fuse: implement NFS-like readdirplus support

This patch implements readdirplus support in FUSE, similar to NFS.
The payload returned in the readdirplus call contains
'fuse_entry_out' structure thereby providing all the necessary inputs
for 'faking' a lookup() operation on the spot.

If the dentry and inode already existed (for e.g. in a re-run of ls -l)
then just the inode attributes timeout and dentry timeout are refreshed.

With a simple client->network->server implementation of a FUSE based
filesystem, the following performance observations were made:

Test: Performing a filesystem crawl over 20,000 files with

sh# time ls -lR /mnt

Without readdirplus:
Run 1: 18.1s
Run 2: 16.0s
Run 3: 16.2s

With readdirplus:
Run 1: 4.1s
Run 2: 3.8s
Run 3: 3.8s

The performance improvement is significant as it avoided 20,000 upcalls
calls (lookup). Cache consistency is no worse than what already is.

Signed-off-by: Anand V. Avati <avati@redhat.com>
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>

5 files changed, 197 insertions, 5 deletions

diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
index e83351aa5bad..05c3eec298f2 100644
--- a/fs/fuse/dev.c
+++ b/fs/fuse/dev.c
@@ -491,6 +491,25 @@ void fuse_request_send_background_locked(struct fuse_conn *fc,
 	fuse_request_send_nowait_locked(fc, req);
 }
 
+void fuse_force_forget(struct file *file, u64 nodeid)
+{
+	struct inode *inode = file->f_path.dentry->d_inode;
+	struct fuse_conn *fc = get_fuse_conn(inode);
+	struct fuse_req *req;
+	struct fuse_forget_in inarg;
+
+	memset(&inarg, 0, sizeof(inarg));
+	inarg.nlookup = 1;
+	req = fuse_get_req_nofail(fc, file);
+	req->in.h.opcode = FUSE_FORGET;
+	req->in.h.nodeid = nodeid;
+	req->in.numargs = 1;
+	req->in.args[0].size = sizeof(inarg);
+	req->in.args[0].value = &inarg;
+	req->isreply = 0;
+	fuse_request_send_nowait(fc, req);
+}
+
 /*
  * Lock the request.  Up to the next unlock_request() there mustn't be
  * anything that could cause a page-fault.  If the request was already
diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index b7c09f9eb40c..dcc1e522c7d4 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -1155,6 +1155,143 @@ static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
 	return 0;
 }
 
+static int fuse_direntplus_link(struct file *file,
+				struct fuse_direntplus *direntplus,
+				u64 attr_version)
+{
+	int err;
+	struct fuse_entry_out *o = &direntplus->entry_out;
+	struct fuse_dirent *dirent = &direntplus->dirent;
+	struct dentry *parent = file->f_path.dentry;
+	struct qstr name = QSTR_INIT(dirent->name, dirent->namelen);
+	struct dentry *dentry;
+	struct dentry *alias;
+	struct inode *dir = parent->d_inode;
+	struct fuse_conn *fc;
+	struct inode *inode;
+
+	if (!o->nodeid) {
+		/*
+		 * Unlike in the case of fuse_lookup, zero nodeid does not mean
+		 * ENOENT. Instead, it only means the userspace filesystem did
+		 * not want to return attributes/handle for this entry.
+		 *
+		 * So do nothing.
+		 */
+		return 0;
+	}
+
+	if (name.name[0] == '.') {
+		/*
+		 * We could potentially refresh the attributes of the directory
+		 * and its parent?
+		 */
+		if (name.len == 1)
+			return 0;
+		if (name.name[1] == '.' && name.len == 2)
+			return 0;
+	}
+	fc = get_fuse_conn(dir);
+
+	name.hash = full_name_hash(name.name, name.len);
+	dentry = d_lookup(parent, &name);
+	if (dentry && dentry->d_inode) {
+		inode = dentry->d_inode;
+		if (get_node_id(inode) == o->nodeid) {
+			struct fuse_inode *fi;
+			fi = get_fuse_inode(inode);
+			spin_lock(&fc->lock);
+			fi->nlookup++;
+			spin_unlock(&fc->lock);
+
+			/*
+			 * The other branch to 'found' comes via fuse_iget()
+			 * which bumps nlookup inside
+			 */
+			goto found;
+		}
+		err = d_invalidate(dentry);
+		if (err)
+			goto out;
+		dput(dentry);
+		dentry = NULL;
+	}
+
+	dentry = d_alloc(parent, &name);
+	err = -ENOMEM;
+	if (!dentry)
+		goto out;
+
+	inode = fuse_iget(dir->i_sb, o->nodeid, o->generation,
+			  &o->attr, entry_attr_timeout(o), attr_version);
+	if (!inode)
+		goto out;
+
+	alias = d_materialise_unique(dentry, inode);
+	err = PTR_ERR(alias);
+	if (IS_ERR(alias))
+		goto out;
+	if (alias) {
+		dput(dentry);
+		dentry = alias;
+	}
+
+found:
+	fuse_change_attributes(inode, &o->attr, entry_attr_timeout(o),
+			       attr_version);
+
+	fuse_change_entry_timeout(dentry, o);
+
+	err = 0;
+out:
+	if (dentry)
+		dput(dentry);
+	return err;
+}
+
+static int parse_dirplusfile(char *buf, size_t nbytes, struct file *file,
+			     void *dstbuf, filldir_t filldir, u64 attr_version)
+{
+	struct fuse_direntplus *direntplus;
+	struct fuse_dirent *dirent;
+	size_t reclen;
+	int over = 0;
+	int ret;
+
+	while (nbytes >= FUSE_NAME_OFFSET_DIRENTPLUS) {
+		direntplus = (struct fuse_direntplus *) buf;
+		dirent = &direntplus->dirent;
+		reclen = FUSE_DIRENTPLUS_SIZE(direntplus);
+
+		if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
+			return -EIO;
+		if (reclen > nbytes)
+			break;
+
+		if (!over) {
+			/* We fill entries into dstbuf only as much as
+			   it can hold. But we still continue iterating
+			   over remaining entries to link them. If not,
+			   we need to send a FORGET for each of those
+			   which we did not link.
+			*/
+			over = filldir(dstbuf, dirent->name, dirent->namelen,
+				       file->f_pos, dirent->ino,
+				       dirent->type);
+			file->f_pos = dirent->off;
+		}
+
+		buf += reclen;
+		nbytes -= reclen;
+
+		ret = fuse_direntplus_link(file, direntplus, attr_version);
+		if (ret)
+			fuse_force_forget(file, direntplus->entry_out.nodeid);
+	}
+
+	return 0;
+}
+
 static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
 {
 	int err;
@@ -1163,6 +1300,7 @@ static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
 	struct inode *inode = file->f_path.dentry->d_inode;
 	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_req *req;
+	u64 attr_version = 0;
 
 	if (is_bad_inode(inode))
 		return -EIO;
@@ -1179,14 +1317,28 @@ static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
 	req->out.argpages = 1;
 	req->num_pages = 1;
 	req->pages[0] = page;
-	fuse_read_fill(req, file, file->f_pos, PAGE_SIZE, FUSE_READDIR);
+	if (fc->do_readdirplus) {
+		attr_version = fuse_get_attr_version(fc);
+		fuse_read_fill(req, file, file->f_pos, PAGE_SIZE,
+			       FUSE_READDIRPLUS);
+	} else {
+		fuse_read_fill(req, file, file->f_pos, PAGE_SIZE,
+			       FUSE_READDIR);
+	}
 	fuse_request_send(fc, req);
 	nbytes = req->out.args[0].size;
 	err = req->out.h.error;
 	fuse_put_request(fc, req);
-	if (!err)
-		err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
-				    filldir);
+	if (!err) {
+		if (fc->do_readdirplus) {
+			err = parse_dirplusfile(page_address(page), nbytes,
+						file, dstbuf, filldir,
+						attr_version);
+		} else {
+			err = parse_dirfile(page_address(page), nbytes, file,
+					    dstbuf, filldir);
+		}
+	}
 
 	__free_page(page);
 	fuse_invalidate_attr(inode); /* atime changed */
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index e105a53fc72d..5c5055306d3c 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -487,6 +487,9 @@ struct fuse_conn {
 	/** Use enhanced/automatic page cache invalidation. */
 	unsigned auto_inval_data:1;
 
+	/** Does the filesystem support readdir-plus? */
+	unsigned do_readdirplus:1;
+
 	/** The number of requests waiting for completion */
 	atomic_t num_waiting;
 
@@ -578,6 +581,9 @@ void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
 
 struct fuse_forget_link *fuse_alloc_forget(void);
 
+/* Used by READDIRPLUS */
+void fuse_force_forget(struct file *file, u64 nodeid);
+
 /**
  * Initialize READ or READDIR request
  */
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index 73ca6b72beaf..6f7d5746bf52 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -863,6 +863,8 @@ static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 				fc->dont_mask = 1;
 			if (arg->flags & FUSE_AUTO_INVAL_DATA)
 				fc->auto_inval_data = 1;
+			if (arg->flags & FUSE_DO_READDIRPLUS)
+				fc->do_readdirplus = 1;
 		} else {
 			ra_pages = fc->max_read / PAGE_CACHE_SIZE;
 			fc->no_lock = 1;
@@ -889,7 +891,8 @@ static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 	arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
-		FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA;
+		FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
+		FUSE_DO_READDIRPLUS;
 	req->in.h.opcode = FUSE_INIT;
 	req->in.numargs = 1;
 	req->in.args[0].size = sizeof(*arg);
diff --git a/include/uapi/linux/fuse.h b/include/uapi/linux/fuse.h
index d8c713e148e3..5dc1fea49ecd 100644
--- a/include/uapi/linux/fuse.h
+++ b/include/uapi/linux/fuse.h
@@ -193,6 +193,7 @@ struct fuse_file_lock {
 #define FUSE_FLOCK_LOCKS	(1 << 10)
 #define FUSE_HAS_IOCTL_DIR	(1 << 11)
 #define FUSE_AUTO_INVAL_DATA	(1 << 12)
+#define FUSE_DO_READDIRPLUS	(1 << 13)
 
 /**
  * CUSE INIT request/reply flags
@@ -299,6 +300,7 @@ enum fuse_opcode {
 	FUSE_NOTIFY_REPLY  = 41,
 	FUSE_BATCH_FORGET  = 42,
 	FUSE_FALLOCATE     = 43,
+	FUSE_READDIRPLUS   = 44,
 
 	/* CUSE specific operations */
 	CUSE_INIT          = 4096,
@@ -630,6 +632,16 @@ struct fuse_dirent {
 #define FUSE_DIRENT_SIZE(d) \
 	FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + (d)->namelen)
 
+struct fuse_direntplus {
+	struct fuse_entry_out entry_out;
+	struct fuse_dirent dirent;
+};
+
+#define FUSE_NAME_OFFSET_DIRENTPLUS \
+	offsetof(struct fuse_direntplus, dirent.name)
+#define FUSE_DIRENTPLUS_SIZE(d) \
+	FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET_DIRENTPLUS + (d)->dirent.namelen)
+
 struct fuse_notify_inval_inode_out {
 	__u64	ino;
 	__s64	off;