Released 2.5

Accessing directories with many (several thousand) files over sshfs is
slow, because most SFTP server implementations only send a fixed amount
of entries per READDIR command (e.g. OpenSSH SFTP: 100 entries). This
patch implements sending several READDIR commands in parallel, in order
to speed up directory listing in these cases.

An option (sync_readdir) is also added so that users can easily switch
on the old behaviour.

The performance improvement is astonishing. Accessing a directory with
30k files in from a remote server that has a RTT of 15ms via OpenSSH
SFTP:

Synchronous readdir:
 $ ./sshfs -o sync_readdir host:/tmp /mnt/temp
 $ time "ls" -1 /mnt/temp/test | wc -l
 30000
 "ls" -1 /mnt/temp/test  0.07s user 0.01s system 1% cpu 6.928 total

Asynchronous readdir:
 $ ./sshfs host:/tmp /mnt/temp
 $ time "ls" -1 /mnt/temp/test | wc -l
 30000
 "ls" -1 /mnt/temp/test  0.07s user 0.01s system 12% cpu 0.605 total

Accessing a directory with 100k files shows even more dramatic
improvement:

Synchronous readdir:
 $ ./sshfs -o sync_readdir host:/tmp /mnt/temp
 $ time "ls" -1 /mnt/temp/test2 | wc -l
 100000
 "ls" -1 /mnt/temp/test2  0.67s user 1.22s system 0% cpu 3:31.56 total

Asynchronous readdir:
 $ ./sshfs host:/tmp /mnt/temp
 $ time "ls" -1 /mnt/temp/test2 | wc -l
 100000
 "ls" -1 /mnt/temp/test2  0.20s user 0.03s system 14% cpu 1.631 total

This can easily be reproduced by creating a directory on a server and
touching a lot of files in it:
 $ mkdir /tmp/test
 $ cd /tmp/test
 $ for i in $(seq 1 30000); do touch $i; done

Signed-off-by: Alexander Neumann <alexander@bumpern.de>

ChangeLog

@@ -1,3 +1,27 @@
+2014-01-14  Miklos Szeredi <miklos@szeredi.hu>
+
+	* Released 2.5
+
+2014-01-08  Miklos Szeredi <miklos@szeredi.hu>
+
+	* Add -o disable_hardlink option (debian bug #670926).  Reported
+	by Louis-David Mitterrand
+
+	* Optimize readdir by sending multiple requests in parallel.  Add
+	-o sync_readdir to restore old behavior.  Patch by Alexander
+	Neumann
+
+2014-01-07  Miklos Szeredi <miklos@szeredi.hu>
+
+	* Map SSH2_FX_FAILURE to ENOTEMPTY for rmdir.  Reported by Ross
+	Lagerwall
+
+2012-05-14  Miklos Szeredi <miklos@szeredi.hu>
+
+	* When checking root directory use LSTAT not STAT.  This prevents
+	I/O error being returned after a successful mount if a symlink is
+	mounted.  Reported by Bart Friederichs
+
 2012-03-08  Miklos Szeredi <miklos@szeredi.hu>
 
 	* Released 2.4

configure.ac

@@ -1,4 +1,4 @@
-AC_INIT(sshfs-fuse, 2.4)
+AC_INIT(sshfs-fuse, 2.5)
 AM_INIT_AUTOMAKE
 AM_CONFIG_HEADER(config.h)
 

sshfs.1

@@ -58,6 +58,9 @@ synchronous writes
 \fB\-o\fR no_readahead
 synchronous reads (no speculative readahead)
 .TP
+\fB\-o\fR sync_readdir
+synchronous readdir
+.TP
 \fB\-o\fR sshfs_debug
 print some debugging information
 .TP
@@ -138,9 +141,15 @@ path to sftp server or subsystem (default: sftp)
 .TP
 \fB\-o\fR directport=PORT
 directly connect to PORT bypassing ssh
+.TP
 \fB\-o\fR slave
 communicate over stdin and stdout bypassing network
 .TP
+\fB\-o\fR disable_hardlink
+link(2) will return with errno set to ENOSYS.  Hard links don't currently work
+perfectly on sshfs, and this confuses some programs.  If that happens try
+disabling hard links with this option.
+.TP
 \fB\-o\fR transform_symlinks
 transform absolute symlinks to relative
 .TP
@@ -256,7 +265,7 @@ perform reads synchronously
 \fB\-o\fR subdir=DIR
 prepend this directory to all paths (mandatory)
 .TP
-\fB\-o\fR [no]rellinksa
+\fB\-o\fR [no]rellinks
 transform absolute symlinks to relative
 .TP
 [iconv]

sshfs.c

@@ -121,6 +121,10 @@
 
 #define SSHNODELAY_SO "sshnodelay.so"
 
+/* Asynchronous readdir parameters */
+#define READDIR_START 2
+#define READDIR_MAX 32
+
 struct buffer {
 	uint8_t *p;
 	size_t len;
@@ -139,6 +143,7 @@ struct request {
 	unsigned int want_reply;
 	sem_t ready;
 	uint8_t reply_type;
+	uint32_t id;
 	int replied;
 	int error;
 	struct buffer reply;
@@ -203,6 +208,7 @@ struct sshfs {
 	int detect_uid;
 	int idmap;
 	int nomap;
+	int disable_hardlink;
 	char *uid_file;
 	char *gid_file;
 	GHashTable *uid_map;
@@ -214,6 +220,7 @@ struct sshfs {
 	unsigned ssh_ver;
 	int sync_write;
 	int sync_read;
+	int sync_readdir;
 	int debug;
 	int foreground;
 	int reconnect;
@@ -350,6 +357,7 @@ static struct fuse_opt sshfs_opts[] = {
 	SSHFS_OPT("nomap=error",       nomap, NOMAP_ERROR),
 	SSHFS_OPT("sshfs_sync",        sync_write, 1),
 	SSHFS_OPT("no_readahead",      sync_read, 1),
+	SSHFS_OPT("sync_readdir",      sync_readdir, 1),
 	SSHFS_OPT("sshfs_debug",       debug, 1),
 	SSHFS_OPT("reconnect",         reconnect, 1),
 	SSHFS_OPT("transform_symlinks", transform_symlinks, 1),
@@ -358,6 +366,7 @@ static struct fuse_opt sshfs_opts[] = {
 	SSHFS_OPT("password_stdin",    password_stdin, 1),
 	SSHFS_OPT("delay_connect",     delay_connect, 1),
 	SSHFS_OPT("slave",             slave, 1),
+	SSHFS_OPT("disable_hardlink",  disable_hardlink, 1),
 
 	FUSE_OPT_KEY("-p ",            KEY_PORT),
 	FUSE_OPT_KEY("-C",             KEY_COMPRESS),
@@ -1620,7 +1629,7 @@ static int sftp_check_root(const char *base_path)
 	buf_init(&buf, 0);
 	buf_add_string(&buf, remote_dir);
 	buf_to_iov(&buf, &iov[0]);
-	if (sftp_send_iov(SSH_FXP_STAT, id, iov, 1) == -1)
+	if (sftp_send_iov(SSH_FXP_LSTAT, id, iov, 1) == -1)
 		goto out;
 	buf_clear(&buf);
 	if (sftp_read(&type, &buf) == -1)
@@ -1790,6 +1799,13 @@ static int sftp_request_wait(struct request *req, uint8_t type,
 				err = -EIO;
 			break;
 
+		case SSH_FX_FAILURE:
+			if (type == SSH_FXP_RMDIR)
+				err = -ENOTEMPTY;
+			else
+				err = -EPERM;
+			break;
+
 		default:
 			err = -sftp_error_to_errno(serr);
 		}
@@ -1827,6 +1843,7 @@ static int sftp_request_send(uint8_t type, struct iovec *iov, size_t count,
 	if (begin_func)
 		begin_func(req);
 	id = sftp_get_id();
+	req->id = id;
 	err = start_processing_thread();
 	if (err) {
 		pthread_mutex_unlock(&sshfs.lock);
@@ -2014,6 +2031,113 @@ static int sshfs_readlink(const char *path, char *linkbuf, size_t size)
 	return err;
 }
 
+static int sftp_readdir_send(struct request **req, struct buffer *handle)
+{
+	struct iovec iov;
+
+	buf_to_iov(handle, &iov);
+	return sftp_request_send(SSH_FXP_READDIR, &iov, 1, NULL, NULL,
+				 SSH_FXP_NAME, NULL, req);
+}
+
+static int sshfs_req_pending(struct request *req)
+{
+	if (g_hash_table_lookup(sshfs.reqtab, GUINT_TO_POINTER(req->id)))
+		return 1;
+	else
+		return 0;
+}
+
+static int sftp_readdir_async(struct buffer *handle, fuse_cache_dirh_t h,
+			      fuse_cache_dirfil_t filler)
+{
+	int err = 0;
+	int outstanding = 0;
+	int max = READDIR_START;
+	GList *list = NULL;
+
+	int done = 0;
+
+	while (!done || outstanding) {
+		struct request *req;
+		struct buffer name;
+		int tmperr;
+
+		while (!done && outstanding < max) {
+			tmperr = sftp_readdir_send(&req, handle);
+
+			if (tmperr && !done) {
+				err = tmperr;
+				done = 1;
+				break;
+			}
+
+			list = g_list_append(list, req);
+			outstanding++;
+		}
+
+		if (outstanding) {
+			GList *first;
+			/* wait for response to next request */
+			first = g_list_first(list);
+			req = first->data;
+			list = g_list_delete_link(list, first);
+			outstanding--;
+
+			if (done) {
+				pthread_mutex_lock(&sshfs.lock);
+				if (sshfs_req_pending(req))
+					req->want_reply = 0;
+				pthread_mutex_unlock(&sshfs.lock);
+				if (!req->want_reply)
+					continue;
+			}
+
+			tmperr = sftp_request_wait(req, SSH_FXP_READDIR,
+						    SSH_FXP_NAME, &name);
+
+			if (tmperr && !done) {
+				err = tmperr;
+				if (err == MY_EOF)
+					err = 0;
+				done = 1;
+			}
+			if (!done) {
+				err = buf_get_entries(&name, h, filler);
+				buf_free(&name);
+
+				/* increase number of outstanding requests */
+				if (max < READDIR_MAX)
+					max++;
+
+				if (err)
+					done = 1;
+			}
+		}
+	}
+	assert(list == NULL);
+
+	return err;
+}
+
+static int sftp_readdir_sync(struct buffer *handle, fuse_cache_dirh_t h,
+			     fuse_cache_dirfil_t filler)
+{
+	int err;
+	do {
+		struct buffer name;
+		err = sftp_request(SSH_FXP_READDIR, handle, SSH_FXP_NAME, &name);
+		if (!err) {
+			err = buf_get_entries(&name, h, filler);
+			buf_free(&name);
+		}
+	} while (!err);
+	if (err == MY_EOF)
+		err = 0;
+
+	return err;
+}
+
 static int sshfs_getdir(const char *path, fuse_cache_dirh_t h,
                         fuse_cache_dirfil_t filler)
 {
@@ -2026,16 +2150,11 @@ static int sshfs_getdir(const char *path, fuse_cache_dirh_t h,
 	if (!err) {
 		int err2;
 		buf_finish(&handle);
-		do {
-			struct buffer name;
-			err = sftp_request(SSH_FXP_READDIR, &handle, SSH_FXP_NAME, &name);
-			if (!err) {
-				err = buf_get_entries(&name, h, filler);
-				buf_free(&name);
-			}
-		} while (!err);
-		if (err == MY_EOF)
-			err = 0;
+
+		if (sshfs.sync_readdir)
+			err = sftp_readdir_sync(&handle, h, filler);
+		else
+			err = sftp_readdir_async(&handle, h, filler);
 
 		err2 = sftp_request(SSH_FXP_CLOSE, &handle, 0, NULL);
 		if (!err)
@@ -2192,7 +2311,7 @@ static int sshfs_link(const char *from, const char *to)
 {
 	int err = -ENOSYS;
 
-	if (sshfs.ext_hardlink) {
+	if (sshfs.ext_hardlink && !sshfs.disable_hardlink) {
 		struct buffer buf;
 
 		buf_init(&buf, 0);
@@ -3162,6 +3281,7 @@ static void usage(const char *progname)
 "    -o delay_connect       delay connection to server\n"
 "    -o sshfs_sync          synchronous writes\n"
 "    -o no_readahead        synchronous reads (no speculative readahead)\n"
+"    -o sync_readdir        synchronous readdir\n"
 "    -o sshfs_debug         print some debugging information\n"
 "    -o cache=BOOL          enable caching {yes,no} (default: yes)\n"
 "    -o cache_timeout=N     sets timeout for caches in seconds (default: 20)\n"
@@ -3190,6 +3310,7 @@ static void usage(const char *progname)
 "    -o sftp_server=SERV    path to sftp server or subsystem (default: sftp)\n"
 "    -o directport=PORT     directly connect to PORT bypassing ssh\n"
 "    -o slave               communicate over stdin and stdout bypassing network\n"
+"    -o disable_hardlink    link(2) will return with errno set to ENOSYS\n"
 "    -o transform_symlinks  transform absolute symlinks to relative\n"
 "    -o follow_symlinks     follow symlinks on the server\n"
 "    -o no_check_root       don't check for existence of 'dir' on server\n"
@@ -3863,6 +3984,10 @@ int main(int argc, char *argv[])
 			exit(1);
 		}
 
+		/*
+		 * FIXME: trim $PATH so it doesn't contain anything inside the
+		 * mountpoint, which would deadlock.
+		 */
 		res = ssh_connect();
 		if (res == -1) {
 			fuse_unmount(mountpoint, ch);