bcachefs: Fix bch2_evict_subvolume_inodes()

This fixes a bug in bch2_evict_subvolume_inodes(): d_mark_dontcache()
doesn't handle the case where i_count is already 0, we need to grab and
put the inode in order for it to be dropped.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

1 files changed, 63 insertions, 30 deletions

diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index 828887abc261..129924dfaf69 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -201,6 +201,10 @@ struct inode *bch2_vfs_inode_get(struct bch_fs *c, subvol_inum inum)
 		return ERR_PTR(ret);
 	}
 
+	mutex_lock(&c->vfs_inodes_lock);
+	list_add(&inode->ei_vfs_inode_list, &c->vfs_inodes_list);
+	mutex_unlock(&c->vfs_inodes_lock);
+
 	unlock_new_inode(&inode->v);
 
 	return &inode->v;
@@ -314,6 +318,9 @@ err_before_quota:
 
 		inode = old;
 	} else {
+		mutex_lock(&c->vfs_inodes_lock);
+		list_add(&inode->ei_vfs_inode_list, &c->vfs_inodes_list);
+		mutex_unlock(&c->vfs_inodes_lock);
 		/*
 		 * we really don't want insert_inode_locked2() to be setting
 		 * I_NEW...
@@ -1370,6 +1377,7 @@ static struct inode *bch2_alloc_inode(struct super_block *sb)
 	inode_init_once(&inode->v);
 	mutex_init(&inode->ei_update_lock);
 	two_state_lock_init(&inode->ei_pagecache_lock);
+	INIT_LIST_HEAD(&inode->ei_vfs_inode_list);
 	mutex_init(&inode->ei_quota_lock);
 
 	return &inode->v;
@@ -1434,53 +1442,78 @@ static void bch2_evict_inode(struct inode *vinode)
 				KEY_TYPE_QUOTA_WARN);
 		bch2_inode_rm(c, inode_inum(inode));
 	}
+
+	mutex_lock(&c->vfs_inodes_lock);
+	list_del_init(&inode->ei_vfs_inode_list);
+	mutex_unlock(&c->vfs_inodes_lock);
 }
 
-void bch2_evict_subvolume_inodes(struct bch_fs *c,
-				 snapshot_id_list *s)
+void bch2_evict_subvolume_inodes(struct bch_fs *c, snapshot_id_list *s)
 {
-	struct super_block *sb = c->vfs_sb;
-	struct inode *inode;
+	struct bch_inode_info *inode, **i;
+	DARRAY(struct bch_inode_info *) grabbed;
+	bool clean_pass = false, this_pass_clean;
 
-	spin_lock(&sb->s_inode_list_lock);
-	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
-		if (!snapshot_list_has_id(s, to_bch_ei(inode)->ei_subvol) ||
-		    (inode->i_state & I_FREEING))
-			continue;
+	/*
+	 * Initially, we scan for inodes without I_DONTCACHE, then mark them to
+	 * be pruned with d_mark_dontcache().
+	 *
+	 * Once we've had a clean pass where we didn't find any inodes without
+	 * I_DONTCACHE, we wait for them to be freed:
+	 */
 
-		d_mark_dontcache(inode);
-		d_prune_aliases(inode);
-	}
-	spin_unlock(&sb->s_inode_list_lock);
+	darray_init(&grabbed);
+	darray_make_room(&grabbed, 1024);
 again:
 	cond_resched();
-	spin_lock(&sb->s_inode_list_lock);
-	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
-		if (!snapshot_list_has_id(s, to_bch_ei(inode)->ei_subvol) ||
-		    (inode->i_state & I_FREEING))
+	this_pass_clean = true;
+
+	mutex_lock(&c->vfs_inodes_lock);
+	list_for_each_entry(inode, &c->vfs_inodes_list, ei_vfs_inode_list) {
+		if (!snapshot_list_has_id(s, inode->ei_subvol))
 			continue;
 
-		if (!(inode->i_state & I_DONTCACHE)) {
-			d_mark_dontcache(inode);
-			d_prune_aliases(inode);
-		}
+		if (!(inode->v.i_state & I_DONTCACHE) &&
+		    !(inode->v.i_state & I_FREEING)) {
+			this_pass_clean = false;
+
+			d_mark_dontcache(&inode->v);
+			d_prune_aliases(&inode->v);
+
+			/*
+			 * If i_count was zero, we have to take and release a
+			 * ref in order for I_DONTCACHE to be noticed and the
+			 * inode to be dropped;
+			 */
+
+			if (!atomic_read(&inode->v.i_count) &&
+			    igrab(&inode->v) &&
+			    darray_push_gfp(&grabbed, inode, GFP_ATOMIC|__GFP_NOWARN))
+				break;
+		} else if (clean_pass && this_pass_clean) {
+			wait_queue_head_t *wq = bit_waitqueue(&inode->v.i_state, __I_NEW);
+			DEFINE_WAIT_BIT(wait, &inode->v.i_state, __I_NEW);
 
-		spin_lock(&inode->i_lock);
-		if (snapshot_list_has_id(s, to_bch_ei(inode)->ei_subvol) &&
-		    !(inode->i_state & I_FREEING)) {
-			wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_NEW);
-			DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
 			prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
-			spin_unlock(&inode->i_lock);
-			spin_unlock(&sb->s_inode_list_lock);
+			mutex_unlock(&c->vfs_inodes_lock);
+
 			schedule();
 			finish_wait(wq, &wait.wq_entry);
 			goto again;
 		}
+	}
+	mutex_unlock(&c->vfs_inodes_lock);
 
-		spin_unlock(&inode->i_lock);
+	darray_for_each(grabbed, i)
+		iput(&(*i)->v);
+	grabbed.nr = 0;
+
+	if (!clean_pass || !this_pass_clean) {
+		clean_pass = this_pass_clean;
+		goto again;
 	}
-	spin_unlock(&sb->s_inode_list_lock);
+
+	darray_exit(&grabbed);
 }
 
 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)