[PATCH] fuse: fix async read for legacy filesystems

While asynchronous reads mean a performance improvement in most cases, if
the filesystem assumed that reads are synchronous, then async reads may
degrade performance (filesystem may receive reads out of order, which can
confuse it's own readahead logic).

With sshfs a 1.5 to 4 times slowdown can be measured.

There's also a need for userspace filesystems to know whether asynchronous
reads are supported by the kernel or not.

To achive these, negotiate in the INIT request whether async reads will be
used and the maximum readahead value.  Update interface version to 7.6

If userspace uses a version earlier than 7.6, then disable async reads, and
set maximum readahead value to the maximum read size, as done in previous
versions.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

3 files changed, 22 insertions, 4 deletions

diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index a7ef5e7..2963516 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -335,9 +335,14 @@ static void fuse_send_readpages(struct fuse_req *req, struct file *file,
 	loff_t pos = page_offset(req->pages[0]);
 	size_t count = req->num_pages << PAGE_CACHE_SHIFT;
 	req->out.page_zeroing = 1;
-	req->end = fuse_readpages_end;
 	fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
-	request_send_background(fc, req);
+	if (fc->async_read) {
+		req->end = fuse_readpages_end;
+		request_send_background(fc, req);
+	} else {
+		request_send(fc, req);
+		fuse_readpages_end(fc, req);
+	}
 }
 
 struct fuse_readpages_data {
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index 46cf933..4a83adf 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -272,6 +272,9 @@ struct fuse_conn {
 	    reply, before any other request, and never cleared */
 	unsigned conn_error : 1;
 
+	/** Do readpages asynchronously?  Only set in INIT */
+	unsigned async_read : 1;
+
 	/*
 	 * The following bitfields are only for optimization purposes
 	 * and hence races in setting them will not cause malfunction
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index c755a04..879e6fb 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -473,6 +473,16 @@ static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 	if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 		fc->conn_error = 1;
 	else {
+		unsigned long ra_pages;
+
+		if (arg->minor >= 6) {
+			ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
+			if (arg->flags & FUSE_ASYNC_READ)
+				fc->async_read = 1;
+		} else
+			ra_pages = fc->max_read / PAGE_CACHE_SIZE;
+
+		fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 		fc->minor = arg->minor;
 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 	}
@@ -496,6 +506,8 @@ static void fuse_send_init(struct fuse_conn *fc)
 
 	arg->major = FUSE_KERNEL_VERSION;
 	arg->minor = FUSE_KERNEL_MINOR_VERSION;
+	arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
+	arg->flags |= FUSE_ASYNC_READ;
 	req->in.h.opcode = FUSE_INIT;
 	req->in.numargs = 1;
 	req->in.args[0].size = sizeof(*arg);
@@ -552,8 +564,6 @@ static int fuse_fill_super(struct super_block *sb, void *data, int silent)
 	fc->user_id = d.user_id;
 	fc->group_id = d.group_id;
 	fc->max_read = d.max_read;
-	if (fc->max_read / PAGE_CACHE_SIZE < fc->bdi.ra_pages)
-		fc->bdi.ra_pages = fc->max_read / PAGE_CACHE_SIZE;
 
 	/* Used by get_root_inode() */
 	sb->s_fs_info = fc;