tipc: convert name table read-write lock to RCU
authorYing Xue <ying.xue@windriver.com>
Tue, 2 Dec 2014 07:00:30 +0000 (15:00 +0800)
committerDavid S. Miller <davem@davemloft.net>
Tue, 9 Dec 2014 01:39:57 +0000 (20:39 -0500)
Convert tipc name table read-write lock to RCU. After this change,
a new spin lock is used to protect name table on write side while
RCU is applied on read side.

Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Erik Hugne <erik.hugne@ericsson.com>
Reviewed-by: Jon Maloy <jon.maloy@ericsson.com>
Tested-by: Erik Hugne <erik.hugne@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/rculist.h
net/tipc/name_distr.c
net/tipc/name_table.c
net/tipc/name_table.h

index 372ad5e..aa79b3c 100644 (file)
@@ -542,6 +542,15 @@ static inline void hlist_add_behind_rcu(struct hlist_node *n,
             pos = hlist_entry_safe(rcu_dereference_bh((pos)->member.next),\
                        typeof(*(pos)), member))
 
+/**
+ * hlist_for_each_entry_from_rcu - iterate over a hlist continuing from current point
+ * @pos:       the type * to use as a loop cursor.
+ * @member:    the name of the hlist_node within the struct.
+ */
+#define hlist_for_each_entry_from_rcu(pos, member)                     \
+       for (; pos;                                                     \
+            pos = hlist_entry_safe(rcu_dereference((pos)->member.next),\
+                       typeof(*(pos)), member))
 
 #endif /* __KERNEL__ */
 #endif
index ed00929..ba6083d 100644 (file)
@@ -113,8 +113,8 @@ struct sk_buff *tipc_named_publish(struct publication *publ)
        struct sk_buff *buf;
        struct distr_item *item;
 
-       list_add_tail(&publ->local_list,
-                     &tipc_nametbl->publ_list[publ->scope]);
+       list_add_tail_rcu(&publ->local_list,
+                         &tipc_nametbl->publ_list[publ->scope]);
 
        if (publ->scope == TIPC_NODE_SCOPE)
                return NULL;
@@ -208,12 +208,12 @@ void tipc_named_node_up(u32 dnode)
 
        __skb_queue_head_init(&head);
 
-       read_lock_bh(&tipc_nametbl_lock);
+       rcu_read_lock();
        named_distribute(&head, dnode,
                         &tipc_nametbl->publ_list[TIPC_CLUSTER_SCOPE]);
        named_distribute(&head, dnode,
                         &tipc_nametbl->publ_list[TIPC_ZONE_SCOPE]);
-       read_unlock_bh(&tipc_nametbl_lock);
+       rcu_read_unlock();
 
        tipc_link_xmit(&head, dnode, dnode);
 }
@@ -260,12 +260,12 @@ static void tipc_publ_purge(struct publication *publ, u32 addr)
 {
        struct publication *p;
 
-       write_lock_bh(&tipc_nametbl_lock);
+       spin_lock_bh(&tipc_nametbl_lock);
        p = tipc_nametbl_remove_publ(publ->type, publ->lower,
                                     publ->node, publ->ref, publ->key);
        if (p)
                tipc_publ_unsubscribe(p, addr);
-       write_unlock_bh(&tipc_nametbl_lock);
+       spin_unlock_bh(&tipc_nametbl_lock);
 
        if (p != publ) {
                pr_err("Unable to remove publication from failed node\n"
@@ -274,7 +274,7 @@ static void tipc_publ_purge(struct publication *publ, u32 addr)
                       publ->key);
        }
 
-       kfree(p);
+       kfree_rcu(p, rcu);
 }
 
 void tipc_publ_notify(struct list_head *nsub_list, u32 addr)
@@ -311,7 +311,7 @@ static bool tipc_update_nametbl(struct distr_item *i, u32 node, u32 dtype)
                                                ntohl(i->key));
                if (publ) {
                        tipc_publ_unsubscribe(publ, node);
-                       kfree(publ);
+                       kfree_rcu(publ, rcu);
                        return true;
                }
        } else {
@@ -376,14 +376,14 @@ void tipc_named_rcv(struct sk_buff *buf)
        u32 count = msg_data_sz(msg) / ITEM_SIZE;
        u32 node = msg_orignode(msg);
 
-       write_lock_bh(&tipc_nametbl_lock);
+       spin_lock_bh(&tipc_nametbl_lock);
        while (count--) {
                if (!tipc_update_nametbl(item, node, msg_type(msg)))
                        tipc_named_add_backlog(item, msg_type(msg), node);
                item++;
        }
        tipc_named_process_backlog();
-       write_unlock_bh(&tipc_nametbl_lock);
+       spin_unlock_bh(&tipc_nametbl_lock);
        kfree_skb(buf);
 }
 
@@ -399,12 +399,12 @@ void tipc_named_reinit(void)
        struct publication *publ;
        int scope;
 
-       write_lock_bh(&tipc_nametbl_lock);
+       spin_lock_bh(&tipc_nametbl_lock);
 
        for (scope = TIPC_ZONE_SCOPE; scope <= TIPC_NODE_SCOPE; scope++)
-               list_for_each_entry(publ, &tipc_nametbl->publ_list[scope],
-                                   local_list)
+               list_for_each_entry_rcu(publ, &tipc_nametbl->publ_list[scope],
+                                       local_list)
                        publ->node = tipc_own_addr;
 
-       write_unlock_bh(&tipc_nametbl_lock);
+       spin_unlock_bh(&tipc_nametbl_lock);
 }
index 3c2e0c3..aafa684 100644 (file)
@@ -92,6 +92,7 @@ struct sub_seq {
  * @ns_list: links to adjacent name sequences in hash chain
  * @subscriptions: list of subscriptions for this 'type'
  * @lock: spinlock controlling access to publication lists of all sub-sequences
+ * @rcu: RCU callback head used for deferred freeing
  */
 struct name_seq {
        u32 type;
@@ -101,10 +102,11 @@ struct name_seq {
        struct hlist_node ns_list;
        struct list_head subscriptions;
        spinlock_t lock;
+       struct rcu_head rcu;
 };
 
 struct name_table *tipc_nametbl;
-DEFINE_RWLOCK(tipc_nametbl_lock);
+DEFINE_SPINLOCK(tipc_nametbl_lock);
 
 static int hash(int x)
 {
@@ -166,7 +168,7 @@ static struct name_seq *tipc_nameseq_create(u32 type, struct hlist_head *seq_hea
        nseq->alloc = 1;
        INIT_HLIST_NODE(&nseq->ns_list);
        INIT_LIST_HEAD(&nseq->subscriptions);
-       hlist_add_head(&nseq->ns_list, seq_head);
+       hlist_add_head_rcu(&nseq->ns_list, seq_head);
        return nseq;
 }
 
@@ -451,7 +453,7 @@ static struct name_seq *nametbl_find_seq(u32 type)
        struct name_seq *ns;
 
        seq_head = &tipc_nametbl->seq_hlist[hash(type)];
-       hlist_for_each_entry(ns, seq_head, ns_list) {
+       hlist_for_each_entry_rcu(ns, seq_head, ns_list) {
                if (ns->type == type)
                        return ns;
        }
@@ -498,10 +500,10 @@ struct publication *tipc_nametbl_remove_publ(u32 type, u32 lower,
        spin_lock_bh(&seq->lock);
        publ = tipc_nameseq_remove_publ(seq, lower, node, ref, key);
        if (!seq->first_free && list_empty(&seq->subscriptions)) {
-               hlist_del_init(&seq->ns_list);
-               spin_unlock_bh(&seq->lock);
+               hlist_del_init_rcu(&seq->ns_list);
                kfree(seq->sseqs);
-               kfree(seq);
+               spin_unlock_bh(&seq->lock);
+               kfree_rcu(seq, rcu);
                return publ;
        }
        spin_unlock_bh(&seq->lock);
@@ -533,7 +535,7 @@ u32 tipc_nametbl_translate(u32 type, u32 instance, u32 *destnode)
        if (!tipc_in_scope(*destnode, tipc_own_addr))
                return 0;
 
-       read_lock_bh(&tipc_nametbl_lock);
+       rcu_read_lock();
        seq = nametbl_find_seq(type);
        if (unlikely(!seq))
                goto not_found;
@@ -590,7 +592,7 @@ u32 tipc_nametbl_translate(u32 type, u32 instance, u32 *destnode)
 no_match:
        spin_unlock_bh(&seq->lock);
 not_found:
-       read_unlock_bh(&tipc_nametbl_lock);
+       rcu_read_unlock();
        *destnode = node;
        return ref;
 }
@@ -616,7 +618,7 @@ int tipc_nametbl_mc_translate(u32 type, u32 lower, u32 upper, u32 limit,
        struct name_info *info;
        int res = 0;
 
-       read_lock_bh(&tipc_nametbl_lock);
+       rcu_read_lock();
        seq = nametbl_find_seq(type);
        if (!seq)
                goto exit;
@@ -641,7 +643,7 @@ int tipc_nametbl_mc_translate(u32 type, u32 lower, u32 upper, u32 limit,
        }
        spin_unlock_bh(&seq->lock);
 exit:
-       read_unlock_bh(&tipc_nametbl_lock);
+       rcu_read_unlock();
        return res;
 }
 
@@ -654,11 +656,11 @@ struct publication *tipc_nametbl_publish(u32 type, u32 lower, u32 upper,
        struct publication *publ;
        struct sk_buff *buf = NULL;
 
-       write_lock_bh(&tipc_nametbl_lock);
+       spin_lock_bh(&tipc_nametbl_lock);
        if (tipc_nametbl->local_publ_count >= TIPC_MAX_PUBLICATIONS) {
                pr_warn("Publication failed, local publication limit reached (%u)\n",
                        TIPC_MAX_PUBLICATIONS);
-               write_unlock_bh(&tipc_nametbl_lock);
+               spin_unlock_bh(&tipc_nametbl_lock);
                return NULL;
        }
 
@@ -670,7 +672,7 @@ struct publication *tipc_nametbl_publish(u32 type, u32 lower, u32 upper,
                /* Any pending external events? */
                tipc_named_process_backlog();
        }
-       write_unlock_bh(&tipc_nametbl_lock);
+       spin_unlock_bh(&tipc_nametbl_lock);
 
        if (buf)
                named_cluster_distribute(buf);
@@ -685,7 +687,7 @@ int tipc_nametbl_withdraw(u32 type, u32 lower, u32 ref, u32 key)
        struct publication *publ;
        struct sk_buff *skb = NULL;
 
-       write_lock_bh(&tipc_nametbl_lock);
+       spin_lock_bh(&tipc_nametbl_lock);
        publ = tipc_nametbl_remove_publ(type, lower, tipc_own_addr, ref, key);
        if (likely(publ)) {
                tipc_nametbl->local_publ_count--;
@@ -693,13 +695,13 @@ int tipc_nametbl_withdraw(u32 type, u32 lower, u32 ref, u32 key)
                /* Any pending external events? */
                tipc_named_process_backlog();
                list_del_init(&publ->pport_list);
-               kfree(publ);
+               kfree_rcu(publ, rcu);
        } else {
                pr_err("Unable to remove local publication\n"
                       "(type=%u, lower=%u, ref=%u, key=%u)\n",
                       type, lower, ref, key);
        }
-       write_unlock_bh(&tipc_nametbl_lock);
+       spin_unlock_bh(&tipc_nametbl_lock);
 
        if (skb) {
                named_cluster_distribute(skb);
@@ -717,7 +719,7 @@ void tipc_nametbl_subscribe(struct tipc_subscription *s)
        int index = hash(type);
        struct name_seq *seq;
 
-       write_lock_bh(&tipc_nametbl_lock);
+       spin_lock_bh(&tipc_nametbl_lock);
        seq = nametbl_find_seq(type);
        if (!seq)
                seq = tipc_nameseq_create(type,
@@ -730,7 +732,7 @@ void tipc_nametbl_subscribe(struct tipc_subscription *s)
                pr_warn("Failed to create subscription for {%u,%u,%u}\n",
                        s->seq.type, s->seq.lower, s->seq.upper);
        }
-       write_unlock_bh(&tipc_nametbl_lock);
+       spin_unlock_bh(&tipc_nametbl_lock);
 }
 
 /**
@@ -740,24 +742,23 @@ void tipc_nametbl_unsubscribe(struct tipc_subscription *s)
 {
        struct name_seq *seq;
 
-       write_lock_bh(&tipc_nametbl_lock);
+       spin_lock_bh(&tipc_nametbl_lock);
        seq = nametbl_find_seq(s->seq.type);
        if (seq != NULL) {
                spin_lock_bh(&seq->lock);
                list_del_init(&s->nameseq_list);
                if (!seq->first_free && list_empty(&seq->subscriptions)) {
-                       hlist_del_init(&seq->ns_list);
-                       spin_unlock_bh(&seq->lock);
+                       hlist_del_init_rcu(&seq->ns_list);
                        kfree(seq->sseqs);
-                       kfree(seq);
+                       spin_unlock_bh(&seq->lock);
+                       kfree_rcu(seq, rcu);
                } else {
                        spin_unlock_bh(&seq->lock);
                }
        }
-       write_unlock_bh(&tipc_nametbl_lock);
+       spin_unlock_bh(&tipc_nametbl_lock);
 }
 
-
 /**
  * subseq_list - print specified sub-sequence contents into the given buffer
  */
@@ -880,7 +881,7 @@ static int nametbl_list(char *buf, int len, u32 depth_info,
                upbound = ~0;
                for (i = 0; i < TIPC_NAMETBL_SIZE; i++) {
                        seq_head = &tipc_nametbl->seq_hlist[i];
-                       hlist_for_each_entry(seq, seq_head, ns_list) {
+                       hlist_for_each_entry_rcu(seq, seq_head, ns_list) {
                                ret += nameseq_list(seq, buf + ret, len - ret,
                                                   depth, seq->type,
                                                   lowbound, upbound, i);
@@ -896,7 +897,7 @@ static int nametbl_list(char *buf, int len, u32 depth_info,
                ret += nametbl_header(buf + ret, len - ret, depth);
                i = hash(type);
                seq_head = &tipc_nametbl->seq_hlist[i];
-               hlist_for_each_entry(seq, seq_head, ns_list) {
+               hlist_for_each_entry_rcu(seq, seq_head, ns_list) {
                        if (seq->type == type) {
                                ret += nameseq_list(seq, buf + ret, len - ret,
                                                   depth, type,
@@ -928,11 +929,11 @@ struct sk_buff *tipc_nametbl_get(const void *req_tlv_area, int req_tlv_space)
        pb = TLV_DATA(rep_tlv);
        pb_len = ULTRA_STRING_MAX_LEN;
        argv = (struct tipc_name_table_query *)TLV_DATA(req_tlv_area);
-       read_lock_bh(&tipc_nametbl_lock);
+       rcu_read_lock();
        str_len = nametbl_list(pb, pb_len, ntohl(argv->depth),
                               ntohl(argv->type),
                               ntohl(argv->lowbound), ntohl(argv->upbound));
-       read_unlock_bh(&tipc_nametbl_lock);
+       rcu_read_unlock();
        str_len += 1;   /* for "\0" */
        skb_put(buf, TLV_SPACE(str_len));
        TLV_SET(rep_tlv, TIPC_TLV_ULTRA_STRING, NULL, str_len);
@@ -974,13 +975,13 @@ static void tipc_purge_publications(struct name_seq *seq)
        list_for_each_entry_safe(publ, safe, &info->zone_list, zone_list) {
                tipc_nametbl_remove_publ(publ->type, publ->lower, publ->node,
                                         publ->ref, publ->key);
-               kfree(publ);
+               kfree_rcu(publ, rcu);
        }
-       hlist_del_init(&seq->ns_list);
+       hlist_del_init_rcu(&seq->ns_list);
+       kfree(seq->sseqs);
        spin_lock_bh(&seq->lock);
 
-       kfree(seq->sseqs);
-       kfree(seq);
+       kfree_rcu(seq, rcu);
 }
 
 void tipc_nametbl_stop(void)
@@ -988,22 +989,22 @@ void tipc_nametbl_stop(void)
        u32 i;
        struct name_seq *seq;
        struct hlist_head *seq_head;
-       struct hlist_node *safe;
 
        /* Verify name table is empty and purge any lingering
         * publications, then release the name table
         */
-       write_lock_bh(&tipc_nametbl_lock);
+       spin_lock_bh(&tipc_nametbl_lock);
        for (i = 0; i < TIPC_NAMETBL_SIZE; i++) {
                if (hlist_empty(&tipc_nametbl->seq_hlist[i]))
                        continue;
                seq_head = &tipc_nametbl->seq_hlist[i];
-               hlist_for_each_entry_safe(seq, safe, seq_head, ns_list) {
+               hlist_for_each_entry_rcu(seq, seq_head, ns_list) {
                        tipc_purge_publications(seq);
                }
        }
-       write_unlock_bh(&tipc_nametbl_lock);
+       spin_unlock_bh(&tipc_nametbl_lock);
 
+       synchronize_net();
        kfree(tipc_nametbl);
 
 }
@@ -1109,7 +1110,7 @@ static int __tipc_nl_seq_list(struct tipc_nl_msg *msg, u32 *last_type,
                              u32 *last_lower, u32 *last_publ)
 {
        struct hlist_head *seq_head;
-       struct name_seq *seq;
+       struct name_seq *seq = NULL;
        int err;
        int i;
 
@@ -1126,13 +1127,13 @@ static int __tipc_nl_seq_list(struct tipc_nl_msg *msg, u32 *last_type,
                        if (!seq)
                                return -EPIPE;
                } else {
-                       seq = hlist_entry_safe((seq_head)->first,
-                                              struct name_seq, ns_list);
+                       hlist_for_each_entry_rcu(seq, seq_head, ns_list)
+                               break;
                        if (!seq)
                                continue;
                }
 
-               hlist_for_each_entry_from(seq, ns_list) {
+               hlist_for_each_entry_from_rcu(seq, ns_list) {
                        spin_lock_bh(&seq->lock);
                        err = __tipc_nl_subseq_list(msg, seq, last_lower,
                                                    last_publ);
@@ -1165,8 +1166,7 @@ int tipc_nl_name_table_dump(struct sk_buff *skb, struct netlink_callback *cb)
        msg.portid = NETLINK_CB(cb->skb).portid;
        msg.seq = cb->nlh->nlmsg_seq;
 
-       read_lock_bh(&tipc_nametbl_lock);
-
+       rcu_read_lock();
        err = __tipc_nl_seq_list(&msg, &last_type, &last_lower, &last_publ);
        if (!err) {
                done = 1;
@@ -1179,8 +1179,7 @@ int tipc_nl_name_table_dump(struct sk_buff *skb, struct netlink_callback *cb)
                 */
                cb->prev_seq = 1;
        }
-
-       read_unlock_bh(&tipc_nametbl_lock);
+       rcu_read_unlock();
 
        cb->args[0] = last_type;
        cb->args[1] = last_lower;
index c1fd734..5f0dee9 100644 (file)
@@ -62,6 +62,7 @@ struct tipc_port_list;
  * @node_list: adjacent matching name seq publications with >= node scope
  * @cluster_list: adjacent matching name seq publications with >= cluster scope
  * @zone_list: adjacent matching name seq publications with >= zone scope
+ * @rcu: RCU callback head used for deferred freeing
  *
  * Note that the node list, cluster list, and zone list are circular lists.
  */
@@ -79,6 +80,7 @@ struct publication {
        struct list_head node_list;
        struct list_head cluster_list;
        struct list_head zone_list;
+       struct rcu_head rcu;
 };
 
 /**
@@ -93,7 +95,7 @@ struct name_table {
        u32 local_publ_count;
 };
 
-extern rwlock_t tipc_nametbl_lock;
+extern spinlock_t tipc_nametbl_lock;
 extern struct name_table *tipc_nametbl;
 
 int tipc_nl_name_table_dump(struct sk_buff *skb, struct netlink_callback *cb);