rculfhash: hold rcu read-side lock in resize
authorMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Sun, 11 Sep 2011 05:09:55 +0000 (22:09 -0700)
committerMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Sun, 11 Sep 2011 05:09:55 +0000 (22:09 -0700)
We need to hold the RCU read-side lock in resize to protect against ABA
caused by re-use of nodes after going through the memory allocator
through a call_rcu performed on another CPU.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
rculfhash.c
tests/test_urcu_hash.c
urcu/rculfhash.h

index 6a80049ddd1f783a21e117d63437eec5771ba514..c3a032e20d26a35c09d1f1b3c466432206ecbd32 100644 (file)
@@ -220,6 +220,8 @@ struct cds_lfht {
        void (*cds_lfht_call_rcu)(struct rcu_head *head,
                      void (*func)(struct rcu_head *head));
        void (*cds_lfht_synchronize_rcu)(void);
+       void (*cds_lfht_rcu_read_lock)(void);
+       void (*cds_lfht_rcu_read_unlock)(void);
        unsigned long count;            /* global approximate item count */
        struct ht_items_count *percpu_count;    /* per-cpu item count */
 };
@@ -862,6 +864,11 @@ end:
                return -ENOENT;
 }
 
+/*
+ * Holding RCU read lock to protect _cds_lfht_add against memory
+ * reclaim that could be performed by other call_rcu worker threads (ABA
+ * problem).
+ */
 static
 void init_table(struct cds_lfht *ht, struct rcu_table *t,
                unsigned long first_order, unsigned long len_order)
@@ -879,6 +886,7 @@ void init_table(struct cds_lfht *ht, struct rcu_table *t,
                dbg_printf("init order %lu len: %lu\n", i, len);
                t->tbl[i] = calloc(1, sizeof(struct rcu_level)
                                + (len * sizeof(struct _cds_lfht_node)));
+               ht->cds_lfht_rcu_read_lock();
                for (j = 0; j < len; j++) {
                        struct cds_lfht_node *new_node =
                                (struct cds_lfht_node *) &t->tbl[i]->nodes[j];
@@ -891,6 +899,7 @@ void init_table(struct cds_lfht *ht, struct rcu_table *t,
                        if (CMM_LOAD_SHARED(ht->in_progress_destroy))
                                break;
                }
+               ht->cds_lfht_rcu_read_unlock();
                /* Update table size */
                t->size = !i ? 1 : (1UL << i);
                dbg_printf("init new size: %lu\n", t->size);
@@ -901,6 +910,11 @@ void init_table(struct cds_lfht *ht, struct rcu_table *t,
        t->resize_initiated = 0;
 }
 
+/*
+ * Holding RCU read lock to protect _cds_lfht_remove against memory
+ * reclaim that could be performed by other call_rcu worker threads (ABA
+ * problem).
+ */
 static
 void fini_table(struct cds_lfht *ht, struct rcu_table *t,
                unsigned long first_order, unsigned long len_order)
@@ -924,6 +938,7 @@ void fini_table(struct cds_lfht *ht, struct rcu_table *t,
                 */
                t->size = 1UL << (i - 1);
                /* Unlink */
+               ht->cds_lfht_rcu_read_lock();
                for (j = 0; j < len; j++) {
                        struct cds_lfht_node *fini_node =
                                (struct cds_lfht_node *) &t->tbl[i]->nodes[j];
@@ -936,6 +951,7 @@ void fini_table(struct cds_lfht *ht, struct rcu_table *t,
                        if (CMM_LOAD_SHARED(ht->in_progress_destroy))
                                break;
                }
+               ht->cds_lfht_rcu_read_unlock();
                ht->cds_lfht_call_rcu(&t->tbl[i]->head, cds_lfht_free_level);
                dbg_printf("fini new size: %lu\n", t->size);
                if (CMM_LOAD_SHARED(ht->in_progress_destroy))
@@ -952,7 +968,9 @@ struct cds_lfht *cds_lfht_new(cds_lfht_hash_fct hash_fct,
                        int flags,
                        void (*cds_lfht_call_rcu)(struct rcu_head *head,
                                        void (*func)(struct rcu_head *head)),
-                       void (*cds_lfht_synchronize_rcu)(void))
+                       void (*cds_lfht_synchronize_rcu)(void),
+                       void (*cds_lfht_rcu_read_lock)(void),
+                       void (*cds_lfht_rcu_read_unlock)(void))
 {
        struct cds_lfht *ht;
        unsigned long order;
@@ -966,6 +984,8 @@ struct cds_lfht *cds_lfht_new(cds_lfht_hash_fct hash_fct,
        ht->hash_seed = hash_seed;
        ht->cds_lfht_call_rcu = cds_lfht_call_rcu;
        ht->cds_lfht_synchronize_rcu = cds_lfht_synchronize_rcu;
+       ht->cds_lfht_rcu_read_lock = cds_lfht_rcu_read_lock;
+       ht->cds_lfht_rcu_read_unlock = cds_lfht_rcu_read_unlock;
        ht->in_progress_resize = 0;
        ht->percpu_count = alloc_per_cpu_items_count();
        /* this mutex should not nest in read-side C.S. */
index 376cd9721094cfac251e37883d85149468798735..414948c8b501251ca1d92cbed6565446e5325cd2 100644 (file)
@@ -719,7 +719,8 @@ int main(int argc, char **argv)
        test_ht = cds_lfht_new(test_hash, test_compare, 0x42UL,
                        init_hash_size,
                        opt_auto_resize ? CDS_LFHT_AUTO_RESIZE : 0,
-                       call_rcu, synchronize_rcu);
+                       call_rcu, synchronize_rcu, rcu_read_lock,
+                       rcu_read_unlock);
        ret = populate_hash();
        assert(!ret);
         err = create_all_cpu_call_rcu_data(0);
index 385a43b56a08c09df194c3c556362ade78a1c6fe..8f501ea29b7b1e549c9b2d87f9ecf2e769251af0 100644 (file)
@@ -92,7 +92,9 @@ struct cds_lfht *cds_lfht_new(cds_lfht_hash_fct hash_fct,
                        int flags,
                        void (*cds_lfht_call_rcu)(struct rcu_head *head,
                                void (*func)(struct rcu_head *head)),
-                       void (*cds_lfht_synchronize_rcu)(void));
+                       void (*cds_lfht_synchronize_rcu)(void),
+                       void (*cds_lfht_rcu_read_lock)(void),
+                       void (*cds_lfht_rcu_read_unlock)(void));
 
 /*
  * cds_lfht_destroy - destroy a hash table.
This page took 0.027779 seconds and 4 git commands to generate.