[Ocfs2-commits] mfasheh commits r1832 - trunk/fs/ocfs2/dlm
svn-commits at oss.oracle.com
svn-commits at oss.oracle.com
Fri Jan 21 19:58:23 CST 2005
Author: mfasheh
Date: 2005-01-21 19:58:21 -0600 (Fri, 21 Jan 2005)
New Revision: 1832
Added:
trunk/fs/ocfs2/dlm/dlmfs.c
trunk/fs/ocfs2/dlm/dlmfs_compat.c
trunk/fs/ocfs2/dlm/dlmfs_compat.h
trunk/fs/ocfs2/dlm/userdlm.c
trunk/fs/ocfs2/dlm/userdlm.h
Modified:
trunk/fs/ocfs2/dlm/Makefile
Log:
* Expose a minimal (blocking, no value blocks) dlm management layer
to userspace via a pseudo file system.
Modified: trunk/fs/ocfs2/dlm/Makefile
===================================================================
--- trunk/fs/ocfs2/dlm/Makefile 2005-01-22 01:47:41 UTC (rev 1831)
+++ trunk/fs/ocfs2/dlm/Makefile 2005-01-22 01:58:21 UTC (rev 1832)
@@ -19,11 +19,18 @@
EXTRA_CFLAGS += $(MODULE_CFLAGS) $(MODULE_DEFINES)
-obj-m := ocfs2_dlm.o
+ifeq ($(KERNELRELEASE),)
+ifeq ($(KERNEL_26),)
+DLMFS_COMPAT := dlmfs_compat.o
+endif
+endif
+obj-m := ocfs2_dlm.o ocfs2_dlmfs.o
+
ocfs2_dlm-objs := dlmmod.o dlmthread.o dlmrecovery.o util.o \
dlmmaster.o dlmast.o dlmconvert.o dlmlock.o dlmunlock.o
+ocfs2_dlmfs-objs := userdlm.o dlmfs.o $(DLMFS_COMPAT)
ifeq ($(KERNELRELEASE),)
@@ -31,21 +38,26 @@
# Called from a regular "make".
#
-SOURCES = \
- dlmmaster.c \
- dlmmod.c \
- dlmast.c \
- dlmconvert.c \
- dlmlock.c \
- dlmunlock.c \
- dlmrecovery.c \
- dlmthread.c \
+SOURCES = \
+ dlmmaster.c \
+ dlmmod.c \
+ dlmast.c \
+ dlmconvert.c \
+ dlmlock.c \
+ dlmunlock.c \
+ dlmrecovery.c \
+ dlmthread.c \
+ userdlm.c \
+ dlmfs.c \
+ dlmfs_compat.c \
util.c
-HEADERS = \
- dlmcommon.h \
- dlmmod.h \
- util.h \
+HEADERS = \
+ dlmcommon.h \
+ dlmmod.h \
+ userdlm.h \
+ dlmfs_compat.h \
+ util.h \
warning_hack.h
DIST_FILES = $(SOURCES) $(HEADERS)
Added: trunk/fs/ocfs2/dlm/dlmfs.c
===================================================================
--- trunk/fs/ocfs2/dlm/dlmfs.c 2005-01-22 01:47:41 UTC (rev 1831)
+++ trunk/fs/ocfs2/dlm/dlmfs.c 2005-01-22 01:58:21 UTC (rev 1832)
@@ -0,0 +1,565 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * userdlmfs.c
+ *
+ * Code which implements the kernel side of a minimal userspace
+ * interface to our DLM. This file handles the virtual file system
+ * used for communication with userspace. Credit should go to ramfs,
+ * which was a template for the fs side of this module.
+ *
+ * Copyright (C) 2003, 2004 Oracle. All rights reserved.
+ *
+ * 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 the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/pagemap.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/init.h>
+#include <linux/string.h>
+#include <linux/smp_lock.h>
+
+#include <asm/uaccess.h>
+
+#include "util.h"
+
+#include "cluster/clcommon.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmcommon.h"
+#include "dlmmod.h"
+
+#include "dlmfs_compat.h"
+#include "userdlm.h"
+
+static struct super_operations dlmfs_ops;
+static struct file_operations dlmfs_file_operations;
+static struct inode_operations dlmfs_dir_inode_operations;
+static struct inode_operations dlmfs_root_inode_operations;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+static struct inode_operations dlmfs_file_inode_operations;
+#endif
+static kmem_cache_t *dlmfs_inode_cache;
+
+struct workqueue_struct *user_dlm_worker;
+
+/*
+ * decodes a set of open flags into a valid lock level and a set of flags.
+ * returns < 0 if we have invalid flags
+ * flags which mean something to us:
+ * O_RDONLY -> PRMODE level
+ * O_WRONLY -> EXMODE level
+ *
+ * O_NONBLOCK -> LKM_NOQUEUE
+ */
+static int dlmfs_decode_open_flags(int open_flags,
+ int *level,
+ int *flags)
+{
+ if (open_flags & O_RDWR)
+ return -EACCES;
+
+ *level = (open_flags & O_WRONLY) ? LKM_EXMODE : LKM_PRMODE;
+ *flags = 0;
+ if (open_flags & O_NONBLOCK)
+ *flags = LKM_NOQUEUE;
+
+ return 0;
+}
+
+static int dlmfs_file_open(struct inode *inode,
+ struct file *file)
+{
+ int status, level, flags;
+ struct dlmfs_filp_private *fp = NULL;
+ struct dlmfs_inode_private *ip;
+
+ if (S_ISDIR(inode->i_mode))
+ BUG();
+
+ dprintk("open called on inode %lu, flags 0x%x\n", inode->i_ino,
+ file->f_flags);
+
+ status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
+ if (status < 0)
+ goto bail;
+
+ fp = kmalloc(sizeof(*fp), GFP_KERNEL);
+ if (!fp) {
+ status = -ENOMEM;
+ goto bail;
+ }
+ fp->fp_lock_level = level;
+
+ ip = DLMFS_I(inode);
+
+ status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
+ if (status < 0) {
+ /* this is a strange error to return here but I want
+ * to be able userspace to be able to distinguish a
+ * valid lock request from one that simply couldn't be
+ * granted. */
+ if (flags & LKM_NOQUEUE && status == -EAGAIN)
+ status = -ETXTBSY;
+ kfree(fp);
+ goto bail;
+ }
+
+ file->private_data = fp;
+bail:
+ return status;
+}
+
+static int dlmfs_file_release(struct inode *inode,
+ struct file *file)
+{
+ int level, status;
+ struct dlmfs_inode_private *ip = DLMFS_I(inode);
+ struct dlmfs_filp_private *fp =
+ (struct dlmfs_filp_private *) file->private_data;
+
+ if (S_ISDIR(inode->i_mode))
+ BUG();
+
+ dprintk("close called on inode %lu\n", inode->i_ino);
+
+ status = 0;
+ if (fp) {
+ level = fp->fp_lock_level;
+ if (level != LKM_IVMODE)
+ user_dlm_cluster_unlock(&ip->ip_lockres, level);
+
+ kfree(fp);
+ file->private_data = NULL;
+ }
+
+ return 0;
+}
+
+static void dlmfs_init_once(void *foo,
+ kmem_cache_t *cachep,
+ unsigned long flags)
+{
+ struct dlmfs_inode_private *ip =
+ (struct dlmfs_inode_private *) foo;
+
+ if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
+ SLAB_CTOR_CONSTRUCTOR) {
+ ip->ip_dlm = NULL;
+ ip->ip_parent = NULL;
+
+ inode_init_once(&ip->ip_vfs_inode);
+ }
+}
+
+static struct inode *dlmfs_alloc_inode(struct super_block *sb)
+{
+ struct dlmfs_inode_private *ip;
+
+ ip = kmem_cache_alloc(dlmfs_inode_cache, SLAB_NOFS);
+ if (!ip)
+ return NULL;
+
+ return &ip->ip_vfs_inode;
+}
+
+static void dlmfs_destroy_inode(struct inode *inode)
+{
+ kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
+}
+
+static void dlmfs_clear_inode(struct inode *inode)
+{
+ int status;
+ struct dlmfs_inode_private *ip;
+
+ if (!inode)
+ return;
+
+ dprintk("inode %lu\n", inode->i_ino);
+
+ ip = DLMFS_I(inode);
+
+ if (S_ISREG(inode->i_mode)) {
+ status = user_dlm_destroy_lock(&ip->ip_lockres);
+ if (status < 0)
+ err_printk("status return of %d from "
+ "user_dlm_destroy_lock!", status);
+ iput(ip->ip_parent);
+ goto clear_fields;
+ }
+
+ dprintk("we're a directory, ip->ip_dlm = 0x%p\n", ip->ip_dlm);
+ /* we must be a directory. If required, lets unregister the
+ * dlm context now. */
+ if (ip->ip_dlm)
+ user_dlm_unregister_context(ip->ip_dlm);
+clear_fields:
+ ip->ip_parent = NULL;
+ ip->ip_dlm = NULL;
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+#include <linux/backing-dev.h>
+
+static struct backing_dev_info dlmfs_backing_dev_info = {
+ .ra_pages = 0, /* No readahead */
+ .memory_backed = 1, /* Does not contribute to dirty memory */
+};
+
+static inline void dlmfs_set_backing_info(struct inode *inode)
+{
+ inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
+}
+#else
+static inline void dlmfs_set_backing_info(struct inode *inode)
+{
+ inode->i_rdev = NODEV;
+}
+#endif
+
+static struct inode *dlmfs_get_root_inode(struct super_block *sb)
+{
+ struct inode *inode = new_inode(sb);
+ int mode = S_IFDIR | 0755;
+ struct dlmfs_inode_private *ip;
+
+ if (inode) {
+ ip = DLMFS_I(inode);
+
+ inode->i_mode = mode;
+ inode->i_uid = current->fsuid;
+ inode->i_gid = current->fsgid;
+ inode->i_blksize = PAGE_CACHE_SIZE;
+ inode->i_blocks = 0;
+ dlmfs_set_backing_info(inode);
+ inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+ inode->i_nlink++;
+
+ inode->i_fop = &simple_dir_operations;
+ inode->i_op = &dlmfs_root_inode_operations;
+ }
+
+ return inode;
+}
+
+static struct inode *dlmfs_get_inode(struct inode *parent,
+ struct dentry *dentry,
+ int mode)
+{
+ struct super_block *sb = parent->i_sb;
+ struct inode * inode = new_inode(sb);
+ struct dlmfs_inode_private *ip;
+
+ if (!inode)
+ return NULL;
+
+ inode->i_mode = mode;
+ inode->i_uid = current->fsuid;
+ inode->i_gid = current->fsgid;
+ inode->i_blksize = PAGE_CACHE_SIZE;
+ inode->i_blocks = 0;
+ dlmfs_set_backing_info(inode);
+ inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+
+ ip = DLMFS_I(inode);
+ ip->ip_dlm = DLMFS_I(parent)->ip_dlm;
+
+ switch (mode & S_IFMT) {
+ default:
+ /* for now we don't support anything other than
+ * directories and regular files. */
+ BUG();
+ break;
+ case S_IFREG:
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+ inode->i_op = &dlmfs_file_inode_operations;
+#endif
+ inode->i_fop = &dlmfs_file_operations;
+
+ user_dlm_lock_res_init(&ip->ip_lockres, dentry);
+
+ /* released at clear_inode time, this insures that we
+ * get to drop the dlm reference on each lock *before*
+ * we call the unregister code for releasing parent
+ * directories. */
+ ip->ip_parent = igrab(parent);
+ BUG_ON(!ip->ip_parent);
+ break;
+ case S_IFDIR:
+ inode->i_op = &dlmfs_dir_inode_operations;
+ inode->i_fop = &simple_dir_operations;
+
+ /* directory inodes start off with i_nlink ==
+ * 2 (for "." entry) */
+ inode->i_nlink++;
+ break;
+ }
+
+ if (parent->i_mode & S_ISGID) {
+ inode->i_gid = parent->i_gid;
+ if (S_ISDIR(mode))
+ inode->i_mode |= S_ISGID;
+ }
+
+ return inode;
+}
+
+/*
+ * File creation. Allocate an inode, and we're done..
+ */
+/* SMP-safe */
+static int dlmfs_mkdir(struct inode * dir,
+ struct dentry * dentry,
+ int mode)
+{
+ int status;
+ struct inode *inode = NULL;
+ struct qstr *domain = &dentry->d_name;
+ struct dlmfs_inode_private *ip;
+
+ dprintk("mkdir %.*s\n", domain->len, domain->name);
+
+ /* verify that we have a proper domain */
+ if (domain->len >= NM_MAX_NAME_LEN) {
+ status = -EINVAL;
+ err_printk("invalid domain name for directory.\n");
+ goto bail;
+ }
+
+ inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
+ if (!inode) {
+ status = -ENOMEM;
+ err_printk("status %d returning from dlmfs_get_inode\n",
+ status);
+ goto bail;
+ }
+
+ ip = DLMFS_I(inode);
+
+ ip->ip_dlm = user_dlm_register_context(domain);
+ if (!ip->ip_dlm) {
+ status = -ENOMEM;
+ err_printk("could not register domain \"%.*s\"\n",
+ domain->len, domain->name);
+ goto bail;
+ }
+
+ dir->i_nlink++;
+ d_instantiate(dentry, inode);
+ dget(dentry); /* Extra count - pin the dentry in core */
+
+ status = 0;
+bail:
+ if (status < 0)
+ iput(inode);
+ return status;
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+static int dlmfs_create(struct inode *dir,
+ struct dentry *dentry,
+ int mode,
+ struct nameidata *nd)
+#else
+static int dlmfs_create(struct inode *dir,
+ struct dentry *dentry,
+ int mode)
+#endif
+{
+ int status = 0;
+ struct inode *inode;
+ struct qstr *name = &dentry->d_name;
+
+ dprintk("create %.*s\n", name->len, name->name);
+
+ /* verify name is valid and doesn't contain any dlm reserved
+ * characters */
+ if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
+ name->name[0] == '$') {
+ status = -EINVAL;
+ err_printk("invalid lock name, %.*s\n", name->len, name->name);
+ goto bail;
+ }
+
+ inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
+ if (!inode) {
+ status = -ENOMEM;
+ err_printk("could not get inode!\n");
+ goto bail;
+ }
+
+ d_instantiate(dentry, inode);
+ dget(dentry); /* Extra count - pin the dentry in core */
+bail:
+ return status;
+}
+
+static int dlmfs_unlink(struct inode *dir,
+ struct dentry *dentry)
+{
+ int status;
+ struct inode *inode = dentry->d_inode;
+
+ dprintk("unlink inode %lu\n", inode->i_ino);
+
+ /* if there are no current holders, or none that are waiting
+ * to acquire a lock, this basically destroys our lockres. */
+ status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
+ if (status < 0) {
+ err_printk("unlink %.*s, error %d from destroy\n",
+ dentry->d_name.len, dentry->d_name.name, status);
+ goto bail;
+ }
+ status = simple_unlink(dir, dentry);
+bail:
+ return status;
+}
+
+static int dlmfs_fill_super(struct super_block * sb,
+ void * data,
+ int silent)
+{
+ struct inode * inode;
+ struct dentry * root;
+
+ sb->s_maxbytes = MAX_LFS_FILESIZE;
+ sb->s_blocksize = PAGE_CACHE_SIZE;
+ sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
+ sb->s_magic = DLMFS_MAGIC;
+ sb->s_op = &dlmfs_ops;
+ inode = dlmfs_get_root_inode(sb);
+ if (!inode)
+ return -ENOMEM;
+
+ root = d_alloc_root(inode);
+ if (!root) {
+ iput(inode);
+ return -ENOMEM;
+ }
+ sb->s_root = root;
+ return 0;
+}
+
+static struct file_operations dlmfs_file_operations = {
+ .open = dlmfs_file_open,
+ .release = dlmfs_file_release,
+};
+
+static struct inode_operations dlmfs_dir_inode_operations = {
+ .create = dlmfs_create,
+ .lookup = simple_lookup,
+ .unlink = dlmfs_unlink,
+};
+
+/* this way we can restrict mkdir to only the toplevel of the fs. */
+static struct inode_operations dlmfs_root_inode_operations = {
+ .lookup = simple_lookup,
+ .mkdir = dlmfs_mkdir,
+ .rmdir = simple_rmdir,
+};
+
+static struct super_operations dlmfs_ops = {
+ .statfs = simple_statfs,
+ .alloc_inode = dlmfs_alloc_inode,
+ .destroy_inode = dlmfs_destroy_inode,
+ .clear_inode = dlmfs_clear_inode,
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+ .drop_inode = generic_delete_inode,
+#endif
+};
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+static struct inode_operations dlmfs_file_inode_operations = {
+ .getattr = simple_getattr,
+};
+
+static struct super_block *dlmfs_get_sb(struct file_system_type *fs_type,
+ int flags, const char *dev_name, void *data)
+{
+ return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super);
+}
+
+static struct file_system_type dlmfs_fs_type = {
+ .name = "ocfs2_dlmfs",
+ .get_sb = dlmfs_get_sb,
+ .kill_sb = kill_litter_super,
+};
+#else
+static struct super_block *dlmfs_read_super(struct super_block *sb,
+ void *data,
+ int silent)
+{
+ return (dlmfs_fill_super(sb, data, silent)) < 0 ? NULL : sb;
+}
+
+static DECLARE_FSTYPE(dlmfs_fs_type, "ocfs2_dlmfs", dlmfs_read_super,
+ FS_LITTER);
+#endif
+
+static int __init init_dlmfs_fs(void)
+{
+ int status;
+ int cleanup_inode = 0, cleanup_worker = 0;
+
+ dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
+ sizeof(struct dlmfs_inode_private),
+ 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
+ dlmfs_init_once, NULL);
+ if (!dlmfs_inode_cache)
+ return -ENOMEM;
+ cleanup_inode = 1;
+
+ user_dlm_worker = create_workqueue("user_dlm");
+ if (!user_dlm_worker) {
+ status = -ENOMEM;
+ goto bail;
+ }
+ cleanup_worker = 1;
+
+ status = register_filesystem(&dlmfs_fs_type);
+bail:
+ if (status) {
+ if (cleanup_inode)
+ kmem_cache_destroy(dlmfs_inode_cache);
+ if (cleanup_worker)
+ destroy_workqueue(user_dlm_worker);
+ } else
+ printk("OCFS2 User DLM kernel interface loaded\n");
+ return status;
+}
+
+static void __exit exit_dlmfs_fs(void)
+{
+ unregister_filesystem(&dlmfs_fs_type);
+
+ flush_workqueue(user_dlm_worker);
+ destroy_workqueue(user_dlm_worker);
+
+ if (kmem_cache_destroy(dlmfs_inode_cache))
+ printk(KERN_INFO "dlmfs_inode_cache: not all structures "
+ "were freed\n");
+}
+
+module_init(init_dlmfs_fs)
+module_exit(exit_dlmfs_fs)
+
+MODULE_LICENSE("GPL");
Added: trunk/fs/ocfs2/dlm/dlmfs_compat.c
===================================================================
--- trunk/fs/ocfs2/dlm/dlmfs_compat.c 2005-01-22 01:47:41 UTC (rev 1831)
+++ trunk/fs/ocfs2/dlm/dlmfs_compat.c 2005-01-22 01:58:21 UTC (rev 1832)
@@ -0,0 +1,85 @@
+#include <linux/module.h>
+#include <linux/pagemap.h>
+#include <linux/mount.h>
+#include <linux/fs.h>
+
+#include "util.h"
+
+#include "cluster/clcommon.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmcommon.h"
+#include "dlmmod.h"
+
+#include "dlmfs_compat.h"
+#include "userdlm.h"
+
+static inline int simple_positive(struct dentry *dentry)
+{
+ return dentry->d_inode && !d_unhashed(dentry);
+}
+
+/*
+ * Check that a directory is empty (this works
+ * for regular files too, they'll just always be
+ * considered empty..).
+ *
+ * Note that an empty directory can still have
+ * children, they just all have to be negative..
+ */
+static int simple_empty(struct dentry *dentry)
+{
+ struct list_head *list;
+
+ spin_lock(&dcache_lock);
+ list = dentry->d_subdirs.next;
+
+ while (list != &dentry->d_subdirs) {
+ struct dentry *de = list_entry(list, struct dentry, d_child);
+
+ if (simple_positive(de)) {
+ spin_unlock(&dcache_lock);
+ return 0;
+ }
+ list = list->next;
+ }
+ spin_unlock(&dcache_lock);
+ return 1;
+}
+
+/*
+ * This works for both directories and regular files.
+ * (non-directories will always have empty subdirs)
+ */
+int simple_unlink(struct inode * dir, struct dentry *dentry)
+{
+ int retval = -ENOTEMPTY;
+
+ if (simple_empty(dentry)) {
+ struct inode *inode = dentry->d_inode;
+
+ inode->i_nlink--;
+ dput(dentry); /* Undo the count from "create" - this does all the work */
+ retval = 0;
+ }
+ return retval;
+}
+
+/*
+ * Lookup the data. This is trivial - if the dentry didn't already
+ * exist, we know it is negative.
+ */
+struct dentry * simple_lookup(struct inode *dir, struct dentry *dentry)
+{
+ d_add(dentry, NULL);
+ return NULL;
+}
+
+int simple_statfs(struct super_block *sb, struct statfs *buf)
+{
+ buf->f_type = DLMFS_MAGIC;
+ buf->f_bsize = PAGE_CACHE_SIZE;
+ buf->f_namelen = 255;
+ return 0;
+}
Added: trunk/fs/ocfs2/dlm/dlmfs_compat.h
===================================================================
--- trunk/fs/ocfs2/dlm/dlmfs_compat.h 2005-01-22 01:47:41 UTC (rev 1831)
+++ trunk/fs/ocfs2/dlm/dlmfs_compat.h 2005-01-22 01:58:21 UTC (rev 1832)
@@ -0,0 +1,20 @@
+#ifndef DLMFS_COMPAT_H
+#define DLMFS_COMPAT_H
+
+#include <linux/version.h>
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)
+
+#include <linux/fs.h>
+
+#define simple_dir_operations dcache_dir_ops
+#define simple_rmdir simple_unlink
+#define SLAB_RECLAIM_ACCOUNT 0
+
+int simple_unlink(struct inode * dir, struct dentry *dentry);
+struct dentry * simple_lookup(struct inode *dir, struct dentry *dentry);
+int simple_statfs(struct super_block *sb, struct statfs *buf);
+
+#endif
+
+#endif /* DLMFS_COMPAT_H */
Added: trunk/fs/ocfs2/dlm/userdlm.c
===================================================================
--- trunk/fs/ocfs2/dlm/userdlm.c 2005-01-22 01:47:41 UTC (rev 1831)
+++ trunk/fs/ocfs2/dlm/userdlm.c 2005-01-22 01:58:21 UTC (rev 1832)
@@ -0,0 +1,587 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * userdlm.c
+ *
+ * Code which implements the kernel side of a minimal userspace
+ * interface to our DLM.
+ *
+ * Many of the functions here are pared down versions of dlmglue.c
+ * functions.
+ *
+ * Copyright (C) 2003, 2004 Oracle. All rights reserved.
+ *
+ * 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 the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <asm/signal.h>
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/version.h>
+#include <linux/workqueue.h>
+
+#include "util.h"
+
+#include "cluster/clcommon.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmcommon.h"
+#include "dlmmod.h"
+
+#include "userdlm.h"
+
+static inline int user_check_wait_flag(struct user_lock_res *lockres,
+ int flag)
+{
+ int ret;
+
+ spin_lock(&lockres->l_lock);
+ ret = lockres->l_flags & flag;
+ spin_unlock(&lockres->l_lock);
+
+ return ret;
+}
+
+static inline void user_wait_on_busy_lock(struct user_lock_res *lockres)
+
+{
+ wait_event_interruptible(lockres->l_event,
+ !user_check_wait_flag(lockres,
+ USER_LOCK_BUSY));
+}
+
+static inline void user_wait_on_blocked_lock(struct user_lock_res *lockres)
+
+{
+ wait_event_interruptible(lockres->l_event,
+ !user_check_wait_flag(lockres,
+ USER_LOCK_BLOCKED));
+}
+
+/* I heart container_of... */
+static inline dlm_ctxt *
+dlm_ctxt_from_user_lockres(struct user_lock_res *lockres)
+{
+ struct dlmfs_inode_private *ip;
+
+ ip = container_of(lockres,
+ struct dlmfs_inode_private,
+ ip_lockres);
+ return ip->ip_dlm;
+}
+
+static struct inode *
+user_dlm_inode_from_user_lockres(struct user_lock_res *lockres)
+{
+ struct dlmfs_inode_private *ip;
+
+ ip = container_of(lockres,
+ struct dlmfs_inode_private,
+ ip_lockres);
+ return &ip->ip_vfs_inode;
+}
+
+static inline void user_recover_from_dlm_error(struct user_lock_res *lockres)
+{
+ spin_lock(&lockres->l_lock);
+ lockres->l_flags &= ~USER_LOCK_BUSY;
+ spin_unlock(&lockres->l_lock);
+}
+
+/* WARNING: This function lives in a world where the only three lock
+ * levels are EX, PR, and NL. It *will* have to be adjusted when more
+ * lock types are added. */
+static inline int user_highest_compat_lock_level(int level)
+{
+ int new_level = LKM_EXMODE;
+
+ if (level == LKM_EXMODE)
+ new_level = LKM_NLMODE;
+ else if (level == LKM_PRMODE)
+ new_level = LKM_PRMODE;
+ return new_level;
+}
+
+static void user_ast(void *opaque)
+{
+ struct user_lock_res *lockres = opaque;
+ dlm_lockstatus *lksb;
+
+ dprintk("AST fired for lockres %s\n", lockres->l_name);
+
+ spin_lock(&lockres->l_lock);
+
+ lksb = &(lockres->l_lksb);
+ if (lksb->status != DLM_NORMAL) {
+ printk("user_ast: lksb status value of %u "
+ "on lockres %s\n", lksb->status, lockres->l_name);
+ spin_unlock(&lockres->l_lock);
+ return;
+ }
+
+ /* we're downconverting. */
+ if (lockres->l_requested < lockres->l_level) {
+ if (lockres->l_requested <=
+ user_highest_compat_lock_level(lockres->l_blocking)) {
+ lockres->l_blocking = LKM_NLMODE;
+ lockres->l_flags &= ~USER_LOCK_BLOCKED;
+ }
+ }
+
+ lockres->l_level = lockres->l_requested;
+ lockres->l_requested = LKM_IVMODE;
+ lockres->l_flags |= USER_LOCK_ATTACHED;
+ lockres->l_flags &= ~USER_LOCK_BUSY;
+
+ spin_unlock(&lockres->l_lock);
+
+ wake_up(&lockres->l_event);
+}
+
+static inline void user_dlm_grab_inode_ref(struct user_lock_res *lockres)
+{
+ struct inode *inode;
+ inode = user_dlm_inode_from_user_lockres(lockres);
+ if (!igrab(inode))
+ BUG();
+}
+
+static void user_dlm_unblock_lock(void *opaque);
+
+static void __user_dlm_queue_lockres(struct user_lock_res *lockres)
+{
+ if (!(lockres->l_flags & USER_LOCK_QUEUED)) {
+ user_dlm_grab_inode_ref(lockres);
+
+ PREPARE_WORK(&lockres->l_work,
+ user_dlm_unblock_lock,
+ &lockres);
+
+ queue_work(user_dlm_worker, &lockres->l_work);
+ lockres->l_flags |= USER_LOCK_QUEUED;
+ }
+}
+
+static void __user_dlm_cond_queue_lockres(struct user_lock_res *lockres)
+{
+ int queue = 0;
+
+ if (!(lockres->l_flags & USER_LOCK_BLOCKED))
+ return;
+
+ switch (lockres->l_blocking) {
+ case LKM_EXMODE:
+ if (!lockres->l_ex_holders && !lockres->l_ro_holders)
+ queue = 1;
+ break;
+ case LKM_PRMODE:
+ if (!lockres->l_ex_holders)
+ queue = 1;
+ break;
+ default:
+ BUG();
+ }
+
+ if (queue)
+ __user_dlm_queue_lockres(lockres);
+}
+
+static void user_bast(void *opaque, int level)
+{
+ struct user_lock_res *lockres = opaque;
+
+ dprintk("Blocking AST fired for lockres %s. Blocking level %d\n",
+ lockres->l_name, level);
+
+ spin_lock(&lockres->l_lock);
+ lockres->l_flags |= USER_LOCK_BLOCKED;
+ if (level > lockres->l_blocking)
+ lockres->l_blocking = level;
+
+ __user_dlm_queue_lockres(lockres);
+ spin_unlock(&lockres->l_lock);
+
+ wake_up(&lockres->l_event);
+}
+
+static void user_unlock_ast(void *opaque, dlm_status status)
+{
+ struct user_lock_res *lockres = opaque;
+
+ dprintk("UNLOCK AST called on lock %s\n", lockres->l_name);
+
+ if (status != DLM_NORMAL)
+ err_printk("Dlm returns status %d\n", status);
+
+ spin_lock(&lockres->l_lock);
+ if (lockres->l_flags & USER_LOCK_IN_TEARDOWN)
+ lockres->l_level = LKM_IVMODE;
+ else {
+ lockres->l_requested = LKM_IVMODE; /* cancel an
+ * upconvert
+ * request. */
+ lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
+ /* we want the unblock thread to look at it again
+ * now. */
+ __user_dlm_queue_lockres(lockres);
+ }
+
+ lockres->l_flags &= ~USER_LOCK_BUSY;
+ spin_unlock(&lockres->l_lock);
+
+ wake_up(&lockres->l_event);
+}
+
+static inline void user_dlm_drop_inode_ref(struct user_lock_res *lockres)
+{
+ struct inode *inode;
+ inode = user_dlm_inode_from_user_lockres(lockres);
+ iput(inode);
+}
+
+static void user_dlm_unblock_lock(void *opaque)
+{
+ int new_level, status;
+ struct user_lock_res *lockres = (struct user_lock_res *) opaque;
+ dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
+
+ dprintk("processing lockres %s\n", lockres->l_name);
+
+ spin_lock(&lockres->l_lock);
+
+ BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));
+ BUG_ON(!(lockres->l_flags & USER_LOCK_QUEUED));
+
+ /* notice that we don't clear USER_LOCK_BLOCKED here. That's
+ * for user_ast to do. */
+ lockres->l_flags &= ~USER_LOCK_QUEUED;
+
+ if (lockres->l_flags & USER_LOCK_BUSY) {
+ if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
+ spin_unlock(&lockres->l_lock);
+ goto drop_ref;
+ }
+
+ lockres->l_flags |= USER_LOCK_IN_CANCEL;
+ spin_unlock(&lockres->l_lock);
+
+ status = dlmunlock(dlm,
+ &lockres->l_lksb,
+ LKM_CANCEL,
+ user_unlock_ast,
+ lockres);
+ if (status == DLM_CANCELGRANT) {
+ /* If we got this, then the ast was fired
+ * before we could cancel. We cleanup our
+ * state, and restart the function. */
+ spin_lock(&lockres->l_lock);
+ lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
+ spin_unlock(&lockres->l_lock);
+ } else if (status != DLM_NORMAL)
+ err_printk("Dlm returns %d\n", status);
+ goto drop_ref;
+ }
+
+ /* If there are still incompat holders, we can exit safely
+ * without worrying about re-queueing this lock as that will
+ * happen on the last call to user_cluster_unlock. */
+ if ((lockres->l_blocking == LKM_EXMODE)
+ && (lockres->l_ex_holders || lockres->l_ro_holders)) {
+ spin_unlock(&lockres->l_lock);
+ goto drop_ref;
+ }
+ if (lockres->l_blocking == LKM_PRMODE
+ && lockres->l_ex_holders) {
+ spin_unlock(&lockres->l_lock);
+ goto drop_ref;
+ }
+
+ /* yay, we can downconvert now. */
+ new_level = user_highest_compat_lock_level(lockres->l_blocking);
+ lockres->l_requested = new_level;
+ lockres->l_flags |= USER_LOCK_BUSY;
+ spin_unlock(&lockres->l_lock);
+
+ /* need lock downconvert request now... */
+ status = dlmlock(dlm,
+ new_level,
+ &lockres->l_lksb,
+ LKM_CONVERT,
+ lockres->l_name,
+ user_ast,
+ lockres,
+ user_bast);
+ if (status != DLM_NORMAL) {
+ err_printk("Dlm returns %d\n", status);
+ user_recover_from_dlm_error(lockres);
+ }
+
+drop_ref:
+ user_dlm_drop_inode_ref(lockres);
+}
+
+static inline void user_dlm_inc_holders(struct user_lock_res *lockres,
+ int level)
+{
+ switch(level) {
+ case LKM_EXMODE:
+ lockres->l_ex_holders++;
+ break;
+ case LKM_PRMODE:
+ lockres->l_ro_holders++;
+ break;
+ default:
+ BUG();
+ }
+}
+
+/* predict what lock level we'll be dropping down to on behalf
+ * of another node, and return true if the currently wanted
+ * level will be compatible with it. */
+static inline int
+user_may_continue_on_blocked_lock(struct user_lock_res *lockres,
+ int wanted)
+{
+ BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));
+
+ return wanted <= user_highest_compat_lock_level(lockres->l_blocking);
+}
+
+int user_dlm_cluster_lock(struct user_lock_res *lockres,
+ int level,
+ int lkm_flags)
+{
+ int status, local_flags;
+ dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
+
+ if (level != LKM_EXMODE &&
+ level != LKM_PRMODE) {
+ err_printk("lockres %s: invalid request!\n", lockres->l_name);
+ status = -EINVAL;
+ goto bail;
+ }
+
+ dprintk("lockres %s: asking for %s lock\n", lockres->l_name,
+ (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE");
+
+again:
+ if (signal_pending(current)) {
+ status = -EINTR;
+ goto bail;
+ }
+
+ spin_lock(&lockres->l_lock);
+
+ /* We only compare against the currently granted level
+ * here. If the lock is blocked waiting on a downconvert,
+ * we'll get caught below. */
+ if (lockres->l_flags & USER_LOCK_BUSY &&
+ level > lockres->l_level) {
+ /* is someone sitting in dlm_lock? If so, wait on
+ * them. */
+ spin_unlock(&lockres->l_lock);
+
+ user_wait_on_busy_lock(lockres);
+ goto again;
+ }
+
+ if (lockres->l_flags & USER_LOCK_BLOCKED &&
+ !user_may_continue_on_blocked_lock(lockres, level)) {
+ /* is the lock is currently blocked on behalf of
+ * another node */
+ spin_unlock(&lockres->l_lock);
+
+ user_wait_on_blocked_lock(lockres);
+ goto again;
+ }
+
+ if (level > lockres->l_level) {
+ local_flags = lkm_flags;
+ if (lockres->l_level != LKM_IVMODE)
+ local_flags |= LKM_CONVERT;
+
+ lockres->l_requested = level;
+ lockres->l_flags |= USER_LOCK_BUSY;
+ spin_unlock(&lockres->l_lock);
+
+ BUG_ON(level == LKM_IVMODE);
+ BUG_ON(level == LKM_NLMODE);
+
+ dprintk("lock %s, get lock from %d to level = %d\n",
+ lockres->l_name, lockres->l_level, level);
+
+ /* call dlm_lock to upgrade lock now */
+ status = dlmlock(dlm,
+ level,
+ &lockres->l_lksb,
+ local_flags,
+ lockres->l_name,
+ user_ast,
+ lockres,
+ user_bast);
+ if (status != DLM_NORMAL) {
+ if ((lkm_flags & LKM_NOQUEUE) &&
+ (status == DLM_NOTQUEUED))
+ status = -EAGAIN;
+ else {
+ err_printk("Dlm returns %d\n", status);
+ status = -ENOENT;
+ }
+ user_recover_from_dlm_error(lockres);
+ goto bail;
+ }
+
+ dprintk("lock %s, successfull return from dlmlock\n",
+ lockres->l_name);
+
+ user_wait_on_busy_lock(lockres);
+ goto again;
+ }
+
+ user_dlm_inc_holders(lockres, level);
+ spin_unlock(&lockres->l_lock);
+
+ status = 0;
+bail:
+ return status;
+}
+
+static inline void user_dlm_dec_holders(struct user_lock_res *lockres,
+ int level)
+{
+ switch(level) {
+ case LKM_EXMODE:
+ BUG_ON(!lockres->l_ex_holders);
+ lockres->l_ex_holders--;
+ break;
+ case LKM_PRMODE:
+ BUG_ON(!lockres->l_ro_holders);
+ lockres->l_ro_holders--;
+ break;
+ default:
+ BUG();
+ }
+}
+
+void user_dlm_cluster_unlock(struct user_lock_res *lockres,
+ int level)
+{
+ if (level != LKM_EXMODE &&
+ level != LKM_PRMODE) {
+ err_printk("lockres %s: invalid request!\n", lockres->l_name);
+ return;
+ }
+
+ dprintk("lockres %s: dropping %s lock\n", lockres->l_name,
+ (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE");
+
+ spin_lock(&lockres->l_lock);
+ user_dlm_dec_holders(lockres, level);
+ __user_dlm_cond_queue_lockres(lockres);
+ spin_unlock(&lockres->l_lock);
+}
+
+void user_dlm_lock_res_init(struct user_lock_res *lockres,
+ struct dentry *dentry)
+{
+ memset(lockres, 0, sizeof(*lockres));
+
+ spin_lock_init(&lockres->l_lock);
+ init_waitqueue_head(&lockres->l_event);
+ lockres->l_level = LKM_IVMODE;
+ lockres->l_requested = LKM_IVMODE;
+ lockres->l_blocking = LKM_IVMODE;
+
+ /* should have been checked before getting here. */
+ BUG_ON(dentry->d_name.len >= USER_DLM_LOCK_ID_MAX_LEN);
+
+ memcpy(lockres->l_name,
+ dentry->d_name.name,
+ USER_DLM_LOCK_ID_MAX_LEN - 1);
+}
+
+int user_dlm_destroy_lock(struct user_lock_res *lockres)
+{
+ int status = -EBUSY;
+ dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
+
+ dprintk("destroying %s\n", lockres->l_name);
+
+ spin_lock(&lockres->l_lock);
+
+ if (lockres->l_flags & USER_LOCK_BUSY) {
+ spin_unlock(&lockres->l_lock);
+ goto bail;
+ }
+
+ if (lockres->l_flags & USER_LOCK_ATTACHED) {
+ if (lockres->l_ro_holders || lockres->l_ex_holders) {
+ spin_unlock(&lockres->l_lock);
+ goto bail;
+ }
+
+ lockres->l_flags &= ~USER_LOCK_ATTACHED;
+ lockres->l_flags |= USER_LOCK_BUSY;
+ lockres->l_flags |= USER_LOCK_IN_TEARDOWN;
+ spin_unlock(&lockres->l_lock);
+
+ status = dlmunlock(dlm,
+ &lockres->l_lksb,
+ 0,
+ user_unlock_ast,
+ lockres);
+ if (status != DLM_NORMAL) {
+ err_printk("dlm returns %d!\n", status);
+ status = -EINVAL;
+ goto bail;
+ }
+
+ user_wait_on_busy_lock(lockres);
+ }
+
+ status = 0;
+bail:
+ return status;
+}
+
+dlm_ctxt *user_dlm_register_context(struct qstr *name)
+{
+ dlm_ctxt *dlm;
+ u32 dlm_key;
+ char *domain;
+
+ domain = kmalloc(name->len + 1, GFP_KERNEL);
+ if (!domain) {
+ err_printk("Out of memory\n");
+ return NULL;
+ }
+
+ snprintf(domain, name->len + 1, "%.*s", name->len, name->name);
+ memcpy(&dlm_key, domain, sizeof(dlm_key));
+
+ dlm = dlm_register_domain(domain, domain, dlm_key);
+
+ kfree(domain);
+ return dlm;
+}
+
+void user_dlm_unregister_context(dlm_ctxt *dlm)
+{
+ dlm_unregister_domain(dlm);
+}
Added: trunk/fs/ocfs2/dlm/userdlm.h
===================================================================
--- trunk/fs/ocfs2/dlm/userdlm.h 2005-01-22 01:47:41 UTC (rev 1831)
+++ trunk/fs/ocfs2/dlm/userdlm.h 2005-01-22 01:58:21 UTC (rev 1832)
@@ -0,0 +1,122 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * userdlm.h
+ *
+ * description here
+ *
+ * Copyright (C) 2002, 2004 Oracle. All rights reserved.
+ *
+ * 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 the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+
+#ifndef USERDLM_H
+#define USERDLM_H
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/version.h>
+#include <linux/workqueue.h>
+
+#define DEBUG_TRACING
+
+#define err_printk(str, args...) \
+ printk(KERN_ERR "(%d) %s:%d " str, current->pid, __FUNCTION__, __LINE__, ##args)
+#define err_printk0(str) \
+ printk(KERN_ERR "(%d) %s:%d " str, current->pid, __FUNCTION__, __LINE__)
+#ifdef DEBUG_TRACING
+#define dprintk(str, args...) \
+ printk("(%d) %s:%d " str, current->pid, __FUNCTION__, __LINE__, ##args)
+#define dprintk0(str) \
+ printk("(%d) %s:%d " str, current->pid, __FUNCTION__, __LINE__)
+#else
+#define dprintk(str, args...)
+#define dprintk0(str)
+#endif
+
+/* user_lock_res->l_flags flags. */
+#define USER_LOCK_ATTACHED (0x00000001) /* have we initialized
+ * the lvb */
+#define USER_LOCK_BUSY (0x00000002) /* we are currently in
+ * dlm_lock */
+#define USER_LOCK_BLOCKED (0x00000004) /* blocked waiting to
+ * downconvert*/
+#define USER_LOCK_IN_TEARDOWN (0x00000008) /* we're currently
+ * destroying this
+ * lock. */
+#define USER_LOCK_QUEUED (0x00000010) /* lock is on the
+ * workqueue */
+#define USER_LOCK_IN_CANCEL (0x00000020)
+
+struct user_lock_res {
+ spinlock_t l_lock;
+
+ int l_flags;
+
+#define USER_DLM_LOCK_ID_MAX_LEN 32
+ char l_name[USER_DLM_LOCK_ID_MAX_LEN];
+ int l_level;
+ unsigned int l_ro_holders;
+ unsigned int l_ex_holders;
+ dlm_lockstatus l_lksb;
+
+ int l_requested;
+ int l_blocking;
+
+ wait_queue_head_t l_event;
+
+ struct work_struct l_work;
+};
+
+extern struct workqueue_struct *user_dlm_worker;
+
+void user_dlm_lock_res_init(struct user_lock_res *lockres,
+ struct dentry *dentry);
+int user_dlm_destroy_lock(struct user_lock_res *lockres);
+int user_dlm_cluster_lock(struct user_lock_res *lockres,
+ int level,
+ int lkm_flags);
+void user_dlm_cluster_unlock(struct user_lock_res *lockres,
+ int level);
+dlm_ctxt *user_dlm_register_context(struct qstr *name);
+void user_dlm_unregister_context(dlm_ctxt *dlm);
+
+struct dlmfs_inode_private {
+ dlm_ctxt *ip_dlm;
+
+ struct user_lock_res ip_lockres; /* unused for directories. */
+ struct inode *ip_parent;
+
+ struct inode ip_vfs_inode;
+};
+
+static inline struct dlmfs_inode_private *
+DLMFS_I(struct inode *inode)
+{
+ return container_of(inode,
+ struct dlmfs_inode_private,
+ ip_vfs_inode);
+}
+
+struct dlmfs_filp_private {
+ int fp_lock_level;
+};
+
+#define DLMFS_MAGIC 0x76a9f425
+
+#endif /* USERDLM_H */
More information about the Ocfs2-commits
mailing list