]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
staging: lustre: convert lov_pool to use rhashtable
authorNeilBrown <neilb@suse.com>
Wed, 11 Apr 2018 21:54:48 +0000 (07:54 +1000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 23 Apr 2018 13:07:40 +0000 (15:07 +0200)
The pools hashtable can be implemented using
the rhashtable implementation in lib.
This has the benefit that lookups are lock-free.

We need to use kfree_rcu() to free a pool so
that a lookup racing with a deletion will not access
freed memory.

rhashtable has no combined lookup-and-delete interface,
but as the lookup is lockless and the chains are short,
this brings little cost.  Even if a lookup finds a pool,
we must be prepared for the delete to fail to find it,
as we might race with another thread doing a delete.

We use atomic_inc_not_zero() after finding a pool in the
hash table and if that fails, we must have raced with a
deletion, so we treat the lookup as a failure.

Use hashlen_string() rather than a hand-crafted hash
function.
Note that the pool_name, and the search key, are
guaranteed to be nul terminated.

Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/lustre/lustre/include/obd.h
drivers/staging/lustre/lustre/include/obd_support.h
drivers/staging/lustre/lustre/lov/lov_internal.h
drivers/staging/lustre/lustre/lov/lov_obd.c
drivers/staging/lustre/lustre/lov/lov_pool.c

index 48cf7ab87a5ed434994910be8205abc0190312ae..7e1de031d6a0c29162d7346d3242bf2c4f1b7a2a 100644 (file)
@@ -46,6 +46,8 @@
 #include <lustre_intent.h>
 #include <cl_object.h>
 
+#include <linux/rhashtable.h>
+
 #define MAX_OBD_DEVICES 8192
 
 struct osc_async_rc {
@@ -383,7 +385,7 @@ struct lov_obd {
        __u32              lov_tgt_size;   /* size of tgts array */
        int                  lov_connects;
        int                  lov_pool_count;
-       struct cfs_hash      *lov_pools_hash_body; /* used for key access */
+       struct rhashtable       lov_pools_hash_body; /* used for key access */
        struct list_head        lov_pool_list; /* used for sequential access */
        struct dentry           *lov_pool_debugfs_entry;
        enum lustre_sec_part    lov_sp_me;
index aebcab191442bb3c255454536d0cf91ba403bf97..730a6ee71565ecf712838590c4940b40ce347f0a 100644 (file)
@@ -61,9 +61,6 @@ extern atomic_long_t obd_dirty_transit_pages;
 extern char obd_jobid_var[];
 
 /* Some hash init argument constants */
-#define HASH_POOLS_BKT_BITS 3
-#define HASH_POOLS_CUR_BITS 3
-#define HASH_POOLS_MAX_BITS 7
 #define HASH_UUID_BKT_BITS 5
 #define HASH_UUID_CUR_BITS 7
 #define HASH_UUID_MAX_BITS 12
index a56d71c2dda2b0f8bec7164d6b79fb01ee0cd851..8dde3828f9353fd147f5ffe053b5814c66b592bc 100644 (file)
@@ -149,11 +149,16 @@ struct pool_desc {
        char                     pool_name[LOV_MAXPOOLNAME + 1];
        struct ost_pool          pool_obds;
        atomic_t                 pool_refcount;
-       struct hlist_node        pool_hash;             /* access by poolname */
-       struct list_head         pool_list;             /* serial access */
+       struct rhash_head        pool_hash;             /* access by poolname */
+       union {
+               struct list_head        pool_list;      /* serial access */
+               struct rcu_head         rcu;            /* delayed free */
+       };
        struct dentry           *pool_debugfs_entry;    /* file in debugfs */
        struct obd_device       *pool_lobd;             /* owner */
 };
+int lov_pool_hash_init(struct rhashtable *tbl);
+void lov_pool_hash_destroy(struct rhashtable *tbl);
 
 struct lov_request {
        struct obd_info   rq_oi;
@@ -241,8 +246,6 @@ void lprocfs_lov_init_vars(struct lprocfs_static_vars *lvars);
 /* lov_cl.c */
 extern struct lu_device_type lov_device_type;
 
-/* pools */
-extern struct cfs_hash_ops pool_hash_operations;
 /* ost_pool methods */
 int lov_ost_pool_init(struct ost_pool *op, unsigned int count);
 int lov_ost_pool_extend(struct ost_pool *op, unsigned int min_count);
index 355e87ecc62d1944341ceb4daecf8c46ef0764d9..94da35e673f73e1a381ef9a6df782a04dd78a37d 100644 (file)
@@ -795,15 +795,11 @@ int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 
        init_rwsem(&lov->lov_notify_lock);
 
-       lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
-                                                  HASH_POOLS_MAX_BITS,
-                                                  HASH_POOLS_BKT_BITS, 0,
-                                                  CFS_HASH_MIN_THETA,
-                                                  CFS_HASH_MAX_THETA,
-                                                  &pool_hash_operations,
-                                                  CFS_HASH_DEFAULT);
        INIT_LIST_HEAD(&lov->lov_pool_list);
        lov->lov_pool_count = 0;
+       rc = lov_pool_hash_init(&lov->lov_pools_hash_body);
+       if (rc)
+               goto out;
        rc = lov_ost_pool_init(&lov->lov_packed, 0);
        if (rc)
                goto out;
@@ -839,7 +835,7 @@ static int lov_cleanup(struct obd_device *obd)
                /* coverity[overrun-buffer-val] */
                lov_pool_del(obd, pool->pool_name);
        }
-       cfs_hash_putref(lov->lov_pools_hash_body);
+       lov_pool_hash_destroy(&lov->lov_pools_hash_body);
        lov_ost_pool_free(&lov->lov_packed);
 
        lprocfs_obd_cleanup(obd);
index ecd9329cd073e9110ad79f8f1eed4ec287311976..b673b4fd305b41b20ee52e559e1abbb0992bd75a 100644 (file)
 #define pool_tgt(_p, _i) \
                _p->pool_lobd->u.lov.lov_tgts[_p->pool_obds.op_array[_i]]
 
+static u32 pool_hashfh(const void *data, u32 len, u32 seed)
+{
+       const char *pool_name = data;
+       return hashlen_hash(hashlen_string((void*)(unsigned long)seed, pool_name));
+}
+
+static int pool_cmpfn(struct rhashtable_compare_arg *arg, const void *obj)
+{
+       const struct pool_desc *pool = obj;
+       const char *pool_name = arg->key;
+       return strcmp(pool_name, pool->pool_name);
+}
+
+static const struct rhashtable_params pools_hash_params = {
+       .key_len        = 1, /* actually variable */
+       .key_offset     = offsetof(struct pool_desc, pool_name),
+       .head_offset    = offsetof(struct pool_desc, pool_hash),
+       .hashfn         = pool_hashfh,
+       .obj_cmpfn      = pool_cmpfn,
+       .automatic_shrinking = true,
+};
+
 static void lov_pool_getref(struct pool_desc *pool)
 {
        CDEBUG(D_INFO, "pool %p\n", pool);
@@ -59,96 +81,13 @@ void lov_pool_putref(struct pool_desc *pool)
 {
        CDEBUG(D_INFO, "pool %p\n", pool);
        if (atomic_dec_and_test(&pool->pool_refcount)) {
-               LASSERT(hlist_unhashed(&pool->pool_hash));
                LASSERT(list_empty(&pool->pool_list));
                LASSERT(!pool->pool_debugfs_entry);
                lov_ost_pool_free(&pool->pool_obds);
-               kfree(pool);
-       }
-}
-
-static void lov_pool_putref_locked(struct pool_desc *pool)
-{
-       CDEBUG(D_INFO, "pool %p\n", pool);
-       LASSERT(atomic_read(&pool->pool_refcount) > 1);
-
-       atomic_dec(&pool->pool_refcount);
-}
-
-/*
- * hash function using a Rotating Hash algorithm
- * Knuth, D. The Art of Computer Programming,
- * Volume 3: Sorting and Searching,
- * Chapter 6.4.
- * Addison Wesley, 1973
- */
-static __u32 pool_hashfn(struct cfs_hash *hash_body, const void *key,
-                        unsigned int mask)
-{
-       int i;
-       __u32 result;
-       char *poolname;
-
-       result = 0;
-       poolname = (char *)key;
-       for (i = 0; i < LOV_MAXPOOLNAME; i++) {
-               if (poolname[i] == '\0')
-                       break;
-               result = (result << 4) ^ (result >> 28) ^  poolname[i];
+               kfree_rcu(pool, rcu);
        }
-       return (result % mask);
-}
-
-static void *pool_key(struct hlist_node *hnode)
-{
-       struct pool_desc *pool;
-
-       pool = hlist_entry(hnode, struct pool_desc, pool_hash);
-       return pool->pool_name;
-}
-
-static int pool_hashkey_keycmp(const void *key, struct hlist_node *compared_hnode)
-{
-       char *pool_name;
-       struct pool_desc *pool;
-
-       pool_name = (char *)key;
-       pool = hlist_entry(compared_hnode, struct pool_desc, pool_hash);
-       return !strncmp(pool_name, pool->pool_name, LOV_MAXPOOLNAME);
-}
-
-static void *pool_hashobject(struct hlist_node *hnode)
-{
-       return hlist_entry(hnode, struct pool_desc, pool_hash);
-}
-
-static void pool_hashrefcount_get(struct cfs_hash *hs, struct hlist_node *hnode)
-{
-       struct pool_desc *pool;
-
-       pool = hlist_entry(hnode, struct pool_desc, pool_hash);
-       lov_pool_getref(pool);
-}
-
-static void pool_hashrefcount_put_locked(struct cfs_hash *hs,
-                                        struct hlist_node *hnode)
-{
-       struct pool_desc *pool;
-
-       pool = hlist_entry(hnode, struct pool_desc, pool_hash);
-       lov_pool_putref_locked(pool);
 }
 
-struct cfs_hash_ops pool_hash_operations = {
-       .hs_hash        = pool_hashfn,
-       .hs_key         = pool_key,
-       .hs_keycmp      = pool_hashkey_keycmp,
-       .hs_object      = pool_hashobject,
-       .hs_get         = pool_hashrefcount_get,
-       .hs_put_locked  = pool_hashrefcount_put_locked,
-
-};
-
 /*
  * pool debugfs seq_file methods
  */
@@ -396,6 +335,23 @@ int lov_ost_pool_free(struct ost_pool *op)
        return 0;
 }
 
+static void
+pools_hash_exit(void *vpool, void *data)
+{
+       struct pool_desc *pool = vpool;
+       lov_pool_putref(pool);
+}
+
+int lov_pool_hash_init(struct rhashtable *tbl)
+{
+       return rhashtable_init(tbl, &pools_hash_params);
+}
+
+void lov_pool_hash_destroy(struct rhashtable *tbl)
+{
+       rhashtable_free_and_destroy(tbl, pools_hash_exit, NULL);
+}
+
 int lov_pool_new(struct obd_device *obd, char *poolname)
 {
        struct lov_obd *lov;
@@ -421,8 +377,6 @@ int lov_pool_new(struct obd_device *obd, char *poolname)
        if (rc)
                goto out_err;
 
-       INIT_HLIST_NODE(&new_pool->pool_hash);
-
        /* get ref for debugfs file */
        lov_pool_getref(new_pool);
        new_pool->pool_debugfs_entry = ldebugfs_add_simple(
@@ -443,11 +397,16 @@ int lov_pool_new(struct obd_device *obd, char *poolname)
        lov->lov_pool_count++;
        spin_unlock(&obd->obd_dev_lock);
 
-       /* add to find only when it fully ready  */
-       rc = cfs_hash_add_unique(lov->lov_pools_hash_body, poolname,
-                                &new_pool->pool_hash);
+       /* Add to hash table only when it is fully ready.  */
+       rc = rhashtable_lookup_insert_fast(&lov->lov_pools_hash_body,
+                                          &new_pool->pool_hash, pools_hash_params);
        if (rc) {
-               rc = -EEXIST;
+               if (rc != -EEXIST)
+                       /*
+                        * Hide -E2BIG and -EBUSY which
+                        * are not helpful.
+                        */
+                       rc = -ENOMEM;
                goto out_err;
        }
 
@@ -476,7 +435,13 @@ int lov_pool_del(struct obd_device *obd, char *poolname)
        lov = &obd->u.lov;
 
        /* lookup and kill hash reference */
-       pool = cfs_hash_del_key(lov->lov_pools_hash_body, poolname);
+       rcu_read_lock();
+       pool = rhashtable_lookup(&lov->lov_pools_hash_body, poolname, pools_hash_params);
+       if (pool)
+               if (rhashtable_remove_fast(&lov->lov_pools_hash_body,
+                                          &pool->pool_hash, pools_hash_params) != 0)
+                       pool = NULL;
+       rcu_read_unlock();
        if (!pool)
                return -ENOENT;
 
@@ -507,7 +472,11 @@ int lov_pool_add(struct obd_device *obd, char *poolname, char *ostname)
 
        lov = &obd->u.lov;
 
-       pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
+       rcu_read_lock();
+       pool = rhashtable_lookup(&lov->lov_pools_hash_body, poolname, pools_hash_params);
+       if (pool && !atomic_inc_not_zero(&pool->pool_refcount))
+               pool = NULL;
+       rcu_read_unlock();
        if (!pool)
                return -ENOENT;
 
@@ -551,7 +520,11 @@ int lov_pool_remove(struct obd_device *obd, char *poolname, char *ostname)
 
        lov = &obd->u.lov;
 
-       pool = cfs_hash_lookup(lov->lov_pools_hash_body, poolname);
+       rcu_read_lock();
+       pool = rhashtable_lookup(&lov->lov_pools_hash_body, poolname, pools_hash_params);
+       if (pool && !atomic_inc_not_zero(&pool->pool_refcount))
+               pool = NULL;
+       rcu_read_unlock();
        if (!pool)
                return -ENOENT;