squashfs: Enhance parallel I/O

Now squashfs have used for only one stream buffer for decompression
so it hurts parallel read performance so this patch supports
multiple decompressor to enhance performance parallel I/O.

Four 1G file dd read on KVM machine which has 2 CPU and 4G memory.

dd if=test/test1.dat of=/dev/null &
dd if=test/test2.dat of=/dev/null &
dd if=test/test3.dat of=/dev/null &
dd if=test/test4.dat of=/dev/null &

old : 1m39s -> new : 9s

* From v1
  * Change comp_strm with decomp_strm - Phillip
  * Change/add comments - Phillip

Signed-off-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Phillip Lougher <phillip@squashfs.org.uk>

3 files changed, 221 insertions, 1 deletions

diff --git a/fs/squashfs/Kconfig b/fs/squashfs/Kconfig
index c70111e..1c6d340 100644
--- a/fs/squashfs/Kconfig
+++ b/fs/squashfs/Kconfig
@@ -63,6 +63,19 @@ config SQUASHFS_LZO
 
 	  If unsure, say N.
 
+config SQUASHFS_MULTI_DECOMPRESSOR
+	bool "Use multiple decompressors for handling parallel I/O"
+	depends on SQUASHFS
+	help
+	  By default Squashfs uses a single decompressor but it gives
+	  poor performance on parallel I/O workloads when using multiple CPU
+	  machines due to waiting on decompressor availability.
+
+	  If you have a parallel I/O workload and your system has enough memory,
+	  using this option may improve overall I/O performance.
+
+	  If unsure, say N.
+
 config SQUASHFS_XZ
 	bool "Include support for XZ compressed file systems"
 	depends on SQUASHFS
diff --git a/fs/squashfs/Makefile b/fs/squashfs/Makefile
index c223c84..dfebc3b 100644
--- a/fs/squashfs/Makefile
+++ b/fs/squashfs/Makefile
@@ -4,8 +4,15 @@
 
 obj-$(CONFIG_SQUASHFS) += squashfs.o
 squashfs-y += block.o cache.o dir.o export.o file.o fragment.o id.o inode.o
-squashfs-y += namei.o super.o symlink.o decompressor.o decompressor_single.o
+squashfs-y += namei.o super.o symlink.o decompressor.o
+
 squashfs-$(CONFIG_SQUASHFS_XATTR) += xattr.o xattr_id.o
 squashfs-$(CONFIG_SQUASHFS_LZO) += lzo_wrapper.o
 squashfs-$(CONFIG_SQUASHFS_XZ) += xz_wrapper.o
 squashfs-$(CONFIG_SQUASHFS_ZLIB) += zlib_wrapper.o
+
+ifdef CONFIG_SQUASHFS_MULTI_DECOMPRESSOR
+	squashfs-y		+= decompressor_multi.o
+else
+	squashfs-y		+= decompressor_single.o
+endif
diff --git a/fs/squashfs/decompressor_multi.c b/fs/squashfs/decompressor_multi.c
new file mode 100644
index 0000000..462731d
--- /dev/null
+++ b/fs/squashfs/decompressor_multi.c
@@ -0,0 +1,200 @@
+/*
+ *  Copyright (c) 2013
+ *  Minchan Kim <minchan@kernel.org>
+ *
+ *  This work is licensed under the terms of the GNU GPL, version 2. See
+ *  the COPYING file in the top-level directory.
+ */
+#include <linux/types.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+#include <linux/buffer_head.h>
+#include <linux/sched.h>
+#include <linux/wait.h>
+#include <linux/cpumask.h>
+
+#include "squashfs_fs.h"
+#include "squashfs_fs_sb.h"
+#include "decompressor.h"
+#include "squashfs.h"
+
+/*
+ * This file implements multi-threaded decompression in the
+ * decompressor framework
+ */
+
+
+/*
+ * The reason that multiply two is that a CPU can request new I/O
+ * while it is waiting previous request.
+ */
+#define MAX_DECOMPRESSOR	(num_online_cpus() * 2)
+
+
+int squashfs_max_decompressors(void)
+{
+	return MAX_DECOMPRESSOR;
+}
+
+
+struct squashfs_stream {
+	void			*comp_opts;
+	struct list_head	strm_list;
+	struct mutex		mutex;
+	int			avail_decomp;
+	wait_queue_head_t	wait;
+};
+
+
+struct decomp_stream {
+	void *stream;
+	struct list_head list;
+};
+
+
+static void put_decomp_stream(struct decomp_stream *decomp_strm,
+				struct squashfs_stream *stream)
+{
+	mutex_lock(&stream->mutex);
+	list_add(&decomp_strm->list, &stream->strm_list);
+	mutex_unlock(&stream->mutex);
+	wake_up(&stream->wait);
+}
+
+void *squashfs_decompressor_create(struct squashfs_sb_info *msblk,
+				void *comp_opts)
+{
+	struct squashfs_stream *stream;
+	struct decomp_stream *decomp_strm = NULL;
+	int err = -ENOMEM;
+
+	stream = kzalloc(sizeof(*stream), GFP_KERNEL);
+	if (!stream)
+		goto out;
+
+	stream->comp_opts = comp_opts;
+	mutex_init(&stream->mutex);
+	INIT_LIST_HEAD(&stream->strm_list);
+	init_waitqueue_head(&stream->wait);
+
+	/*
+	 * We should have a decompressor at least as default
+	 * so if we fail to allocate new decompressor dynamically,
+	 * we could always fall back to default decompressor and
+	 * file system works.
+	 */
+	decomp_strm = kmalloc(sizeof(*decomp_strm), GFP_KERNEL);
+	if (!decomp_strm)
+		goto out;
+
+	decomp_strm->stream = msblk->decompressor->init(msblk,
+						stream->comp_opts);
+	if (IS_ERR(decomp_strm->stream)) {
+		err = PTR_ERR(decomp_strm->stream);
+		goto out;
+	}
+
+	list_add(&decomp_strm->list, &stream->strm_list);
+	stream->avail_decomp = 1;
+	return stream;
+
+out:
+	kfree(decomp_strm);
+	kfree(stream);
+	return ERR_PTR(err);
+}
+
+
+void squashfs_decompressor_destroy(struct squashfs_sb_info *msblk)
+{
+	struct squashfs_stream *stream = msblk->stream;
+	if (stream) {
+		struct decomp_stream *decomp_strm;
+
+		while (!list_empty(&stream->strm_list)) {
+			decomp_strm = list_entry(stream->strm_list.prev,
+						struct decomp_stream, list);
+			list_del(&decomp_strm->list);
+			msblk->decompressor->free(decomp_strm->stream);
+			kfree(decomp_strm);
+			stream->avail_decomp--;
+		}
+	}
+
+	WARN_ON(stream->avail_decomp);
+	kfree(stream->comp_opts);
+	kfree(stream);
+}
+
+
+static struct decomp_stream *get_decomp_stream(struct squashfs_sb_info *msblk,
+					struct squashfs_stream *stream)
+{
+	struct decomp_stream *decomp_strm;
+
+	while (1) {
+		mutex_lock(&stream->mutex);
+
+		/* There is available decomp_stream */
+		if (!list_empty(&stream->strm_list)) {
+			decomp_strm = list_entry(stream->strm_list.prev,
+				struct decomp_stream, list);
+			list_del(&decomp_strm->list);
+			mutex_unlock(&stream->mutex);
+			break;
+		}
+
+		/*
+		 * If there is no available decomp and already full,
+		 * let's wait for releasing decomp from other users.
+		 */
+		if (stream->avail_decomp >= MAX_DECOMPRESSOR)
+			goto wait;
+
+		/* Let's allocate new decomp */
+		decomp_strm = kmalloc(sizeof(*decomp_strm), GFP_KERNEL);
+		if (!decomp_strm)
+			goto wait;
+
+		decomp_strm->stream = msblk->decompressor->init(msblk,
+						stream->comp_opts);
+		if (IS_ERR(decomp_strm->stream)) {
+			kfree(decomp_strm);
+			goto wait;
+		}
+
+		stream->avail_decomp++;
+		WARN_ON(stream->avail_decomp > MAX_DECOMPRESSOR);
+
+		mutex_unlock(&stream->mutex);
+		break;
+wait:
+		/*
+		 * If system memory is tough, let's for other's
+		 * releasing instead of hurting VM because it could
+		 * make page cache thrashing.
+		 */
+		mutex_unlock(&stream->mutex);
+		wait_event(stream->wait,
+			!list_empty(&stream->strm_list));
+	}
+
+	return decomp_strm;
+}
+
+
+int squashfs_decompress(struct squashfs_sb_info *msblk,
+	void **buffer, struct buffer_head **bh, int b, int offset, int length,
+	int srclength, int pages)
+{
+	int res;
+	struct squashfs_stream *stream = msblk->stream;
+	struct decomp_stream *decomp_stream = get_decomp_stream(msblk, stream);
+	res = msblk->decompressor->decompress(msblk, decomp_stream->stream,
+		buffer, bh, b, offset, length, srclength, pages);
+	put_decomp_stream(decomp_stream, stream);
+	if (res < 0)
+		ERROR("%s decompression failed, data probably corrupt\n",
+			msblk->decompressor->name);
+	return res;
+}