X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=net%2Fipv4%2Fipcomp.c;h=a0c28b2b756e0c1e8dab144eba3440f3c67d60fe;hb=16c70f8c1b54b61c3b951b6fb220df250fe09b32;hp=2a5a7dfcf28af5e474d0e25228155efe0fb6815e;hpb=5273a3df6485dc2ad6aa7ddd441b9a21970f003b;p=linux-2.6.git

diff --git a/net/ipv4/ipcomp.c b/net/ipv4/ipcomp.c
index 2a5a7dfcf..a0c28b2b7 100644
--- a/net/ipv4/ipcomp.c
+++ b/net/ipv4/ipcomp.c
@@ -13,29 +13,51 @@
  *   - Compression stats.
  *   - Adaptive compression.
  */
-#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 <linux/mutex.h>
 #include <net/ip.h>
 #include <net/xfrm.h>
 #include <net/icmp.h>
 #include <net/ipcomp.h>
+#include <net/protocol.h>
+
+struct ipcomp_tfms {
+	struct list_head list;
+	struct crypto_tfm **tfms;
+	int users;
+};
+
+static DEFINE_MUTEX(ipcomp_resource_mutex);
+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;
 
@@ -48,45 +70,31 @@ static int ipcomp_decompress(struct xfrm_state *x, struct sk_buff *skb)
 	if (err)
 		goto out;
 		
-	skb_put(skb, dlen - plen);
+	skb->truesize += dlen - plen;
+	__skb_put(skb, dlen - plen);
 	memcpy(skb->data, scratch, dlen);
-	iph = skb->nh.iph;
-	iph->tot_len = htons(dlen + iph->ihl * 4);
 out:	
+	put_cpu();
 	return err;
 }
 
-static int ipcomp_input(struct xfrm_state *x,
-                        struct xfrm_decap_state *decap, struct sk_buff *skb)
+static int ipcomp_input(struct xfrm_state *x, struct sk_buff *skb)
 {
-	u8 nexthdr;
-	int err = 0;
+	int err = -ENOMEM;
 	struct iphdr *iph;
-	union {
-		struct iphdr	iph;
-		char 		buf[60];
-	} tmp_iph;
-
+	struct ip_comp_hdr *ipch;
 
-	if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
-	    skb_linearize(skb, GFP_ATOMIC) != 0) {
-	    	err = -ENOMEM;
+	if (skb_linearize_cow(skb))
 	    	goto out;
-	}
 
 	skb->ip_summed = CHECKSUM_NONE;
 
 	/* Remove ipcomp header and decompress original payload */	
 	iph = skb->nh.iph;
-	memcpy(&tmp_iph, iph, iph->ihl * 4);
-	nexthdr = *(u8 *)skb->data;
-	skb_pull(skb, sizeof(struct ip_comp_hdr));
-	skb->nh.raw += sizeof(struct ip_comp_hdr);
-	memcpy(skb->nh.raw, &tmp_iph, tmp_iph.iph.ihl * 4);
-	iph = skb->nh.iph;
-	iph->tot_len = htons(ntohs(iph->tot_len) - sizeof(struct ip_comp_hdr));
-	iph->protocol = nexthdr;
-	skb->h.raw = skb->data;
+	ipch = (void *)skb->data;
+	iph->protocol = ipch->nexthdr;
+	skb->h.raw = skb->nh.raw + sizeof(*ipch);
+	__skb_pull(skb, sizeof(*ipch));
 	err = ipcomp_decompress(x, skb);
 
 out:	
@@ -98,14 +106,20 @@ 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;
 
@@ -114,132 +128,57 @@ 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)
+static int ipcomp_output(struct xfrm_state *x, 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) {
-		if (x->props.mode) {
-			ipcomp_tunnel_encap(x, skb);
-			iph = skb->nh.iph;
-			iph->protocol = IPPROTO_IPIP;
-			ip_send_check(iph);
-		}
+		/* Don't bother compressing */
 		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;
-	    	goto error;
-	}
+	if (skb_linearize_cow(skb))
+		goto out_ok;
 	
 	err = ipcomp_compress(x, skb);
+	iph = skb->nh.iph;
+
 	if (err) {
-		if (err == -EMSGSIZE) {
-			if (x->props.mode) {
-				iph = skb->nh.iph;
-				iph->protocol = IPPROTO_IPIP;
-				ip_send_check(iph);
-			}
-			goto out_ok;
-		}
-		goto error;
+		goto out_ok;
 	}
 
 	/* 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;
+	return 0;
 
 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;
-
-out_exit:
-	return err;
-error:
-	spin_unlock_bh(&x->lock);
-error_nolock:
-	kfree_skb(skb);
-	goto out_exit;
+	if (x->props.mode)
+		ip_send_check(iph);
+	return 0;
 }
 
 static void ipcomp4_err(struct sk_buff *skb, u32 info)
@@ -253,13 +192,13 @@ static void ipcomp4_err(struct sk_buff *skb, u32 info)
 	    skb->h.icmph->code != ICMP_FRAG_NEEDED)
 		return;
 
-	spi = ntohl(ntohs(ipch->cpi));
+	spi = htonl(ntohs(ipch->cpi));
 	x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
 	                      spi, IPPROTO_COMP, AF_INET);
 	if (!x)
 		return;
-	NETDEBUG(printk(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
-	       spi, NIPQUAD(iph->daddr)));
+	NETDEBUG(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
+		 spi, NIPQUAD(iph->daddr));
 	xfrm_state_put(x);
 }
 
@@ -280,15 +219,10 @@ static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
 	t->props.mode = 1;
 	t->props.saddr.a4 = x->props.saddr.a4;
 	t->props.flags = x->props.flags;
-	
-	t->type = xfrm_get_type(IPPROTO_IPIP, t->props.family);
-	if (t->type == NULL)
-		goto error;
-		
-	if (t->type->init_state(t, NULL))
+
+	if (xfrm_init_state(t))
 		goto error;
 
-	t->km.state = XFRM_STATE_VALID;
 	atomic_set(&t->tunnel_users, 1);
 out:
 	return t;
@@ -301,7 +235,7 @@ error:
 }
 
 /*
- * Must be protected by xfrm_cfg_sem.  State and tunnel user references are
+ * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
  * always incremented on success.
  */
 static int ipcomp_tunnel_attach(struct xfrm_state *x)
@@ -326,12 +260,128 @@ 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_possible_cpu(i)
+		vfree(*per_cpu_ptr(scratches, i));
+
+	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_possible_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_possible_cpu(cpu) {
+		struct crypto_tfm *tfm = *per_cpu_ptr(tfms, cpu);
+		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 = raw_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_possible_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)
@@ -339,11 +389,14 @@ static void ipcomp_destroy(struct xfrm_state *x)
 	struct ipcomp_data *ipcd = x->data;
 	if (!ipcd)
 		return;
+	xfrm_state_delete_tunnel(x);
+	mutex_lock(&ipcomp_resource_mutex);
 	ipcomp_free_data(ipcd);
+	mutex_unlock(&ipcomp_resource_mutex);
 	kfree(ipcd);
 }
 
-static int ipcomp_init_state(struct xfrm_state *x, void *args)
+static int ipcomp_init_state(struct xfrm_state *x)
 {
 	int err;
 	struct ipcomp_data *ipcd;
@@ -353,31 +406,34 @@ 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);
+	ipcd = kzalloc(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)
+	mutex_lock(&ipcomp_resource_mutex);
+	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;
+	mutex_unlock(&ipcomp_resource_mutex);
 
 	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 +441,12 @@ static int ipcomp_init_state(struct xfrm_state *x, void *args)
 out:
 	return err;
 
+error_tunnel:
+	mutex_lock(&ipcomp_resource_mutex);
 error:
-	if (ipcd) {
-		ipcomp_free_data(ipcd);
-		kfree(ipcd);
-	}
+	ipcomp_free_data(ipcd);
+	mutex_unlock(&ipcomp_resource_mutex);
+	kfree(ipcd);
 	goto out;
 }
 
@@ -403,7 +460,7 @@ 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,