rpc: add an rpc_pipe_open method
[safe/jmp/linux-2.6] / net / sunrpc / rpc_pipe.c
index 700c6e0..c9b57f4 100644 (file)
@@ -14,7 +14,7 @@
 #include <linux/pagemap.h>
 #include <linux/mount.h>
 #include <linux/namei.h>
-#include <linux/dnotify.h>
+#include <linux/fsnotify.h>
 #include <linux/kernel.h>
 
 #include <asm/ioctls.h>
@@ -33,7 +33,7 @@ static int rpc_mount_count;
 static struct file_system_type rpc_pipe_fs_type;
 
 
-static kmem_cache_t *rpc_inode_cachep __read_mostly;
+static struct kmem_cache *rpc_inode_cachep __read_mostly;
 
 #define RPC_UPCALL_TIMEOUT (30*HZ)
 
@@ -54,10 +54,11 @@ static void rpc_purge_list(struct rpc_inode *rpci, struct list_head *head,
 }
 
 static void
-rpc_timeout_upcall_queue(void *data)
+rpc_timeout_upcall_queue(struct work_struct *work)
 {
        LIST_HEAD(free_list);
-       struct rpc_inode *rpci = (struct rpc_inode *)data;
+       struct rpc_inode *rpci =
+               container_of(work, struct rpc_inode, queue_timeout.work);
        struct inode *inode = &rpci->vfs_inode;
        void (*destroy_msg)(struct rpc_pipe_msg *);
 
@@ -75,6 +76,16 @@ rpc_timeout_upcall_queue(void *data)
        rpc_purge_list(rpci, &free_list, destroy_msg, -ETIMEDOUT);
 }
 
+/**
+ * rpc_queue_upcall
+ * @inode: inode of upcall pipe on which to queue given message
+ * @msg: message to queue
+ *
+ * Call with an @inode created by rpc_mkpipe() to queue an upcall.
+ * A userspace process may then later read the upcall by performing a
+ * read on an open file for this inode.  It is up to the caller to
+ * initialize the fields of @msg (other than @msg->list) appropriately.
+ */
 int
 rpc_queue_upcall(struct inode *inode, struct rpc_pipe_msg *msg)
 {
@@ -102,6 +113,7 @@ out:
        wake_up(&rpci->waitq);
        return res;
 }
+EXPORT_SYMBOL_GPL(rpc_queue_upcall);
 
 static inline void
 rpc_inode_setowner(struct inode *inode, void *private)
@@ -131,8 +143,7 @@ rpc_close_pipes(struct inode *inode)
                rpci->nwriters = 0;
                if (ops->release_pipe)
                        ops->release_pipe(inode);
-               cancel_delayed_work(&rpci->queue_timeout);
-               flush_workqueue(rpciod_workqueue);
+               cancel_delayed_work_sync(&rpci->queue_timeout);
        }
        rpc_inode_setowner(inode, NULL);
        mutex_unlock(&inode->i_mutex);
@@ -142,7 +153,7 @@ static struct inode *
 rpc_alloc_inode(struct super_block *sb)
 {
        struct rpc_inode *rpci;
-       rpci = (struct rpc_inode *)kmem_cache_alloc(rpc_inode_cachep, SLAB_KERNEL);
+       rpci = (struct rpc_inode *)kmem_cache_alloc(rpc_inode_cachep, GFP_KERNEL);
        if (!rpci)
                return NULL;
        return &rpci->vfs_inode;
@@ -158,16 +169,24 @@ static int
 rpc_pipe_open(struct inode *inode, struct file *filp)
 {
        struct rpc_inode *rpci = RPC_I(inode);
+       int first_open;
        int res = -ENXIO;
 
        mutex_lock(&inode->i_mutex);
-       if (rpci->ops != NULL) {
-               if (filp->f_mode & FMODE_READ)
-                       rpci->nreaders ++;
-               if (filp->f_mode & FMODE_WRITE)
-                       rpci->nwriters ++;
-               res = 0;
+       if (rpci->ops == NULL)
+               goto out;
+       first_open = rpci->nreaders == 0 && rpci->nwriters == 0;
+       if (first_open && rpci->ops->open_pipe) {
+               res = rpci->ops->open_pipe(inode);
+               if (res)
+                       goto out;
        }
+       if (filp->f_mode & FMODE_READ)
+               rpci->nreaders++;
+       if (filp->f_mode & FMODE_WRITE)
+               rpci->nwriters++;
+       res = 0;
+out:
        mutex_unlock(&inode->i_mutex);
        return res;
 }
@@ -213,7 +232,7 @@ out:
 static ssize_t
 rpc_pipe_read(struct file *filp, char __user *buf, size_t len, loff_t *offset)
 {
-       struct inode *inode = filp->f_dentry->d_inode;
+       struct inode *inode = filp->f_path.dentry->d_inode;
        struct rpc_inode *rpci = RPC_I(inode);
        struct rpc_pipe_msg *msg;
        int res = 0;
@@ -256,7 +275,7 @@ out_unlock:
 static ssize_t
 rpc_pipe_write(struct file *filp, const char __user *buf, size_t len, loff_t *offset)
 {
-       struct inode *inode = filp->f_dentry->d_inode;
+       struct inode *inode = filp->f_path.dentry->d_inode;
        struct rpc_inode *rpci = RPC_I(inode);
        int res;
 
@@ -274,13 +293,13 @@ rpc_pipe_poll(struct file *filp, struct poll_table_struct *wait)
        struct rpc_inode *rpci;
        unsigned int mask = 0;
 
-       rpci = RPC_I(filp->f_dentry->d_inode);
+       rpci = RPC_I(filp->f_path.dentry->d_inode);
        poll_wait(filp, &rpci->waitq, wait);
 
        mask = POLLOUT | POLLWRNORM;
        if (rpci->ops == NULL)
                mask |= POLLERR | POLLHUP;
-       if (!list_empty(&rpci->pipe))
+       if (filp->private_data || !list_empty(&rpci->pipe))
                mask |= POLLIN | POLLRDNORM;
        return mask;
 }
@@ -289,7 +308,7 @@ static int
 rpc_pipe_ioctl(struct inode *ino, struct file *filp,
                unsigned int cmd, unsigned long arg)
 {
-       struct rpc_inode *rpci = RPC_I(filp->f_dentry->d_inode);
+       struct rpc_inode *rpci = RPC_I(filp->f_path.dentry->d_inode);
        int len;
 
        switch (cmd) {
@@ -308,7 +327,7 @@ rpc_pipe_ioctl(struct inode *ino, struct file *filp,
        }
 }
 
-static struct file_operations rpc_pipe_fops = {
+static const struct file_operations rpc_pipe_fops = {
        .owner          = THIS_MODULE,
        .llseek         = no_llseek,
        .read           = rpc_pipe_read,
@@ -329,6 +348,7 @@ rpc_show_info(struct seq_file *m, void *v)
                        clnt->cl_prog, clnt->cl_vers);
        seq_printf(m, "address: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_ADDR));
        seq_printf(m, "protocol: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PROTO));
+       seq_printf(m, "port: %s\n", rpc_peeraddr2str(clnt, RPC_DISPLAY_PORT));
        return 0;
 }
 
@@ -343,7 +363,7 @@ rpc_info_open(struct inode *inode, struct file *file)
                mutex_lock(&inode->i_mutex);
                clnt = RPC_I(inode)->private;
                if (clnt) {
-                       atomic_inc(&clnt->cl_users);
+                       kref_get(&clnt->cl_kref);
                        m->private = clnt;
                } else {
                        single_release(inode, file);
@@ -365,7 +385,7 @@ rpc_info_release(struct inode *inode, struct file *file)
        return single_release(inode, file);
 }
 
-static struct file_operations rpc_info_operations = {
+static const struct file_operations rpc_info_operations = {
        .owner          = THIS_MODULE,
        .open           = rpc_info_open,
        .read           = seq_read,
@@ -447,26 +467,33 @@ void rpc_put_mount(void)
        simple_release_fs(&rpc_mount, &rpc_mount_count);
 }
 
+static int rpc_delete_dentry(struct dentry *dentry)
+{
+       return 1;
+}
+
+static struct dentry_operations rpc_dentry_operations = {
+       .d_delete = rpc_delete_dentry,
+};
+
 static int
 rpc_lookup_parent(char *path, struct nameidata *nd)
 {
+       struct vfsmount *mnt;
+
        if (path[0] == '\0')
                return -ENOENT;
-       nd->mnt = rpc_get_mount();
-       if (IS_ERR(nd->mnt)) {
+
+       mnt = rpc_get_mount();
+       if (IS_ERR(mnt)) {
                printk(KERN_WARNING "%s: %s failed to mount "
-                              "pseudofilesystem \n", __FILE__, __FUNCTION__);
-               return PTR_ERR(nd->mnt);
+                              "pseudofilesystem \n", __FILE__, __func__);
+               return PTR_ERR(mnt);
        }
-       mntget(nd->mnt);
-       nd->dentry = dget(rpc_mount->mnt_root);
-       nd->last_type = LAST_ROOT;
-       nd->flags = LOOKUP_PARENT;
-       nd->depth = 0;
 
-       if (path_walk(path, nd)) {
+       if (vfs_path_lookup(mnt->mnt_root, mnt, path, LOOKUP_PARENT, nd)) {
                printk(KERN_WARNING "%s: %s failed to find path %s\n",
-                               __FILE__, __FUNCTION__, path);
+                               __FILE__, __func__, path);
                rpc_put_mount();
                return -ENOENT;
        }
@@ -476,7 +503,7 @@ rpc_lookup_parent(char *path, struct nameidata *nd)
 static void
 rpc_release_path(struct nameidata *nd)
 {
-       path_release(nd);
+       path_put(&nd->path);
        rpc_put_mount();
 }
 
@@ -494,7 +521,7 @@ rpc_get_inode(struct super_block *sb, int mode)
                case S_IFDIR:
                        inode->i_fop = &simple_dir_operations;
                        inode->i_op = &simple_dir_inode_operations;
-                       inode->i_nlink++;
+                       inc_nlink(inode);
                default:
                        break;
        }
@@ -504,8 +531,8 @@ rpc_get_inode(struct super_block *sb, int mode)
 /*
  * FIXME: This probably has races.
  */
-static void
-rpc_depopulate(struct dentry *parent)
+static void rpc_depopulate(struct dentry *parent,
+                          unsigned long start, unsigned long eof)
 {
        struct inode *dir = parent->d_inode;
        struct list_head *pos, *next;
@@ -517,6 +544,10 @@ repeat:
        spin_lock(&dcache_lock);
        list_for_each_safe(pos, next, &parent->d_subdirs) {
                dentry = list_entry(pos, struct dentry, d_u.d_child);
+               if (!dentry->d_inode ||
+                               dentry->d_inode->i_ino < start ||
+                               dentry->d_inode->i_ino >= eof)
+                       continue;
                spin_lock(&dentry->d_lock);
                if (!d_unhashed(dentry)) {
                        dget_locked(dentry);
@@ -532,11 +563,11 @@ repeat:
        if (n) {
                do {
                        dentry = dvec[--n];
-                       if (dentry->d_inode) {
-                               rpc_close_pipes(dentry->d_inode);
+                       if (S_ISREG(dentry->d_inode->i_mode))
                                simple_unlink(dir, dentry);
-                       }
-                       inode_dir_notify(dir, DN_DELETE);
+                       else if (S_ISDIR(dentry->d_inode->i_mode))
+                               simple_rmdir(dir, dentry);
+                       d_delete(dentry);
                        dput(dentry);
                } while (n);
                goto repeat;
@@ -559,6 +590,7 @@ rpc_populate(struct dentry *parent,
                dentry = d_alloc_name(parent, files[i].name);
                if (!dentry)
                        goto out_bad;
+               dentry->d_op = &rpc_dentry_operations;
                mode = files[i].mode;
                inode = rpc_get_inode(dir->i_sb, mode);
                if (!inode) {
@@ -571,15 +603,16 @@ rpc_populate(struct dentry *parent,
                if (private)
                        rpc_inode_setowner(inode, private);
                if (S_ISDIR(mode))
-                       dir->i_nlink++;
+                       inc_nlink(dir);
                d_add(dentry, inode);
+               fsnotify_create(dir, dentry);
        }
        mutex_unlock(&dir->i_mutex);
        return 0;
 out_bad:
        mutex_unlock(&dir->i_mutex);
        printk(KERN_WARNING "%s: %s failed to populate directory %s\n",
-                       __FILE__, __FUNCTION__, parent->d_name.name);
+                       __FILE__, __func__, parent->d_name.name);
        return -ENOMEM;
 }
 
@@ -588,17 +621,17 @@ __rpc_mkdir(struct inode *dir, struct dentry *dentry)
 {
        struct inode *inode;
 
-       inode = rpc_get_inode(dir->i_sb, S_IFDIR | S_IRUSR | S_IXUSR);
+       inode = rpc_get_inode(dir->i_sb, S_IFDIR | S_IRUGO | S_IXUGO);
        if (!inode)
                goto out_err;
        inode->i_ino = iunique(dir->i_sb, 100);
        d_instantiate(dentry, inode);
-       dir->i_nlink++;
-       inode_dir_notify(dir, DN_CREATE);
+       inc_nlink(dir);
+       fsnotify_mkdir(dir, dentry);
        return 0;
 out_err:
        printk(KERN_WARNING "%s: %s failed to allocate inode for dentry %s\n",
-                       __FILE__, __FUNCTION__, dentry->d_name.name);
+                       __FILE__, __func__, dentry->d_name.name);
        return -ENOMEM;
 }
 
@@ -606,21 +639,14 @@ static int
 __rpc_rmdir(struct inode *dir, struct dentry *dentry)
 {
        int error;
-
-       shrink_dcache_parent(dentry);
-       if (d_unhashed(dentry))
-               return 0;
-       if ((error = simple_rmdir(dir, dentry)) != 0)
-               return error;
-       if (!error) {
-               inode_dir_notify(dir, DN_DELETE);
-               d_drop(dentry);
-       }
-       return 0;
+       error = simple_rmdir(dir, dentry);
+       if (!error)
+               d_delete(dentry);
+       return error;
 }
 
 static struct dentry *
-rpc_lookup_create(struct dentry *parent, const char *name, int len)
+rpc_lookup_create(struct dentry *parent, const char *name, int len, int exclusive)
 {
        struct inode *dir = parent->d_inode;
        struct dentry *dentry;
@@ -629,7 +655,9 @@ rpc_lookup_create(struct dentry *parent, const char *name, int len)
        dentry = lookup_one_len(name, parent, len);
        if (IS_ERR(dentry))
                goto out_err;
-       if (dentry->d_inode) {
+       if (!dentry->d_inode)
+               dentry->d_op = &rpc_dentry_operations;
+       else if (exclusive) {
                dput(dentry);
                dentry = ERR_PTR(-EEXIST);
                goto out_err;
@@ -648,13 +676,23 @@ rpc_lookup_negative(char *path, struct nameidata *nd)
 
        if ((error = rpc_lookup_parent(path, nd)) != 0)
                return ERR_PTR(error);
-       dentry = rpc_lookup_create(nd->dentry, nd->last.name, nd->last.len);
+       dentry = rpc_lookup_create(nd->path.dentry, nd->last.name, nd->last.len,
+                                  1);
        if (IS_ERR(dentry))
                rpc_release_path(nd);
        return dentry;
 }
 
-
+/**
+ * rpc_mkdir - Create a new directory in rpc_pipefs
+ * @path: path from the rpc_pipefs root to the new directory
+ * @rpc_client: rpc client to associate with this directory
+ *
+ * This creates a directory at the given @path associated with
+ * @rpc_clnt, which will contain a file named "info" with some basic
+ * information about the client, together with any "pipes" that may
+ * later be created using rpc_mkpipe().
+ */
 struct dentry *
 rpc_mkdir(char *path, struct rpc_clnt *rpc_client)
 {
@@ -666,7 +704,7 @@ rpc_mkdir(char *path, struct rpc_clnt *rpc_client)
        dentry = rpc_lookup_negative(path, &nd);
        if (IS_ERR(dentry))
                return dentry;
-       dir = nd.dentry->d_inode;
+       dir = nd.path.dentry->d_inode;
        if ((error = __rpc_mkdir(dir, dentry)) != 0)
                goto err_dput;
        RPC_I(dentry->d_inode)->private = rpc_client;
@@ -680,16 +718,20 @@ out:
        rpc_release_path(&nd);
        return dentry;
 err_depopulate:
-       rpc_depopulate(dentry);
+       rpc_depopulate(dentry, RPCAUTH_info, RPCAUTH_EOF);
        __rpc_rmdir(dir, dentry);
 err_dput:
        dput(dentry);
        printk(KERN_WARNING "%s: %s() failed to create directory %s (errno = %d)\n",
-                       __FILE__, __FUNCTION__, path, error);
+                       __FILE__, __func__, path, error);
        dentry = ERR_PTR(error);
        goto out;
 }
 
+/**
+ * rpc_rmdir - Remove a directory created with rpc_mkdir()
+ * @dentry: directory to remove
+ */
 int
 rpc_rmdir(struct dentry *dentry)
 {
@@ -700,7 +742,7 @@ rpc_rmdir(struct dentry *dentry)
        parent = dget_parent(dentry);
        dir = parent->d_inode;
        mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
-       rpc_depopulate(dentry);
+       rpc_depopulate(dentry, RPCAUTH_info, RPCAUTH_EOF);
        error = __rpc_rmdir(dir, dentry);
        dput(dentry);
        mutex_unlock(&dir->i_mutex);
@@ -708,6 +750,26 @@ rpc_rmdir(struct dentry *dentry)
        return error;
 }
 
+/**
+ * rpc_mkpipe - make an rpc_pipefs file for kernel<->userspace communication
+ * @parent: dentry of directory to create new "pipe" in
+ * @name: name of pipe
+ * @private: private data to associate with the pipe, for the caller's use
+ * @ops: operations defining the behavior of the pipe: upcall, downcall,
+ *     release_pipe, open_pipe, and destroy_msg.
+ * @flags: rpc_inode flags
+ *
+ * Data is made available for userspace to read by calls to
+ * rpc_queue_upcall().  The actual reads will result in calls to
+ * @ops->upcall, which will be called with the file pointer,
+ * message, and userspace buffer to copy to.
+ *
+ * Writes can come at any time, and do not necessarily have to be
+ * responses to upcalls.  They will result in calls to @msg->downcall.
+ *
+ * The @private argument passed here will be available to all these methods
+ * from the file pointer, via RPC_I(file->f_dentry->d_inode)->private.
+ */
 struct dentry *
 rpc_mkpipe(struct dentry *parent, const char *name, void *private, struct rpc_pipe_ops *ops, int flags)
 {
@@ -715,10 +777,21 @@ rpc_mkpipe(struct dentry *parent, const char *name, void *private, struct rpc_pi
        struct inode *dir, *inode;
        struct rpc_inode *rpci;
 
-       dentry = rpc_lookup_create(parent, name, strlen(name));
+       dentry = rpc_lookup_create(parent, name, strlen(name), 0);
        if (IS_ERR(dentry))
                return dentry;
        dir = parent->d_inode;
+       if (dentry->d_inode) {
+               rpci = RPC_I(dentry->d_inode);
+               if (rpci->private != private ||
+                               rpci->ops != ops ||
+                               rpci->flags != flags) {
+                       dput (dentry);
+                       dentry = ERR_PTR(-EBUSY);
+               }
+               rpci->nkern_readwriters++;
+               goto out;
+       }
        inode = rpc_get_inode(dir->i_sb, S_IFIFO | S_IRUSR | S_IWUSR);
        if (!inode)
                goto err_dput;
@@ -729,7 +802,8 @@ rpc_mkpipe(struct dentry *parent, const char *name, void *private, struct rpc_pi
        rpci->private = private;
        rpci->flags = flags;
        rpci->ops = ops;
-       inode_dir_notify(dir, DN_CREATE);
+       rpci->nkern_readwriters = 1;
+       fsnotify_create(dir, dentry);
        dget(dentry);
 out:
        mutex_unlock(&dir->i_mutex);
@@ -738,11 +812,20 @@ err_dput:
        dput(dentry);
        dentry = ERR_PTR(-ENOMEM);
        printk(KERN_WARNING "%s: %s() failed to create pipe %s/%s (errno = %d)\n",
-                       __FILE__, __FUNCTION__, parent->d_name.name, name,
+                       __FILE__, __func__, parent->d_name.name, name,
                        -ENOMEM);
        goto out;
 }
+EXPORT_SYMBOL_GPL(rpc_mkpipe);
 
+/**
+ * rpc_unlink - remove a pipe
+ * @dentry: dentry for the pipe, as returned from rpc_mkpipe
+ *
+ * After this call, lookups will no longer find the pipe, and any
+ * attempts to read or write using preexisting opens of the pipe will
+ * return -EPIPE.
+ */
 int
 rpc_unlink(struct dentry *dentry)
 {
@@ -753,19 +836,18 @@ rpc_unlink(struct dentry *dentry)
        parent = dget_parent(dentry);
        dir = parent->d_inode;
        mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
-       if (!d_unhashed(dentry)) {
-               d_drop(dentry);
-               if (dentry->d_inode) {
-                       rpc_close_pipes(dentry->d_inode);
-                       error = simple_unlink(dir, dentry);
-               }
-               inode_dir_notify(dir, DN_DELETE);
+       if (--RPC_I(dentry->d_inode)->nkern_readwriters == 0) {
+               rpc_close_pipes(dentry->d_inode);
+               error = simple_unlink(dir, dentry);
+               if (!error)
+                       d_delete(dentry);
        }
        dput(dentry);
        mutex_unlock(&dir->i_mutex);
        dput(parent);
        return error;
 }
+EXPORT_SYMBOL_GPL(rpc_unlink);
 
 /*
  * populate the filesystem
@@ -823,35 +905,41 @@ static struct file_system_type rpc_pipe_fs_type = {
 };
 
 static void
-init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
+init_once(void *foo)
 {
        struct rpc_inode *rpci = (struct rpc_inode *) foo;
 
-       if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
-           SLAB_CTOR_CONSTRUCTOR) {
-               inode_init_once(&rpci->vfs_inode);
-               rpci->private = NULL;
-               rpci->nreaders = 0;
-               rpci->nwriters = 0;
-               INIT_LIST_HEAD(&rpci->in_upcall);
-               INIT_LIST_HEAD(&rpci->pipe);
-               rpci->pipelen = 0;
-               init_waitqueue_head(&rpci->waitq);
-               INIT_WORK(&rpci->queue_timeout, rpc_timeout_upcall_queue, rpci);
-               rpci->ops = NULL;
-       }
+       inode_init_once(&rpci->vfs_inode);
+       rpci->private = NULL;
+       rpci->nreaders = 0;
+       rpci->nwriters = 0;
+       INIT_LIST_HEAD(&rpci->in_upcall);
+       INIT_LIST_HEAD(&rpci->in_downcall);
+       INIT_LIST_HEAD(&rpci->pipe);
+       rpci->pipelen = 0;
+       init_waitqueue_head(&rpci->waitq);
+       INIT_DELAYED_WORK(&rpci->queue_timeout,
+                           rpc_timeout_upcall_queue);
+       rpci->ops = NULL;
 }
 
 int register_rpc_pipefs(void)
 {
+       int err;
+
        rpc_inode_cachep = kmem_cache_create("rpc_inode_cache",
                                sizeof(struct rpc_inode),
                                0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
                                                SLAB_MEM_SPREAD),
-                               init_once, NULL);
+                               init_once);
        if (!rpc_inode_cachep)
                return -ENOMEM;
-       register_filesystem(&rpc_pipe_fs_type);
+       err = register_filesystem(&rpc_pipe_fs_type);
+       if (err) {
+               kmem_cache_destroy(rpc_inode_cachep);
+               return err;
+       }
+
        return 0;
 }