xen: add xenfs to allow usermode <-> Xen interaction
authorAlex Zeffertt <alex.zeffertt@eu.citrix.com>
Thu, 8 Jan 2009 02:07:11 +0000 (18:07 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 8 Jan 2009 16:30:59 +0000 (08:30 -0800)
The xenfs filesystem exports various interfaces to usermode.  Initially
this exports a file to allow usermode to interact with xenbus/xenstore.

Traditionally this appeared in /proc/xen.  Rather than extending procfs,
this patch adds a backward-compat mountpoint on /proc/xen, and provides
a xenfs filesystem which can be mounted there.

Signed-off-by: Alex Zeffertt <alex.zeffertt@eu.citrix.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
drivers/xen/Kconfig
drivers/xen/Makefile
drivers/xen/xenbus/xenbus_probe.c
drivers/xen/xenbus/xenbus_xs.c
drivers/xen/xenfs/Makefile [new file with mode: 0644]
drivers/xen/xenfs/super.c [new file with mode: 0644]
drivers/xen/xenfs/xenbus.c [new file with mode: 0644]
drivers/xen/xenfs/xenfs.h [new file with mode: 0644]
include/linux/magic.h
include/xen/xenbus.h

index 4b75a16de0094b3ea36f38d4d2b157ac3a20641d..526187c8a12de73ccbab828eb815eac4c2abd52e 100644 (file)
@@ -17,3 +17,27 @@ config XEN_SCRUB_PAGES
          is not accidentally visible to other domains.  Is it more
          secure, but slightly less efficient.
          If in doubt, say yes.
+
+config XENFS
+       tristate "Xen filesystem"
+       depends on XEN
+       default y
+       help
+         The xen filesystem provides a way for domains to share
+         information with each other and with the hypervisor.
+         For example, by reading and writing the "xenbus" file, guests
+         may pass arbitrary information to the initial domain.
+         If in doubt, say yes.
+
+config XEN_COMPAT_XENFS
+       bool "Create compatibility mount point /proc/xen"
+       depends on XENFS
+       default y
+       help
+         The old xenstore userspace tools expect to find "xenbus"
+         under /proc/xen, but "xenbus" is now found at the root of the
+         xenfs filesystem.  Selecting this causes the kernel to create
+         the compatibilty mount point /proc/xen if it is running on
+         a xen platform.
+         If in doubt, say yes.
+
index d2a8fdf0e1918c7ada49c7a1c5c369647ddbeafd..ff8accc9e103f9d2a4fc21dcbcc2e88d479a0ade 100644 (file)
@@ -1,5 +1,7 @@
 obj-y  += grant-table.o features.o events.o manage.o
 obj-y  += xenbus/
+
 obj-$(CONFIG_HOTPLUG_CPU)      += cpu_hotplug.o
 obj-$(CONFIG_XEN_XENCOMM)      += xencomm.o
 obj-$(CONFIG_XEN_BALLOON)      += balloon.o
+obj-$(CONFIG_XENFS)            += xenfs/
\ No newline at end of file
index b2a03184a2461af2544bd686a0ed4148d76f3d17..773d1cf2328334b9c62cc52fdf90c7107df02778 100644 (file)
@@ -40,6 +40,7 @@
 #include <linux/ctype.h>
 #include <linux/fcntl.h>
 #include <linux/mm.h>
+#include <linux/proc_fs.h>
 #include <linux/notifier.h>
 #include <linux/kthread.h>
 #include <linux/mutex.h>
 #include "xenbus_comms.h"
 #include "xenbus_probe.h"
 
+
 int xen_store_evtchn;
+EXPORT_SYMBOL(xen_store_evtchn);
+
 struct xenstore_domain_interface *xen_store_interface;
 static unsigned long xen_store_mfn;
 
@@ -166,6 +170,9 @@ static int read_backend_details(struct xenbus_device *xendev)
        return read_otherend_details(xendev, "backend-id", "backend");
 }
 
+static struct device_attribute xenbus_dev_attrs[] = {
+       __ATTR_NULL
+};
 
 /* Bus type for frontend drivers. */
 static struct xen_bus_type xenbus_frontend = {
@@ -174,12 +181,13 @@ static struct xen_bus_type xenbus_frontend = {
        .get_bus_id = frontend_bus_id,
        .probe = xenbus_probe_frontend,
        .bus = {
-               .name     = "xen",
-               .match    = xenbus_match,
-               .uevent   = xenbus_uevent,
-               .probe    = xenbus_dev_probe,
-               .remove   = xenbus_dev_remove,
-               .shutdown = xenbus_dev_shutdown,
+               .name      = "xen",
+               .match     = xenbus_match,
+               .uevent    = xenbus_uevent,
+               .probe     = xenbus_dev_probe,
+               .remove    = xenbus_dev_remove,
+               .shutdown  = xenbus_dev_shutdown,
+               .dev_attrs = xenbus_dev_attrs,
        },
 };
 
@@ -852,6 +860,14 @@ static int __init xenbus_probe_init(void)
        if (!xen_initial_domain())
                xenbus_probe(NULL);
 
+#ifdef CONFIG_XEN_COMPAT_XENFS
+       /*
+        * Create xenfs mountpoint in /proc for compatibility with
+        * utilities that expect to find "xenbus" under "/proc/xen".
+        */
+       proc_mkdir("xen", NULL);
+#endif
+
        return 0;
 
   out_unreg_back:
index 7f2f91c0e11dc0c8e5a343c6400ec664faadd824..e325eab4724d8cb0fab667b26f8c5b104c965c05 100644 (file)
@@ -184,6 +184,7 @@ void *xenbus_dev_request_and_reply(struct xsd_sockmsg *msg)
 
        return ret;
 }
+EXPORT_SYMBOL(xenbus_dev_request_and_reply);
 
 /* Send message to xs, get kmalloc'ed reply.  ERR_PTR() on error. */
 static void *xs_talkv(struct xenbus_transaction t,
diff --git a/drivers/xen/xenfs/Makefile b/drivers/xen/xenfs/Makefile
new file mode 100644 (file)
index 0000000..25275c3
--- /dev/null
@@ -0,0 +1,3 @@
+obj-$(CONFIG_XENFS) += xenfs.o
+
+xenfs-objs = super.o xenbus.o
\ No newline at end of file
diff --git a/drivers/xen/xenfs/super.c b/drivers/xen/xenfs/super.c
new file mode 100644 (file)
index 0000000..515741a
--- /dev/null
@@ -0,0 +1,64 @@
+/*
+ *  xenfs.c - a filesystem for passing info between the a domain and
+ *  the hypervisor.
+ *
+ * 2008-10-07  Alex Zeffertt    Replaced /proc/xen/xenbus with xenfs filesystem
+ *                              and /proc/xen compatibility mount point.
+ *                              Turned xenfs into a loadable module.
+ */
+
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/magic.h>
+
+#include "xenfs.h"
+
+#include <asm/xen/hypervisor.h>
+
+MODULE_DESCRIPTION("Xen filesystem");
+MODULE_LICENSE("GPL");
+
+static int xenfs_fill_super(struct super_block *sb, void *data, int silent)
+{
+       static struct tree_descr xenfs_files[] = {
+               [2] = {"xenbus", &xenbus_file_ops, S_IRUSR|S_IWUSR},
+               {""},
+       };
+
+       return simple_fill_super(sb, XENFS_SUPER_MAGIC, xenfs_files);
+}
+
+static int xenfs_get_sb(struct file_system_type *fs_type,
+                       int flags, const char *dev_name,
+                       void *data, struct vfsmount *mnt)
+{
+       return get_sb_single(fs_type, flags, data, xenfs_fill_super, mnt);
+}
+
+static struct file_system_type xenfs_type = {
+       .owner =        THIS_MODULE,
+       .name =         "xenfs",
+       .get_sb =       xenfs_get_sb,
+       .kill_sb =      kill_litter_super,
+};
+
+static int __init xenfs_init(void)
+{
+       if (xen_pv_domain())
+               return register_filesystem(&xenfs_type);
+
+       printk(KERN_INFO "XENFS: not registering filesystem on non-xen platform\n");
+       return 0;
+}
+
+static void __exit xenfs_exit(void)
+{
+       if (xen_pv_domain())
+               unregister_filesystem(&xenfs_type);
+}
+
+module_init(xenfs_init);
+module_exit(xenfs_exit);
+
diff --git a/drivers/xen/xenfs/xenbus.c b/drivers/xen/xenfs/xenbus.c
new file mode 100644 (file)
index 0000000..875a4c5
--- /dev/null
@@ -0,0 +1,593 @@
+/*
+ * Driver giving user-space access to the kernel's xenbus connection
+ * to xenstore.
+ *
+ * Copyright (c) 2005, Christian Limpach
+ * Copyright (c) 2005, Rusty Russell, IBM Corporation
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version 2
+ * as published by the Free Software Foundation; or, when distributed
+ * separately from the Linux kernel or incorporated into other
+ * software packages, subject to the following license:
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this source file (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use, copy, modify,
+ * merge, publish, distribute, sublicense, and/or sell copies of the Software,
+ * and to permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ * Changes:
+ * 2008-10-07  Alex Zeffertt    Replaced /proc/xen/xenbus with xenfs filesystem
+ *                              and /proc/xen compatibility mount point.
+ *                              Turned xenfs into a loadable module.
+ */
+
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/uio.h>
+#include <linux/notifier.h>
+#include <linux/wait.h>
+#include <linux/fs.h>
+#include <linux/poll.h>
+#include <linux/mutex.h>
+#include <linux/spinlock.h>
+#include <linux/mount.h>
+#include <linux/pagemap.h>
+#include <linux/uaccess.h>
+#include <linux/init.h>
+#include <linux/namei.h>
+#include <linux/string.h>
+
+#include "xenfs.h"
+#include "../xenbus/xenbus_comms.h"
+
+#include <xen/xenbus.h>
+#include <asm/xen/hypervisor.h>
+
+/*
+ * An element of a list of outstanding transactions, for which we're
+ * still waiting a reply.
+ */
+struct xenbus_transaction_holder {
+       struct list_head list;
+       struct xenbus_transaction handle;
+};
+
+/*
+ * A buffer of data on the queue.
+ */
+struct read_buffer {
+       struct list_head list;
+       unsigned int cons;
+       unsigned int len;
+       char msg[];
+};
+
+struct xenbus_file_priv {
+       /*
+        * msgbuffer_mutex is held while partial requests are built up
+        * and complete requests are acted on.  It therefore protects
+        * the "transactions" and "watches" lists, and the partial
+        * request length and buffer.
+        *
+        * reply_mutex protects the reply being built up to return to
+        * usermode.  It nests inside msgbuffer_mutex but may be held
+        * alone during a watch callback.
+        */
+       struct mutex msgbuffer_mutex;
+
+       /* In-progress transactions */
+       struct list_head transactions;
+
+       /* Active watches. */
+       struct list_head watches;
+
+       /* Partial request. */
+       unsigned int len;
+       union {
+               struct xsd_sockmsg msg;
+               char buffer[PAGE_SIZE];
+       } u;
+
+       /* Response queue. */
+       struct mutex reply_mutex;
+       struct list_head read_buffers;
+       wait_queue_head_t read_waitq;
+
+};
+
+/* Read out any raw xenbus messages queued up. */
+static ssize_t xenbus_file_read(struct file *filp,
+                              char __user *ubuf,
+                              size_t len, loff_t *ppos)
+{
+       struct xenbus_file_priv *u = filp->private_data;
+       struct read_buffer *rb;
+       unsigned i;
+       int ret;
+
+       mutex_lock(&u->reply_mutex);
+       while (list_empty(&u->read_buffers)) {
+               mutex_unlock(&u->reply_mutex);
+               ret = wait_event_interruptible(u->read_waitq,
+                                              !list_empty(&u->read_buffers));
+               if (ret)
+                       return ret;
+               mutex_lock(&u->reply_mutex);
+       }
+
+       rb = list_entry(u->read_buffers.next, struct read_buffer, list);
+       i = 0;
+       while (i < len) {
+               unsigned sz = min((unsigned)len - i, rb->len - rb->cons);
+
+               ret = copy_to_user(ubuf + i, &rb->msg[rb->cons], sz);
+
+               i += sz - ret;
+               rb->cons += sz - ret;
+
+               if (ret != sz) {
+                       if (i == 0)
+                               i = -EFAULT;
+                       goto out;
+               }
+
+               /* Clear out buffer if it has been consumed */
+               if (rb->cons == rb->len) {
+                       list_del(&rb->list);
+                       kfree(rb);
+                       if (list_empty(&u->read_buffers))
+                               break;
+                       rb = list_entry(u->read_buffers.next,
+                                       struct read_buffer, list);
+               }
+       }
+
+out:
+       mutex_unlock(&u->reply_mutex);
+       return i;
+}
+
+/*
+ * Add a buffer to the queue.  Caller must hold the appropriate lock
+ * if the queue is not local.  (Commonly the caller will build up
+ * multiple queued buffers on a temporary local list, and then add it
+ * to the appropriate list under lock once all the buffers have een
+ * successfully allocated.)
+ */
+static int queue_reply(struct list_head *queue, const void *data, size_t len)
+{
+       struct read_buffer *rb;
+
+       if (len == 0)
+               return 0;
+
+       rb = kmalloc(sizeof(*rb) + len, GFP_KERNEL);
+       if (rb == NULL)
+               return -ENOMEM;
+
+       rb->cons = 0;
+       rb->len = len;
+
+       memcpy(rb->msg, data, len);
+
+       list_add_tail(&rb->list, queue);
+       return 0;
+}
+
+/*
+ * Free all the read_buffer s on a list.
+ * Caller must have sole reference to list.
+ */
+static void queue_cleanup(struct list_head *list)
+{
+       struct read_buffer *rb;
+
+       while (!list_empty(list)) {
+               rb = list_entry(list->next, struct read_buffer, list);
+               list_del(list->next);
+               kfree(rb);
+       }
+}
+
+struct watch_adapter {
+       struct list_head list;
+       struct xenbus_watch watch;
+       struct xenbus_file_priv *dev_data;
+       char *token;
+};
+
+static void free_watch_adapter(struct watch_adapter *watch)
+{
+       kfree(watch->watch.node);
+       kfree(watch->token);
+       kfree(watch);
+}
+
+static struct watch_adapter *alloc_watch_adapter(const char *path,
+                                                const char *token)
+{
+       struct watch_adapter *watch;
+
+       watch = kzalloc(sizeof(*watch), GFP_KERNEL);
+       if (watch == NULL)
+               goto out_fail;
+
+       watch->watch.node = kstrdup(path, GFP_KERNEL);
+       if (watch->watch.node == NULL)
+               goto out_free;
+
+       watch->token = kstrdup(token, GFP_KERNEL);
+       if (watch->token == NULL)
+               goto out_free;
+
+       return watch;
+
+out_free:
+       free_watch_adapter(watch);
+
+out_fail:
+       return NULL;
+}
+
+static void watch_fired(struct xenbus_watch *watch,
+                       const char **vec,
+                       unsigned int len)
+{
+       struct watch_adapter *adap;
+       struct xsd_sockmsg hdr;
+       const char *path, *token;
+       int path_len, tok_len, body_len, data_len = 0;
+       int ret;
+       LIST_HEAD(staging_q);
+
+       adap = container_of(watch, struct watch_adapter, watch);
+
+       path = vec[XS_WATCH_PATH];
+       token = adap->token;
+
+       path_len = strlen(path) + 1;
+       tok_len = strlen(token) + 1;
+       if (len > 2)
+               data_len = vec[len] - vec[2] + 1;
+       body_len = path_len + tok_len + data_len;
+
+       hdr.type = XS_WATCH_EVENT;
+       hdr.len = body_len;
+
+       mutex_lock(&adap->dev_data->reply_mutex);
+
+       ret = queue_reply(&staging_q, &hdr, sizeof(hdr));
+       if (!ret)
+               ret = queue_reply(&staging_q, path, path_len);
+       if (!ret)
+               ret = queue_reply(&staging_q, token, tok_len);
+       if (!ret && len > 2)
+               ret = queue_reply(&staging_q, vec[2], data_len);
+
+       if (!ret) {
+               /* success: pass reply list onto watcher */
+               list_splice_tail(&staging_q, &adap->dev_data->read_buffers);
+               wake_up(&adap->dev_data->read_waitq);
+       } else
+               queue_cleanup(&staging_q);
+
+       mutex_unlock(&adap->dev_data->reply_mutex);
+}
+
+static int xenbus_write_transaction(unsigned msg_type,
+                                   struct xenbus_file_priv *u)
+{
+       int rc, ret;
+       void *reply;
+       struct xenbus_transaction_holder *trans = NULL;
+       LIST_HEAD(staging_q);
+
+       if (msg_type == XS_TRANSACTION_START) {
+               trans = kmalloc(sizeof(*trans), GFP_KERNEL);
+               if (!trans) {
+                       rc = -ENOMEM;
+                       goto out;
+               }
+       }
+
+       reply = xenbus_dev_request_and_reply(&u->u.msg);
+       if (IS_ERR(reply)) {
+               kfree(trans);
+               rc = PTR_ERR(reply);
+               goto out;
+       }
+
+       if (msg_type == XS_TRANSACTION_START) {
+               trans->handle.id = simple_strtoul(reply, NULL, 0);
+
+               list_add(&trans->list, &u->transactions);
+       } else if (msg_type == XS_TRANSACTION_END) {
+               list_for_each_entry(trans, &u->transactions, list)
+                       if (trans->handle.id == u->u.msg.tx_id)
+                               break;
+               BUG_ON(&trans->list == &u->transactions);
+               list_del(&trans->list);
+
+               kfree(trans);
+       }
+
+       mutex_lock(&u->reply_mutex);
+       ret = queue_reply(&staging_q, &u->u.msg, sizeof(u->u.msg));
+       if (!ret)
+               ret = queue_reply(&staging_q, reply, u->u.msg.len);
+       if (!ret) {
+               list_splice_tail(&staging_q, &u->read_buffers);
+               wake_up(&u->read_waitq);
+       } else {
+               queue_cleanup(&staging_q);
+               rc = ret;
+       }
+       mutex_unlock(&u->reply_mutex);
+
+       kfree(reply);
+
+out:
+       return rc;
+}
+
+static int xenbus_write_watch(unsigned msg_type, struct xenbus_file_priv *u)
+{
+       struct watch_adapter *watch, *tmp_watch;
+       char *path, *token;
+       int err, rc;
+       LIST_HEAD(staging_q);
+
+       path = u->u.buffer + sizeof(u->u.msg);
+       token = memchr(path, 0, u->u.msg.len);
+       if (token == NULL) {
+               rc = -EILSEQ;
+               goto out;
+       }
+       token++;
+
+       if (msg_type == XS_WATCH) {
+               watch = alloc_watch_adapter(path, token);
+               if (watch == NULL) {
+                       rc = -ENOMEM;
+                       goto out;
+               }
+
+               watch->watch.callback = watch_fired;
+               watch->dev_data = u;
+
+               err = register_xenbus_watch(&watch->watch);
+               if (err) {
+                       free_watch_adapter(watch);
+                       rc = err;
+                       goto out;
+               }
+               list_add(&watch->list, &u->watches);
+       } else {
+               list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
+                       if (!strcmp(watch->token, token) &&
+                           !strcmp(watch->watch.node, path)) {
+                               unregister_xenbus_watch(&watch->watch);
+                               list_del(&watch->list);
+                               free_watch_adapter(watch);
+                               break;
+                       }
+               }
+       }
+
+       /* Success.  Synthesize a reply to say all is OK. */
+       {
+               struct {
+                       struct xsd_sockmsg hdr;
+                       char body[3];
+               } __packed reply = {
+                       {
+                               .type = msg_type,
+                               .len = sizeof(reply.body)
+                       },
+                       "OK"
+               };
+
+               mutex_lock(&u->reply_mutex);
+               rc = queue_reply(&u->read_buffers, &reply, sizeof(reply));
+               mutex_unlock(&u->reply_mutex);
+       }
+
+out:
+       return rc;
+}
+
+static ssize_t xenbus_file_write(struct file *filp,
+                               const char __user *ubuf,
+                               size_t len, loff_t *ppos)
+{
+       struct xenbus_file_priv *u = filp->private_data;
+       uint32_t msg_type;
+       int rc = len;
+       int ret;
+       LIST_HEAD(staging_q);
+
+       /*
+        * We're expecting usermode to be writing properly formed
+        * xenbus messages.  If they write an incomplete message we
+        * buffer it up.  Once it is complete, we act on it.
+        */
+
+       /*
+        * Make sure concurrent writers can't stomp all over each
+        * other's messages and make a mess of our partial message
+        * buffer.  We don't make any attemppt to stop multiple
+        * writers from making a mess of each other's incomplete
+        * messages; we're just trying to guarantee our own internal
+        * consistency and make sure that single writes are handled
+        * atomically.
+        */
+       mutex_lock(&u->msgbuffer_mutex);
+
+       /* Get this out of the way early to avoid confusion */
+       if (len == 0)
+               goto out;
+
+       /* Can't write a xenbus message larger we can buffer */
+       if ((len + u->len) > sizeof(u->u.buffer)) {
+               /* On error, dump existing buffer */
+               u->len = 0;
+               rc = -EINVAL;
+               goto out;
+       }
+
+       ret = copy_from_user(u->u.buffer + u->len, ubuf, len);
+
+       if (ret == len) {
+               rc = -EFAULT;
+               goto out;
+       }
+
+       /* Deal with a partial copy. */
+       len -= ret;
+       rc = len;
+
+       u->len += len;
+
+       /* Return if we haven't got a full message yet */
+       if (u->len < sizeof(u->u.msg))
+               goto out;       /* not even the header yet */
+
+       /* If we're expecting a message that's larger than we can
+          possibly send, dump what we have and return an error. */
+       if ((sizeof(u->u.msg) + u->u.msg.len) > sizeof(u->u.buffer)) {
+               rc = -E2BIG;
+               u->len = 0;
+               goto out;
+       }
+
+       if (u->len < (sizeof(u->u.msg) + u->u.msg.len))
+               goto out;       /* incomplete data portion */
+
+       /*
+        * OK, now we have a complete message.  Do something with it.
+        */
+
+       msg_type = u->u.msg.type;
+
+       switch (msg_type) {
+       case XS_TRANSACTION_START:
+       case XS_TRANSACTION_END:
+       case XS_DIRECTORY:
+       case XS_READ:
+       case XS_GET_PERMS:
+       case XS_RELEASE:
+       case XS_GET_DOMAIN_PATH:
+       case XS_WRITE:
+       case XS_MKDIR:
+       case XS_RM:
+       case XS_SET_PERMS:
+               /* Send out a transaction */
+               ret = xenbus_write_transaction(msg_type, u);
+               break;
+
+       case XS_WATCH:
+       case XS_UNWATCH:
+               /* (Un)Ask for some path to be watched for changes */
+               ret = xenbus_write_watch(msg_type, u);
+               break;
+
+       default:
+               ret = -EINVAL;
+               break;
+       }
+       if (ret != 0)
+               rc = ret;
+
+       /* Buffered message consumed */
+       u->len = 0;
+
+ out:
+       mutex_unlock(&u->msgbuffer_mutex);
+       return rc;
+}
+
+static int xenbus_file_open(struct inode *inode, struct file *filp)
+{
+       struct xenbus_file_priv *u;
+
+       if (xen_store_evtchn == 0)
+               return -ENOENT;
+
+       nonseekable_open(inode, filp);
+
+       u = kzalloc(sizeof(*u), GFP_KERNEL);
+       if (u == NULL)
+               return -ENOMEM;
+
+       INIT_LIST_HEAD(&u->transactions);
+       INIT_LIST_HEAD(&u->watches);
+       INIT_LIST_HEAD(&u->read_buffers);
+       init_waitqueue_head(&u->read_waitq);
+
+       mutex_init(&u->reply_mutex);
+       mutex_init(&u->msgbuffer_mutex);
+
+       filp->private_data = u;
+
+       return 0;
+}
+
+static int xenbus_file_release(struct inode *inode, struct file *filp)
+{
+       struct xenbus_file_priv *u = filp->private_data;
+       struct xenbus_transaction_holder *trans, *tmp;
+       struct watch_adapter *watch, *tmp_watch;
+
+       /*
+        * No need for locking here because there are no other users,
+        * by definition.
+        */
+
+       list_for_each_entry_safe(trans, tmp, &u->transactions, list) {
+               xenbus_transaction_end(trans->handle, 1);
+               list_del(&trans->list);
+               kfree(trans);
+       }
+
+       list_for_each_entry_safe(watch, tmp_watch, &u->watches, list) {
+               unregister_xenbus_watch(&watch->watch);
+               list_del(&watch->list);
+               free_watch_adapter(watch);
+       }
+
+       kfree(u);
+
+       return 0;
+}
+
+static unsigned int xenbus_file_poll(struct file *file, poll_table *wait)
+{
+       struct xenbus_file_priv *u = file->private_data;
+
+       poll_wait(file, &u->read_waitq, wait);
+       if (!list_empty(&u->read_buffers))
+               return POLLIN | POLLRDNORM;
+       return 0;
+}
+
+const struct file_operations xenbus_file_ops = {
+       .read = xenbus_file_read,
+       .write = xenbus_file_write,
+       .open = xenbus_file_open,
+       .release = xenbus_file_release,
+       .poll = xenbus_file_poll,
+};
diff --git a/drivers/xen/xenfs/xenfs.h b/drivers/xen/xenfs/xenfs.h
new file mode 100644 (file)
index 0000000..51f08b2
--- /dev/null
@@ -0,0 +1,6 @@
+#ifndef _XENFS_XENBUS_H
+#define _XENFS_XENBUS_H
+
+extern const struct file_operations xenbus_file_ops;
+
+#endif /* _XENFS_XENBUS_H */
index f7f3fdddbef0b5f4f599cd506c524e4c9bf0b911..439f6f3cb0c445a18d5ee0036f500f06f836bafe 100644 (file)
@@ -13,6 +13,7 @@
 #define EFS_SUPER_MAGIC                0x414A53
 #define EXT2_SUPER_MAGIC       0xEF53
 #define EXT3_SUPER_MAGIC       0xEF53
+#define XENFS_SUPER_MAGIC      0xabba1974
 #define EXT4_SUPER_MAGIC       0xEF53
 #define HPFS_SUPER_MAGIC       0xf995e849
 #define ISOFS_SUPER_MAGIC      0x9660
index 6369d89c25d5346a6e8dcc4948218c883d900f37..f87f9614844db69e7eeacb2784646f5458ed954a 100644 (file)
@@ -136,8 +136,6 @@ struct xenbus_transaction
 /* Nil transaction ID. */
 #define XBT_NIL ((struct xenbus_transaction) { 0 })
 
-int __init xenbus_dev_init(void);
-
 char **xenbus_directory(struct xenbus_transaction t,
                        const char *dir, const char *node, unsigned int *num);
 void *xenbus_read(struct xenbus_transaction t,