = x->props.mode ? IPPROTO_IPIP : tmp_iph.iph.protocol;
+       ipch->nexthdr = iph->protocol;
        ipch->flags = 0;
        ipch->cpi = htons((u16 )ntohl(x->id.spi));
+       iph->protocol = IPPROTO_COMP;
        ip_send_check(iph);
-       skb->nh.raw = skb->data;
 
 out_ok:
-       x->curlft.bytes += skb->len;
-       x->curlft.packets++;
-       spin_unlock_bh(&x->lock);
-       
-       if ((skb->dst = dst_pop(dst)) == NULL) {
-               err = -EHOSTUNREACH;
-               goto error_nolock;
-       }
-       err = NET_XMIT_BYPASS;
+       err = 0;
 
-out_exit:
-       return err;
 error:
-       spin_unlock_bh(&x->lock);
-error_nolock:
-       kfree_skb(skb);
-       goto out_exit;
+       return err;
 }
 
 static void ipcomp4_err(struct sk_buff *skb, u32 info)
@@ -326,12 +294,132 @@ out:
        return err;
 }
 
+static void ipcomp_free_scratches(void)
+{
+       int i;
+       void **scratches;
+
+       if (--ipcomp_scratch_users)
+               return;
+
+       scratches = ipcomp_scratches;
+       if (!scratches)
+               return;
+
+       for_each_cpu(i) {
+               void *scratch = *per_cpu_ptr(scratches, i);
+               if (scratch)
+                       vfree(scratch);
+       }
+
+       free_percpu(scratches);
+}
+
+static void **ipcomp_alloc_scratches(void)
+{
+       int i;
+       void **scratches;
+
+       if (ipcomp_scratch_users++)
+               return ipcomp_scratches;
+
+       scratches = alloc_percpu(void *);
+       if (!scratches)
+               return NULL;
+
+       ipcomp_scratches = scratches;
+
+       for_each_cpu(i) {
+               void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
+               if (!scratch)
+                       return NULL;
+               *per_cpu_ptr(scratches, i) = scratch;
+       }
+
+       return scratches;
+}
+
+static void ipcomp_free_tfms(struct crypto_tfm **tfms)
+{
+       struct ipcomp_tfms *pos;
+       int cpu;
+
+       list_for_each_entry(pos, &ipcomp_tfms_list, list) {
+               if (pos->tfms == tfms)
+                       break;
+       }
+
+       BUG_TRAP(pos);
+
+       if (--pos->users)
+               return;
+
+       list_del(&pos->list);
+       kfree(pos);
+
+       if (!tfms)
+               return;
+
+       for_each_cpu(cpu) {
+               struct crypto_tfm *tfm = *per_cpu_ptr(tfms, cpu);
+               if (tfm)
+                       crypto_free_tfm(tfm);
+       }
+       free_percpu(tfms);
+}
+
+static struct crypto_tfm **ipcomp_alloc_tfms(const char *alg_name)
+{
+       struct ipcomp_tfms *pos;
+       struct crypto_tfm **tfms;
+       int cpu;
+
+       /* This can be any valid CPU ID so we don't need locking. */
+       cpu = smp_processor_id();
+
+       list_for_each_entry(pos, &ipcomp_tfms_list, list) {
+               struct crypto_tfm *tfm;
+
+               tfms = pos->tfms;
+               tfm = *per_cpu_ptr(tfms, cpu);
+
+               if (!strcmp(crypto_tfm_alg_name(tfm), alg_name)) {
+                       pos->users++;
+                       return tfms;
+               }
+       }
+
+       pos = kmalloc(sizeof(*pos), GFP_KERNEL);
+       if (!pos)
+               return NULL;
+
+       pos->users = 1;
+       INIT_LIST_HEAD(&pos->list);
+       list_add(&pos->list, &ipcomp_tfms_list);
+
+       pos->tfms = tfms = alloc_percpu(struct crypto_tfm *);
+       if (!tfms)
+               goto error;
+
+       for_each_cpu(cpu) {
+               struct crypto_tfm *tfm = crypto_alloc_tfm(alg_name, 0);
+               if (!tfm)
+                       goto error;
+               *per_cpu_ptr(tfms, cpu) = tfm;
+       }
+
+       return tfms;
+
+error:
+       ipcomp_free_tfms(tfms);
+       return NULL;
+}
+
 static void ipcomp_free_data(struct ipcomp_data *ipcd)
 {
-       if (ipcd->tfm)
-               crypto_free_tfm(ipcd->tfm);
-       if (ipcd->scratch)
-               kfree(ipcd->scratch);   
+       if (ipcd->tfms)
+               ipcomp_free_tfms(ipcd->tfms);
+       ipcomp_free_scratches();
 }
 
 static void ipcomp_destroy(struct xfrm_state *x)
@@ -339,7 +427,10 @@ static void ipcomp_destroy(struct xfrm_state *x)
        struct ipcomp_data *ipcd = x->data;
        if (!ipcd)
                return;
+       xfrm_state_delete_tunnel(x);
+       down(&ipcomp_resource_sem);
        ipcomp_free_data(ipcd);
+       up(&ipcomp_resource_sem);
        kfree(ipcd);
 }
 
@@ -353,31 +444,35 @@ static int ipcomp_init_state(struct xfrm_state *x, void *args)
        if (!x->calg)
                goto out;
 
+       if (x->encap)
+               goto out;
+
        err = -ENOMEM;
        ipcd = kmalloc(sizeof(*ipcd), GFP_KERNEL);
        if (!ipcd)
-               goto error;
+               goto out;
 
        memset(ipcd, 0, sizeof(*ipcd));
-       x->props.header_len = sizeof(struct ip_comp_hdr);
+       x->props.header_len = 0;
        if (x->props.mode)
                x->props.header_len += sizeof(struct iphdr);
 
-       ipcd->scratch = kmalloc(IPCOMP_SCRATCH_SIZE, GFP_KERNEL);
-       if (!ipcd->scratch)
+       down(&ipcomp_resource_sem);
+       if (!ipcomp_alloc_scratches())
                goto error;
-       
-       ipcd->tfm = crypto_alloc_tfm(x->calg->alg_name, 0);
-       if (!ipcd->tfm)
+
+       ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
+       if (!ipcd->tfms)
                goto error;
+       up(&ipcomp_resource_sem);
 
        if (x->props.mode) {
                err = ipcomp_tunnel_attach(x);
                if (err)
-                       goto error;
+                       goto error_tunnel;
        }
 
-       calg_desc = xfrm_calg_get_byname(x->calg->alg_name);
+       calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
        BUG_ON(!calg_desc);
        ipcd->threshold = calg_desc->uinfo.comp.threshold;
        x->data = ipcd;
@@ -385,11 +480,12 @@ static int ipcomp_init_state(struct xfrm_state *x, void *args)
 out:
        return err;
 
+error_tunnel:
+       down(&ipcomp_resource_sem);
 error:
-       if (ipcd) {
-               ipcomp_free_data(ipcd);
-               kfree(ipcd);
-       }
+       ipcomp_free_data(ipcd);
+       up(&ipcomp_resource_sem);
+       kfree(ipcd);
        goto out;
 }
 
@@ -403,7 +499,7 @@ static struct xfrm_type ipcomp_type = {
        .output         = ipcomp_output
 };
 
-static struct inet_protocol ipcomp4_protocol = {
+static struct net_protocol ipcomp4_protocol = {
        .handler        =       xfrm4_rcv,
        .err_handler    =       ipcomp4_err,
        .no_policy      =       1,