+#if CFS_HASH_DEBUG_LEVEL >= CFS_HASH_DEBUG_1
+static unsigned int warn_on_depth = 8;
+module_param(warn_on_depth, uint, 0644);
+MODULE_PARM_DESC(warn_on_depth, "warning when hash depth is high.");
+#endif
+
+struct cfs_wi_sched *cfs_sched_rehash;
+
+static inline void
+cfs_hash_nl_lock(union cfs_hash_lock *lock, int exclusive) {}
+
+static inline void
+cfs_hash_nl_unlock(union cfs_hash_lock *lock, int exclusive) {}
+
+static inline void
+cfs_hash_spin_lock(union cfs_hash_lock *lock, int exclusive)
+ __acquires(&lock->spin)
+{
+ spin_lock(&lock->spin);
+}
+
+static inline void
+cfs_hash_spin_unlock(union cfs_hash_lock *lock, int exclusive)
+ __releases(&lock->spin)
+{
+ spin_unlock(&lock->spin);
+}
+
+static inline void
+cfs_hash_rw_lock(union cfs_hash_lock *lock, int exclusive)
+ __acquires(&lock->rw)
+{
+ if (!exclusive)
+ read_lock(&lock->rw);
+ else
+ write_lock(&lock->rw);
+}
+
+static inline void
+cfs_hash_rw_unlock(union cfs_hash_lock *lock, int exclusive)
+ __releases(&lock->rw)
+{
+ if (!exclusive)
+ read_unlock(&lock->rw);
+ else
+ write_unlock(&lock->rw);
+}
+
+/** No lock hash */
+static struct cfs_hash_lock_ops cfs_hash_nl_lops = {
+ .hs_lock = cfs_hash_nl_lock,
+ .hs_unlock = cfs_hash_nl_unlock,
+ .hs_bkt_lock = cfs_hash_nl_lock,
+ .hs_bkt_unlock = cfs_hash_nl_unlock,
+};
+
+/** no bucket lock, one spinlock to protect everything */
+static struct cfs_hash_lock_ops cfs_hash_nbl_lops = {
+ .hs_lock = cfs_hash_spin_lock,
+ .hs_unlock = cfs_hash_spin_unlock,
+ .hs_bkt_lock = cfs_hash_nl_lock,
+ .hs_bkt_unlock = cfs_hash_nl_unlock,
+};
+
+/** spin bucket lock, rehash is enabled */
+static struct cfs_hash_lock_ops cfs_hash_bkt_spin_lops = {
+ .hs_lock = cfs_hash_rw_lock,
+ .hs_unlock = cfs_hash_rw_unlock,
+ .hs_bkt_lock = cfs_hash_spin_lock,
+ .hs_bkt_unlock = cfs_hash_spin_unlock,
+};
+
+/** rw bucket lock, rehash is enabled */
+static struct cfs_hash_lock_ops cfs_hash_bkt_rw_lops = {
+ .hs_lock = cfs_hash_rw_lock,
+ .hs_unlock = cfs_hash_rw_unlock,
+ .hs_bkt_lock = cfs_hash_rw_lock,
+ .hs_bkt_unlock = cfs_hash_rw_unlock,
+};
+
+/** spin bucket lock, rehash is disabled */
+static struct cfs_hash_lock_ops cfs_hash_nr_bkt_spin_lops = {
+ .hs_lock = cfs_hash_nl_lock,
+ .hs_unlock = cfs_hash_nl_unlock,
+ .hs_bkt_lock = cfs_hash_spin_lock,
+ .hs_bkt_unlock = cfs_hash_spin_unlock,
+};
+
+/** rw bucket lock, rehash is disabled */
+static struct cfs_hash_lock_ops cfs_hash_nr_bkt_rw_lops = {
+ .hs_lock = cfs_hash_nl_lock,
+ .hs_unlock = cfs_hash_nl_unlock,
+ .hs_bkt_lock = cfs_hash_rw_lock,
+ .hs_bkt_unlock = cfs_hash_rw_unlock,
+};
+
+static void
+cfs_hash_lock_setup(struct cfs_hash *hs)
+{
+ if (cfs_hash_with_no_lock(hs)) {
+ hs->hs_lops = &cfs_hash_nl_lops;
+
+ } else if (cfs_hash_with_no_bktlock(hs)) {
+ hs->hs_lops = &cfs_hash_nbl_lops;
+ spin_lock_init(&hs->hs_lock.spin);
+
+ } else if (cfs_hash_with_rehash(hs)) {
+ rwlock_init(&hs->hs_lock.rw);
+
+ if (cfs_hash_with_rw_bktlock(hs))
+ hs->hs_lops = &cfs_hash_bkt_rw_lops;
+ else if (cfs_hash_with_spin_bktlock(hs))
+ hs->hs_lops = &cfs_hash_bkt_spin_lops;
+ else
+ LBUG();
+ } else {
+ if (cfs_hash_with_rw_bktlock(hs))
+ hs->hs_lops = &cfs_hash_nr_bkt_rw_lops;
+ else if (cfs_hash_with_spin_bktlock(hs))
+ hs->hs_lops = &cfs_hash_nr_bkt_spin_lops;
+ else
+ LBUG();
+ }
+}
+
+/**
+ * Simple hash head without depth tracking
+ * new element is always added to head of hlist
+ */
+struct cfs_hash_head {
+ struct hlist_head hh_head; /**< entries list */
+};
+
+static int
+cfs_hash_hh_hhead_size(struct cfs_hash *hs)
+{
+ return sizeof(struct cfs_hash_head);
+}
+
+static struct hlist_head *
+cfs_hash_hh_hhead(struct cfs_hash *hs, struct cfs_hash_bd *bd)
+{
+ struct cfs_hash_head *head;
+
+ head = (struct cfs_hash_head *)&bd->bd_bucket->hsb_head[0];
+ return &head[bd->bd_offset].hh_head;
+}
+
+static int
+cfs_hash_hh_hnode_add(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnode)
+{
+ hlist_add_head(hnode, cfs_hash_hh_hhead(hs, bd));
+ return -1; /* unknown depth */
+}
+
+static int
+cfs_hash_hh_hnode_del(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnode)
+{
+ hlist_del_init(hnode);
+ return -1; /* unknown depth */
+}
+
+/**
+ * Simple hash head with depth tracking
+ * new element is always added to head of hlist
+ */
+struct cfs_hash_head_dep {
+ struct hlist_head hd_head; /**< entries list */
+ unsigned int hd_depth; /**< list length */
+};
+
+static int
+cfs_hash_hd_hhead_size(struct cfs_hash *hs)
+{
+ return sizeof(struct cfs_hash_head_dep);
+}
+
+static struct hlist_head *
+cfs_hash_hd_hhead(struct cfs_hash *hs, struct cfs_hash_bd *bd)
+{
+ struct cfs_hash_head_dep *head;
+
+ head = (struct cfs_hash_head_dep *)&bd->bd_bucket->hsb_head[0];
+ return &head[bd->bd_offset].hd_head;
+}
+
+static int
+cfs_hash_hd_hnode_add(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnode)
+{
+ struct cfs_hash_head_dep *hh;
+
+ hh = container_of(cfs_hash_hd_hhead(hs, bd),
+ struct cfs_hash_head_dep, hd_head);
+ hlist_add_head(hnode, &hh->hd_head);
+ return ++hh->hd_depth;
+}
+
+static int
+cfs_hash_hd_hnode_del(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnode)
+{
+ struct cfs_hash_head_dep *hh;
+
+ hh = container_of(cfs_hash_hd_hhead(hs, bd),
+ struct cfs_hash_head_dep, hd_head);
+ hlist_del_init(hnode);
+ return --hh->hd_depth;
+}
+
+/**
+ * double links hash head without depth tracking
+ * new element is always added to tail of hlist
+ */
+struct cfs_hash_dhead {
+ struct hlist_head dh_head; /**< entries list */
+ struct hlist_node *dh_tail; /**< the last entry */
+};
+
+static int
+cfs_hash_dh_hhead_size(struct cfs_hash *hs)
+{
+ return sizeof(struct cfs_hash_dhead);
+}
+
+static struct hlist_head *
+cfs_hash_dh_hhead(struct cfs_hash *hs, struct cfs_hash_bd *bd)
+{
+ struct cfs_hash_dhead *head;
+
+ head = (struct cfs_hash_dhead *)&bd->bd_bucket->hsb_head[0];
+ return &head[bd->bd_offset].dh_head;
+}
+
+static int
+cfs_hash_dh_hnode_add(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnode)
+{
+ struct cfs_hash_dhead *dh;
+
+ dh = container_of(cfs_hash_dh_hhead(hs, bd),
+ struct cfs_hash_dhead, dh_head);
+ if (dh->dh_tail != NULL) /* not empty */
+ hlist_add_behind(hnode, dh->dh_tail);
+ else /* empty list */
+ hlist_add_head(hnode, &dh->dh_head);
+ dh->dh_tail = hnode;
+ return -1; /* unknown depth */
+}
+
+static int
+cfs_hash_dh_hnode_del(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnd)
+{
+ struct cfs_hash_dhead *dh;
+
+ dh = container_of(cfs_hash_dh_hhead(hs, bd),
+ struct cfs_hash_dhead, dh_head);
+ if (hnd->next == NULL) { /* it's the tail */
+ dh->dh_tail = (hnd->pprev == &dh->dh_head.first) ? NULL :
+ container_of(hnd->pprev, struct hlist_node, next);
+ }
+ hlist_del_init(hnd);
+ return -1; /* unknown depth */
+}
+
+/**
+ * double links hash head with depth tracking
+ * new element is always added to tail of hlist
+ */
+struct cfs_hash_dhead_dep {
+ struct hlist_head dd_head; /**< entries list */
+ struct hlist_node *dd_tail; /**< the last entry */
+ unsigned int dd_depth; /**< list length */
+};
+
+static int
+cfs_hash_dd_hhead_size(struct cfs_hash *hs)
+{
+ return sizeof(struct cfs_hash_dhead_dep);
+}
+
+static struct hlist_head *
+cfs_hash_dd_hhead(struct cfs_hash *hs, struct cfs_hash_bd *bd)
+{
+ struct cfs_hash_dhead_dep *head;
+
+ head = (struct cfs_hash_dhead_dep *)&bd->bd_bucket->hsb_head[0];
+ return &head[bd->bd_offset].dd_head;
+}
+
+static int
+cfs_hash_dd_hnode_add(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnode)
+{
+ struct cfs_hash_dhead_dep *dh;
+
+ dh = container_of(cfs_hash_dd_hhead(hs, bd),
+ struct cfs_hash_dhead_dep, dd_head);
+ if (dh->dd_tail != NULL) /* not empty */
+ hlist_add_behind(hnode, dh->dd_tail);
+ else /* empty list */
+ hlist_add_head(hnode, &dh->dd_head);
+ dh->dd_tail = hnode;
+ return ++dh->dd_depth;
+}
+
+static int
+cfs_hash_dd_hnode_del(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnd)
+{
+ struct cfs_hash_dhead_dep *dh;
+
+ dh = container_of(cfs_hash_dd_hhead(hs, bd),
+ struct cfs_hash_dhead_dep, dd_head);
+ if (hnd->next == NULL) { /* it's the tail */
+ dh->dd_tail = (hnd->pprev == &dh->dd_head.first) ? NULL :
+ container_of(hnd->pprev, struct hlist_node, next);
+ }
+ hlist_del_init(hnd);
+ return --dh->dd_depth;
+}
+
+static struct cfs_hash_hlist_ops cfs_hash_hh_hops = {
+ .hop_hhead = cfs_hash_hh_hhead,
+ .hop_hhead_size = cfs_hash_hh_hhead_size,
+ .hop_hnode_add = cfs_hash_hh_hnode_add,
+ .hop_hnode_del = cfs_hash_hh_hnode_del,
+};
+
+static struct cfs_hash_hlist_ops cfs_hash_hd_hops = {
+ .hop_hhead = cfs_hash_hd_hhead,
+ .hop_hhead_size = cfs_hash_hd_hhead_size,
+ .hop_hnode_add = cfs_hash_hd_hnode_add,
+ .hop_hnode_del = cfs_hash_hd_hnode_del,
+};
+
+static struct cfs_hash_hlist_ops cfs_hash_dh_hops = {
+ .hop_hhead = cfs_hash_dh_hhead,
+ .hop_hhead_size = cfs_hash_dh_hhead_size,
+ .hop_hnode_add = cfs_hash_dh_hnode_add,
+ .hop_hnode_del = cfs_hash_dh_hnode_del,
+};
+
+static struct cfs_hash_hlist_ops cfs_hash_dd_hops = {
+ .hop_hhead = cfs_hash_dd_hhead,
+ .hop_hhead_size = cfs_hash_dd_hhead_size,
+ .hop_hnode_add = cfs_hash_dd_hnode_add,
+ .hop_hnode_del = cfs_hash_dd_hnode_del,
+};
+
+static void
+cfs_hash_hlist_setup(struct cfs_hash *hs)
+{
+ if (cfs_hash_with_add_tail(hs)) {
+ hs->hs_hops = cfs_hash_with_depth(hs) ?
+ &cfs_hash_dd_hops : &cfs_hash_dh_hops;
+ } else {
+ hs->hs_hops = cfs_hash_with_depth(hs) ?
+ &cfs_hash_hd_hops : &cfs_hash_hh_hops;
+ }
+}
+
+static void
+cfs_hash_bd_from_key(struct cfs_hash *hs, struct cfs_hash_bucket **bkts,
+ unsigned int bits, const void *key, struct cfs_hash_bd *bd)
+{
+ unsigned int index = cfs_hash_id(hs, key, (1U << bits) - 1);
+
+ LASSERT(bits == hs->hs_cur_bits || bits == hs->hs_rehash_bits);
+
+ bd->bd_bucket = bkts[index & ((1U << (bits - hs->hs_bkt_bits)) - 1)];
+ bd->bd_offset = index >> (bits - hs->hs_bkt_bits);
+}
+
+void
+cfs_hash_bd_get(struct cfs_hash *hs, const void *key, struct cfs_hash_bd *bd)
+{
+ /* NB: caller should hold hs->hs_rwlock if REHASH is set */
+ if (likely(hs->hs_rehash_buckets == NULL)) {
+ cfs_hash_bd_from_key(hs, hs->hs_buckets,
+ hs->hs_cur_bits, key, bd);
+ } else {
+ LASSERT(hs->hs_rehash_bits != 0);
+ cfs_hash_bd_from_key(hs, hs->hs_rehash_buckets,
+ hs->hs_rehash_bits, key, bd);
+ }
+}
+EXPORT_SYMBOL(cfs_hash_bd_get);
+
+static inline void
+cfs_hash_bd_dep_record(struct cfs_hash *hs, struct cfs_hash_bd *bd, int dep_cur)
+{
+ if (likely(dep_cur <= bd->bd_bucket->hsb_depmax))
+ return;
+
+ bd->bd_bucket->hsb_depmax = dep_cur;
+# if CFS_HASH_DEBUG_LEVEL >= CFS_HASH_DEBUG_1
+ if (likely(warn_on_depth == 0 ||
+ max(warn_on_depth, hs->hs_dep_max) >= dep_cur))
+ return;
+
+ spin_lock(&hs->hs_dep_lock);
+ hs->hs_dep_max = dep_cur;
+ hs->hs_dep_bkt = bd->bd_bucket->hsb_index;
+ hs->hs_dep_off = bd->bd_offset;
+ hs->hs_dep_bits = hs->hs_cur_bits;
+ spin_unlock(&hs->hs_dep_lock);
+
+ cfs_wi_schedule(cfs_sched_rehash, &hs->hs_dep_wi);
+# endif
+}
+
+void
+cfs_hash_bd_add_locked(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnode)
+{
+ int rc;
+
+ rc = hs->hs_hops->hop_hnode_add(hs, bd, hnode);
+ cfs_hash_bd_dep_record(hs, bd, rc);
+ bd->bd_bucket->hsb_version++;
+ if (unlikely(bd->bd_bucket->hsb_version == 0))
+ bd->bd_bucket->hsb_version++;
+ bd->bd_bucket->hsb_count++;
+
+ if (cfs_hash_with_counter(hs))
+ atomic_inc(&hs->hs_count);
+ if (!cfs_hash_with_no_itemref(hs))
+ cfs_hash_get(hs, hnode);
+}
+EXPORT_SYMBOL(cfs_hash_bd_add_locked);
+
+void
+cfs_hash_bd_del_locked(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnode)
+{
+ hs->hs_hops->hop_hnode_del(hs, bd, hnode);
+
+ LASSERT(bd->bd_bucket->hsb_count > 0);
+ bd->bd_bucket->hsb_count--;
+ bd->bd_bucket->hsb_version++;
+ if (unlikely(bd->bd_bucket->hsb_version == 0))
+ bd->bd_bucket->hsb_version++;
+
+ if (cfs_hash_with_counter(hs)) {
+ LASSERT(atomic_read(&hs->hs_count) > 0);
+ atomic_dec(&hs->hs_count);
+ }
+ if (!cfs_hash_with_no_itemref(hs))
+ cfs_hash_put_locked(hs, hnode);
+}
+EXPORT_SYMBOL(cfs_hash_bd_del_locked);
+
+void
+cfs_hash_bd_move_locked(struct cfs_hash *hs, struct cfs_hash_bd *bd_old,
+ struct cfs_hash_bd *bd_new, struct hlist_node *hnode)
+{
+ struct cfs_hash_bucket *obkt = bd_old->bd_bucket;
+ struct cfs_hash_bucket *nbkt = bd_new->bd_bucket;
+ int rc;
+
+ if (cfs_hash_bd_compare(bd_old, bd_new) == 0)
+ return;
+
+ /* use cfs_hash_bd_hnode_add/del, to avoid atomic & refcount ops
+ * in cfs_hash_bd_del/add_locked */
+ hs->hs_hops->hop_hnode_del(hs, bd_old, hnode);
+ rc = hs->hs_hops->hop_hnode_add(hs, bd_new, hnode);
+ cfs_hash_bd_dep_record(hs, bd_new, rc);
+
+ LASSERT(obkt->hsb_count > 0);
+ obkt->hsb_count--;
+ obkt->hsb_version++;
+ if (unlikely(obkt->hsb_version == 0))
+ obkt->hsb_version++;
+ nbkt->hsb_count++;
+ nbkt->hsb_version++;
+ if (unlikely(nbkt->hsb_version == 0))
+ nbkt->hsb_version++;
+}
+
+enum {
+ /** always set, for sanity (avoid ZERO intent) */
+ CFS_HS_LOOKUP_MASK_FIND = 1 << 0,
+ /** return entry with a ref */
+ CFS_HS_LOOKUP_MASK_REF = 1 << 1,
+ /** add entry if not existing */
+ CFS_HS_LOOKUP_MASK_ADD = 1 << 2,
+ /** delete entry, ignore other masks */
+ CFS_HS_LOOKUP_MASK_DEL = 1 << 3,
+};
+
+enum cfs_hash_lookup_intent {
+ /** return item w/o refcount */
+ CFS_HS_LOOKUP_IT_PEEK = CFS_HS_LOOKUP_MASK_FIND,
+ /** return item with refcount */
+ CFS_HS_LOOKUP_IT_FIND = (CFS_HS_LOOKUP_MASK_FIND |
+ CFS_HS_LOOKUP_MASK_REF),
+ /** return item w/o refcount if existed, otherwise add */
+ CFS_HS_LOOKUP_IT_ADD = (CFS_HS_LOOKUP_MASK_FIND |
+ CFS_HS_LOOKUP_MASK_ADD),
+ /** return item with refcount if existed, otherwise add */
+ CFS_HS_LOOKUP_IT_FINDADD = (CFS_HS_LOOKUP_IT_FIND |
+ CFS_HS_LOOKUP_MASK_ADD),
+ /** delete if existed */
+ CFS_HS_LOOKUP_IT_FINDDEL = (CFS_HS_LOOKUP_MASK_FIND |
+ CFS_HS_LOOKUP_MASK_DEL)
+};
+
+static struct hlist_node *
+cfs_hash_bd_lookup_intent(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ const void *key, struct hlist_node *hnode,
+ enum cfs_hash_lookup_intent intent)
+
+{
+ struct hlist_head *hhead = cfs_hash_bd_hhead(hs, bd);
+ struct hlist_node *ehnode;
+ struct hlist_node *match;
+ int intent_add = (intent & CFS_HS_LOOKUP_MASK_ADD) != 0;
+
+ /* with this function, we can avoid a lot of useless refcount ops,
+ * which are expensive atomic operations most time. */
+ match = intent_add ? NULL : hnode;
+ hlist_for_each(ehnode, hhead) {
+ if (!cfs_hash_keycmp(hs, key, ehnode))
+ continue;
+
+ if (match != NULL && match != ehnode) /* can't match */
+ continue;
+
+ /* match and ... */
+ if ((intent & CFS_HS_LOOKUP_MASK_DEL) != 0) {
+ cfs_hash_bd_del_locked(hs, bd, ehnode);
+ return ehnode;
+ }
+
+ /* caller wants refcount? */
+ if ((intent & CFS_HS_LOOKUP_MASK_REF) != 0)
+ cfs_hash_get(hs, ehnode);
+ return ehnode;
+ }
+ /* no match item */
+ if (!intent_add)
+ return NULL;
+
+ LASSERT(hnode != NULL);
+ cfs_hash_bd_add_locked(hs, bd, hnode);
+ return hnode;
+}
+
+struct hlist_node *
+cfs_hash_bd_lookup_locked(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ const void *key)
+{
+ return cfs_hash_bd_lookup_intent(hs, bd, key, NULL,
+ CFS_HS_LOOKUP_IT_FIND);
+}
+EXPORT_SYMBOL(cfs_hash_bd_lookup_locked);
+
+struct hlist_node *
+cfs_hash_bd_peek_locked(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ const void *key)
+{
+ return cfs_hash_bd_lookup_intent(hs, bd, key, NULL,
+ CFS_HS_LOOKUP_IT_PEEK);
+}
+EXPORT_SYMBOL(cfs_hash_bd_peek_locked);