Submit max 32k reads and writes to the sftp server

Also don't limit the kernel to 64k reads and writes, rather split into
32k sized chunks and send them to the server all at once.  This is
more efficient and less demanding from the server.

Reported by Ludovic Courtès.  Fix suggested by Niels Möller

1 files changed, 261 insertions, 114 deletions

diff --git a/sshfs.c b/sshfs.c
index 08c6447..89b3974 100644
--- a/sshfs.c
+++ b/sshfs.c
@@ -169,14 +169,27 @@ struct request {
 	struct list_head list;
 };
 
+struct sshfs_io {
+	int num_reqs;
+	pthread_cond_t finished;
+	int error;
+};
+
+struct read_req {
+	struct sshfs_io *sio;
+	struct list_head list;
+	struct buffer data;
+	size_t size;
+	ssize_t res;
+};
+
 struct read_chunk {
-	sem_t ready;
 	off_t offset;
 	size_t size;
-	struct buffer data;
 	int refs;
-	int res;
 	long modifver;
+	struct list_head reqs;
+	struct sshfs_io sio;
 };
 
 struct sshfs_file {
@@ -1263,8 +1276,14 @@ static void request_free(struct request *req)
 
 static void chunk_free(struct read_chunk *chunk)
 {
-	buf_free(&chunk->data);
-	sem_destroy(&chunk->ready);
+	while (!list_empty(&chunk->reqs)) {
+		struct read_req *rreq;
+
+		rreq = list_entry(chunk->reqs.prev, struct read_req, list);
+		list_del(&rreq->list);
+		buf_free(&rreq->data);
+		g_free(rreq);
+	}
 	g_free(chunk);
 }
 
@@ -2463,131 +2482,178 @@ static int sshfs_release(const char *path, struct fuse_file_info *fi)
 	return 0;
 }
 
-static int sshfs_sync_read(struct sshfs_file *sf, char *rbuf, size_t size,
-                           off_t offset)
-{
-	int err;
-	struct buffer buf;
-	struct buffer data;
-	struct buffer *handle = &sf->handle;
-	buf_init(&buf, 0);
-	buf_add_buf(&buf, handle);
-	buf_add_uint64(&buf, offset);
-	buf_add_uint32(&buf, size);
-	err = sftp_request(SSH_FXP_READ, &buf, SSH_FXP_DATA, &data);
-	if (!err) {
-		uint32_t retsize;
-		err = -EIO;
-		if (buf_get_uint32(&data, &retsize) != -1) {
-			if (retsize > size)
-				fprintf(stderr, "long read\n");
-			else {
-				buf_get_mem(&data, rbuf, retsize);
-				err = retsize;
-			}
-		}
-		buf_free(&data);
-	} else if (err == MY_EOF)
-		err = 0;
-	buf_free(&buf);
-	return err;
-}
-
 static void sshfs_read_end(struct request *req)
 {
-	struct read_chunk *chunk = (struct read_chunk *) req->data;
+	struct read_req *rreq = (struct read_req *) req->data;
 	if (req->error)
-		chunk->res = req->error;
+		rreq->res = req->error;
 	else if (req->replied) {
-		chunk->res = -EIO;
+		rreq->res = -EIO;
 
 		if (req->reply_type == SSH_FXP_STATUS) {
 			uint32_t serr;
 			if (buf_get_uint32(&req->reply, &serr) != -1) {
 				if (serr == SSH_FX_EOF)
-					chunk->res = 0;
+					rreq->res = 0;
+			} else {
+				rreq->res = -sftp_error_to_errno(serr);
 			}
 		} else if (req->reply_type == SSH_FXP_DATA) {
 			uint32_t retsize;
 			if (buf_get_uint32(&req->reply, &retsize) != -1) {
-				if (retsize > chunk->size)
+				if (retsize > rreq->size) {
 					fprintf(stderr, "long read\n");
-				else {
-					chunk->res = retsize;
-					chunk->data = req->reply;
+				} else if (buf_check_get(&req->reply, retsize) != -1) {
+					rreq->res = retsize;
+					rreq->data = req->reply;
 					buf_init(&req->reply, 0);
 				}
 			}
-		} else
+		} else {
 			fprintf(stderr, "protocol error\n");
-	} else
-		chunk->res = -EIO;
+		}
+	} else {
+		rreq->res = -EIO;
+	}
 
-	sem_post(&chunk->ready);
-	chunk_put(chunk);
+	rreq->sio->num_reqs--;
+	if (!rreq->sio->num_reqs)
+		pthread_cond_broadcast(&rreq->sio->finished);
 }
 
 static void sshfs_read_begin(struct request *req)
 {
-	struct read_chunk *chunk = (struct read_chunk *) req->data;
-	chunk->refs++;
+	struct read_req *rreq = (struct read_req *) req->data;
+	rreq->sio->num_reqs++;
 }
 
-static void sshfs_send_async_read(struct sshfs_file *sf,
-                                  struct read_chunk *chunk)
+static struct read_chunk *sshfs_send_read(struct sshfs_file *sf, size_t size,
+					  off_t offset)
 {
-	struct buffer buf;
+	struct read_chunk *chunk = g_new0(struct read_chunk, 1);
 	struct buffer *handle = &sf->handle;
-	struct iovec iov;
 
-	buf_init(&buf, 0);
-	buf_add_buf(&buf, handle);
-	buf_add_uint64(&buf, chunk->offset);
-	buf_add_uint32(&buf, chunk->size);
-	buf_to_iov(&buf, &iov);
-	sftp_request_send(SSH_FXP_READ, &iov, 1, sshfs_read_begin,
-			  sshfs_read_end, 0, chunk, NULL);
-	buf_free(&buf);
+	pthread_cond_init(&chunk->sio.finished, NULL);
+	list_init(&chunk->reqs);
+	chunk->size = size;
+	chunk->offset = offset;
+	chunk->refs = 1;
+
+	while (size) {
+		int err;
+		struct buffer buf;
+		struct iovec iov[1];
+		struct read_req *rreq;
+		size_t bsize = size < sshfs.max_read ? size : sshfs.max_read;
+
+		rreq = g_new0(struct read_req, 1);
+		rreq->sio = &chunk->sio;
+		rreq->size = bsize;
+		buf_init(&rreq->data, 0);
+		list_add(&rreq->list, &chunk->reqs);
+
+		buf_init(&buf, 0);
+		buf_add_buf(&buf, handle);
+		buf_add_uint64(&buf, offset);
+		buf_add_uint32(&buf, bsize);
+		buf_to_iov(&buf, &iov[0]);
+		err = sftp_request_send(SSH_FXP_READ, iov, 1, 
+					sshfs_read_begin,
+					sshfs_read_end,
+					0, rreq, NULL);
+
+		buf_free(&buf);
+		if (err)
+			break;
+
+		size -= bsize;
+		offset += bsize;
+	}
+
+	return chunk;
 }
 
-static void submit_read(struct sshfs_file *sf, size_t size, off_t offset,
-                        struct read_chunk **chunkp)
+static int wait_chunk(struct read_chunk *chunk, char *buf, size_t size)
 {
-	struct read_chunk *chunk = g_new0(struct read_chunk, 1);
+	int res = 0;
+	struct read_req *rreq;
 
-	sem_init(&chunk->ready, 0, 0);
-	buf_init(&chunk->data, 0);
-	chunk->offset = offset;
-	chunk->size = size;
-	chunk->refs = 1;
-	pthread_mutex_lock(&sshfs.lock);	
-	chunk->modifver = sshfs.modifver;
-	pthread_mutex_unlock(&sshfs.lock);	
-	sshfs_send_async_read(sf, chunk);
 	pthread_mutex_lock(&sshfs.lock);
-	chunk_put(*chunkp);
-	*chunkp = chunk;
+	while (chunk->sio.num_reqs)
+	       pthread_cond_wait(&chunk->sio.finished, &sshfs.lock);
 	pthread_mutex_unlock(&sshfs.lock);
-}
 
-static int wait_chunk(struct read_chunk *chunk, char *buf, size_t size)
-{
-	int res;
-	while (sem_wait(&chunk->ready));
-	res = chunk->res;
+
+	if (chunk->sio.error) {
+		if (chunk->sio.error != MY_EOF)
+			res = chunk->sio.error;
+
+		goto out;
+	}
+
+	while (!list_empty(&chunk->reqs) && size) {
+		rreq = list_entry(chunk->reqs.prev, struct read_req, list);
+
+		if (rreq->res < 0) {
+			chunk->sio.error = rreq->res;
+			break;
+		} if (rreq->res == 0) {
+			chunk->sio.error = MY_EOF;
+			break;
+		} else if (size < (size_t) rreq->res) {
+			buf_get_mem(&rreq->data, buf, size);
+			rreq->res -= size;
+			rreq->size -= size;
+			res += size;
+			break;
+		} else {
+			buf_get_mem(&rreq->data, buf, rreq->res);
+			res += rreq->res;
+			if ((size_t) rreq->res < rreq->size) {
+				chunk->sio.error = MY_EOF;
+				break;
+			}
+			buf += rreq->res;
+			size -= rreq->res;
+			list_del(&rreq->list);
+			buf_free(&rreq->data);
+			g_free(rreq);
+		}
+	}
+
 	if (res > 0) {
-		if ((size_t) res > size)
-			res = size;
-		buf_get_mem(&chunk->data, buf, res);
 		chunk->offset += res;
 		chunk->size -= res;
-		chunk->res -= res;
 	}
-	sem_post(&chunk->ready);
+
+out:
 	chunk_put_locked(chunk);
 	return res;
 }
 
+static int sshfs_sync_read(struct sshfs_file *sf, char *buf, size_t size,
+                           off_t offset)
+{
+	struct read_chunk *chunk;
+
+	chunk = sshfs_send_read(sf, size, offset);
+	return wait_chunk(chunk, buf, size);
+}
+
+static void submit_read(struct sshfs_file *sf, size_t size, off_t offset,
+                        struct read_chunk **chunkp)
+{
+	struct read_chunk *chunk;
+
+	chunk = sshfs_send_read(sf, size, offset);
+	pthread_mutex_lock(&sshfs.lock);	
+	chunk->modifver = sshfs.modifver;
+	chunk_put(*chunkp);
+	*chunkp = chunk;
+	chunk->refs++;
+	pthread_mutex_unlock(&sshfs.lock);
+}
+
 static struct read_chunk *search_read_chunk(struct sshfs_file *sf, off_t offset)
 {
 	struct read_chunk *ch = sf->readahead;
@@ -2691,14 +2757,110 @@ static void sshfs_write_end(struct request *req)
 	sshfs_file_put(sf);
 }
 
+static int sshfs_async_write(struct sshfs_file *sf, const char *wbuf,
+			     size_t size, off_t offset)
+{
+	int err = 0;
+	struct buffer *handle = &sf->handle;
+
+	while (!err && size) {
+		struct buffer buf;
+		struct iovec iov[2];
+		size_t bsize = size < sshfs.max_write ? size : sshfs.max_write;
+
+		buf_init(&buf, 0);
+		buf_add_buf(&buf, handle);
+		buf_add_uint64(&buf, offset);
+		buf_add_uint32(&buf, bsize);
+		buf_to_iov(&buf, &iov[0]);
+		iov[1].iov_base = (void *) wbuf;
+		iov[1].iov_len = bsize;
+		err = sftp_request_send(SSH_FXP_WRITE, iov, 2,
+					sshfs_write_begin, sshfs_write_end,
+					0, sf, NULL);
+		buf_free(&buf);
+		size -= bsize;
+		wbuf += bsize;
+		offset += bsize;
+	}
+
+	return err;
+}
+
+static void sshfs_sync_write_begin(struct request *req)
+{
+	struct sshfs_io *sio = (struct sshfs_io *) req->data;
+	sio->num_reqs++;
+}
+
+static void sshfs_sync_write_end(struct request *req)
+{
+	uint32_t serr;
+	struct sshfs_io *sio = (struct sshfs_io *) req->data;
+
+	if (req->error) {
+		sio->error = req->error;
+	} else if (req->replied) {
+		if (req->reply_type != SSH_FXP_STATUS) {
+			fprintf(stderr, "protocol error\n");
+		} else if (buf_get_uint32(&req->reply, &serr) != -1 &&
+			 serr != SSH_FX_OK) {
+			sio->error = -EIO;
+		}
+	}
+	sio->num_reqs--;
+	if (!sio->num_reqs)
+		pthread_cond_broadcast(&sio->finished);
+}
+
+
+static int sshfs_sync_write(struct sshfs_file *sf, const char *wbuf,
+			    size_t size, off_t offset)
+{
+	int err = 0;
+	struct buffer *handle = &sf->handle;
+	struct sshfs_io sio = { .error = 0, .num_reqs = 0 };
+
+	pthread_cond_init(&sio.finished, NULL);
+
+	while (!err && size) {
+		struct buffer buf;
+		struct iovec iov[2];
+		size_t bsize = size < sshfs.max_write ? size : sshfs.max_write;
+
+		buf_init(&buf, 0);
+		buf_add_buf(&buf, handle);
+		buf_add_uint64(&buf, offset);
+		buf_add_uint32(&buf, bsize);
+		buf_to_iov(&buf, &iov[0]);
+		iov[1].iov_base = (void *) wbuf;
+		iov[1].iov_len = bsize;
+		err = sftp_request_send(SSH_FXP_WRITE, iov, 2,
+					sshfs_sync_write_begin,
+					sshfs_sync_write_end,
+					0, &sio, NULL);
+		buf_free(&buf);
+		size -= bsize;
+		wbuf += bsize;
+		offset += bsize;
+	}
+
+	pthread_mutex_lock(&sshfs.lock);
+	while (sio.num_reqs)
+	       pthread_cond_wait(&sio.finished, &sshfs.lock);
+	pthread_mutex_unlock(&sshfs.lock);
+
+	if (!err)
+		err = sio.error;
+
+	return err;
+}
+
 static int sshfs_write(const char *path, const char *wbuf, size_t size,
                        off_t offset, struct fuse_file_info *fi)
 {
 	int err;
-	struct buffer buf;
 	struct sshfs_file *sf = get_sshfs_file(fi);
-	struct buffer *handle = &sf->handle;
-	struct iovec iov[2];
 
 	(void) path;
 
@@ -2706,22 +2868,12 @@ static int sshfs_write(const char *path, const char *wbuf, size_t size,
 		return -EIO;
 
 	sshfs_inc_modifver();
-	buf_init(&buf, 0);
-	buf_add_buf(&buf, handle);
-	buf_add_uint64(&buf, offset);
-	buf_add_uint32(&buf, size);
-	buf_to_iov(&buf, &iov[0]);
-	iov[1].iov_base = (void *) wbuf;
-	iov[1].iov_len = size;
-	if (!sshfs.sync_write && !sf->write_error) {
-		err = sftp_request_send(SSH_FXP_WRITE, iov, 2,
-					sshfs_write_begin, sshfs_write_end,
-					0, sf, NULL);
-	} else {
-		err = sftp_request_iov(SSH_FXP_WRITE, iov, 2, SSH_FXP_STATUS,
-				       NULL);
-	}
-	buf_free(&buf);
+
+	if (!sshfs.sync_write && !sf->write_error)
+		err = sshfs_async_write(sf, wbuf, size, offset);
+	else
+		err = sshfs_sync_write(sf, wbuf, size, offset);
+
 	return err ? err : (int) size;
 }
 
@@ -3413,8 +3565,9 @@ int main(int argc, char *argv[])
 	g_thread_init(NULL);
 
 	sshfs.blksize = 4096;
-	sshfs.max_read = 65536;
-	sshfs.max_write = 65536;
+	/* SFTP spec says all servers should allow at least 32k I/O */
+	sshfs.max_read = 32768;
+	sshfs.max_write = 32768;
 	sshfs.nodelay_workaround = 1;
 	sshfs.nodelaysrv_workaround = 0;
 #if __APPLE__
@@ -3501,12 +3654,6 @@ int main(int argc, char *argv[])
 
 	if (fuse_is_lib_option("ac_attr_timeout="))
 		fuse_opt_insert_arg(&args, 1, "-oauto_cache,ac_attr_timeout=0");
-	tmp = g_strdup_printf("-omax_read=%u", sshfs.max_read);
-	fuse_opt_insert_arg(&args, 1, tmp);
-	g_free(tmp);
-	tmp = g_strdup_printf("-omax_write=%u", sshfs.max_write);
-	fuse_opt_insert_arg(&args, 1, tmp);
-	g_free(tmp);
 #if FUSE_VERSION >= 27
 	libver = fuse_version();
 	assert(libver >= 27);