nfsd: Containerise filecache laundrette

Ensure that if the filecache laundrette gets stuck, it only affects
the knfsd instances of one container.

The notifier callbacks can be called from various contexts so avoid
using synchonous filesystem operations that might deadlock.

Signed-off-by: Trond Myklebust <[e-mail]>
Signed-off-by: J. Bruce Fields <[e-mail]>

fs/nfsd/filecache.c

···
 
 static DEFINE_PER_CPU(unsigned long, nfsd_file_cache_hits);
 
+struct nfsd_fcache_disposal {
+	struct list_head list;
+	struct work_struct work;
+	struct net *net;
+	spinlock_t lock;
+	struct list_head freeme;
+	struct rcu_head rcu;
+};
+
+struct workqueue_struct *nfsd_filecache_wq __read_mostly;
+
 static struct kmem_cache		*nfsd_file_slab;
 static struct kmem_cache		*nfsd_file_mark_slab;
 static struct nfsd_fcache_bucket	*nfsd_file_hashtbl;
···
 static struct fsnotify_group		*nfsd_file_fsnotify_group;
 static atomic_long_t			nfsd_filecache_count;
 static struct delayed_work		nfsd_filecache_laundrette;
+static DEFINE_SPINLOCK(laundrette_lock);
+static LIST_HEAD(laundrettes);
 
-enum nfsd_file_laundrette_ctl {
-	NFSD_FILE_LAUNDRETTE_NOFLUSH = 0,
-	NFSD_FILE_LAUNDRETTE_MAY_FLUSH
-};
+static void nfsd_file_gc(void);
 
 static void
-nfsd_file_schedule_laundrette(enum nfsd_file_laundrette_ctl ctl)
+nfsd_file_schedule_laundrette(void)
 {
 	long count = atomic_long_read(&nfsd_filecache_count);
 
 	if (count == 0 || test_bit(NFSD_FILE_SHUTDOWN, &nfsd_file_lru_flags))
 		return;
 
-	/* Be more aggressive about scanning if over the threshold */
-	if (count > NFSD_FILE_LRU_THRESHOLD)
-		mod_delayed_work(system_wq, &nfsd_filecache_laundrette, 0);
-	else
-		schedule_delayed_work(&nfsd_filecache_laundrette, NFSD_LAUNDRETTE_DELAY);
-
-	if (ctl == NFSD_FILE_LAUNDRETTE_NOFLUSH)
-		return;
-
-	/* ...and don't delay flushing if we're out of control */
-	if (count >= NFSD_FILE_LRU_LIMIT)
-		flush_delayed_work(&nfsd_filecache_laundrette);
+	queue_delayed_work(system_wq, &nfsd_filecache_laundrette,
+			NFSD_LAUNDRETTE_DELAY);
 }
 
 static void
···
 
 	set_bit(NFSD_FILE_REFERENCED, &nf->nf_flags);
 	if (nfsd_file_put_noref(nf) == 1 && is_hashed && unused)
-		nfsd_file_schedule_laundrette(NFSD_FILE_LAUNDRETTE_MAY_FLUSH);
+		nfsd_file_schedule_laundrette();
+	if (atomic_long_read(&nfsd_filecache_count) >= NFSD_FILE_LRU_LIMIT)
+		nfsd_file_gc();
 }
 
 struct nfsd_file *
···
 	}
 	if (flush)
 		flush_delayed_fput();
+}
+
+static void
+nfsd_file_list_remove_disposal(struct list_head *dst,
+		struct nfsd_fcache_disposal *l)
+{
+	spin_lock(&l->lock);
+	list_splice_init(&l->freeme, dst);
+	spin_unlock(&l->lock);
+}
+
+static void
+nfsd_file_list_add_disposal(struct list_head *files, struct net *net)
+{
+	struct nfsd_fcache_disposal *l;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(l, &laundrettes, list) {
+		if (l->net == net) {
+			spin_lock(&l->lock);
+			list_splice_tail_init(files, &l->freeme);
+			spin_unlock(&l->lock);
+			queue_work(nfsd_filecache_wq, &l->work);
+			break;
+		}
+	}
+	rcu_read_unlock();
+}
+
+static void
+nfsd_file_list_add_pernet(struct list_head *dst, struct list_head *src,
+		struct net *net)
+{
+	struct nfsd_file *nf, *tmp;
+
+	list_for_each_entry_safe(nf, tmp, src, nf_lru) {
+		if (nf->nf_net == net)
+			list_move_tail(&nf->nf_lru, dst);
+	}
+}
+
+static void
+nfsd_file_dispose_list_delayed(struct list_head *dispose)
+{
+	LIST_HEAD(list);
+	struct nfsd_file *nf;
+
+	while(!list_empty(dispose)) {
+		nf = list_first_entry(dispose, struct nfsd_file, nf_lru);
+		nfsd_file_list_add_pernet(&list, dispose, nf->nf_net);
+		nfsd_file_list_add_disposal(&list, nf->nf_net);
+	}
 }
 
 /*
···
 	return LRU_SKIP;
 }
 
-static void
-nfsd_file_lru_dispose(struct list_head *head)
+static unsigned long
+nfsd_file_lru_walk_list(struct shrink_control *sc)
 {
+	LIST_HEAD(head);
 	struct nfsd_file *nf;
+	unsigned long ret;
 
-	list_for_each_entry(nf, head, nf_lru) {
+	if (sc)
+		ret = list_lru_shrink_walk(&nfsd_file_lru, sc,
+				nfsd_file_lru_cb, &head);
+	else
+		ret = list_lru_walk(&nfsd_file_lru,
+				nfsd_file_lru_cb,
+				&head, LONG_MAX);
+	list_for_each_entry(nf, &head, nf_lru) {
 		spin_lock(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
 		nfsd_file_do_unhash(nf);
 		spin_unlock(&nfsd_file_hashtbl[nf->nf_hashval].nfb_lock);
 	}
-	nfsd_file_dispose_list(head);
+	nfsd_file_dispose_list_delayed(&head);
+	return ret;
+}
+
+static void
+nfsd_file_gc(void)
+{
+	nfsd_file_lru_walk_list(NULL);
+}
+
+static void
+nfsd_file_gc_worker(struct work_struct *work)
+{
+	nfsd_file_gc();
+	nfsd_file_schedule_laundrette();
 }
 
 static unsigned long
···
 static unsigned long
 nfsd_file_lru_scan(struct shrinker *s, struct shrink_control *sc)
 {
-	LIST_HEAD(head);
-	unsigned long ret;
-
-	ret = list_lru_shrink_walk(&nfsd_file_lru, sc, nfsd_file_lru_cb, &head);
-	nfsd_file_lru_dispose(&head);
-	return ret;
+	return nfsd_file_lru_walk_list(sc);
 }
 
 static struct shrinker	nfsd_file_shrinker = {
···
 
 	__nfsd_file_close_inode(inode, hashval, &dispose);
 	trace_nfsd_file_close_inode(inode, hashval, !list_empty(&dispose));
-	nfsd_file_dispose_list(&dispose);
+	nfsd_file_dispose_list_delayed(&dispose);
 }
 
 /**
···
 nfsd_file_delayed_close(struct work_struct *work)
 {
 	LIST_HEAD(head);
+	struct nfsd_fcache_disposal *l = container_of(work,
+			struct nfsd_fcache_disposal, work);
 
-	list_lru_walk(&nfsd_file_lru, nfsd_file_lru_cb, &head, LONG_MAX);
-
-	if (test_and_clear_bit(NFSD_FILE_LRU_RESCAN, &nfsd_file_lru_flags))
-		nfsd_file_schedule_laundrette(NFSD_FILE_LAUNDRETTE_NOFLUSH);
-
-	if (!list_empty(&head)) {
-		nfsd_file_lru_dispose(&head);
-		flush_delayed_fput();
-	}
+	nfsd_file_list_remove_disposal(&head, l);
+	nfsd_file_dispose_list(&head);
 }
 
 static int
···
 	if (nfsd_file_hashtbl)
 		return 0;
 
+	nfsd_filecache_wq = alloc_workqueue("nfsd_filecache", 0, 0);
+	if (!nfsd_filecache_wq)
+		goto out;
+
 	nfsd_file_hashtbl = kcalloc(NFSD_FILE_HASH_SIZE,
 				sizeof(*nfsd_file_hashtbl), GFP_KERNEL);
 	if (!nfsd_file_hashtbl) {
···
 		spin_lock_init(&nfsd_file_hashtbl[i].nfb_lock);
 	}
 
-	INIT_DELAYED_WORK(&nfsd_filecache_laundrette, nfsd_file_delayed_close);
+	INIT_DELAYED_WORK(&nfsd_filecache_laundrette, nfsd_file_gc_worker);
 out:
 	return ret;
 out_notifier:
···
 	nfsd_file_mark_slab = NULL;
 	kfree(nfsd_file_hashtbl);
 	nfsd_file_hashtbl = NULL;
+	destroy_workqueue(nfsd_filecache_wq);
+	nfsd_filecache_wq = NULL;
 	goto out;
 }
 
···
 	}
 }
 
+static struct nfsd_fcache_disposal *
+nfsd_alloc_fcache_disposal(struct net *net)
+{
+	struct nfsd_fcache_disposal *l;
+
+	l = kmalloc(sizeof(*l), GFP_KERNEL);
+	if (!l)
+		return NULL;
+	INIT_WORK(&l->work, nfsd_file_delayed_close);
+	l->net = net;
+	spin_lock_init(&l->lock);
+	INIT_LIST_HEAD(&l->freeme);
+	return l;
+}
+
+static void
+nfsd_free_fcache_disposal(struct nfsd_fcache_disposal *l)
+{
+	rcu_assign_pointer(l->net, NULL);
+	cancel_work_sync(&l->work);
+	nfsd_file_dispose_list(&l->freeme);
+	kfree_rcu(l, rcu);
+}
+
+static void
+nfsd_add_fcache_disposal(struct nfsd_fcache_disposal *l)
+{
+	spin_lock(&laundrette_lock);
+	list_add_tail_rcu(&l->list, &laundrettes);
+	spin_unlock(&laundrette_lock);
+}
+
+static void
+nfsd_del_fcache_disposal(struct nfsd_fcache_disposal *l)
+{
+	spin_lock(&laundrette_lock);
+	list_del_rcu(&l->list);
+	spin_unlock(&laundrette_lock);
+}
+
+static int
+nfsd_alloc_fcache_disposal_net(struct net *net)
+{
+	struct nfsd_fcache_disposal *l;
+
+	l = nfsd_alloc_fcache_disposal(net);
+	if (!l)
+		return -ENOMEM;
+	nfsd_add_fcache_disposal(l);
+	return 0;
+}
+
+static void
+nfsd_free_fcache_disposal_net(struct net *net)
+{
+	struct nfsd_fcache_disposal *l;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(l, &laundrettes, list) {
+		if (l->net != net)
+			continue;
+		nfsd_del_fcache_disposal(l);
+		rcu_read_unlock();
+		nfsd_free_fcache_disposal(l);
+		return;
+	}
+	rcu_read_unlock();
+}
+
+int
+nfsd_file_cache_start_net(struct net *net)
+{
+	return nfsd_alloc_fcache_disposal_net(net);
+}
+
+void
+nfsd_file_cache_shutdown_net(struct net *net)
+{
+	nfsd_file_cache_purge(net);
+	nfsd_free_fcache_disposal_net(net);
+}
+
 void
 nfsd_file_cache_shutdown(void)
 {
···
 	nfsd_file_mark_slab = NULL;
 	kfree(nfsd_file_hashtbl);
 	nfsd_file_hashtbl = NULL;
+	destroy_workqueue(nfsd_filecache_wq);
+	nfsd_filecache_wq = NULL;
 }
 
 static bool
···
 	nfsd_file_hashtbl[hashval].nfb_maxcount = max(nfsd_file_hashtbl[hashval].nfb_maxcount,
 			nfsd_file_hashtbl[hashval].nfb_count);
 	spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
-	atomic_long_inc(&nfsd_filecache_count);
+	if (atomic_long_inc_return(&nfsd_filecache_count) >= NFSD_FILE_LRU_THRESHOLD)
+		nfsd_file_gc();
 
 	nf->nf_mark = nfsd_file_mark_find_or_create(nf);
 	if (nf->nf_mark)

fs/nfsd/filecache.h

···
 int nfsd_file_cache_init(void);
 void nfsd_file_cache_purge(struct net *);
 void nfsd_file_cache_shutdown(void);
+int nfsd_file_cache_start_net(struct net *net);
+void nfsd_file_cache_shutdown_net(struct net *net);
 void nfsd_file_put(struct nfsd_file *nf);
 struct nfsd_file *nfsd_file_get(struct nfsd_file *nf);
 void nfsd_file_close_inode_sync(struct inode *inode);

fs/nfsd/nfssvc.c

···
 		nn->lockd_up = true;
 	}
 
-	ret = nfs4_state_start_net(net);
+	ret = nfsd_file_cache_start_net(net);
 	if (ret)
 		goto out_lockd;
+	ret = nfs4_state_start_net(net);
+	if (ret)
+		goto out_filecache;
 
 	nn->nfsd_net_up = true;
 	return 0;
 
+out_filecache:
+	nfsd_file_cache_shutdown_net(net);
 out_lockd:
 	if (nn->lockd_up) {
 		lockd_down(net);
···
 {
 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 
-	nfsd_file_cache_purge(net);
+	nfsd_file_cache_shutdown_net(net);
 	nfs4_state_shutdown_net(net);
 	if (nn->lockd_up) {
 		lockd_down(net);