[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZnwSrjqHmOzSjShI@casper.infradead.org>
Date: Wed, 26 Jun 2024 14:07:58 +0100
From: Matthew Wilcox <willy@...radead.org>
To: linux-kernel@...r.kernel.org
Cc: linux-fsdevel@...r.kernel.org, maple-tree@...ts.infradead.org
Subject: Re: [PATCH v2 2/5] rosebush: Add new data structure
On Tue, Jun 25, 2024 at 10:17:57PM +0100, Matthew Wilcox (Oracle) wrote:
> Rosebush is a resizing hash table. See
> Docuemntation/core-api/rosebush.rst for details.
I thought I had more debugging enabled than I actually did, and
there's some unbalanced RCU locking. I'll fold this fix in:
diff --git a/lib/rosebush.c b/lib/rosebush.c
index 47106a04d11d..ab2d314cecec 100644
--- a/lib/rosebush.c
+++ b/lib/rosebush.c
@@ -305,13 +305,14 @@ static int rbh_split_bucket(struct rbh *rbh, struct rbh_bucket *bucket,
rcu_read_unlock();
/* XXX: use slab */
+ err = -ENOMEM;
buckets[0] = kmalloc(sizeof(*bucket), GFP_KERNEL);
if (!buckets[0])
- return -ENOMEM;
+ goto nomem;
buckets[1] = kmalloc(sizeof(*bucket), GFP_KERNEL);
if (!buckets[1]) {
kfree(buckets[0]);
- return -ENOMEM;
+ goto nomem;
}
//printk("%s: adding buckets %p %p for hash %d\n", __func__, buckets[0], buckets[1], hash);
@@ -320,6 +321,8 @@ static int rbh_split_bucket(struct rbh *rbh, struct rbh_bucket *bucket,
table = (struct rbh_table *)(tagged & (tagged + 1));
mask = tagged - (unsigned long)table;
hash &= mask;
+
+ err = 0;
if (rbh_dereference_protected(table->buckets[hash], rbh) != bucket)
goto free;
@@ -354,14 +357,17 @@ static int rbh_split_bucket(struct rbh *rbh, struct rbh_bucket *bucket,
rbh_resize_unlock(rbh);
kvfree_rcu_mightsleep(bucket);
+ rcu_read_lock();
return 0;
free:
rbh_resize_unlock(rbh);
//printk("%s: freeing bucket %p\n", __func__, bucket);
- kfree(buckets[0]);
kfree(buckets[1]);
+nomem:
+ kfree(buckets[0]);
- return 0;
+ rcu_read_lock();
+ return err;
}
static int __rbh_insert(struct rbh *rbh, u32 hash, void *p)
Powered by blists - more mailing lists