datapath: Clarify meaning of n_buckets argument to tbl_create().
[sliver-openvswitch.git] / datapath / datapath.c
index 7a70163..a2d395c 100644 (file)
@@ -67,7 +67,7 @@ EXPORT_SYMBOL(dp_ioctl_hook);
  * It is safe to access the datapath and vport structures with just
  * dp_mutex.
  */
-static struct datapath *dps[ODP_MAX];
+static struct datapath __rcu *dps[ODP_MAX];
 static DEFINE_MUTEX(dp_mutex);
 
 static int new_vport(struct datapath *, struct odp_port *, int port_no);
@@ -77,7 +77,8 @@ struct datapath *get_dp(int dp_idx)
 {
        if (dp_idx < 0 || dp_idx >= ODP_MAX)
                return NULL;
-       return rcu_dereference(dps[dp_idx]);
+       return rcu_dereference_check(dps[dp_idx], rcu_read_lock_held() ||
+                                        lockdep_is_held(&dp_mutex));
 }
 EXPORT_SYMBOL_GPL(get_dp);
 
@@ -93,10 +94,15 @@ static struct datapath *get_dp_locked(int dp_idx)
        return dp;
 }
 
+static struct tbl *get_table_protected(struct datapath *dp)
+{
+       return rcu_dereference_protected(dp->table, lockdep_is_held(&dp->mutex));
+}
+
 /* Must be called with rcu_read_lock or RTNL lock. */
 const char *dp_name(const struct datapath *dp)
 {
-       return vport_get_name(dp->ports[ODPP_LOCAL]);
+       return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
 }
 
 static inline size_t br_nlmsg_size(void)
@@ -139,7 +145,8 @@ static int dp_fill_ifinfo(struct sk_buff *skb,
        hdr->ifi_change = 0;
 
        NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
-       NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]));
+       NLA_PUT_U32(skb, IFLA_MASTER,
+               vport_get_ifindex(rtnl_dereference(dp->ports[ODPP_LOCAL])));
        NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 #ifdef IFLA_OPERSTATE
        NLA_PUT_U8(skb, IFLA_OPERSTATE,
@@ -245,7 +252,7 @@ static int create_dp(int dp_idx, const char __user *devnamep)
 
        /* Allocate table. */
        err = -ENOMEM;
-       rcu_assign_pointer(dp->table, tbl_create(0));
+       rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
        if (!dp->table)
                goto err_free_dp;
 
@@ -263,8 +270,10 @@ static int create_dp(int dp_idx, const char __user *devnamep)
 
        dp->drop_frags = 0;
        dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
-       if (!dp->stats_percpu)
+       if (!dp->stats_percpu) {
+               err = -ENOMEM;
                goto err_destroy_local_port;
+       }
 
        rcu_assign_pointer(dps[dp_idx], dp);
        dp_sysfs_add_dp(dp);
@@ -493,7 +502,8 @@ void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
                flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
                                        flow_hash(&key), flow_cmp);
                if (unlikely(!flow_node)) {
-                       dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id);
+                       dp_output_control(dp, skb, _ODPL_MISS_NR,
+                                        (__force u64)OVS_CB(skb)->tun_id);
                        stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
                        goto out;
                }
@@ -501,6 +511,7 @@ void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
                OVS_CB(skb)->flow = flow_cast(flow_node);
        }
 
+       stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
        flow_used(OVS_CB(skb)->flow, skb);
 
        acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
@@ -511,13 +522,13 @@ void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
                loop->looping = true;
        if (unlikely(loop->looping)) {
                loop_suppress(dp, acts);
+               kfree_skb(skb);
                goto out_loop;
        }
 
        /* Execute actions. */
        execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
                        acts->actions_len);
-       stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
 
        /* Check whether sub-actions looped too much. */
        if (unlikely(loop->looping))
@@ -544,7 +555,7 @@ out:
 /* Append each packet in 'skb' list to 'queue'.  There will be only one packet
  * unless we broke up a GSO packet. */
 static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
-                                int queue_no, u32 arg)
+                                int queue_no, u64 arg)
 {
        struct sk_buff *nskb;
        int port_no;
@@ -569,7 +580,6 @@ static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue
                header->type = queue_no;
                header->length = skb->len;
                header->port = port_no;
-               header->reserved = 0;
                header->arg = arg;
                skb_queue_tail(queue, skb);
 
@@ -587,7 +597,7 @@ err_kfree_skbs:
 }
 
 int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
-                     u32 arg)
+                     u64 arg)
 {
        struct dp_stats_percpu *stats;
        struct sk_buff_head *queue;
@@ -610,16 +620,12 @@ int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
         * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
        if (skb_is_gso(skb)) {
                struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
-               if (nskb) {
-                       kfree_skb(skb);
-                       skb = nskb;
-                       if (unlikely(IS_ERR(skb))) {
-                               err = PTR_ERR(skb);
-                               goto err;
-                       }
-               } else {
-                       /* XXX This case might not be possible.  It's hard to
-                        * tell from the skb_gso_segment() code and comment. */
+               
+               kfree_skb(skb);
+               skb = nskb;
+               if (IS_ERR(skb)) {
+                       err = PTR_ERR(skb);
+                       goto err;
                }
        }
 
@@ -644,10 +650,10 @@ err:
 
 static int flush_flows(struct datapath *dp)
 {
-       struct tbl *old_table = rcu_dereference(dp->table);
+       struct tbl *old_table = get_table_protected(dp);
        struct tbl *new_table;
 
-       new_table = tbl_create(0);
+       new_table = tbl_create(TBL_MIN_BUCKETS);
        if (!new_table)
                return -ENOMEM;
 
@@ -666,7 +672,7 @@ static int validate_actions(const struct nlattr *actions, u32 actions_len)
         nla_for_each_attr(a, actions, actions_len, rem) {
                 static const u32 action_lens[ODPAT_MAX + 1] = {
                         [ODPAT_OUTPUT] = 4,
-                        [ODPAT_CONTROLLER] = 4,
+                        [ODPAT_CONTROLLER] = 8,
                         [ODPAT_SET_DL_TCI] = 2,
                         [ODPAT_STRIP_VLAN] = 0,
                         [ODPAT_SET_DL_SRC] = ETH_ALEN,
@@ -676,7 +682,7 @@ static int validate_actions(const struct nlattr *actions, u32 actions_len)
                         [ODPAT_SET_NW_TOS] = 1,
                         [ODPAT_SET_TP_SRC] = 2,
                         [ODPAT_SET_TP_DST] = 2,
-                        [ODPAT_SET_TUNNEL] = 4,
+                        [ODPAT_SET_TUNNEL] = 8,
                         [ODPAT_SET_PRIORITY] = 4,
                         [ODPAT_POP_PRIORITY] = 0,
                         [ODPAT_DROP_SPOOFED_ARP] = 0,
@@ -741,7 +747,9 @@ static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
                goto error;
 
        error = -EFAULT;
-       if (copy_from_user(actions->actions, flow->actions, flow->actions_len))
+       if (copy_from_user(actions->actions,
+                          (struct nlattr __user __force *)flow->actions,
+                          flow->actions_len))
                goto error_free_actions;
        error = validate_actions(actions->actions, actions->actions_len);
        if (error)
@@ -789,7 +797,7 @@ static void clear_stats(struct sw_flow *flow)
 
 static int expand_table(struct datapath *dp)
 {
-       struct tbl *old_table = rcu_dereference(dp->table);
+       struct tbl *old_table = get_table_protected(dp);
        struct tbl *new_table;
 
        new_table = tbl_expand(old_table);
@@ -809,9 +817,11 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
        struct sw_flow *flow;
        struct tbl *table;
        int error;
+       u32 hash;
 
-       table = rcu_dereference(dp->table);
-       flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
+       hash = flow_hash(&uf->flow.key);
+       table = get_table_protected(dp);
+       flow_node = tbl_lookup(table, &uf->flow.key, hash, flow_cmp);
        if (!flow_node) {
                /* No such flow. */
                struct sw_flow_actions *acts;
@@ -825,7 +835,7 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
                        error = expand_table(dp);
                        if (error)
                                goto error;
-                       table = rcu_dereference(dp->table);
+                       table = get_table_protected(dp);
                }
 
                /* Allocate flow. */
@@ -845,7 +855,7 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
                rcu_assign_pointer(flow->sf_acts, acts);
 
                /* Put flow in bucket. */
-               error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
+               error = tbl_insert(table, &flow->tbl_node, hash);
                if (error)
                        goto error_free_flow_acts;
 
@@ -866,7 +876,9 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
                error = PTR_ERR(new_acts);
                if (IS_ERR(new_acts))
                        goto error;
-               old_acts = rcu_dereference(flow->sf_acts);
+
+               old_acts = rcu_dereference_protected(flow->sf_acts,
+                                                    lockdep_is_held(&dp->mutex));
                if (old_acts->actions_len != new_acts->actions_len ||
                    memcmp(old_acts->actions, new_acts->actions,
                           old_acts->actions_len)) {
@@ -915,7 +927,8 @@ static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
        return 0;
 }
 
-static int do_answer_query(struct sw_flow *flow, u32 query_flags,
+static int do_answer_query(struct datapath *dp, struct sw_flow *flow,
+                          u32 query_flags,
                           struct odp_flow_stats __user *ustats,
                           struct nlattr __user *actions,
                           u32 __user *actions_lenp)
@@ -938,7 +951,8 @@ static int do_answer_query(struct sw_flow *flow, u32 query_flags,
        if (!actions_len)
                return 0;
 
-       sf_acts = rcu_dereference(flow->sf_acts);
+       sf_acts = rcu_dereference_protected(flow->sf_acts,
+                                           lockdep_is_held(&dp->mutex));
        if (put_user(sf_acts->actions_len, actions_lenp) ||
            (actions && copy_to_user(actions, sf_acts->actions,
                                     min(sf_acts->actions_len, actions_len))))
@@ -947,21 +961,21 @@ static int do_answer_query(struct sw_flow *flow, u32 query_flags,
        return 0;
 }
 
-static int answer_query(struct sw_flow *flow, u32 query_flags,
-                       struct odp_flow __user *ufp)
+static int answer_query(struct datapath *dp, struct sw_flow *flow,
+                       u32 query_flags, struct odp_flow __user *ufp)
 {
-       struct nlattr *actions;
+       struct nlattr __user *actions;
 
-       if (get_user(actions, &ufp->actions))
+       if (get_user(actions, (struct nlattr __user * __user *)&ufp->actions))
                return -EFAULT;
 
-       return do_answer_query(flow, query_flags, 
+       return do_answer_query(dp, flow, query_flags, 
                               &ufp->stats, actions, &ufp->actions_len);
 }
 
 static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
 {
-       struct tbl *table = rcu_dereference(dp->table);
+       struct tbl *table = get_table_protected(dp);
        struct tbl_node *flow_node;
        int error;
 
@@ -993,18 +1007,18 @@ static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
        if (IS_ERR(flow))
                return PTR_ERR(flow);
 
-       error = answer_query(flow, 0, ufp);
+       error = answer_query(dp, flow, 0, ufp);
        flow_deferred_free(flow);
        return error;
 }
 
 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
 {
-       struct tbl *table = rcu_dereference(dp->table);
+       struct tbl *table = get_table_protected(dp);
        u32 i;
 
        for (i = 0; i < flowvec->n_flows; i++) {
-               struct odp_flow __user *ufp = &flowvec->flows[i];
+               struct odp_flow __user *ufp = (struct odp_flow __user __force *)&flowvec->flows[i];
                struct odp_flow uf;
                struct tbl_node *flow_node;
                int error;
@@ -1016,7 +1030,7 @@ static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec
                if (!flow_node)
                        error = put_user(ENOENT, &ufp->stats.error);
                else
-                       error = answer_query(flow_cast(flow_node), uf.flags, ufp);
+                       error = answer_query(dp, flow_cast(flow_node), uf.flags, ufp);
                if (error)
                        return -EFAULT;
        }
@@ -1024,6 +1038,7 @@ static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec
 }
 
 struct list_flows_cbdata {
+       struct datapath *dp;
        struct odp_flow __user *uflows;
        u32 n_flows;
        u32 listed_flows;
@@ -1038,7 +1053,7 @@ static int list_flow(struct tbl_node *node, void *cbdata_)
 
        if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
                return -EFAULT;
-       error = answer_query(flow, 0, ufp);
+       error = answer_query(cbdata->dp, flow, 0, ufp);
        if (error)
                return error;
 
@@ -1055,11 +1070,12 @@ static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
        if (!flowvec->n_flows)
                return 0;
 
-       cbdata.uflows = flowvec->flows;
+       cbdata.dp = dp;
+       cbdata.uflows = (struct odp_flow __user __force*)flowvec->flows;
        cbdata.n_flows = flowvec->n_flows;
        cbdata.listed_flows = 0;
 
-       error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata);
+       error = tbl_foreach(get_table_protected(dp), list_flow, &cbdata);
        return error ? error : cbdata.listed_flows;
 }
 
@@ -1104,7 +1120,8 @@ static int do_execute(struct datapath *dp, const struct odp_execute *execute)
        }
 
        err = -EFAULT;
-       if (copy_from_user(actions->actions, execute->actions, execute->actions_len))
+       if (copy_from_user(actions->actions,
+           (struct nlattr __user __force *)execute->actions, execute->actions_len))
                goto error_free_actions;
 
        err = validate_actions(actions->actions, execute->actions_len);
@@ -1117,7 +1134,8 @@ static int do_execute(struct datapath *dp, const struct odp_execute *execute)
                goto error_free_actions;
 
        err = -EFAULT;
-       if (copy_from_user(skb_put(skb, execute->length), execute->data,
+       if (copy_from_user(skb_put(skb, execute->length),
+                          (const void __user __force *)execute->data,
                           execute->length))
                goto error_free_skb;
 
@@ -1163,7 +1181,7 @@ static int execute_packet(struct datapath *dp, const struct odp_execute __user *
 
 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
 {
-       struct tbl *table = rcu_dereference(dp->table);
+       struct tbl *table = get_table_protected(dp);
        struct odp_stats stats;
        int i;
 
@@ -1321,7 +1339,8 @@ static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
        if (copy_from_user(&pv, upv, sizeof pv))
                return -EFAULT;
 
-       retval = do_list_ports(dp, pv.ports, pv.n_ports);
+       retval = do_list_ports(dp, (struct odp_port __user __force *)pv.ports,
+                              pv.n_ports);
        if (retval < 0)
                return retval;
 
@@ -1525,7 +1544,7 @@ static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow _
            __get_user(flow->flags, &compat->flags))
                return -EFAULT;
 
-       flow->actions = compat_ptr(actions);
+       flow->actions = (struct nlattr __force *)compat_ptr(actions);
        return 0;
 }
 
@@ -1550,7 +1569,8 @@ static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __use
        return 0;
 }
 
-static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
+static int compat_answer_query(struct datapath *dp, struct sw_flow *flow,
+                              u32 query_flags,
                               struct compat_odp_flow __user *ufp)
 {
        compat_uptr_t actions;
@@ -1558,7 +1578,7 @@ static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
        if (get_user(actions, &ufp->actions))
                return -EFAULT;
 
-       return do_answer_query(flow, query_flags, &ufp->stats,
+       return do_answer_query(dp, flow, query_flags, &ufp->stats,
                               compat_ptr(actions), &ufp->actions_len);
 }
 
@@ -1575,14 +1595,16 @@ static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *u
        if (IS_ERR(flow))
                return PTR_ERR(flow);
 
-       error = compat_answer_query(flow, 0, ufp);
+       error = compat_answer_query(dp, flow, 0, ufp);
        flow_deferred_free(flow);
        return error;
 }
 
-static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
+static int compat_query_flows(struct datapath *dp,
+                             struct compat_odp_flow __user *flows,
+                             u32 n_flows)
 {
-       struct tbl *table = rcu_dereference(dp->table);
+       struct tbl *table = get_table_protected(dp);
        u32 i;
 
        for (i = 0; i < n_flows; i++) {
@@ -1598,7 +1620,8 @@ static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows
                if (!flow_node)
                        error = put_user(ENOENT, &ufp->stats.error);
                else
-                       error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp);
+                       error = compat_answer_query(dp, flow_cast(flow_node),
+                                                   uf.flags, ufp);
                if (error)
                        return -EFAULT;
        }
@@ -1606,6 +1629,7 @@ static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows
 }
 
 struct compat_list_flows_cbdata {
+       struct datapath *dp;
        struct compat_odp_flow __user *uflows;
        u32 n_flows;
        u32 listed_flows;
@@ -1620,7 +1644,7 @@ static int compat_list_flow(struct tbl_node *node, void *cbdata_)
 
        if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
                return -EFAULT;
-       error = compat_answer_query(flow, 0, ufp);
+       error = compat_answer_query(cbdata->dp, flow, 0, ufp);
        if (error)
                return error;
 
@@ -1629,7 +1653,8 @@ static int compat_list_flow(struct tbl_node *node, void *cbdata_)
        return 0;
 }
 
-static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
+static int compat_list_flows(struct datapath *dp,
+                            struct compat_odp_flow __user *flows, u32 n_flows)
 {
        struct compat_list_flows_cbdata cbdata;
        int error;
@@ -1637,17 +1662,18 @@ static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows,
        if (!n_flows)
                return 0;
 
+       cbdata.dp = dp;
        cbdata.uflows = flows;
        cbdata.n_flows = n_flows;
        cbdata.listed_flows = 0;
 
-       error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
+       error = tbl_foreach(get_table_protected(dp), compat_list_flow, &cbdata);
        return error ? error : cbdata.listed_flows;
 }
 
 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
                                int (*function)(struct datapath *,
-                                               struct compat_odp_flow *,
+                                               struct compat_odp_flow __user *,
                                                u32 n_flows))
 {
        struct compat_odp_flowvec __user *uflowvec;
@@ -1687,8 +1713,8 @@ static int compat_execute(struct datapath *dp, const struct compat_odp_execute _
            __get_user(execute.length, &uexecute->length))
                return -EFAULT;
 
-       execute.actions = compat_ptr(actions);
-       execute.data = compat_ptr(data);
+       execute.actions = (struct nlattr __force *)compat_ptr(actions);
+       execute.data = (const void __force *)compat_ptr(data);
 
        return do_execute(dp, &execute);
 }
@@ -1914,6 +1940,8 @@ success:
                        u16 csum_start, csum_offset;
 
                        get_skb_csum_pointers(skb, &csum_start, &csum_offset);
+                       csum_start -= skb_headroom(skb);
+
                        BUG_ON(csum_start >= skb_headlen(skb));
                        retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
                                                            copy_bytes - csum_start, &csum);
@@ -1923,7 +1951,8 @@ success:
                                copy_bytes = csum_start;
                                csump = (__sum16 __user *)(buf + csum_start + csum_offset);
 
-                               BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes);
+                               BUG_ON((char __user *)csump + sizeof(__sum16) >
+                                       buf + nbytes);
                                put_user(csum_fold(csum), csump);
                        }
                } else
@@ -1931,7 +1960,7 @@ success:
        }
 
        if (!retval) {
-               struct iovec __user iov;
+               struct iovec iov;
 
                iov.iov_base = buf;
                iov.iov_len = copy_bytes;