[PATCH] v9fs: new multiplexer implementation

New multiplexer implementation. Decreases the number of kernel threads
required. Better handling when the user process receives a signal.

Signed-off-by: Latchesar Ionkov <lucho@ionkov.net>
Cc: Eric Van Hensbergen <ericvh@ericvh.myip.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

1 files changed, 794 insertions, 328 deletions

diff --git a/fs/9p/mux.c b/fs/9p/mux.c
index 8835b576f74..62b6ad0767e 100644
--- a/fs/9p/mux.c
+++ b/fs/9p/mux.c
@@ -4,7 +4,7 @@
  * Protocol Multiplexer
  *
  *  Copyright (C) 2004 by Eric Van Hensbergen <ericvh@gmail.com>
- *  Copyright (C) 2004 by Latchesar Ionkov <lucho@ionkov.net>
+ *  Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
  *
  *  This program is free software; you can redistribute it and/or modify
  *  it under the terms of the GNU General Public License as published by
@@ -28,6 +28,7 @@
 #include <linux/module.h>
 #include <linux/errno.h>
 #include <linux/fs.h>
+#include <linux/poll.h>
 #include <linux/kthread.h>
 #include <linux/idr.h>
 
@@ -38,438 +39,903 @@
 #include "conv.h"
 #include "mux.h"
 
-/**
- * dprintcond - print condition of session info
- * @v9ses: session info structure
- * @req: RPC request structure
- *
- */
+#define ERREQFLUSH	1
+#define SCHED_TIMEOUT	10
+#define MAXPOLLWADDR	2
+
+enum {
+	Rworksched = 1,		/* read work scheduled or running */
+	Rpending = 2,		/* can read */
+	Wworksched = 4,		/* write work scheduled or running */
+	Wpending = 8,		/* can write */
+};
+
+struct v9fs_mux_poll_task;
+
+struct v9fs_req {
+	int tag;
+	struct v9fs_fcall *tcall;
+	struct v9fs_fcall *rcall;
+	int err;
+	v9fs_mux_req_callback cb;
+	void *cba;
+	struct list_head req_list;
+};
+
+struct v9fs_mux_data {
+	spinlock_t lock;
+	struct list_head mux_list;
+	struct v9fs_mux_poll_task *poll_task;
+	int msize;
+	unsigned char *extended;
+	struct v9fs_transport *trans;
+	struct v9fs_idpool tidpool;
+	int err;
+	wait_queue_head_t equeue;
+	struct list_head req_list;
+	struct list_head unsent_req_list;
+	int rpos;
+	char *rbuf;
+	int wpos;
+	int wsize;
+	char *wbuf;
+	wait_queue_t poll_wait[MAXPOLLWADDR];
+	wait_queue_head_t *poll_waddr[MAXPOLLWADDR];
+	poll_table pt;
+	struct work_struct rq;
+	struct work_struct wq;
+	unsigned long wsched;
+};
+
+struct v9fs_mux_poll_task {
+	struct task_struct *task;
+	struct list_head mux_list;
+	int muxnum;
+};
+
+struct v9fs_mux_rpc {
+	struct v9fs_mux_data *m;
+	struct v9fs_req *req;
+	int err;
+	struct v9fs_fcall *rcall;
+	wait_queue_head_t wqueue;
+};
+
+static int v9fs_poll_proc(void *);
+static void v9fs_read_work(void *);
+static void v9fs_write_work(void *);
+static void v9fs_pollwait(struct file *filp, wait_queue_head_t * wait_address,
+			  poll_table * p);
+
+static DECLARE_MUTEX(v9fs_mux_task_lock);
+static struct workqueue_struct *v9fs_mux_wq;
+
+static int v9fs_mux_num;
+static int v9fs_mux_poll_task_num;
+static struct v9fs_mux_poll_task v9fs_mux_poll_tasks[100];
+
+void v9fs_mux_global_init(void)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(v9fs_mux_poll_tasks); i++)
+		v9fs_mux_poll_tasks[i].task = NULL;
+
+	v9fs_mux_wq = create_workqueue("v9fs");
+}
 
-static inline int
-dprintcond(struct v9fs_session_info *v9ses, struct v9fs_rpcreq *req)
+void v9fs_mux_global_exit(void)
 {
-	dprintk(DEBUG_MUX, "condition: %d, %p\n", v9ses->transport->status,
-		req->rcall);
-	return 0;
+	destroy_workqueue(v9fs_mux_wq);
 }
 
 /**
- * xread - force read of a certain number of bytes
- * @v9ses: session info structure
- * @ptr: pointer to buffer
- * @sz: number of bytes to read
+ * v9fs_mux_calc_poll_procs - calculates the number of polling procs
+ * based on the number of mounted v9fs filesystems.
  *
- * Chuck Cranor CS-533 project1
+ * The current implementation returns sqrt of the number of mounts.
  */
+inline int v9fs_mux_calc_poll_procs(int muxnum)
+{
+	int n;
+
+	if (v9fs_mux_poll_task_num)
+		n = muxnum / v9fs_mux_poll_task_num +
+		    (muxnum % v9fs_mux_poll_task_num ? 1 : 0);
+	else
+		n = 1;
+
+	if (n > ARRAY_SIZE(v9fs_mux_poll_tasks))
+		n = ARRAY_SIZE(v9fs_mux_poll_tasks);
 
-static int xread(struct v9fs_session_info *v9ses, void *ptr, unsigned long sz)
+	return n;
+}
+
+static void v9fs_mux_poll_start(struct v9fs_mux_data *m)
 {
-	int rd = 0;
-	int ret = 0;
-	while (rd < sz) {
-		ret = v9ses->transport->read(v9ses->transport, ptr, sz - rd);
-		if (ret <= 0) {
-			dprintk(DEBUG_ERROR, "xread errno %d\n", ret);
-			return ret;
+	int i, n;
+	struct v9fs_mux_poll_task *vpt, *vptlast;
+
+	dprintk(DEBUG_MUX, "mux %p muxnum %d procnum %d\n", m, v9fs_mux_num,
+		v9fs_mux_poll_task_num);
+	up(&v9fs_mux_task_lock);
+
+	n = v9fs_mux_calc_poll_procs(v9fs_mux_num + 1);
+	if (n > v9fs_mux_poll_task_num) {
+		for (i = 0; i < ARRAY_SIZE(v9fs_mux_poll_tasks); i++) {
+			if (v9fs_mux_poll_tasks[i].task == NULL) {
+				vpt = &v9fs_mux_poll_tasks[i];
+				dprintk(DEBUG_MUX, "create proc %p\n", vpt);
+				vpt->task = kthread_create(v9fs_poll_proc,
+					vpt, "v9fs-poll");
+				INIT_LIST_HEAD(&vpt->mux_list);
+				vpt->muxnum = 0;
+				v9fs_mux_poll_task_num++;
+				wake_up_process(vpt->task);
+				break;
+			}
 		}
-		rd += ret;
-		ptr += ret;
-	}
-	return (rd);
-}
 
-/**
- * read_message - read a full 9P2000 fcall packet
- * @v9ses: session info structure
- * @rcall: fcall structure to read into
- * @rcalllen: size of fcall buffer
- *
- */
+		if (i >= ARRAY_SIZE(v9fs_mux_poll_tasks))
+			dprintk(DEBUG_ERROR, "warning: no free poll slots\n");
+	}
 
-static int
-read_message(struct v9fs_session_info *v9ses,
-	     struct v9fs_fcall *rcall, int rcalllen)
-{
-	unsigned char buf[4];
-	void *data;
-	int size = 0;
-	int res = 0;
-
-	res = xread(v9ses, buf, sizeof(buf));
-	if (res < 0) {
-		dprintk(DEBUG_ERROR,
-			"Reading of count field failed returned: %d\n", res);
-		return res;
+	n = (v9fs_mux_num + 1) / v9fs_mux_poll_task_num +
+	    ((v9fs_mux_num + 1) % v9fs_mux_poll_task_num ? 1 : 0);
+
+	vptlast = NULL;
+	for (i = 0; i < ARRAY_SIZE(v9fs_mux_poll_tasks); i++) {
+		vpt = &v9fs_mux_poll_tasks[i];
+		if (vpt->task != NULL) {
+			vptlast = vpt;
+			if (vpt->muxnum < n) {
+				dprintk(DEBUG_MUX, "put in proc %d\n", i);
+				list_add(&m->mux_list, &vpt->mux_list);
+				vpt->muxnum++;
+				m->poll_task = vpt;
+				memset(&m->poll_waddr, 0, sizeof(m->poll_waddr));
+				init_poll_funcptr(&m->pt, v9fs_pollwait);
+				break;
+			}
+		}
 	}
 
-	if (res < 4) {
-		dprintk(DEBUG_ERROR,
-			"Reading of count field failed returned: %d\n", res);
-		return -EIO;
+	if (i >= ARRAY_SIZE(v9fs_mux_poll_tasks)) {
+		dprintk(DEBUG_MUX, "put in proc %d\n", i);
+		list_add(&m->mux_list, &vptlast->mux_list);
+		vptlast->muxnum++;
+		m->poll_task = vpt;
+		memset(&m->poll_waddr, 0, sizeof(m->poll_waddr));
+		init_poll_funcptr(&m->pt, v9fs_pollwait);
 	}
 
-	size = buf[0] | (buf[1] << 8) | (buf[2] << 16) | (buf[3] << 24);
-	dprintk(DEBUG_MUX, "got a packet count: %d\n", size);
+	v9fs_mux_num++;
+	down(&v9fs_mux_task_lock);
+}
 
-	/* adjust for the four bytes of size */
-	size -= 4;
+static void v9fs_mux_poll_stop(struct v9fs_mux_data *m)
+{
+	int i;
+	struct v9fs_mux_poll_task *vpt;
+
+	up(&v9fs_mux_task_lock);
+	vpt = m->poll_task;
+	list_del(&m->mux_list);
+	for(i = 0; i < ARRAY_SIZE(m->poll_waddr); i++) {
+		if (m->poll_waddr[i] != NULL) {
+			remove_wait_queue(m->poll_waddr[i], &m->poll_wait[i]);
+			m->poll_waddr[i] = NULL;
+		}
+	}
+	vpt->muxnum--;
+	if (!vpt->muxnum) {
+		dprintk(DEBUG_MUX, "destroy proc %p\n", vpt);
+		send_sig(SIGKILL, vpt->task, 1);
+		vpt->task = NULL;
+		v9fs_mux_poll_task_num--;
+	}
+	v9fs_mux_num--;
+	down(&v9fs_mux_task_lock);
+}
 
-	if (size > v9ses->maxdata) {
-		dprintk(DEBUG_ERROR, "packet too big: %d\n", size);
-		return -E2BIG;
+/**
+ * v9fs_mux_init - allocate and initialize the per-session mux data
+ * Creates the polling task if this is the first session.
+ *
+ * @trans - transport structure
+ * @msize - maximum message size
+ * @extended - pointer to the extended flag
+ */
+struct v9fs_mux_data *v9fs_mux_init(struct v9fs_transport *trans, int msize,
+				    unsigned char *extended)
+{
+	int i, n;
+	struct v9fs_mux_data *m, *mtmp;
+
+	dprintk(DEBUG_MUX, "transport %p msize %d\n", trans, msize);
+	m = kmalloc(sizeof(struct v9fs_mux_data) + 2 * msize, GFP_KERNEL);
+	if (!m)
+		return ERR_PTR(-ENOMEM);
+
+	spin_lock_init(&m->lock);
+	INIT_LIST_HEAD(&m->mux_list);
+	m->msize = msize;
+	m->extended = extended;
+	m->trans = trans;
+	idr_init(&m->tidpool.pool);
+	init_MUTEX(&m->tidpool.lock);
+	m->err = 0;
+	init_waitqueue_head(&m->equeue);
+	INIT_LIST_HEAD(&m->req_list);
+	INIT_LIST_HEAD(&m->unsent_req_list);
+	m->rpos = 0;
+	m->rbuf = (char *)m + sizeof(struct v9fs_mux_data);
+	m->wpos = m->wsize = 0;
+	m->wbuf = m->rbuf + msize;
+	INIT_WORK(&m->rq, v9fs_read_work, m);
+	INIT_WORK(&m->wq, v9fs_write_work, m);
+	m->wsched = 0;
+	memset(&m->poll_waddr, 0, sizeof(m->poll_waddr));
+	v9fs_mux_poll_start(m);
+
+	n = trans->poll(trans, &m->pt);
+	if (n & POLLIN) {
+		dprintk(DEBUG_MUX, "mux %p can read\n", m);
+		set_bit(Rpending, &m->wsched);
 	}
 
-	data = kmalloc(size, GFP_KERNEL);
-	if (!data) {
-		eprintk(KERN_WARNING, "out of memory\n");
-		return -ENOMEM;
+	if (n & POLLOUT) {
+		dprintk(DEBUG_MUX, "mux %p can write\n", m);
+		set_bit(Wpending, &m->wsched);
 	}
 
-	res = xread(v9ses, data, size);
-	if (res < size) {
-		dprintk(DEBUG_ERROR, "Reading of fcall failed returned: %d\n",
-			res);
-		kfree(data);
-		return res;
+	for(i = 0; i < ARRAY_SIZE(m->poll_waddr); i++) {
+		if (IS_ERR(m->poll_waddr[i])) {
+			v9fs_mux_poll_stop(m);
+			mtmp = (void *)m->poll_waddr;	/* the error code */
+			kfree(m);
+			m = mtmp;
+			break;
+		}
 	}
 
-	/* we now have an in-memory string that is the reply.
-	 * deserialize it. There is very little to go wrong at this point
-	 * save for v9fs_alloc errors.
-	 */
-	res = v9fs_deserialize_fcall(v9ses, size, data, v9ses->maxdata,
-				     rcall, rcalllen);
+	return m;
+}
 
-	kfree(data);
+/**
+ * v9fs_mux_destroy - cancels all pending requests and frees mux resources
+ */
+void v9fs_mux_destroy(struct v9fs_mux_data *m)
+{
+	dprintk(DEBUG_MUX, "mux %p prev %p next %p\n", m,
+		m->mux_list.prev, m->mux_list.next);
+	v9fs_mux_cancel(m, -ECONNRESET);
+
+	if (!list_empty(&m->req_list)) {
+		/* wait until all processes waiting on this session exit */
+		dprintk(DEBUG_MUX, "mux %p waiting for empty request queue\n",
+			m);
+		wait_event_timeout(m->equeue, (list_empty(&m->req_list)), 5000);
+		dprintk(DEBUG_MUX, "mux %p request queue empty: %d\n", m,
+			list_empty(&m->req_list));
+	}
 
-	if (res < 0)
-		return res;
+	v9fs_mux_poll_stop(m);
+	m->trans = NULL;
 
-	return 0;
+	kfree(m);
 }
 
 /**
- * v9fs_recv - receive an RPC response for a particular tag
- * @v9ses: session info structure
- * @req: RPC request structure
- *
+ * v9fs_pollwait - called by files poll operation to add v9fs-poll task
+ * 	to files wait queue
  */
-
-static int v9fs_recv(struct v9fs_session_info *v9ses, struct v9fs_rpcreq *req)
+static void
+v9fs_pollwait(struct file *filp, wait_queue_head_t * wait_address,
+	      poll_table * p)
 {
-	int ret = 0;
-
-	dprintk(DEBUG_MUX, "waiting for response: %d\n", req->tcall->tag);
-	ret = wait_event_interruptible(v9ses->read_wait,
-		       ((v9ses->transport->status != Connected) ||
-			(req->rcall != 0) || (req->err < 0) ||
-			dprintcond(v9ses, req)));
+	int i;
+	struct v9fs_mux_data *m;
 
-	dprintk(DEBUG_MUX, "got it: rcall %p\n", req->rcall);
+	m = container_of(p, struct v9fs_mux_data, pt);
+	for(i = 0; i < ARRAY_SIZE(m->poll_waddr); i++)
+		if (m->poll_waddr[i] == NULL)
+			break;
 
-	spin_lock(&v9ses->muxlock);
-	list_del(&req->next);
-	spin_unlock(&v9ses->muxlock);
+	if (i >= ARRAY_SIZE(m->poll_waddr)) {
+		dprintk(DEBUG_ERROR, "not enough wait_address slots\n");
+		return;
+	}
 
-	if (req->err < 0)
-		return req->err;
+	m->poll_waddr[i] = wait_address;
 
-	if (v9ses->transport->status == Disconnected)
-		return -ECONNRESET;
+	if (!wait_address) {
+		dprintk(DEBUG_ERROR, "no wait_address\n");
+		m->poll_waddr[i] = ERR_PTR(-EIO);
+		return;
+	}
 
-	return ret;
+	init_waitqueue_entry(&m->poll_wait[i], m->poll_task->task);
+	add_wait_queue(wait_address, &m->poll_wait[i]);
 }
 
 /**
- * v9fs_send - send a 9P request
- * @v9ses: session info structure
- * @req: RPC request to send
- *
+ * v9fs_poll_mux - polls a mux and schedules read or write works if necessary
  */
-
-static int v9fs_send(struct v9fs_session_info *v9ses, struct v9fs_rpcreq *req)
+static inline void v9fs_poll_mux(struct v9fs_mux_data *m)
 {
-	int ret = -1;
-	void *data = NULL;
-	struct v9fs_fcall *tcall = req->tcall;
-
-	data = kmalloc(v9ses->maxdata + V9FS_IOHDRSZ, GFP_KERNEL);
-	if (!data)
-		return -ENOMEM;
-
-	tcall->size = 0;	/* enforce size recalculation */
-	ret =
-	    v9fs_serialize_fcall(v9ses, tcall, data,
-				 v9ses->maxdata + V9FS_IOHDRSZ);
-	if (ret < 0)
-		goto free_data;
-
-	spin_lock(&v9ses->muxlock);
-	list_add(&req->next, &v9ses->mux_fcalls);
-	spin_unlock(&v9ses->muxlock);
-
-	dprintk(DEBUG_MUX, "sending message: tag %d size %d\n", tcall->tag,
-		tcall->size);
-	ret = v9ses->transport->write(v9ses->transport, data, tcall->size);
-
-	if (ret != tcall->size) {
-		spin_lock(&v9ses->muxlock);
-		list_del(&req->next);
-		kfree(req->rcall);
+	int n;
 
-		spin_unlock(&v9ses->muxlock);
-		if (ret >= 0)
-			ret = -EREMOTEIO;
-	} else
-		ret = 0;
+	if (m->err < 0)
+		return;
+
+	n = m->trans->poll(m->trans, NULL);
+	if (n < 0 || n & (POLLERR | POLLHUP | POLLNVAL)) {
+		dprintk(DEBUG_MUX, "error mux %p err %d\n", m, n);
+		if (n >= 0)
+			n = -ECONNRESET;
+		v9fs_mux_cancel(m, n);
+	}
+
+	if (n & POLLIN) {
+		set_bit(Rpending, &m->wsched);
+		dprintk(DEBUG_MUX, "mux %p can read\n", m);
+		if (!test_and_set_bit(Rworksched, &m->wsched)) {
+			dprintk(DEBUG_MUX, "schedule read work mux %p\n", m);
+			queue_work(v9fs_mux_wq, &m->rq);
+		}
+	}
 
-      free_data:
-	kfree(data);
-	return ret;
+	if (n & POLLOUT) {
+		set_bit(Wpending, &m->wsched);
+		dprintk(DEBUG_MUX, "mux %p can write\n", m);
+		if ((m->wsize || !list_empty(&m->unsent_req_list))
+		    && !test_and_set_bit(Wworksched, &m->wsched)) {
+			dprintk(DEBUG_MUX, "schedule write work mux %p\n", m);
+			queue_work(v9fs_mux_wq, &m->wq);
+		}
+	}
 }
 
 /**
- * v9fs_mux_rpc - send a request, receive a response
- * @v9ses: session info structure
- * @tcall: fcall to send
- * @rcall: buffer to place response into
- *
+ * v9fs_poll_proc - polls all v9fs transports for new events and queues
+ * 	the appropriate work to the work queue
  */
-
-long
-v9fs_mux_rpc(struct v9fs_session_info *v9ses, struct v9fs_fcall *tcall,
-	     struct v9fs_fcall **rcall)
+static int v9fs_poll_proc(void *a)
 {
-	int tid = -1;
-	struct v9fs_fcall *fcall = NULL;
-	struct v9fs_rpcreq req;
-	int ret = -1;
+	struct v9fs_mux_data *m, *mtmp;
+	struct v9fs_mux_poll_task *vpt;
 
-	if (!v9ses)
-		return -EINVAL;
+	vpt = a;
+	dprintk(DEBUG_MUX, "start %p %p\n", current, vpt);
+	allow_signal(SIGKILL);
+	while (!kthread_should_stop()) {
+		set_current_state(TASK_INTERRUPTIBLE);
+		if (signal_pending(current))
+			break;
 
-	if (!v9ses->transport || v9ses->transport->status != Connected)
-		return -EIO;
+		list_for_each_entry_safe(m, mtmp, &vpt->mux_list, mux_list) {
+			v9fs_poll_mux(m);
+		}
+
+		dprintk(DEBUG_MUX, "sleeping...\n");
+		schedule_timeout(SCHED_TIMEOUT * HZ);
+	}
 
-	if (rcall)
-		*rcall = NULL;
+	__set_current_state(TASK_RUNNING);
+	dprintk(DEBUG_MUX, "finish\n");
+	return 0;
+}
 
-	if (tcall->id != TVERSION) {
-		tid = v9fs_get_idpool(&v9ses->tidpool);
-		if (tid < 0)
-			return -ENOMEM;
+static inline int v9fs_write_req(struct v9fs_mux_data *m, struct v9fs_req *req)
+{
+	int n;
+
+	list_move_tail(&req->req_list, &m->req_list);
+	n = v9fs_serialize_fcall(req->tcall, m->wbuf, m->msize, *m->extended);
+	if (n < 0) {
+		req->err = n;
+		list_del(&req->req_list);
+		if (req->cb) {
+			spin_unlock(&m->lock);
+			(*req->cb) (req->cba, req->tcall, req->rcall, req->err);
+			req->cb = NULL;
+			spin_lock(&m->lock);
+		} else
+			kfree(req->rcall);
+
+		kfree(req);
 	}
 
-	tcall->tag = tid;
+	return n;
+}
 
-	req.tcall = tcall;
-	req.err = 0;
-	req.rcall = NULL;
+/**
+ * v9fs_write_work - called when a transport can send some data
+ */
+static void v9fs_write_work(void *a)
+{
+	int n, err;
+	struct v9fs_mux_data *m;
+	struct v9fs_req *req, *rtmp;
 
-	ret = v9fs_send(v9ses, &req);
+	m = a;
 
-	if (ret < 0) {
-		if (tcall->id != TVERSION)
-			v9fs_put_idpool(tid, &v9ses->tidpool);
-		dprintk(DEBUG_MUX, "error %d\n", ret);
-		return ret;
+	if (m->err < 0) {
+		clear_bit(Wworksched, &m->wsched);
+		return;
 	}
 
-	ret = v9fs_recv(v9ses, &req);
-
-	fcall = req.rcall;
-
-	dprintk(DEBUG_MUX, "received: tag=%x, ret=%d\n", tcall->tag, ret);
-	if (ret == -ERESTARTSYS) {
-		if (v9ses->transport->status != Disconnected
-		    && tcall->id != TFLUSH) {
-			unsigned long flags;
-
-			dprintk(DEBUG_MUX, "flushing the tag: %d\n",
-				tcall->tag);
-			clear_thread_flag(TIF_SIGPENDING);
-			v9fs_t_flush(v9ses, tcall->tag);
-			spin_lock_irqsave(&current->sighand->siglock, flags);
-			recalc_sigpending();
-			spin_unlock_irqrestore(&current->sighand->siglock,
-					       flags);
-			dprintk(DEBUG_MUX, "flushing done\n");
+	if (!m->wsize) {
+		if (list_empty(&m->unsent_req_list)) {
+			clear_bit(Wworksched, &m->wsched);
+			return;
 		}
 
-		goto release_req;
-	} else if (ret < 0)
-		goto release_req;
-
-	if (!fcall)
-		ret = -EIO;
-	else {
-		if (fcall->id == RERROR) {
-			ret = v9fs_errstr2errno(fcall->params.rerror.error);
-			if (ret == 0) {	/* string match failed */
-				if (fcall->params.rerror.errno)
-					ret = -(fcall->params.rerror.errno);
-				else
-					ret = -ESERVERFAULT;
-			}
-		} else if (fcall->id != tcall->id + 1) {
-			dprintk(DEBUG_ERROR,
-				"fcall mismatch: expected %d, got %d\n",
-				tcall->id + 1, fcall->id);
-			ret = -EIO;
+		err = 0;
+		spin_lock(&m->lock);
+		list_for_each_entry_safe(req, rtmp, &m->unsent_req_list,
+					 req_list) {
+			err = v9fs_write_req(m, req);
+			if (err > 0)
+				break;
 		}
+
+		m->wsize = err;
+		m->wpos = 0;
+		spin_unlock(&m->lock);
 	}
 
-      release_req:
-	if (tcall->id != TVERSION)
-		v9fs_put_idpool(tid, &v9ses->tidpool);
-	if (rcall)
-		*rcall = fcall;
-	else
-		kfree(fcall);
+	dprintk(DEBUG_MUX, "mux %p pos %d size %d\n", m, m->wpos, m->wsize);
+	clear_bit(Wpending, &m->wsched);
+	err = m->trans->write(m->trans, m->wbuf + m->wpos, m->wsize - m->wpos);
+	dprintk(DEBUG_MUX, "mux %p sent %d bytes\n", m, err);
+	if (err == -EAGAIN) {
+		clear_bit(Wworksched, &m->wsched);
+		return;
+	}
+
+	if (err <= 0)
+		goto error;
+
+	m->wpos += err;
+	if (m->wpos == m->wsize)
+		m->wpos = m->wsize = 0;
+
+	if (m->wsize == 0 && !list_empty(&m->unsent_req_list)) {
+		if (test_and_clear_bit(Wpending, &m->wsched))
+			n = POLLOUT;
+		else
+			n = m->trans->poll(m->trans, NULL);
+
+		if (n & POLLOUT) {
+			dprintk(DEBUG_MUX, "schedule write work mux %p\n", m);
+			queue_work(v9fs_mux_wq, &m->wq);
+		} else
+			clear_bit(Wworksched, &m->wsched);
+	} else
+		clear_bit(Wworksched, &m->wsched);
 
-	return ret;
+	return;
+
+      error:
+	v9fs_mux_cancel(m, err);
+	clear_bit(Wworksched, &m->wsched);
 }
 
-/**
- * v9fs_mux_cancel_requests - cancels all pending requests
- *
- * @v9ses: session info structure
- * @err: error code to return to the requests
- */
-void v9fs_mux_cancel_requests(struct v9fs_session_info *v9ses, int err)
+static void process_request(struct v9fs_mux_data *m, struct v9fs_req *req)
 {
-	struct v9fs_rpcreq *rptr;
-	struct v9fs_rpcreq *rreq;
+	int ecode, tag;
+	char *ename;
+
+	tag = req->tag;
+	if (req->rcall->id == RERROR && !req->err) {
+		ecode = req->rcall->params.rerror.errno;
+		ename = req->rcall->params.rerror.error;
 
-	dprintk(DEBUG_MUX, " %d\n", err);
-	spin_lock(&v9ses->muxlock);
-	list_for_each_entry_safe(rreq, rptr, &v9ses->mux_fcalls, next) {
-		rreq->err = err;
+		dprintk(DEBUG_MUX, "Rerror %s\n", ename);
+
+		if (*m->extended)
+			req->err = -ecode;
+
+		if (!req->err) {
+			req->err = v9fs_errstr2errno(ename);
+
+			if (!req->err) {	/* string match failed */
+				dprintk(DEBUG_ERROR, "unknown error: %s\n",
+					ename);
+			}
+
+			if (!req->err)
+				req->err = -ESERVERFAULT;
+		}
+	} else if (req->tcall && req->rcall->id != req->tcall->id + 1) {
+		dprintk(DEBUG_ERROR, "fcall mismatch: expected %d, got %d\n",
+			req->tcall->id + 1, req->rcall->id);
+		if (!req->err)
+			req->err = -EIO;
 	}
-	spin_unlock(&v9ses->muxlock);
-	wake_up_all(&v9ses->read_wait);
+
+	if (req->cb && req->err != ERREQFLUSH) {
+		dprintk(DEBUG_MUX, "calling callback tcall %p rcall %p\n",
+			req->tcall, req->rcall);
+
+		(*req->cb) (req->cba, req->tcall, req->rcall, req->err);
+		req->cb = NULL;
+	} else
+		kfree(req->rcall);
+
+	if (tag != V9FS_NOTAG)
+		v9fs_put_idpool(tag, &m->tidpool);
+
+	wake_up(&m->equeue);
+	kfree(req);
 }
 
 /**
- * v9fs_recvproc - kproc to handle demultiplexing responses
- * @data: session info structure
- *
+ * v9fs_read_work - called when there is some data to be read from a transport
  */
-
-static int v9fs_recvproc(void *data)
+static void v9fs_read_work(void *a)
 {
-	struct v9fs_session_info *v9ses = (struct v9fs_session_info *)data;
-	struct v9fs_fcall *rcall = NULL;
-	struct v9fs_rpcreq *rptr;
-	struct v9fs_rpcreq *req;
-	struct v9fs_rpcreq *rreq;
-	int err = 0;
+	int n, err, rcallen;
+	struct v9fs_mux_data *m;
+	struct v9fs_req *req, *rptr, *rreq;
+	struct v9fs_fcall *rcall;
+
+	m = a;
+
+	if (m->err < 0)
+		return;
+
+	rcall = NULL;
+	dprintk(DEBUG_MUX, "start mux %p pos %d\n", m, m->rpos);
+	clear_bit(Rpending, &m->wsched);
+	err = m->trans->read(m->trans, m->rbuf + m->rpos, m->msize - m->rpos);
+	dprintk(DEBUG_MUX, "mux %p got %d bytes\n", m, err);
+	if (err == -EAGAIN) {
+		clear_bit(Rworksched, &m->wsched);
+		return;
+	}
 
-	allow_signal(SIGKILL);
-	set_current_state(TASK_INTERRUPTIBLE);
-	complete(&v9ses->proccmpl);
-	while (!kthread_should_stop() && err >= 0) {
-		req = rptr = rreq = NULL;
+	if (err <= 0)
+		goto error;
 
-		rcall = kmalloc(v9ses->maxdata + V9FS_IOHDRSZ, GFP_KERNEL);
-		if (!rcall) {
-			eprintk(KERN_ERR, "no memory for buffers\n");
+	m->rpos += err;
+	while (m->rpos > 4) {
+		n = le32_to_cpu(*(__le32 *) m->rbuf);
+		if (n >= m->msize) {
+			dprintk(DEBUG_ERROR,
+				"requested packet size too big: %d\n", n);
+			err = -EIO;
+			goto error;
+		}
+
+		if (m->rpos < n)
 			break;
+
+		rcallen = n + V9FS_FCALLHDRSZ;
+		rcall = kmalloc(rcallen, GFP_KERNEL);
+		if (!rcall) {
+			err = -ENOMEM;
+			goto error;
 		}
 
-		err = read_message(v9ses, rcall, v9ses->maxdata + V9FS_IOHDRSZ);
-		spin_lock(&v9ses->muxlock);
+		dump_data(m->rbuf, n);
+		err = v9fs_deserialize_fcall(m->rbuf, n, rcall, rcallen,
+					     *m->extended);
 		if (err < 0) {
-			list_for_each_entry_safe(rreq, rptr, &v9ses->mux_fcalls, next) {
-				rreq->err = err;
-			}
-			if(err != -ERESTARTSYS)
-				eprintk(KERN_ERR,
-					"Transport error while reading message %d\n", err);
-		} else {
-			list_for_each_entry_safe(rreq, rptr, &v9ses->mux_fcalls, next) {
-				if (rreq->tcall->tag == rcall->tag) {
-					req = rreq;
-					req->rcall = rcall;
-					break;
-				}
-			}
+			kfree(rcall);
+			goto error;
 		}
 
-		if (req && (req->tcall->id == TFLUSH)) {
-			struct v9fs_rpcreq *treq = NULL;
-			list_for_each_entry_safe(treq, rptr, &v9ses->mux_fcalls, next) {
-				if (treq->tcall->tag ==
-				    req->tcall->params.tflush.oldtag) {
-					list_del(&rptr->next);
-					kfree(treq->rcall);
-					break;
-				}
+		dprintk(DEBUG_MUX, "mux %p fcall id %d tag %d\n", m, rcall->id,
+			rcall->tag);
+
+		req = NULL;
+		spin_lock(&m->lock);
+		list_for_each_entry_safe(rreq, rptr, &m->req_list, req_list) {
+			if (rreq->tag == rcall->tag) {
+				req = rreq;
+				req->rcall = rcall;
+				list_del(&req->req_list);
+				spin_unlock(&m->lock);
+				process_request(m, req);
+				break;
 			}
 		}
 
-		spin_unlock(&v9ses->muxlock);
-
 		if (!req) {
-			if (err >= 0)
+			spin_unlock(&m->lock);
+			if (err >= 0 && rcall->id != RFLUSH)
 				dprintk(DEBUG_ERROR,
-					"unexpected response: id %d tag %d\n",
-					rcall->id, rcall->tag);
-
+					"unexpected response mux %p id %d tag %d\n",
+					m, rcall->id, rcall->tag);
 			kfree(rcall);
 		}
 
-		wake_up_all(&v9ses->read_wait);
-		set_current_state(TASK_INTERRUPTIBLE);
+		if (m->rpos > n)
+			memmove(m->rbuf, m->rbuf + n, m->rpos - n);
+		m->rpos -= n;
 	}
 
-	v9ses->transport->close(v9ses->transport);
-
-	/* Inform all pending processes about the failure */
-	wake_up_all(&v9ses->read_wait);
-
-	if (signal_pending(current))
-		complete(&v9ses->proccmpl);
+	if (!list_empty(&m->req_list)) {
+		if (test_and_clear_bit(Rpending, &m->wsched))
+			n = POLLIN;
+		else
+			n = m->trans->poll(m->trans, NULL);
+
+		if (n & POLLIN) {
+			dprintk(DEBUG_MUX, "schedule read work mux %p\n", m);
+			queue_work(v9fs_mux_wq, &m->rq);
+		} else
+			clear_bit(Rworksched, &m->wsched);
+	} else
+		clear_bit(Rworksched, &m->wsched);
 
-	dprintk(DEBUG_MUX, "recvproc: end\n");
-	v9ses->recvproc = NULL;
+	return;
 
-	return err >= 0;
+      error:
+	v9fs_mux_cancel(m, err);
+	clear_bit(Rworksched, &m->wsched);
 }
 
 /**
- * v9fs_mux_init - initialize multiplexer (spawn kproc)
- * @v9ses: session info structure
- * @dev_name: mount device information (to create unique kproc)
+ * v9fs_send_request - send 9P request
+ * The function can sleep until the request is scheduled for sending.
+ * The function can be interrupted. Return from the function is not
+ * a guarantee that the request is sent succesfully. Can return errors
+ * that can be retrieved by PTR_ERR macros.
  *
+ * @m: mux data
+ * @tc: request to be sent
+ * @cb: callback function to call when response is received
+ * @cba: parameter to pass to the callback function
  */
+static struct v9fs_req *v9fs_send_request(struct v9fs_mux_data *m,
+					  struct v9fs_fcall *tc,
+					  v9fs_mux_req_callback cb, void *cba)
+{
+	int n;
+	struct v9fs_req *req;
+
+	dprintk(DEBUG_MUX, "mux %p task %p tcall %p id %d\n", m, current,
+		tc, tc->id);
+	if (m->err < 0)
+		return ERR_PTR(m->err);
+
+	req = kmalloc(sizeof(struct v9fs_req), GFP_KERNEL);
+	if (!req)
+		return ERR_PTR(-ENOMEM);
 
-int v9fs_mux_init(struct v9fs_session_info *v9ses, const char *dev_name)
+	if (tc->id == TVERSION)
+		n = V9FS_NOTAG;
+	else
+		n = v9fs_get_idpool(&m->tidpool);
+
+	if (n < 0)
+		return ERR_PTR(-ENOMEM);
+
+	tc->tag = n;
+	req->tag = n;
+	req->tcall = tc;
+	req->rcall = NULL;
+	req->err = 0;
+	req->cb = cb;
+	req->cba = cba;
+
+	spin_lock(&m->lock);
+	list_add_tail(&req->req_list, &m->unsent_req_list);
+	spin_unlock(&m->lock);
+
+	if (test_and_clear_bit(Wpending, &m->wsched))
+		n = POLLOUT;
+	else
+		n = m->trans->poll(m->trans, NULL);
+
+	if (n & POLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
+		queue_work(v9fs_mux_wq, &m->wq);
+
+	return req;
+}
+
+static inline void
+v9fs_mux_flush_cb(void *a, struct v9fs_fcall *tc, struct v9fs_fcall *rc,
+		  int err)
 {
-	char procname[60];
-
-	strncpy(procname, dev_name, sizeof(procname));
-	procname[sizeof(procname) - 1] = 0;
-
-	init_waitqueue_head(&v9ses->read_wait);
-	init_completion(&v9ses->fcread);
-	init_completion(&v9ses->proccmpl);
-	spin_lock_init(&v9ses->muxlock);
-	INIT_LIST_HEAD(&v9ses->mux_fcalls);
-	v9ses->recvproc = NULL;
-	v9ses->curfcall = NULL;
-
-	v9ses->recvproc = kthread_create(v9fs_recvproc, v9ses,
-					 "v9fs_recvproc %s", procname);
-
-	if (IS_ERR(v9ses->recvproc)) {
-		eprintk(KERN_ERR, "cannot create receiving thread\n");
-		v9fs_session_close(v9ses);
-		return -ECONNABORTED;
+	v9fs_mux_req_callback cb;
+	int tag;
+	struct v9fs_mux_data *m;
+	struct v9fs_req *req, *rptr;
+
+	m = a;
+	dprintk(DEBUG_MUX, "mux %p tc %p rc %p err %d oldtag %d\n", m, tc,
+		rc, err, tc->params.tflush.oldtag);
+
+	spin_lock(&m->lock);
+	cb = NULL;
+	tag = tc->params.tflush.oldtag;
+	list_for_each_entry_safe(req, rptr, &m->req_list, req_list) {
+		if (req->tag == tag) {
+			list_del(&req->req_list);
+			if (req->cb) {
+				cb = req->cb;
+				req->cb = NULL;
+				spin_unlock(&m->lock);
+				(*cb) (req->cba, req->tcall, req->rcall,
+				       req->err);
+			}
+			kfree(req);
+			wake_up(&m->equeue);
+			break;
+		}
+	}
+
+	if (!cb)
+		spin_unlock(&m->lock);
+
+	if (v9fs_check_idpool(tag, &m->tidpool))
+		v9fs_put_idpool(tag, &m->tidpool);
+
+	kfree(tc);
+	kfree(rc);
+}
+
+static void
+v9fs_mux_flush_request(struct v9fs_mux_data *m, struct v9fs_req *req)
+{
+	struct v9fs_fcall *fc;
+
+	dprintk(DEBUG_MUX, "mux %p req %p tag %d\n", m, req, req->tag);
+
+	fc = kmalloc(sizeof(struct v9fs_fcall), GFP_KERNEL);
+	fc->id = TFLUSH;
+	fc->params.tflush.oldtag = req->tag;
+
+	v9fs_send_request(m, fc, v9fs_mux_flush_cb, m);
+}
+
+static void
+v9fs_mux_rpc_cb(void *a, struct v9fs_fcall *tc, struct v9fs_fcall *rc, int err)
+{
+	struct v9fs_mux_rpc *r;
+
+	if (err == ERREQFLUSH) {
+		dprintk(DEBUG_MUX, "err req flush\n");
+		return;
+	}
+
+	r = a;
+	dprintk(DEBUG_MUX, "mux %p req %p tc %p rc %p err %d\n", r->m, r->req,
+		tc, rc, err);
+	r->rcall = rc;
+	r->err = err;
+	wake_up(&r->wqueue);
+}
+
+/**
+ * v9fs_mux_rpc - sends 9P request and waits until a response is available.
+ *	The function can be interrupted.
+ * @m: mux data
+ * @tc: request to be sent
+ * @rc: pointer where a pointer to the response is stored
+ */
+int
+v9fs_mux_rpc(struct v9fs_mux_data *m, struct v9fs_fcall *tc,
+	     struct v9fs_fcall **rc)
+{
+	int err;
+	unsigned long flags;
+	struct v9fs_req *req;
+	struct v9fs_mux_rpc r;
+
+	r.err = 0;
+	r.rcall = NULL;
+	r.m = m;
+	init_waitqueue_head(&r.wqueue);
+
+	if (rc)
+		*rc = NULL;
+
+	req = v9fs_send_request(m, tc, v9fs_mux_rpc_cb, &r);
+	if (IS_ERR(req)) {
+		err = PTR_ERR(req);
+		dprintk(DEBUG_MUX, "error %d\n", err);
+		return PTR_ERR(req);
+	}
+
+	r.req = req;
+	dprintk(DEBUG_MUX, "mux %p tc %p tag %d rpc %p req %p\n", m, tc,
+		req->tag, &r, req);
+	err = wait_event_interruptible(r.wqueue, r.rcall != NULL || r.err < 0);
+	if (r.err < 0)
+		err = r.err;
+
+	if (err == -ERESTARTSYS && m->trans->status == Connected && m->err == 0) {
+		spin_lock(&m->lock);
+		req->tcall = NULL;
+		req->err = ERREQFLUSH;
+		spin_unlock(&m->lock);
+
+		clear_thread_flag(TIF_SIGPENDING);
+		v9fs_mux_flush_request(m, req);
+		spin_lock_irqsave(&current->sighand->siglock, flags);
+		recalc_sigpending();
+		spin_unlock_irqrestore(&current->sighand->siglock, flags);
 	}
 
-	wake_up_process(v9ses->recvproc);
-	wait_for_completion(&v9ses->proccmpl);
+	if (!err) {
+		if (r.rcall)
+			dprintk(DEBUG_MUX, "got response id %d tag %d\n",
+				r.rcall->id, r.rcall->tag);
+
+		if (rc)
+			*rc = r.rcall;
+		else
+			kfree(r.rcall);
+	} else {
+		kfree(r.rcall);
+		dprintk(DEBUG_MUX, "got error %d\n", err);
+		if (err > 0)
+			err = -EIO;
+	}
+
+	return err;
+}
+
+/**
+ * v9fs_mux_rpcnb - sends 9P request without waiting for response.
+ * @m: mux data
+ * @tc: request to be sent
+ * @cb: callback function to be called when response arrives
+ * @cba: value to pass to the callback function
+ */
+int v9fs_mux_rpcnb(struct v9fs_mux_data *m, struct v9fs_fcall *tc,
+		   v9fs_mux_req_callback cb, void *a)
+{
+	int err;
+	struct v9fs_req *req;
+
+	req = v9fs_send_request(m, tc, cb, a);
+	if (IS_ERR(req)) {
+		err = PTR_ERR(req);
+		dprintk(DEBUG_MUX, "error %d\n", err);
+		return PTR_ERR(req);
+	}
 
+	dprintk(DEBUG_MUX, "mux %p tc %p tag %d\n", m, tc, req->tag);
 	return 0;
 }
+
+/**
+ * v9fs_mux_cancel - cancel all pending requests with error
+ * @m: mux data
+ * @err: error code
+ */
+void v9fs_mux_cancel(struct v9fs_mux_data *m, int err)
+{
+	struct v9fs_req *req, *rtmp;
+	LIST_HEAD(cancel_list);
+
+	dprintk(DEBUG_MUX, "mux %p err %d\n", m, err);
+	m->err = err;
+	spin_lock(&m->lock);
+	list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
+		list_move(&req->req_list, &cancel_list);
+	}
+	spin_unlock(&m->lock);
+
+	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
+		list_del(&req->req_list);
+		if (!req->err)
+			req->err = err;
+
+		if (req->cb)
+			(*req->cb) (req->cba, req->tcall, req->rcall, req->err);
+		else
+			kfree(req->rcall);
+
+		kfree(req);
+	}
+
+	wake_up(&m->equeue);
+}