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>
47 #include <asm/uaccess.h>
50 #include "cluster/nodemanager.h"
51 #include "cluster/heartbeat.h"
52 #include "cluster/tcp.h"
60 #define MLOG_MASK_PREFIX ML_DLMFS
61 #include "cluster/masklog.h"
63 static const struct super_operations dlmfs_ops;
64 static const struct file_operations dlmfs_file_operations;
65 static const struct inode_operations dlmfs_dir_inode_operations;
66 static const struct inode_operations dlmfs_root_inode_operations;
67 static const struct inode_operations dlmfs_file_inode_operations;
68 static struct kmem_cache *dlmfs_inode_cache;
70 struct workqueue_struct *user_dlm_worker;
73 * decodes a set of open flags into a valid lock level and a set of flags.
74 * returns < 0 if we have invalid flags
75 * flags which mean something to us:
76 * O_RDONLY -> PRMODE level
77 * O_WRONLY -> EXMODE level
79 * O_NONBLOCK -> LKM_NOQUEUE
81 static int dlmfs_decode_open_flags(int open_flags,
85 if (open_flags & (O_WRONLY|O_RDWR))
91 if (open_flags & O_NONBLOCK)
92 *flags |= LKM_NOQUEUE;
97 static int dlmfs_file_open(struct inode *inode,
100 int status, level, flags;
101 struct dlmfs_filp_private *fp = NULL;
102 struct dlmfs_inode_private *ip;
104 if (S_ISDIR(inode->i_mode))
107 mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
110 status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
114 /* We don't want to honor O_APPEND at read/write time as it
115 * doesn't make sense for LVB writes. */
116 file->f_flags &= ~O_APPEND;
118 fp = kmalloc(sizeof(*fp), GFP_NOFS);
123 fp->fp_lock_level = level;
127 status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
129 /* this is a strange error to return here but I want
130 * to be able userspace to be able to distinguish a
131 * valid lock request from one that simply couldn't be
133 if (flags & LKM_NOQUEUE && status == -EAGAIN)
139 file->private_data = fp;
144 static int dlmfs_file_release(struct inode *inode,
148 struct dlmfs_inode_private *ip = DLMFS_I(inode);
149 struct dlmfs_filp_private *fp =
150 (struct dlmfs_filp_private *) file->private_data;
152 if (S_ISDIR(inode->i_mode))
155 mlog(0, "close called on inode %lu\n", inode->i_ino);
159 level = fp->fp_lock_level;
160 if (level != LKM_IVMODE)
161 user_dlm_cluster_unlock(&ip->ip_lockres, level);
164 file->private_data = NULL;
170 static ssize_t dlmfs_file_read(struct file *filp,
178 struct inode *inode = filp->f_path.dentry->d_inode;
180 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
181 inode->i_ino, count, *ppos);
183 if (*ppos >= i_size_read(inode))
189 if (!access_ok(VERIFY_WRITE, buf, count))
192 /* don't read past the lvb */
193 if ((count + *ppos) > i_size_read(inode))
194 readlen = i_size_read(inode) - *ppos;
196 readlen = count - *ppos;
198 lvb_buf = kmalloc(readlen, GFP_NOFS);
202 user_dlm_read_lvb(inode, lvb_buf, readlen);
203 bytes_left = __copy_to_user(buf, lvb_buf, readlen);
204 readlen -= bytes_left;
208 *ppos = *ppos + readlen;
210 mlog(0, "read %zd bytes\n", readlen);
214 static ssize_t dlmfs_file_write(struct file *filp,
215 const char __user *buf,
222 struct inode *inode = filp->f_path.dentry->d_inode;
224 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
225 inode->i_ino, count, *ppos);
227 if (*ppos >= i_size_read(inode))
233 if (!access_ok(VERIFY_READ, buf, count))
236 /* don't write past the lvb */
237 if ((count + *ppos) > i_size_read(inode))
238 writelen = i_size_read(inode) - *ppos;
240 writelen = count - *ppos;
242 lvb_buf = kmalloc(writelen, GFP_NOFS);
246 bytes_left = copy_from_user(lvb_buf, buf, writelen);
247 writelen -= bytes_left;
249 user_dlm_write_lvb(inode, lvb_buf, writelen);
253 *ppos = *ppos + writelen;
254 mlog(0, "wrote %zd bytes\n", writelen);
258 static void dlmfs_init_once(void *foo,
259 struct kmem_cache *cachep,
262 struct dlmfs_inode_private *ip =
263 (struct dlmfs_inode_private *) foo;
265 if (flags & SLAB_CTOR_CONSTRUCTOR) {
267 ip->ip_parent = NULL;
269 inode_init_once(&ip->ip_vfs_inode);
273 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
275 struct dlmfs_inode_private *ip;
277 ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
281 return &ip->ip_vfs_inode;
284 static void dlmfs_destroy_inode(struct inode *inode)
286 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
289 static void dlmfs_clear_inode(struct inode *inode)
292 struct dlmfs_inode_private *ip;
297 mlog(0, "inode %lu\n", inode->i_ino);
301 if (S_ISREG(inode->i_mode)) {
302 status = user_dlm_destroy_lock(&ip->ip_lockres);
309 mlog(0, "we're a directory, ip->ip_dlm = 0x%p\n", ip->ip_dlm);
310 /* we must be a directory. If required, lets unregister the
311 * dlm context now. */
313 user_dlm_unregister_context(ip->ip_dlm);
315 ip->ip_parent = NULL;
319 static struct backing_dev_info dlmfs_backing_dev_info = {
320 .ra_pages = 0, /* No readahead */
321 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
324 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
326 struct inode *inode = new_inode(sb);
327 int mode = S_IFDIR | 0755;
328 struct dlmfs_inode_private *ip;
333 inode->i_mode = mode;
334 inode->i_uid = current->fsuid;
335 inode->i_gid = current->fsgid;
337 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
338 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
341 inode->i_fop = &simple_dir_operations;
342 inode->i_op = &dlmfs_root_inode_operations;
348 static struct inode *dlmfs_get_inode(struct inode *parent,
349 struct dentry *dentry,
352 struct super_block *sb = parent->i_sb;
353 struct inode * inode = new_inode(sb);
354 struct dlmfs_inode_private *ip;
359 inode->i_mode = mode;
360 inode->i_uid = current->fsuid;
361 inode->i_gid = current->fsgid;
363 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
364 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
367 ip->ip_dlm = DLMFS_I(parent)->ip_dlm;
369 switch (mode & S_IFMT) {
371 /* for now we don't support anything other than
372 * directories and regular files. */
376 inode->i_op = &dlmfs_file_inode_operations;
377 inode->i_fop = &dlmfs_file_operations;
379 i_size_write(inode, DLM_LVB_LEN);
381 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
383 /* released at clear_inode time, this insures that we
384 * get to drop the dlm reference on each lock *before*
385 * we call the unregister code for releasing parent
387 ip->ip_parent = igrab(parent);
388 BUG_ON(!ip->ip_parent);
391 inode->i_op = &dlmfs_dir_inode_operations;
392 inode->i_fop = &simple_dir_operations;
394 /* directory inodes start off with i_nlink ==
395 * 2 (for "." entry) */
400 if (parent->i_mode & S_ISGID) {
401 inode->i_gid = parent->i_gid;
403 inode->i_mode |= S_ISGID;
410 * File creation. Allocate an inode, and we're done..
413 static int dlmfs_mkdir(struct inode * dir,
414 struct dentry * dentry,
418 struct inode *inode = NULL;
419 struct qstr *domain = &dentry->d_name;
420 struct dlmfs_inode_private *ip;
421 struct dlm_ctxt *dlm;
423 mlog(0, "mkdir %.*s\n", domain->len, domain->name);
425 /* verify that we have a proper domain */
426 if (domain->len >= O2NM_MAX_NAME_LEN) {
428 mlog(ML_ERROR, "invalid domain name for directory.\n");
432 inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
441 dlm = user_dlm_register_context(domain);
443 status = PTR_ERR(dlm);
444 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
445 status, domain->len, domain->name);
451 d_instantiate(dentry, inode);
452 dget(dentry); /* Extra count - pin the dentry in core */
461 static int dlmfs_create(struct inode *dir,
462 struct dentry *dentry,
464 struct nameidata *nd)
468 struct qstr *name = &dentry->d_name;
470 mlog(0, "create %.*s\n", name->len, name->name);
472 /* verify name is valid and doesn't contain any dlm reserved
474 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
475 name->name[0] == '$') {
477 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
482 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
489 d_instantiate(dentry, inode);
490 dget(dentry); /* Extra count - pin the dentry in core */
495 static int dlmfs_unlink(struct inode *dir,
496 struct dentry *dentry)
499 struct inode *inode = dentry->d_inode;
501 mlog(0, "unlink inode %lu\n", inode->i_ino);
503 /* if there are no current holders, or none that are waiting
504 * to acquire a lock, this basically destroys our lockres. */
505 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
507 mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
508 dentry->d_name.len, dentry->d_name.name, status);
511 status = simple_unlink(dir, dentry);
516 static int dlmfs_fill_super(struct super_block * sb,
520 struct inode * inode;
521 struct dentry * root;
523 sb->s_maxbytes = MAX_LFS_FILESIZE;
524 sb->s_blocksize = PAGE_CACHE_SIZE;
525 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
526 sb->s_magic = DLMFS_MAGIC;
527 sb->s_op = &dlmfs_ops;
528 inode = dlmfs_get_root_inode(sb);
532 root = d_alloc_root(inode);
541 static const struct file_operations dlmfs_file_operations = {
542 .open = dlmfs_file_open,
543 .release = dlmfs_file_release,
544 .read = dlmfs_file_read,
545 .write = dlmfs_file_write,
548 static const struct inode_operations dlmfs_dir_inode_operations = {
549 .create = dlmfs_create,
550 .lookup = simple_lookup,
551 .unlink = dlmfs_unlink,
554 /* this way we can restrict mkdir to only the toplevel of the fs. */
555 static const struct inode_operations dlmfs_root_inode_operations = {
556 .lookup = simple_lookup,
557 .mkdir = dlmfs_mkdir,
558 .rmdir = simple_rmdir,
561 static const struct super_operations dlmfs_ops = {
562 .statfs = simple_statfs,
563 .alloc_inode = dlmfs_alloc_inode,
564 .destroy_inode = dlmfs_destroy_inode,
565 .clear_inode = dlmfs_clear_inode,
566 .drop_inode = generic_delete_inode,
569 static const struct inode_operations dlmfs_file_inode_operations = {
570 .getattr = simple_getattr,
573 static int dlmfs_get_sb(struct file_system_type *fs_type,
574 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
576 return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super, mnt);
579 static struct file_system_type dlmfs_fs_type = {
580 .owner = THIS_MODULE,
581 .name = "ocfs2_dlmfs",
582 .get_sb = dlmfs_get_sb,
583 .kill_sb = kill_litter_super,
586 static int __init init_dlmfs_fs(void)
589 int cleanup_inode = 0, cleanup_worker = 0;
591 dlmfs_print_version();
593 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
594 sizeof(struct dlmfs_inode_private),
595 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
597 dlmfs_init_once, NULL);
598 if (!dlmfs_inode_cache)
602 user_dlm_worker = create_singlethread_workqueue("user_dlm");
603 if (!user_dlm_worker) {
609 status = register_filesystem(&dlmfs_fs_type);
613 kmem_cache_destroy(dlmfs_inode_cache);
615 destroy_workqueue(user_dlm_worker);
617 printk("OCFS2 User DLM kernel interface loaded\n");
621 static void __exit exit_dlmfs_fs(void)
623 unregister_filesystem(&dlmfs_fs_type);
625 flush_workqueue(user_dlm_worker);
626 destroy_workqueue(user_dlm_worker);
628 kmem_cache_destroy(dlmfs_inode_cache);
631 MODULE_AUTHOR("Oracle");
632 MODULE_LICENSE("GPL");
634 module_init(init_dlmfs_fs)
635 module_exit(exit_dlmfs_fs)