[PATCH 14/19] fs: introduce a per-cpu last_ino allocator

[Dave Chinner]

From: Eric Dumazet <eric.dumazet@xxxxxxxxx>

new_inode() dirties a contended cache line to get increasing
inode numbers. This limits performance on workloads that cause
significant parallel inode allocation.

Solve this problem by using a per_cpu variable fed by the shared
last_ino in batches of 1024 allocations.  This reduces contention on
the shared last_ino, and give same spreading ino numbers than before
(i.e. same wraparound after 2^32 allocations).

Signed-off-by: Eric Dumazet <eric.dumazet@xxxxxxxxx>
Signed-off-by: Nick Piggin <npiggin@xxxxxxx>
Signed-off-by: Dave Chinner <dchinner@xxxxxxxxxx>
Reviewed-by: Christoph Hellwig <hch@xxxxxx>
---
 fs/inode.c |   45 ++++++++++++++++++++++++++++++++++++++-------
 1 files changed, 38 insertions(+), 7 deletions(-)

diff --git a/fs/inode.c b/fs/inode.c
index 1ac9a61..05852a1 100644
--- a/fs/inode.c
+++ b/fs/inode.c
@@ -850,6 +850,43 @@ void inode_add_to_lists(struct super_block *sb, struct inode *inode)
 }
 EXPORT_SYMBOL_GPL(inode_add_to_lists);
 
+/*
+ * Each cpu owns a range of LAST_INO_BATCH numbers.
+ * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
+ * to renew the exhausted range.
+ *
+ * This does not significantly increase overflow rate because every CPU can
+ * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
+ * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
+ * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
+ * overflow rate by 2x, which does not seem too significant.
+ *
+ * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
+ * error if st_ino won't fit in target struct field. Use 32bit counter
+ * here to attempt to avoid that.
+ */
+#define LAST_INO_BATCH 1024
+static DEFINE_PER_CPU(unsigned int, last_ino);
+
+static unsigned int get_next_ino(void)
+{
+	unsigned int *p = &get_cpu_var(last_ino);
+	unsigned int res = *p;
+
+#ifdef CONFIG_SMP
+	if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
+		static atomic_t shared_last_ino;
+		int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
+
+		res = next - LAST_INO_BATCH;
+	}
+#endif
+
+	*p = ++res;
+	put_cpu_var(last_ino);
+	return res;
+}
+
 /**
  *	new_inode 	- obtain an inode
  *	@sb: superblock
@@ -864,12 +901,6 @@ EXPORT_SYMBOL_GPL(inode_add_to_lists);
  */
 struct inode *new_inode(struct super_block *sb)
 {
-	/*
-	 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
-	 * error if st_ino won't fit in target struct field. Use 32bit counter
-	 * here to attempt to avoid that.
-	 */
-	static unsigned int last_ino;
 	struct inode *inode;
 
 	spin_lock_prefetch(&inode_lock);
@@ -882,7 +913,7 @@ struct inode *new_inode(struct super_block *sb)
 		 * set the inode state before we make the inode accessible to
 		 * the outside world.
 		 */
-		inode->i_ino = ++last_ino;
+		inode->i_ino = get_next_ino();
 		inode->i_state = 0;
 		__inode_add_to_lists(sb, NULL, inode);
 		spin_unlock(&inode_lock);
-- 
1.7.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/