Merge branch 'for-linus' of git://gitorious.org/linux-omap-dss2/linux
[safe/jmp/linux-2.6] / net / xfrm / xfrm_policy.c
index 96895ef..4725a54 100644 (file)
 
 #include "xfrm_hash.h"
 
-int sysctl_xfrm_larval_drop __read_mostly = 1;
-
-#ifdef CONFIG_XFRM_STATISTICS
-DEFINE_SNMP_STAT(struct linux_xfrm_mib, xfrm_statistics) __read_mostly;
-EXPORT_SYMBOL(xfrm_statistics);
-#endif
-
 DEFINE_MUTEX(xfrm_cfg_mutex);
 EXPORT_SYMBOL(xfrm_cfg_mutex);
 
@@ -58,6 +51,9 @@ static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
 static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
 static void xfrm_init_pmtu(struct dst_entry *dst);
 
+static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
+                                               int dir);
+
 static inline int
 __xfrm4_selector_match(struct xfrm_selector *sel, struct flowi *fl)
 {
@@ -92,7 +88,7 @@ int xfrm_selector_match(struct xfrm_selector *sel, struct flowi *fl,
        return 0;
 }
 
-static inline struct dst_entry *__xfrm_dst_lookup(int tos,
+static inline struct dst_entry *__xfrm_dst_lookup(struct net *net, int tos,
                                                  xfrm_address_t *saddr,
                                                  xfrm_address_t *daddr,
                                                  int family)
@@ -104,7 +100,7 @@ static inline struct dst_entry *__xfrm_dst_lookup(int tos,
        if (unlikely(afinfo == NULL))
                return ERR_PTR(-EAFNOSUPPORT);
 
-       dst = afinfo->dst_lookup(tos, saddr, daddr);
+       dst = afinfo->dst_lookup(net, tos, saddr, daddr);
 
        xfrm_policy_put_afinfo(afinfo);
 
@@ -116,6 +112,7 @@ static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
                                                xfrm_address_t *prev_daddr,
                                                int family)
 {
+       struct net *net = xs_net(x);
        xfrm_address_t *saddr = &x->props.saddr;
        xfrm_address_t *daddr = &x->id.daddr;
        struct dst_entry *dst;
@@ -129,7 +126,7 @@ static inline struct dst_entry *xfrm_dst_lookup(struct xfrm_state *x, int tos,
                daddr = x->coaddr;
        }
 
-       dst = __xfrm_dst_lookup(tos, saddr, daddr, family);
+       dst = __xfrm_dst_lookup(net, tos, saddr, daddr, family);
 
        if (!IS_ERR(dst)) {
                if (prev_saddr != saddr)
@@ -590,12 +587,8 @@ int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
        xfrm_pol_hold(policy);
        net->xfrm.policy_count[dir]++;
        atomic_inc(&flow_cache_genid);
-       if (delpol) {
-               hlist_del(&delpol->bydst);
-               hlist_del(&delpol->byidx);
-               list_del(&delpol->walk.all);
-               net->xfrm.policy_count[dir]--;
-       }
+       if (delpol)
+               __xfrm_policy_unlink(delpol, dir);
        policy->index = delpol ? delpol->index : xfrm_gen_index(net, dir);
        hlist_add_head(&policy->byidx, net->xfrm.policy_byidx+idx_hash(net, policy->index));
        policy->curlft.add_time = get_seconds();
@@ -667,10 +660,7 @@ struct xfrm_policy *xfrm_policy_bysel_ctx(struct net *net, u8 type, int dir,
                                        write_unlock_bh(&xfrm_policy_lock);
                                        return pol;
                                }
-                               hlist_del(&pol->bydst);
-                               hlist_del(&pol->byidx);
-                               list_del(&pol->walk.all);
-                               net->xfrm.policy_count[dir]--;
+                               __xfrm_policy_unlink(pol, dir);
                        }
                        ret = pol;
                        break;
@@ -711,10 +701,7 @@ struct xfrm_policy *xfrm_policy_byid(struct net *net, u8 type, int dir, u32 id,
                                        write_unlock_bh(&xfrm_policy_lock);
                                        return pol;
                                }
-                               hlist_del(&pol->bydst);
-                               hlist_del(&pol->byidx);
-                               list_del(&pol->walk.all);
-                               net->xfrm.policy_count[dir]--;
+                               __xfrm_policy_unlink(pol, dir);
                        }
                        ret = pol;
                        break;
@@ -795,16 +782,14 @@ int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
        for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
                struct xfrm_policy *pol;
                struct hlist_node *entry;
-               int i, killed;
+               int i;
 
-               killed = 0;
        again1:
                hlist_for_each_entry(pol, entry,
                                     &net->xfrm.policy_inexact[dir], bydst) {
                        if (pol->type != type)
                                continue;
-                       hlist_del(&pol->bydst);
-                       hlist_del(&pol->byidx);
+                       __xfrm_policy_unlink(pol, dir);
                        write_unlock_bh(&xfrm_policy_lock);
 
                        xfrm_audit_policy_delete(pol, 1, audit_info->loginuid,
@@ -812,7 +797,6 @@ int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
                                                 audit_info->secid);
 
                        xfrm_policy_kill(pol);
-                       killed++;
 
                        write_lock_bh(&xfrm_policy_lock);
                        goto again1;
@@ -825,9 +809,7 @@ int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
                                             bydst) {
                                if (pol->type != type)
                                        continue;
-                               hlist_del(&pol->bydst);
-                               hlist_del(&pol->byidx);
-                               list_del(&pol->walk.all);
+                               __xfrm_policy_unlink(pol, dir);
                                write_unlock_bh(&xfrm_policy_lock);
 
                                xfrm_audit_policy_delete(pol, 1,
@@ -835,14 +817,12 @@ int xfrm_policy_flush(struct net *net, u8 type, struct xfrm_audit *audit_info)
                                                         audit_info->sessionid,
                                                         audit_info->secid);
                                xfrm_policy_kill(pol);
-                               killed++;
 
                                write_lock_bh(&xfrm_policy_lock);
                                goto again2;
                        }
                }
 
-               net->xfrm.policy_count[dir] -= killed;
        }
        atomic_inc(&flow_cache_genid);
 out:
@@ -851,7 +831,7 @@ out:
 }
 EXPORT_SYMBOL(xfrm_policy_flush);
 
-int xfrm_policy_walk(struct xfrm_policy_walk *walk,
+int xfrm_policy_walk(struct net *net, struct xfrm_policy_walk *walk,
                     int (*func)(struct xfrm_policy *, int, int, void*),
                     void *data)
 {
@@ -868,10 +848,10 @@ int xfrm_policy_walk(struct xfrm_policy_walk *walk,
 
        write_lock_bh(&xfrm_policy_lock);
        if (list_empty(&walk->walk.all))
-               x = list_first_entry(&init_net.xfrm.policy_all, struct xfrm_policy_walk_entry, all);
+               x = list_first_entry(&net->xfrm.policy_all, struct xfrm_policy_walk_entry, all);
        else
                x = list_entry(&walk->walk.all, struct xfrm_policy_walk_entry, all);
-       list_for_each_entry_from(x, &init_net.xfrm.policy_all, all) {
+       list_for_each_entry_from(x, &net->xfrm.policy_all, all) {
                if (x->dead)
                        continue;
                pol = container_of(x, struct xfrm_policy, walk);
@@ -940,7 +920,8 @@ static int xfrm_policy_match(struct xfrm_policy *pol, struct flowi *fl,
        return ret;
 }
 
-static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
+static struct xfrm_policy *xfrm_policy_lookup_bytype(struct net *net, u8 type,
+                                                    struct flowi *fl,
                                                     u16 family, u8 dir)
 {
        int err;
@@ -956,7 +937,7 @@ static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
                return NULL;
 
        read_lock_bh(&xfrm_policy_lock);
-       chain = policy_hash_direct(&init_net, daddr, saddr, family, dir);
+       chain = policy_hash_direct(net, daddr, saddr, family, dir);
        ret = NULL;
        hlist_for_each_entry(pol, entry, chain, bydst) {
                err = xfrm_policy_match(pol, fl, type, family, dir);
@@ -973,7 +954,7 @@ static struct xfrm_policy *xfrm_policy_lookup_bytype(u8 type, struct flowi *fl,
                        break;
                }
        }
-       chain = &init_net.xfrm.policy_inexact[dir];
+       chain = &net->xfrm.policy_inexact[dir];
        hlist_for_each_entry(pol, entry, chain, bydst) {
                err = xfrm_policy_match(pol, fl, type, family, dir);
                if (err) {
@@ -996,14 +977,14 @@ fail:
        return ret;
 }
 
-static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
-                              void **objp, atomic_t **obj_refp)
+static int xfrm_policy_lookup(struct net *net, struct flowi *fl, u16 family,
+                             u8 dir, void **objp, atomic_t **obj_refp)
 {
        struct xfrm_policy *pol;
        int err = 0;
 
 #ifdef CONFIG_XFRM_SUB_POLICY
-       pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_SUB, fl, family, dir);
+       pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_SUB, fl, family, dir);
        if (IS_ERR(pol)) {
                err = PTR_ERR(pol);
                pol = NULL;
@@ -1011,7 +992,7 @@ static int xfrm_policy_lookup(struct flowi *fl, u16 family, u8 dir,
        if (pol || err)
                goto end;
 #endif
-       pol = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN, fl, family, dir);
+       pol = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN, fl, family, dir);
        if (IS_ERR(pol)) {
                err = PTR_ERR(pol);
                pol = NULL;
@@ -1185,7 +1166,7 @@ int __xfrm_sk_clone_policy(struct sock *sk)
 }
 
 static int
-xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
+xfrm_get_saddr(struct net *net, xfrm_address_t *local, xfrm_address_t *remote,
               unsigned short family)
 {
        int err;
@@ -1193,7 +1174,7 @@ xfrm_get_saddr(xfrm_address_t *local, xfrm_address_t *remote,
 
        if (unlikely(afinfo == NULL))
                return -EINVAL;
-       err = afinfo->get_saddr(local, remote);
+       err = afinfo->get_saddr(net, local, remote);
        xfrm_policy_put_afinfo(afinfo);
        return err;
 }
@@ -1205,6 +1186,7 @@ xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
                      struct xfrm_state **xfrm,
                      unsigned short family)
 {
+       struct net *net = xp_net(policy);
        int nx;
        int i, error;
        xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
@@ -1223,7 +1205,7 @@ xfrm_tmpl_resolve_one(struct xfrm_policy *policy, struct flowi *fl,
                        local = &tmpl->saddr;
                        family = tmpl->encap_family;
                        if (xfrm_addr_any(local, family)) {
-                               error = xfrm_get_saddr(&tmp, remote, family);
+                               error = xfrm_get_saddr(net, &tmp, remote, family);
                                if (error)
                                        goto fail;
                                local = &tmp;
@@ -1463,7 +1445,7 @@ static struct dst_entry *xfrm_bundle_create(struct xfrm_policy *policy,
        if (!dev)
                goto free_dst;
 
-       /* Copy neighbout for reachability confirmation */
+       /* Copy neighbour for reachability confirmation */
        dst0->neighbour = neigh_clone(dst->neighbour);
 
        xfrm_init_path((struct xfrm_dst *)dst0, dst, nfheader_len);
@@ -1537,7 +1519,7 @@ static int stale_bundle(struct dst_entry *dst);
  * At the moment we eat a raw IP route. Mostly to speed up lookups
  * on interfaces with disabled IPsec.
  */
-int __xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
+int __xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
                  struct sock *sk, int flags)
 {
        struct xfrm_policy *policy;
@@ -1567,7 +1549,7 @@ restart:
                policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl);
                err = PTR_ERR(policy);
                if (IS_ERR(policy)) {
-                       XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
+                       XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
                        goto dropdst;
                }
        }
@@ -1575,14 +1557,14 @@ restart:
        if (!policy) {
                /* To accelerate a bit...  */
                if ((dst_orig->flags & DST_NOXFRM) ||
-                   !init_net.xfrm.policy_count[XFRM_POLICY_OUT])
+                   !net->xfrm.policy_count[XFRM_POLICY_OUT])
                        goto nopol;
 
-               policy = flow_cache_lookup(fl, dst_orig->ops->family,
+               policy = flow_cache_lookup(net, fl, dst_orig->ops->family,
                                           dir, xfrm_policy_lookup);
                err = PTR_ERR(policy);
                if (IS_ERR(policy)) {
-                       XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
+                       XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
                        goto dropdst;
                }
        }
@@ -1605,7 +1587,7 @@ restart:
        default:
        case XFRM_POLICY_BLOCK:
                /* Prohibit the flow */
-               XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
+               XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
                err = -EPERM;
                goto error;
 
@@ -1625,7 +1607,7 @@ restart:
                 */
                dst = xfrm_find_bundle(fl, policy, family);
                if (IS_ERR(dst)) {
-                       XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
+                       XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
                        err = PTR_ERR(dst);
                        goto error;
                }
@@ -1635,17 +1617,18 @@ restart:
 
 #ifdef CONFIG_XFRM_SUB_POLICY
                if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
-                       pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
+                       pols[1] = xfrm_policy_lookup_bytype(net,
+                                                           XFRM_POLICY_TYPE_MAIN,
                                                            fl, family,
                                                            XFRM_POLICY_OUT);
                        if (pols[1]) {
                                if (IS_ERR(pols[1])) {
-                                       XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLERROR);
+                                       XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLERROR);
                                        err = PTR_ERR(pols[1]);
                                        goto error;
                                }
                                if (pols[1]->action == XFRM_POLICY_BLOCK) {
-                                       XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLBLOCK);
+                                       XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLBLOCK);
                                        err = -EPERM;
                                        goto error;
                                }
@@ -1672,27 +1655,27 @@ restart:
 
                if (unlikely(nx<0)) {
                        err = nx;
-                       if (err == -EAGAIN && sysctl_xfrm_larval_drop) {
+                       if (err == -EAGAIN && net->xfrm.sysctl_larval_drop) {
                                /* EREMOTE tells the caller to generate
                                 * a one-shot blackhole route.
                                 */
-                               XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
+                               XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
                                xfrm_pol_put(policy);
                                return -EREMOTE;
                        }
                        if (err == -EAGAIN && (flags & XFRM_LOOKUP_WAIT)) {
                                DECLARE_WAITQUEUE(wait, current);
 
-                               add_wait_queue(&init_net.xfrm.km_waitq, &wait);
+                               add_wait_queue(&net->xfrm.km_waitq, &wait);
                                set_current_state(TASK_INTERRUPTIBLE);
                                schedule();
                                set_current_state(TASK_RUNNING);
-                               remove_wait_queue(&init_net.xfrm.km_waitq, &wait);
+                               remove_wait_queue(&net->xfrm.km_waitq, &wait);
 
                                nx = xfrm_tmpl_resolve(pols, npols, fl, xfrm, family);
 
                                if (nx == -EAGAIN && signal_pending(current)) {
-                                       XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
+                                       XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
                                        err = -ERESTART;
                                        goto error;
                                }
@@ -1704,7 +1687,7 @@ restart:
                                err = nx;
                        }
                        if (err < 0) {
-                               XFRM_INC_STATS(LINUX_MIB_XFRMOUTNOSTATES);
+                               XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTNOSTATES);
                                goto error;
                        }
                }
@@ -1717,7 +1700,7 @@ restart:
                dst = xfrm_bundle_create(policy, xfrm, nx, fl, dst_orig);
                err = PTR_ERR(dst);
                if (IS_ERR(dst)) {
-                       XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLEGENERROR);
+                       XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLEGENERROR);
                        goto error;
                }
 
@@ -1738,9 +1721,9 @@ restart:
                        dst_free(dst);
 
                        if (pol_dead)
-                               XFRM_INC_STATS(LINUX_MIB_XFRMOUTPOLDEAD);
+                               XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTPOLDEAD);
                        else
-                               XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
+                               XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
                        err = -EHOSTUNREACH;
                        goto error;
                }
@@ -1752,7 +1735,7 @@ restart:
                if (unlikely(err)) {
                        write_unlock_bh(&policy->lock);
                        dst_free(dst);
-                       XFRM_INC_STATS(LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
+                       XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTBUNDLECHECKERROR);
                        goto error;
                }
 
@@ -1781,10 +1764,10 @@ nopol:
 }
 EXPORT_SYMBOL(__xfrm_lookup);
 
-int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
+int xfrm_lookup(struct net *net, struct dst_entry **dst_p, struct flowi *fl,
                struct sock *sk, int flags)
 {
-       int err = __xfrm_lookup(dst_p, fl, sk, flags);
+       int err = __xfrm_lookup(net, dst_p, fl, sk, flags);
 
        if (err == -EREMOTE) {
                dst_release(*dst_p);
@@ -1892,6 +1875,7 @@ static inline int secpath_has_nontransport(struct sec_path *sp, int k, int *idxp
 int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
                        unsigned short family)
 {
+       struct net *net = dev_net(skb->dev);
        struct xfrm_policy *pol;
        struct xfrm_policy *pols[XFRM_POLICY_TYPE_MAX];
        int npols = 0;
@@ -1907,7 +1891,7 @@ int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
        fl_dir = policy_to_flow_dir(dir);
 
        if (__xfrm_decode_session(skb, &fl, family, reverse) < 0) {
-               XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
+               XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
                return 0;
        }
 
@@ -1920,7 +1904,7 @@ int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
                for (i=skb->sp->len-1; i>=0; i--) {
                        struct xfrm_state *x = skb->sp->xvec[i];
                        if (!xfrm_selector_match(&x->sel, &fl, family)) {
-                               XFRM_INC_STATS(LINUX_MIB_XFRMINSTATEMISMATCH);
+                               XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
                                return 0;
                        }
                }
@@ -1930,24 +1914,24 @@ int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
        if (sk && sk->sk_policy[dir]) {
                pol = xfrm_sk_policy_lookup(sk, dir, &fl);
                if (IS_ERR(pol)) {
-                       XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
+                       XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
                        return 0;
                }
        }
 
        if (!pol)
-               pol = flow_cache_lookup(&fl, family, fl_dir,
+               pol = flow_cache_lookup(net, &fl, family, fl_dir,
                                        xfrm_policy_lookup);
 
        if (IS_ERR(pol)) {
-               XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
+               XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
                return 0;
        }
 
        if (!pol) {
                if (skb->sp && secpath_has_nontransport(skb->sp, 0, &xerr_idx)) {
                        xfrm_secpath_reject(xerr_idx, skb, &fl);
-                       XFRM_INC_STATS(LINUX_MIB_XFRMINNOPOLS);
+                       XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOPOLS);
                        return 0;
                }
                return 1;
@@ -1959,12 +1943,12 @@ int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
        npols ++;
 #ifdef CONFIG_XFRM_SUB_POLICY
        if (pols[0]->type != XFRM_POLICY_TYPE_MAIN) {
-               pols[1] = xfrm_policy_lookup_bytype(XFRM_POLICY_TYPE_MAIN,
+               pols[1] = xfrm_policy_lookup_bytype(net, XFRM_POLICY_TYPE_MAIN,
                                                    &fl, family,
                                                    XFRM_POLICY_IN);
                if (pols[1]) {
                        if (IS_ERR(pols[1])) {
-                               XFRM_INC_STATS(LINUX_MIB_XFRMINPOLERROR);
+                               XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLERROR);
                                return 0;
                        }
                        pols[1]->curlft.use_time = get_seconds();
@@ -1988,11 +1972,11 @@ int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
                for (pi = 0; pi < npols; pi++) {
                        if (pols[pi] != pol &&
                            pols[pi]->action != XFRM_POLICY_ALLOW) {
-                               XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
+                               XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
                                goto reject;
                        }
                        if (ti + pols[pi]->xfrm_nr >= XFRM_MAX_DEPTH) {
-                               XFRM_INC_STATS(LINUX_MIB_XFRMINBUFFERERROR);
+                               XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
                                goto reject_error;
                        }
                        for (i = 0; i < pols[pi]->xfrm_nr; i++)
@@ -2016,20 +2000,20 @@ int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
                                if (k < -1)
                                        /* "-2 - errored_index" returned */
                                        xerr_idx = -(2+k);
-                               XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
+                               XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
                                goto reject;
                        }
                }
 
                if (secpath_has_nontransport(sp, k, &xerr_idx)) {
-                       XFRM_INC_STATS(LINUX_MIB_XFRMINTMPLMISMATCH);
+                       XFRM_INC_STATS(net, LINUX_MIB_XFRMINTMPLMISMATCH);
                        goto reject;
                }
 
                xfrm_pols_put(pols, npols);
                return 1;
        }
-       XFRM_INC_STATS(LINUX_MIB_XFRMINPOLBLOCK);
+       XFRM_INC_STATS(net, LINUX_MIB_XFRMINPOLBLOCK);
 
 reject:
        xfrm_secpath_reject(xerr_idx, skb, &fl);
@@ -2041,15 +2025,22 @@ EXPORT_SYMBOL(__xfrm_policy_check);
 
 int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
 {
+       struct net *net = dev_net(skb->dev);
        struct flowi fl;
+       struct dst_entry *dst;
+       int res;
 
        if (xfrm_decode_session(skb, &fl, family) < 0) {
                /* XXX: we should have something like FWDHDRERROR here. */
-               XFRM_INC_STATS(LINUX_MIB_XFRMINHDRERROR);
+               XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
                return 0;
        }
 
-       return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
+       dst = skb_dst(skb);
+
+       res = xfrm_lookup(net, &dst, &fl, NULL, 0) == 0;
+       skb_dst_set(skb, dst);
+       return res;
 }
 EXPORT_SYMBOL(__xfrm_route_forward);
 
@@ -2133,7 +2124,7 @@ static void prune_one_bundle(struct xfrm_policy *pol, int (*func)(struct dst_ent
        write_unlock(&pol->lock);
 }
 
-static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
+static void xfrm_prune_bundles(struct net *net, int (*func)(struct dst_entry *))
 {
        struct dst_entry *gc_list = NULL;
        int dir;
@@ -2146,11 +2137,11 @@ static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
                int i;
 
                hlist_for_each_entry(pol, entry,
-                                    &init_net.xfrm.policy_inexact[dir], bydst)
+                                    &net->xfrm.policy_inexact[dir], bydst)
                        prune_one_bundle(pol, func, &gc_list);
 
-               table = init_net.xfrm.policy_bydst[dir].table;
-               for (i = init_net.xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
+               table = net->xfrm.policy_bydst[dir].table;
+               for (i = net->xfrm.policy_bydst[dir].hmask; i >= 0; i--) {
                        hlist_for_each_entry(pol, entry, table + i, bydst)
                                prune_one_bundle(pol, func, &gc_list);
                }
@@ -2169,14 +2160,14 @@ static int unused_bundle(struct dst_entry *dst)
        return !atomic_read(&dst->__refcnt);
 }
 
-static void __xfrm_garbage_collect(void)
+static void __xfrm_garbage_collect(struct net *net)
 {
-       xfrm_prune_bundles(unused_bundle);
+       xfrm_prune_bundles(net, unused_bundle);
 }
 
-static int xfrm_flush_bundles(void)
+static int xfrm_flush_bundles(struct net *net)
 {
-       xfrm_prune_bundles(stale_bundle);
+       xfrm_prune_bundles(net, stale_bundle);
        return 0;
 }
 
@@ -2362,12 +2353,9 @@ static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void
 {
        struct net_device *dev = ptr;
 
-       if (!net_eq(dev_net(dev), &init_net))
-               return NOTIFY_DONE;
-
        switch (event) {
        case NETDEV_DOWN:
-               xfrm_flush_bundles();
+               xfrm_flush_bundles(dev_net(dev));
        }
        return NOTIFY_DONE;
 }
@@ -2377,13 +2365,33 @@ static struct notifier_block xfrm_dev_notifier = {
 };
 
 #ifdef CONFIG_XFRM_STATISTICS
-static int __init xfrm_statistics_init(void)
+static int __net_init xfrm_statistics_init(struct net *net)
 {
-       if (snmp_mib_init((void **)xfrm_statistics,
+       int rv;
+
+       if (snmp_mib_init((void **)net->mib.xfrm_statistics,
                          sizeof(struct linux_xfrm_mib)) < 0)
                return -ENOMEM;
+       rv = xfrm_proc_init(net);
+       if (rv < 0)
+               snmp_mib_free((void **)net->mib.xfrm_statistics);
+       return rv;
+}
+
+static void xfrm_statistics_fini(struct net *net)
+{
+       xfrm_proc_fini(net);
+       snmp_mib_free((void **)net->mib.xfrm_statistics);
+}
+#else
+static int __net_init xfrm_statistics_init(struct net *net)
+{
        return 0;
 }
+
+static void xfrm_statistics_fini(struct net *net)
+{
+}
 #endif
 
 static int __net_init xfrm_policy_init(struct net *net)
@@ -2438,9 +2446,23 @@ out_byidx:
 
 static void xfrm_policy_fini(struct net *net)
 {
+       struct xfrm_audit audit_info;
        unsigned int sz;
        int dir;
 
+       flush_work(&net->xfrm.policy_hash_work);
+#ifdef CONFIG_XFRM_SUB_POLICY
+       audit_info.loginuid = -1;
+       audit_info.sessionid = -1;
+       audit_info.secid = 0;
+       xfrm_policy_flush(net, XFRM_POLICY_TYPE_SUB, &audit_info);
+#endif
+       audit_info.loginuid = -1;
+       audit_info.sessionid = -1;
+       audit_info.secid = 0;
+       xfrm_policy_flush(net, XFRM_POLICY_TYPE_MAIN, &audit_info);
+       flush_work(&xfrm_policy_gc_work);
+
        WARN_ON(!list_empty(&net->xfrm.policy_all));
 
        for (dir = 0; dir < XFRM_POLICY_MAX * 2; dir++) {
@@ -2463,24 +2485,36 @@ static int __net_init xfrm_net_init(struct net *net)
 {
        int rv;
 
+       rv = xfrm_statistics_init(net);
+       if (rv < 0)
+               goto out_statistics;
        rv = xfrm_state_init(net);
        if (rv < 0)
                goto out_state;
        rv = xfrm_policy_init(net);
        if (rv < 0)
                goto out_policy;
+       rv = xfrm_sysctl_init(net);
+       if (rv < 0)
+               goto out_sysctl;
        return 0;
 
+out_sysctl:
+       xfrm_policy_fini(net);
 out_policy:
        xfrm_state_fini(net);
 out_state:
+       xfrm_statistics_fini(net);
+out_statistics:
        return rv;
 }
 
 static void __net_exit xfrm_net_exit(struct net *net)
 {
+       xfrm_sysctl_fini(net);
        xfrm_policy_fini(net);
        xfrm_state_fini(net);
+       xfrm_statistics_fini(net);
 }
 
 static struct pernet_operations __net_initdata xfrm_net_ops = {
@@ -2491,13 +2525,7 @@ static struct pernet_operations __net_initdata xfrm_net_ops = {
 void __init xfrm_init(void)
 {
        register_pernet_subsys(&xfrm_net_ops);
-#ifdef CONFIG_XFRM_STATISTICS
-       xfrm_statistics_init();
-#endif
        xfrm_input_init();
-#ifdef CONFIG_XFRM_STATISTICS
-       xfrm_proc_init();
-#endif
 }
 
 #ifdef CONFIG_AUDITSYSCALL