1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Code which implements the kernel side of a minimal userspace
7 * interface to our DLM. This file handles the virtual file system
8 * used for communication with userspace. Credit should go to ramfs,
9 * which was a template for the fs side of this module.
11 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public
15 * License as published by the Free Software Foundation; either
16 * version 2 of the License, or (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
21 * General Public License for more details.
23 * You should have received a copy of the GNU General Public
24 * License along with this program; if not, write to the
25 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
26 * Boston, MA 021110-1307, USA.
29 /* Simple VFS hooks based on: */
31 * Resizable simple ram filesystem for Linux.
33 * Copyright (C) 2000 Linus Torvalds.
34 * 2000 Transmeta Corp.
37 #include <linux/module.h>
39 #include <linux/pagemap.h>
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43 #include <linux/init.h>
44 #include <linux/string.h>
45 #include <linux/backing-dev.h>
46 #include <linux/poll.h>
48 #include <linux/uaccess.h>
50 #include "stackglue.h"
53 #define MLOG_MASK_PREFIX ML_DLMFS
54 #include "cluster/masklog.h"
57 static const struct super_operations dlmfs_ops;
58 static const struct file_operations dlmfs_file_operations;
59 static const struct inode_operations dlmfs_dir_inode_operations;
60 static const struct inode_operations dlmfs_root_inode_operations;
61 static const struct inode_operations dlmfs_file_inode_operations;
62 static struct kmem_cache *dlmfs_inode_cache;
64 struct workqueue_struct *user_dlm_worker;
69 * These are the ABI capabilities of dlmfs.
71 * Over time, dlmfs has added some features that were not part of the
72 * initial ABI. Unfortunately, some of these features are not detectable
73 * via standard usage. For example, Linux's default poll always returns
74 * EPOLLIN, so there is no way for a caller of poll(2) to know when dlmfs
75 * added poll support. Instead, we provide this list of new capabilities.
77 * Capabilities is a read-only attribute. We do it as a module parameter
78 * so we can discover it whether dlmfs is built in, loaded, or even not
81 * The ABI features are local to this machine's dlmfs mount. This is
82 * distinct from the locking protocol, which is concerned with inter-node
86 * - bast : EPOLLIN against the file descriptor of a held lock
87 * signifies a bast fired on the lock.
89 #define DLMFS_CAPABILITIES "bast stackglue"
90 static int param_set_dlmfs_capabilities(const char *val,
91 const struct kernel_param *kp)
93 printk(KERN_ERR "%s: readonly parameter\n", kp->name);
96 static int param_get_dlmfs_capabilities(char *buffer,
97 const struct kernel_param *kp)
99 return strlcpy(buffer, DLMFS_CAPABILITIES,
100 strlen(DLMFS_CAPABILITIES) + 1);
102 module_param_call(capabilities, param_set_dlmfs_capabilities,
103 param_get_dlmfs_capabilities, NULL, 0444);
104 MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES);
108 * decodes a set of open flags into a valid lock level and a set of flags.
109 * returns < 0 if we have invalid flags
110 * flags which mean something to us:
111 * O_RDONLY -> PRMODE level
112 * O_WRONLY -> EXMODE level
114 * O_NONBLOCK -> NOQUEUE
116 static int dlmfs_decode_open_flags(int open_flags,
120 if (open_flags & (O_WRONLY|O_RDWR))
121 *level = DLM_LOCK_EX;
123 *level = DLM_LOCK_PR;
126 if (open_flags & O_NONBLOCK)
127 *flags |= DLM_LKF_NOQUEUE;
132 static int dlmfs_file_open(struct inode *inode,
135 int status, level, flags;
136 struct dlmfs_filp_private *fp = NULL;
137 struct dlmfs_inode_private *ip;
139 if (S_ISDIR(inode->i_mode))
142 mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
145 status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
149 /* We don't want to honor O_APPEND at read/write time as it
150 * doesn't make sense for LVB writes. */
151 file->f_flags &= ~O_APPEND;
153 fp = kmalloc(sizeof(*fp), GFP_NOFS);
158 fp->fp_lock_level = level;
162 status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
164 /* this is a strange error to return here but I want
165 * to be able userspace to be able to distinguish a
166 * valid lock request from one that simply couldn't be
168 if (flags & DLM_LKF_NOQUEUE && status == -EAGAIN)
174 file->private_data = fp;
179 static int dlmfs_file_release(struct inode *inode,
183 struct dlmfs_inode_private *ip = DLMFS_I(inode);
184 struct dlmfs_filp_private *fp = file->private_data;
186 if (S_ISDIR(inode->i_mode))
189 mlog(0, "close called on inode %lu\n", inode->i_ino);
192 level = fp->fp_lock_level;
193 if (level != DLM_LOCK_IV)
194 user_dlm_cluster_unlock(&ip->ip_lockres, level);
197 file->private_data = NULL;
204 * We do ->setattr() just to override size changes. Our size is the size
205 * of the LVB and nothing else.
207 static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr)
210 struct inode *inode = d_inode(dentry);
212 attr->ia_valid &= ~ATTR_SIZE;
213 error = setattr_prepare(dentry, attr);
217 setattr_copy(inode, attr);
218 mark_inode_dirty(inode);
222 static __poll_t dlmfs_file_poll(struct file *file, poll_table *wait)
225 struct inode *inode = file_inode(file);
226 struct dlmfs_inode_private *ip = DLMFS_I(inode);
228 poll_wait(file, &ip->ip_lockres.l_event, wait);
230 spin_lock(&ip->ip_lockres.l_lock);
231 if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED)
232 event = EPOLLIN | EPOLLRDNORM;
233 spin_unlock(&ip->ip_lockres.l_lock);
238 static ssize_t dlmfs_file_read(struct file *filp,
244 ssize_t readlen, got;
246 struct inode *inode = file_inode(filp);
248 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
249 inode->i_ino, count, *ppos);
251 if (*ppos >= i_size_read(inode))
257 if (!access_ok(buf, count))
260 /* don't read past the lvb */
261 if ((count + *ppos) > i_size_read(inode))
262 readlen = i_size_read(inode) - *ppos;
266 lvb_buf = kmalloc(readlen, GFP_NOFS);
270 got = user_dlm_read_lvb(inode, lvb_buf, readlen);
272 BUG_ON(got != readlen);
273 bytes_left = __copy_to_user(buf, lvb_buf, readlen);
274 readlen -= bytes_left;
280 *ppos = *ppos + readlen;
282 mlog(0, "read %zd bytes\n", readlen);
286 static ssize_t dlmfs_file_write(struct file *filp,
287 const char __user *buf,
294 struct inode *inode = file_inode(filp);
296 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
297 inode->i_ino, count, *ppos);
299 if (*ppos >= i_size_read(inode))
305 if (!access_ok(buf, count))
308 /* don't write past the lvb */
309 if ((count + *ppos) > i_size_read(inode))
310 writelen = i_size_read(inode) - *ppos;
312 writelen = count - *ppos;
314 lvb_buf = kmalloc(writelen, GFP_NOFS);
318 bytes_left = copy_from_user(lvb_buf, buf, writelen);
319 writelen -= bytes_left;
321 user_dlm_write_lvb(inode, lvb_buf, writelen);
325 *ppos = *ppos + writelen;
326 mlog(0, "wrote %zd bytes\n", writelen);
330 static void dlmfs_init_once(void *foo)
332 struct dlmfs_inode_private *ip =
333 (struct dlmfs_inode_private *) foo;
336 ip->ip_parent = NULL;
338 inode_init_once(&ip->ip_vfs_inode);
341 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
343 struct dlmfs_inode_private *ip;
345 ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
349 return &ip->ip_vfs_inode;
352 static void dlmfs_i_callback(struct rcu_head *head)
354 struct inode *inode = container_of(head, struct inode, i_rcu);
355 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
358 static void dlmfs_destroy_inode(struct inode *inode)
360 call_rcu(&inode->i_rcu, dlmfs_i_callback);
363 static void dlmfs_evict_inode(struct inode *inode)
366 struct dlmfs_inode_private *ip;
370 mlog(0, "inode %lu\n", inode->i_ino);
374 if (S_ISREG(inode->i_mode)) {
375 status = user_dlm_destroy_lock(&ip->ip_lockres);
382 mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn);
383 /* we must be a directory. If required, lets unregister the
384 * dlm context now. */
386 user_dlm_unregister(ip->ip_conn);
388 ip->ip_parent = NULL;
392 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
394 struct inode *inode = new_inode(sb);
395 umode_t mode = S_IFDIR | 0755;
398 inode->i_ino = get_next_ino();
399 inode_init_owner(inode, NULL, mode);
400 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
403 inode->i_fop = &simple_dir_operations;
404 inode->i_op = &dlmfs_root_inode_operations;
410 static struct inode *dlmfs_get_inode(struct inode *parent,
411 struct dentry *dentry,
414 struct super_block *sb = parent->i_sb;
415 struct inode * inode = new_inode(sb);
416 struct dlmfs_inode_private *ip;
421 inode->i_ino = get_next_ino();
422 inode_init_owner(inode, parent, mode);
423 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
426 ip->ip_conn = DLMFS_I(parent)->ip_conn;
428 switch (mode & S_IFMT) {
430 /* for now we don't support anything other than
431 * directories and regular files. */
435 inode->i_op = &dlmfs_file_inode_operations;
436 inode->i_fop = &dlmfs_file_operations;
438 i_size_write(inode, DLM_LVB_LEN);
440 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
442 /* released at clear_inode time, this insures that we
443 * get to drop the dlm reference on each lock *before*
444 * we call the unregister code for releasing parent
446 ip->ip_parent = igrab(parent);
447 BUG_ON(!ip->ip_parent);
450 inode->i_op = &dlmfs_dir_inode_operations;
451 inode->i_fop = &simple_dir_operations;
453 /* directory inodes start off with i_nlink ==
454 * 2 (for "." entry) */
462 * File creation. Allocate an inode, and we're done..
465 static int dlmfs_mkdir(struct inode * dir,
466 struct dentry * dentry,
470 struct inode *inode = NULL;
471 const struct qstr *domain = &dentry->d_name;
472 struct dlmfs_inode_private *ip;
473 struct ocfs2_cluster_connection *conn;
475 mlog(0, "mkdir %.*s\n", domain->len, domain->name);
477 /* verify that we have a proper domain */
478 if (domain->len >= GROUP_NAME_MAX) {
480 mlog(ML_ERROR, "invalid domain name for directory.\n");
484 inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
493 conn = user_dlm_register(domain);
495 status = PTR_ERR(conn);
496 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
497 status, domain->len, domain->name);
503 d_instantiate(dentry, inode);
504 dget(dentry); /* Extra count - pin the dentry in core */
513 static int dlmfs_create(struct inode *dir,
514 struct dentry *dentry,
520 const struct qstr *name = &dentry->d_name;
522 mlog(0, "create %.*s\n", name->len, name->name);
524 /* verify name is valid and doesn't contain any dlm reserved
526 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
527 name->name[0] == '$') {
529 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
534 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
541 d_instantiate(dentry, inode);
542 dget(dentry); /* Extra count - pin the dentry in core */
547 static int dlmfs_unlink(struct inode *dir,
548 struct dentry *dentry)
551 struct inode *inode = d_inode(dentry);
553 mlog(0, "unlink inode %lu\n", inode->i_ino);
555 /* if there are no current holders, or none that are waiting
556 * to acquire a lock, this basically destroys our lockres. */
557 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
559 mlog(ML_ERROR, "unlink %pd, error %d from destroy\n",
563 status = simple_unlink(dir, dentry);
568 static int dlmfs_fill_super(struct super_block * sb,
572 sb->s_maxbytes = MAX_LFS_FILESIZE;
573 sb->s_blocksize = PAGE_SIZE;
574 sb->s_blocksize_bits = PAGE_SHIFT;
575 sb->s_magic = DLMFS_MAGIC;
576 sb->s_op = &dlmfs_ops;
577 sb->s_root = d_make_root(dlmfs_get_root_inode(sb));
583 static const struct file_operations dlmfs_file_operations = {
584 .open = dlmfs_file_open,
585 .release = dlmfs_file_release,
586 .poll = dlmfs_file_poll,
587 .read = dlmfs_file_read,
588 .write = dlmfs_file_write,
589 .llseek = default_llseek,
592 static const struct inode_operations dlmfs_dir_inode_operations = {
593 .create = dlmfs_create,
594 .lookup = simple_lookup,
595 .unlink = dlmfs_unlink,
598 /* this way we can restrict mkdir to only the toplevel of the fs. */
599 static const struct inode_operations dlmfs_root_inode_operations = {
600 .lookup = simple_lookup,
601 .mkdir = dlmfs_mkdir,
602 .rmdir = simple_rmdir,
605 static const struct super_operations dlmfs_ops = {
606 .statfs = simple_statfs,
607 .alloc_inode = dlmfs_alloc_inode,
608 .destroy_inode = dlmfs_destroy_inode,
609 .evict_inode = dlmfs_evict_inode,
610 .drop_inode = generic_delete_inode,
613 static const struct inode_operations dlmfs_file_inode_operations = {
614 .getattr = simple_getattr,
615 .setattr = dlmfs_file_setattr,
618 static struct dentry *dlmfs_mount(struct file_system_type *fs_type,
619 int flags, const char *dev_name, void *data)
621 return mount_nodev(fs_type, flags, data, dlmfs_fill_super);
624 static struct file_system_type dlmfs_fs_type = {
625 .owner = THIS_MODULE,
626 .name = "ocfs2_dlmfs",
627 .mount = dlmfs_mount,
628 .kill_sb = kill_litter_super,
630 MODULE_ALIAS_FS("ocfs2_dlmfs");
632 static int __init init_dlmfs_fs(void)
635 int cleanup_inode = 0, cleanup_worker = 0;
637 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
638 sizeof(struct dlmfs_inode_private),
639 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
640 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
642 if (!dlmfs_inode_cache) {
648 user_dlm_worker = alloc_workqueue("user_dlm", WQ_MEM_RECLAIM, 0);
649 if (!user_dlm_worker) {
655 user_dlm_set_locking_protocol();
656 status = register_filesystem(&dlmfs_fs_type);
660 kmem_cache_destroy(dlmfs_inode_cache);
662 destroy_workqueue(user_dlm_worker);
664 printk("OCFS2 User DLM kernel interface loaded\n");
668 static void __exit exit_dlmfs_fs(void)
670 unregister_filesystem(&dlmfs_fs_type);
672 destroy_workqueue(user_dlm_worker);
675 * Make sure all delayed rcu free inodes are flushed before we
679 kmem_cache_destroy(dlmfs_inode_cache);
683 MODULE_AUTHOR("Oracle");
684 MODULE_LICENSE("GPL");
685 MODULE_DESCRIPTION("OCFS2 DLM-Filesystem");
687 module_init(init_dlmfs_fs)
688 module_exit(exit_dlmfs_fs)