[AF_RXRPC]: Make the in-kernel AFS filesystem use AF_RXRPC.

Make the in-kernel AFS filesystem use AF_RXRPC instead of the old RxRPC code.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 373 insertions, 96 deletions

diff --git a/fs/afs/callback.c b/fs/afs/callback.c
index 26a48fea42f..61121554714 100644
--- a/fs/afs/callback.c
+++ b/fs/afs/callback.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2002 Red Hat, Inc. All rights reserved.
+ * Copyright (c) 2002, 2007 Red Hat, Inc. All rights reserved.
  *
  * This software may be freely redistributed under the terms of the
  * GNU General Public License.
@@ -16,83 +16,182 @@
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/init.h>
-#include "server.h"
-#include "vnode.h"
+#include <linux/circ_buf.h>
 #include "internal.h"
-#include "cmservice.h"
+
+unsigned afs_vnode_update_timeout = 10;
+
+#define afs_breakring_space(server) \
+	CIRC_SPACE((server)->cb_break_head, (server)->cb_break_tail,	\
+		   ARRAY_SIZE((server)->cb_break))
+
+//static void afs_callback_updater(struct work_struct *);
+
+static struct workqueue_struct *afs_callback_update_worker;
 
 /*
  * allow the fileserver to request callback state (re-)initialisation
  */
-int SRXAFSCM_InitCallBackState(struct afs_server *server)
+void afs_init_callback_state(struct afs_server *server)
 {
-	struct list_head callbacks;
+	struct afs_vnode *vnode;
 
-	_enter("%p", server);
+	_enter("{%p}", server);
 
-	INIT_LIST_HEAD(&callbacks);
-
-	/* transfer the callback list from the server to a temp holding area */
 	spin_lock(&server->cb_lock);
 
-	list_add(&callbacks, &server->cb_promises);
-	list_del_init(&server->cb_promises);
+	/* kill all the promises on record from this server */
+	while (!RB_EMPTY_ROOT(&server->cb_promises)) {
+		vnode = rb_entry(server->cb_promises.rb_node,
+				 struct afs_vnode, cb_promise);
+		printk("\nUNPROMISE on %p\n", vnode);
+		rb_erase(&vnode->cb_promise, &server->cb_promises);
+		vnode->cb_promised = false;
+	}
 
-	/* munch our way through the list, grabbing the inode, dropping all the
-	 * locks and regetting them in the right order
-	 */
-	while (!list_empty(&callbacks)) {
-		struct afs_vnode *vnode;
-		struct inode *inode;
+	spin_unlock(&server->cb_lock);
+	_leave("");
+}
 
-		vnode = list_entry(callbacks.next, struct afs_vnode, cb_link);
-		list_del_init(&vnode->cb_link);
+/*
+ * handle the data invalidation side of a callback being broken
+ */
+void afs_broken_callback_work(struct work_struct *work)
+{
+	struct afs_vnode *vnode =
+		container_of(work, struct afs_vnode, cb_broken_work);
 
-		/* try and grab the inode - may fail */
-		inode = igrab(AFS_VNODE_TO_I(vnode));
-		if (inode) {
-			int release = 0;
+	_enter("");
 
-			spin_unlock(&server->cb_lock);
-			spin_lock(&vnode->lock);
+	if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
+		return;
 
-			if (vnode->cb_server == server) {
-				vnode->cb_server = NULL;
-				afs_kafstimod_del_timer(&vnode->cb_timeout);
-				spin_lock(&afs_cb_hash_lock);
-				list_del_init(&vnode->cb_hash_link);
-				spin_unlock(&afs_cb_hash_lock);
-				release = 1;
-			}
+	/* we're only interested in dealing with a broken callback on *this*
+	 * vnode and only if no-one else has dealt with it yet */
+	if (!mutex_trylock(&vnode->cb_broken_lock))
+		return; /* someone else is dealing with it */
 
-			spin_unlock(&vnode->lock);
+	if (test_bit(AFS_VNODE_CB_BROKEN, &vnode->flags)) {
+		if (afs_vnode_fetch_status(vnode) < 0)
+			goto out;
 
-			iput(inode);
-			afs_put_server(server);
+		if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
+			goto out;
 
-			spin_lock(&server->cb_lock);
+		/* if the vnode's data version number changed then its contents
+		 * are different */
+		if (test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags)) {
+			_debug("zap data");
+			invalidate_remote_inode(&vnode->vfs_inode);
 		}
 	}
 
-	spin_unlock(&server->cb_lock);
+out:
+	mutex_unlock(&vnode->cb_broken_lock);
 
-	_leave(" = 0");
-	return 0;
+	/* avoid the potential race whereby the mutex_trylock() in this
+	 * function happens again between the clear_bit() and the
+	 * mutex_unlock() */
+	if (test_bit(AFS_VNODE_CB_BROKEN, &vnode->flags)) {
+		_debug("requeue");
+		queue_work(afs_callback_update_worker, &vnode->cb_broken_work);
+	}
+	_leave("");
+}
+
+/*
+ * actually break a callback
+ */
+static void afs_break_callback(struct afs_server *server,
+			       struct afs_vnode *vnode)
+{
+	_enter("");
+
+	set_bit(AFS_VNODE_CB_BROKEN, &vnode->flags);
+
+	if (vnode->cb_promised) {
+		spin_lock(&vnode->lock);
+
+		_debug("break callback");
+
+		spin_lock(&server->cb_lock);
+		if (vnode->cb_promised) {
+			rb_erase(&vnode->cb_promise, &server->cb_promises);
+			vnode->cb_promised = false;
+		}
+		spin_unlock(&server->cb_lock);
+
+		queue_work(afs_callback_update_worker, &vnode->cb_broken_work);
+		spin_unlock(&vnode->lock);
+	}
+}
+
+/*
+ * allow the fileserver to explicitly break one callback
+ * - happens when
+ *   - the backing file is changed
+ *   - a lock is released
+ */
+static void afs_break_one_callback(struct afs_server *server,
+				   struct afs_fid *fid)
+{
+	struct afs_vnode *vnode;
+	struct rb_node *p;
+
+	_debug("find");
+	spin_lock(&server->fs_lock);
+	p = server->fs_vnodes.rb_node;
+	while (p) {
+		vnode = rb_entry(p, struct afs_vnode, server_rb);
+		if (fid->vid < vnode->fid.vid)
+			p = p->rb_left;
+		else if (fid->vid > vnode->fid.vid)
+			p = p->rb_right;
+		else if (fid->vnode < vnode->fid.vnode)
+			p = p->rb_left;
+		else if (fid->vnode > vnode->fid.vnode)
+			p = p->rb_right;
+		else if (fid->unique < vnode->fid.unique)
+			p = p->rb_left;
+		else if (fid->unique > vnode->fid.unique)
+			p = p->rb_right;
+		else
+			goto found;
+	}
+
+	/* not found so we just ignore it (it may have moved to another
+	 * server) */
+not_available:
+	_debug("not avail");
+	spin_unlock(&server->fs_lock);
+	_leave("");
+	return;
+
+found:
+	_debug("found");
+	ASSERTCMP(server, ==, vnode->server);
+
+	if (!igrab(AFS_VNODE_TO_I(vnode)))
+		goto not_available;
+	spin_unlock(&server->fs_lock);
+
+	afs_break_callback(server, vnode);
+	iput(&vnode->vfs_inode);
+	_leave("");
 }
 
 /*
  * allow the fileserver to break callback promises
  */
-int SRXAFSCM_CallBack(struct afs_server *server, size_t count,
-		      struct afs_callback callbacks[])
+void afs_break_callbacks(struct afs_server *server, size_t count,
+			 struct afs_callback callbacks[])
 {
-	_enter("%p,%u,", server, count);
+	_enter("%p,%zu,", server, count);
 
-	for (; count > 0; callbacks++, count--) {
-		struct afs_vnode *vnode = NULL;
-		struct inode *inode = NULL;
-		int valid = 0;
+	ASSERT(server != NULL);
+	ASSERTCMP(count, <=, AFSCBMAX);
 
+	for (; count > 0; callbacks++, count--) {
 		_debug("- Fid { vl=%08x n=%u u=%u }  CB { v=%u x=%u t=%u }",
 		       callbacks->fid.vid,
 		       callbacks->fid.vnode,
@@ -101,66 +200,244 @@ int SRXAFSCM_CallBack(struct afs_server *server, size_t count,
 		       callbacks->expiry,
 		       callbacks->type
 		       );
+		afs_break_one_callback(server, &callbacks->fid);
+	}
+
+	_leave("");
+	return;
+}
 
-		/* find the inode for this fid */
-		spin_lock(&afs_cb_hash_lock);
+/*
+ * record the callback for breaking
+ * - the caller must hold server->cb_lock
+ */
+static void afs_do_give_up_callback(struct afs_server *server,
+				    struct afs_vnode *vnode)
+{
+	struct afs_callback *cb;
 
-		list_for_each_entry(vnode,
-				    &afs_cb_hash(server, &callbacks->fid),
-				    cb_hash_link) {
-			if (memcmp(&vnode->fid, &callbacks->fid,
-				   sizeof(struct afs_fid)) != 0)
-				continue;
+	_enter("%p,%p", server, vnode);
 
-			/* right vnode, but is it same server? */
-			if (vnode->cb_server != server)
-				break; /* no */
+	cb = &server->cb_break[server->cb_break_head];
+	cb->fid		= vnode->fid;
+	cb->version	= vnode->cb_version;
+	cb->expiry	= vnode->cb_expiry;
+	cb->type	= vnode->cb_type;
+	smp_wmb();
+	server->cb_break_head =
+		(server->cb_break_head + 1) &
+		(ARRAY_SIZE(server->cb_break) - 1);
 
-			/* try and nail the inode down */
-			inode = igrab(AFS_VNODE_TO_I(vnode));
-			break;
+	/* defer the breaking of callbacks to try and collect as many as
+	 * possible to ship in one operation */
+	switch (atomic_inc_return(&server->cb_break_n)) {
+	case 1 ... AFSCBMAX - 1:
+		queue_delayed_work(afs_callback_update_worker,
+				   &server->cb_break_work, HZ * 2);
+		break;
+	case AFSCBMAX:
+		afs_flush_callback_breaks(server);
+		break;
+	default:
+		break;
+	}
+
+	ASSERT(server->cb_promises.rb_node != NULL);
+	rb_erase(&vnode->cb_promise, &server->cb_promises);
+	vnode->cb_promised = false;
+	_leave("");
+}
+
+/*
+ * give up the callback registered for a vnode on the file server when the
+ * inode is being cleared
+ */
+void afs_give_up_callback(struct afs_vnode *vnode)
+{
+	struct afs_server *server = vnode->server;
+
+	DECLARE_WAITQUEUE(myself, current);
+
+	_enter("%d", vnode->cb_promised);
+
+	_debug("GIVE UP INODE %p", &vnode->vfs_inode);
+
+	if (!vnode->cb_promised) {
+		_leave(" [not promised]");
+		return;
+	}
+
+	ASSERT(server != NULL);
+
+	spin_lock(&server->cb_lock);
+	if (vnode->cb_promised && afs_breakring_space(server) == 0) {
+		add_wait_queue(&server->cb_break_waitq, &myself);
+		for (;;) {
+			set_current_state(TASK_UNINTERRUPTIBLE);
+			if (!vnode->cb_promised ||
+			    afs_breakring_space(server) != 0)
+				break;
+			spin_unlock(&server->cb_lock);
+			schedule();
+			spin_lock(&server->cb_lock);
 		}
+		remove_wait_queue(&server->cb_break_waitq, &myself);
+		__set_current_state(TASK_RUNNING);
+	}
+
+	/* of course, it's always possible for the server to break this vnode's
+	 * callback first... */
+	if (vnode->cb_promised)
+		afs_do_give_up_callback(server, vnode);
+
+	spin_unlock(&server->cb_lock);
+	_leave("");
+}
+
+/*
+ * dispatch a deferred give up callbacks operation
+ */
+void afs_dispatch_give_up_callbacks(struct work_struct *work)
+{
+	struct afs_server *server =
+		container_of(work, struct afs_server, cb_break_work.work);
+
+	_enter("");
+
+	/* tell the fileserver to discard the callback promises it has
+	 * - in the event of ENOMEM or some other error, we just forget that we
+	 *   had callbacks entirely, and the server will call us later to break
+	 *   them
+	 */
+	afs_fs_give_up_callbacks(server, &afs_async_call);
+}
+
+/*
+ * flush the outstanding callback breaks on a server
+ */
+void afs_flush_callback_breaks(struct afs_server *server)
+{
+	cancel_delayed_work(&server->cb_break_work);
+	queue_delayed_work(afs_callback_update_worker,
+			   &server->cb_break_work, 0);
+}
+
+#if 0
+/*
+ * update a bunch of callbacks
+ */
+static void afs_callback_updater(struct work_struct *work)
+{
+	struct afs_server *server;
+	struct afs_vnode *vnode, *xvnode;
+	time_t now;
+	long timeout;
+	int ret;
+
+	server = container_of(work, struct afs_server, updater);
+
+	_enter("");
 
-		spin_unlock(&afs_cb_hash_lock);
-
-		if (inode) {
-			/* we've found the record for this vnode */
-			spin_lock(&vnode->lock);
-			if (vnode->cb_server == server) {
-				/* the callback _is_ on the calling server */
-				vnode->cb_server = NULL;
-				valid = 1;
-
-				afs_kafstimod_del_timer(&vnode->cb_timeout);
-				vnode->flags |= AFS_VNODE_CHANGED;
-
-				spin_lock(&server->cb_lock);
-				list_del_init(&vnode->cb_link);
-				spin_unlock(&server->cb_lock);
-
-				spin_lock(&afs_cb_hash_lock);
-				list_del_init(&vnode->cb_hash_link);
-				spin_unlock(&afs_cb_hash_lock);
-			}
-			spin_unlock(&vnode->lock);
-
-			if (valid) {
-				invalidate_remote_inode(inode);
-				afs_put_server(server);
-			}
-			iput(inode);
+	now = get_seconds();
+
+	/* find the first vnode to update */
+	spin_lock(&server->cb_lock);
+	for (;;) {
+		if (RB_EMPTY_ROOT(&server->cb_promises)) {
+			spin_unlock(&server->cb_lock);
+			_leave(" [nothing]");
+			return;
 		}
+
+		vnode = rb_entry(rb_first(&server->cb_promises),
+				 struct afs_vnode, cb_promise);
+		if (atomic_read(&vnode->usage) > 0)
+			break;
+		rb_erase(&vnode->cb_promise, &server->cb_promises);
+		vnode->cb_promised = false;
+	}
+
+	timeout = vnode->update_at - now;
+	if (timeout > 0) {
+		queue_delayed_work(afs_vnode_update_worker,
+				   &afs_vnode_update, timeout * HZ);
+		spin_unlock(&server->cb_lock);
+		_leave(" [nothing]");
+		return;
+	}
+
+	list_del_init(&vnode->update);
+	atomic_inc(&vnode->usage);
+	spin_unlock(&server->cb_lock);
+
+	/* we can now perform the update */
+	_debug("update %s", vnode->vldb.name);
+	vnode->state = AFS_VL_UPDATING;
+	vnode->upd_rej_cnt = 0;
+	vnode->upd_busy_cnt = 0;
+
+	ret = afs_vnode_update_record(vl, &vldb);
+	switch (ret) {
+	case 0:
+		afs_vnode_apply_update(vl, &vldb);
+		vnode->state = AFS_VL_UPDATING;
+		break;
+	case -ENOMEDIUM:
+		vnode->state = AFS_VL_VOLUME_DELETED;
+		break;
+	default:
+		vnode->state = AFS_VL_UNCERTAIN;
+		break;
+	}
+
+	/* and then reschedule */
+	_debug("reschedule");
+	vnode->update_at = get_seconds() + afs_vnode_update_timeout;
+
+	spin_lock(&server->cb_lock);
+
+	if (!list_empty(&server->cb_promises)) {
+		/* next update in 10 minutes, but wait at least 1 second more
+		 * than the newest record already queued so that we don't spam
+		 * the VL server suddenly with lots of requests
+		 */
+		xvnode = list_entry(server->cb_promises.prev,
+				    struct afs_vnode, update);
+		if (vnode->update_at <= xvnode->update_at)
+			vnode->update_at = xvnode->update_at + 1;
+		xvnode = list_entry(server->cb_promises.next,
+				    struct afs_vnode, update);
+		timeout = xvnode->update_at - now;
+		if (timeout < 0)
+			timeout = 0;
+	} else {
+		timeout = afs_vnode_update_timeout;
 	}
 
-	_leave(" = 0");
-	return 0;
+	list_add_tail(&vnode->update, &server->cb_promises);
+
+	_debug("timeout %ld", timeout);
+	queue_delayed_work(afs_vnode_update_worker,
+			   &afs_vnode_update, timeout * HZ);
+	spin_unlock(&server->cb_lock);
+	afs_put_vnode(vl);
+}
+#endif
+
+/*
+ * initialise the callback update process
+ */
+int __init afs_callback_update_init(void)
+{
+	afs_callback_update_worker =
+		create_singlethread_workqueue("kafs_callbackd");
+	return afs_callback_update_worker ? 0 : -ENOMEM;
 }
 
 /*
- * allow the fileserver to see if the cache manager is still alive
+ * shut down the callback update process
  */
-int SRXAFSCM_Probe(struct afs_server *server)
+void __exit afs_callback_update_kill(void)
 {
-	_debug("SRXAFSCM_Probe(%p)\n", server);
-	return 0;
+	destroy_workqueue(afs_callback_update_worker);
 }