]> git.itanic.dy.fi Git - linux-stable/commitdiff
staging: lustre: lu_object: move retry logic inside htable_lookup
authorNeilBrown <neilb@suse.com>
Mon, 7 May 2018 00:54:48 +0000 (10:54 +1000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 8 May 2018 11:09:20 +0000 (13:09 +0200)
The current retry logic, to wait when a 'dying' object is found,
spans multiple functions.  The process is attached to a waitqueue
and set TASK_UNINTERRUPTIBLE in htable_lookup, and this status
is passed back through lu_object_find_try() to lu_object_find_at()
where schedule() is called and the process is removed from the queue.

This can be simplified by moving all the logic (including
hashtable locking) inside htable_lookup(), which now never returns
EAGAIN.

Note that htable_lookup() is called with the hash bucket lock
held, and will drop and retake it if it needs to schedule.

I made this a 'goto' loop rather than a 'while(1)' loop as the
diff is easier to read.

Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/lustre/lustre/obdclass/lu_object.c

index b862c5aba7a61badafc151309cb14b219dcb4b08..32a593a0858a6bee683d987a0be975447b2f8dcc 100644 (file)
@@ -584,16 +584,24 @@ void lu_object_print(const struct lu_env *env, void *cookie,
 }
 EXPORT_SYMBOL(lu_object_print);
 
+/*
+ * NOTE: htable_lookup() is called with the relevant
+ * hash bucket locked, but might drop and re-acquire the lock.
+ */
 static struct lu_object *htable_lookup(struct lu_site *s,
                                       struct cfs_hash_bd *bd,
                                       const struct lu_fid *f,
-                                      wait_queue_entry_t *waiter,
                                       __u64 *version)
 {
+       struct cfs_hash         *hs = s->ls_obj_hash;
        struct lu_site_bkt_data *bkt;
        struct lu_object_header *h;
        struct hlist_node       *hnode;
-       __u64  ver = cfs_hash_bd_version_get(bd);
+       __u64 ver;
+       wait_queue_entry_t waiter;
+
+retry:
+       ver = cfs_hash_bd_version_get(bd);
 
        if (*version == ver)
                return ERR_PTR(-ENOENT);
@@ -626,11 +634,15 @@ static struct lu_object *htable_lookup(struct lu_site *s,
         * drained), and moreover, lookup has to wait until object is freed.
         */
 
-       init_waitqueue_entry(waiter, current);
-       add_wait_queue(&bkt->lsb_marche_funebre, waiter);
+       init_waitqueue_entry(&waiter, current);
+       add_wait_queue(&bkt->lsb_marche_funebre, &waiter);
        set_current_state(TASK_UNINTERRUPTIBLE);
        lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_DEATH_RACE);
-       return ERR_PTR(-EAGAIN);
+       cfs_hash_bd_unlock(hs, bd, 1);
+       schedule();
+       remove_wait_queue(&bkt->lsb_marche_funebre, &waiter);
+       cfs_hash_bd_lock(hs, bd, 1);
+       goto retry;
 }
 
 /**
@@ -694,13 +706,14 @@ static struct lu_object *lu_object_new(const struct lu_env *env,
 }
 
 /**
- * Core logic of lu_object_find*() functions.
+ * Much like lu_object_find(), but top level device of object is specifically
+ * \a dev rather than top level device of the site. This interface allows
+ * objects of different "stacking" to be created within the same site.
  */
-static struct lu_object *lu_object_find_try(const struct lu_env *env,
-                                           struct lu_device *dev,
-                                           const struct lu_fid *f,
-                                           const struct lu_object_conf *conf,
-                                           wait_queue_entry_t *waiter)
+struct lu_object *lu_object_find_at(const struct lu_env *env,
+                                   struct lu_device *dev,
+                                   const struct lu_fid *f,
+                                   const struct lu_object_conf *conf)
 {
        struct lu_object      *o;
        struct lu_object      *shadow;
@@ -726,8 +739,6 @@ static struct lu_object *lu_object_find_try(const struct lu_env *env,
         * It is unnecessary to perform lookup-alloc-lookup-insert, instead,
         * just alloc and insert directly.
         *
-        * If dying object is found during index search, add @waiter to the
-        * site wait-queue and return ERR_PTR(-EAGAIN).
         */
        if (conf && conf->loc_flags & LOC_F_NEW)
                return lu_object_new(env, dev, f, conf);
@@ -735,8 +746,9 @@ static struct lu_object *lu_object_find_try(const struct lu_env *env,
        s  = dev->ld_site;
        hs = s->ls_obj_hash;
        cfs_hash_bd_get_and_lock(hs, (void *)f, &bd, 1);
-       o = htable_lookup(s, &bd, f, waiter, &version);
+       o = htable_lookup(s, &bd, f, &version);
        cfs_hash_bd_unlock(hs, &bd, 1);
+
        if (!IS_ERR(o) || PTR_ERR(o) != -ENOENT)
                return o;
 
@@ -752,7 +764,7 @@ static struct lu_object *lu_object_find_try(const struct lu_env *env,
 
        cfs_hash_bd_lock(hs, &bd, 1);
 
-       shadow = htable_lookup(s, &bd, f, waiter, &version);
+       shadow = htable_lookup(s, &bd, f, &version);
        if (likely(PTR_ERR(shadow) == -ENOENT)) {
                cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
                cfs_hash_bd_unlock(hs, &bd, 1);
@@ -767,34 +779,6 @@ static struct lu_object *lu_object_find_try(const struct lu_env *env,
        lu_object_free(env, o);
        return shadow;
 }
-
-/**
- * Much like lu_object_find(), but top level device of object is specifically
- * \a dev rather than top level device of the site. This interface allows
- * objects of different "stacking" to be created within the same site.
- */
-struct lu_object *lu_object_find_at(const struct lu_env *env,
-                                   struct lu_device *dev,
-                                   const struct lu_fid *f,
-                                   const struct lu_object_conf *conf)
-{
-       wait_queue_head_t       *wq;
-       struct lu_object        *obj;
-       wait_queue_entry_t         wait;
-
-       while (1) {
-               obj = lu_object_find_try(env, dev, f, conf, &wait);
-               if (obj != ERR_PTR(-EAGAIN))
-                       return obj;
-               /*
-                * lu_object_find_try() already added waiter into the
-                * wait queue.
-                */
-               schedule();
-               wq = lu_site_wq_from_fid(dev->ld_site, (void *)f);
-               remove_wait_queue(wq, &wait);
-       }
-}
 EXPORT_SYMBOL(lu_object_find_at);
 
 /**