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/smp_lock.h>
46 #include <linux/backing-dev.h>
48 #include <asm/uaccess.h>
51 #include "cluster/nodemanager.h"
52 #include "cluster/heartbeat.h"
53 #include "cluster/tcp.h"
61 #define MLOG_MASK_PREFIX ML_DLMFS
62 #include "cluster/masklog.h"
64 static struct super_operations dlmfs_ops;
65 static struct file_operations dlmfs_file_operations;
66 static struct inode_operations dlmfs_dir_inode_operations;
67 static struct inode_operations dlmfs_root_inode_operations;
68 static struct inode_operations dlmfs_file_inode_operations;
69 static struct kmem_cache *dlmfs_inode_cache;
71 struct workqueue_struct *user_dlm_worker;
74 * decodes a set of open flags into a valid lock level and a set of flags.
75 * returns < 0 if we have invalid flags
76 * flags which mean something to us:
77 * O_RDONLY -> PRMODE level
78 * O_WRONLY -> EXMODE level
80 * O_NONBLOCK -> LKM_NOQUEUE
82 static int dlmfs_decode_open_flags(int open_flags,
86 if (open_flags & (O_WRONLY|O_RDWR))
92 if (open_flags & O_NONBLOCK)
93 *flags |= LKM_NOQUEUE;
98 static int dlmfs_file_open(struct inode *inode,
101 int status, level, flags;
102 struct dlmfs_filp_private *fp = NULL;
103 struct dlmfs_inode_private *ip;
105 if (S_ISDIR(inode->i_mode))
108 mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
111 status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
115 /* We don't want to honor O_APPEND at read/write time as it
116 * doesn't make sense for LVB writes. */
117 file->f_flags &= ~O_APPEND;
119 fp = kmalloc(sizeof(*fp), GFP_NOFS);
124 fp->fp_lock_level = level;
128 status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
130 /* this is a strange error to return here but I want
131 * to be able userspace to be able to distinguish a
132 * valid lock request from one that simply couldn't be
134 if (flags & LKM_NOQUEUE && status == -EAGAIN)
140 file->private_data = fp;
145 static int dlmfs_file_release(struct inode *inode,
149 struct dlmfs_inode_private *ip = DLMFS_I(inode);
150 struct dlmfs_filp_private *fp =
151 (struct dlmfs_filp_private *) file->private_data;
153 if (S_ISDIR(inode->i_mode))
156 mlog(0, "close called on inode %lu\n", inode->i_ino);
160 level = fp->fp_lock_level;
161 if (level != LKM_IVMODE)
162 user_dlm_cluster_unlock(&ip->ip_lockres, level);
165 file->private_data = NULL;
171 static ssize_t dlmfs_file_read(struct file *filp,
179 struct inode *inode = filp->f_dentry->d_inode;
181 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
182 inode->i_ino, count, *ppos);
184 if (*ppos >= i_size_read(inode))
190 if (!access_ok(VERIFY_WRITE, buf, count))
193 /* don't read past the lvb */
194 if ((count + *ppos) > i_size_read(inode))
195 readlen = i_size_read(inode) - *ppos;
197 readlen = count - *ppos;
199 lvb_buf = kmalloc(readlen, GFP_NOFS);
203 user_dlm_read_lvb(inode, lvb_buf, readlen);
204 bytes_left = __copy_to_user(buf, lvb_buf, readlen);
205 readlen -= bytes_left;
209 *ppos = *ppos + readlen;
211 mlog(0, "read %zd bytes\n", readlen);
215 static ssize_t dlmfs_file_write(struct file *filp,
216 const char __user *buf,
223 struct inode *inode = filp->f_dentry->d_inode;
225 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
226 inode->i_ino, count, *ppos);
228 if (*ppos >= i_size_read(inode))
234 if (!access_ok(VERIFY_READ, buf, count))
237 /* don't write past the lvb */
238 if ((count + *ppos) > i_size_read(inode))
239 writelen = i_size_read(inode) - *ppos;
241 writelen = count - *ppos;
243 lvb_buf = kmalloc(writelen, GFP_NOFS);
247 bytes_left = copy_from_user(lvb_buf, buf, writelen);
248 writelen -= bytes_left;
250 user_dlm_write_lvb(inode, lvb_buf, writelen);
254 *ppos = *ppos + writelen;
255 mlog(0, "wrote %zd bytes\n", writelen);
259 static void dlmfs_init_once(void *foo,
260 struct kmem_cache *cachep,
263 struct dlmfs_inode_private *ip =
264 (struct dlmfs_inode_private *) foo;
266 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
267 SLAB_CTOR_CONSTRUCTOR) {
269 ip->ip_parent = NULL;
271 inode_init_once(&ip->ip_vfs_inode);
275 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
277 struct dlmfs_inode_private *ip;
279 ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
283 return &ip->ip_vfs_inode;
286 static void dlmfs_destroy_inode(struct inode *inode)
288 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
291 static void dlmfs_clear_inode(struct inode *inode)
294 struct dlmfs_inode_private *ip;
299 mlog(0, "inode %lu\n", inode->i_ino);
303 if (S_ISREG(inode->i_mode)) {
304 status = user_dlm_destroy_lock(&ip->ip_lockres);
311 mlog(0, "we're a directory, ip->ip_dlm = 0x%p\n", ip->ip_dlm);
312 /* we must be a directory. If required, lets unregister the
313 * dlm context now. */
315 user_dlm_unregister_context(ip->ip_dlm);
317 ip->ip_parent = NULL;
321 static struct backing_dev_info dlmfs_backing_dev_info = {
322 .ra_pages = 0, /* No readahead */
323 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
326 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
328 struct inode *inode = new_inode(sb);
329 int mode = S_IFDIR | 0755;
330 struct dlmfs_inode_private *ip;
335 inode->i_mode = mode;
336 inode->i_uid = current->fsuid;
337 inode->i_gid = current->fsgid;
339 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
340 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
343 inode->i_fop = &simple_dir_operations;
344 inode->i_op = &dlmfs_root_inode_operations;
350 static struct inode *dlmfs_get_inode(struct inode *parent,
351 struct dentry *dentry,
354 struct super_block *sb = parent->i_sb;
355 struct inode * inode = new_inode(sb);
356 struct dlmfs_inode_private *ip;
361 inode->i_mode = mode;
362 inode->i_uid = current->fsuid;
363 inode->i_gid = current->fsgid;
365 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
366 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
369 ip->ip_dlm = DLMFS_I(parent)->ip_dlm;
371 switch (mode & S_IFMT) {
373 /* for now we don't support anything other than
374 * directories and regular files. */
378 inode->i_op = &dlmfs_file_inode_operations;
379 inode->i_fop = &dlmfs_file_operations;
381 i_size_write(inode, DLM_LVB_LEN);
383 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
385 /* released at clear_inode time, this insures that we
386 * get to drop the dlm reference on each lock *before*
387 * we call the unregister code for releasing parent
389 ip->ip_parent = igrab(parent);
390 BUG_ON(!ip->ip_parent);
393 inode->i_op = &dlmfs_dir_inode_operations;
394 inode->i_fop = &simple_dir_operations;
396 /* directory inodes start off with i_nlink ==
397 * 2 (for "." entry) */
402 if (parent->i_mode & S_ISGID) {
403 inode->i_gid = parent->i_gid;
405 inode->i_mode |= S_ISGID;
412 * File creation. Allocate an inode, and we're done..
415 static int dlmfs_mkdir(struct inode * dir,
416 struct dentry * dentry,
420 struct inode *inode = NULL;
421 struct qstr *domain = &dentry->d_name;
422 struct dlmfs_inode_private *ip;
423 struct dlm_ctxt *dlm;
425 mlog(0, "mkdir %.*s\n", domain->len, domain->name);
427 /* verify that we have a proper domain */
428 if (domain->len >= O2NM_MAX_NAME_LEN) {
430 mlog(ML_ERROR, "invalid domain name for directory.\n");
434 inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
443 dlm = user_dlm_register_context(domain);
445 status = PTR_ERR(dlm);
446 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
447 status, domain->len, domain->name);
453 d_instantiate(dentry, inode);
454 dget(dentry); /* Extra count - pin the dentry in core */
463 static int dlmfs_create(struct inode *dir,
464 struct dentry *dentry,
466 struct nameidata *nd)
470 struct qstr *name = &dentry->d_name;
472 mlog(0, "create %.*s\n", name->len, name->name);
474 /* verify name is valid and doesn't contain any dlm reserved
476 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
477 name->name[0] == '$') {
479 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
484 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
491 d_instantiate(dentry, inode);
492 dget(dentry); /* Extra count - pin the dentry in core */
497 static int dlmfs_unlink(struct inode *dir,
498 struct dentry *dentry)
501 struct inode *inode = dentry->d_inode;
503 mlog(0, "unlink inode %lu\n", inode->i_ino);
505 /* if there are no current holders, or none that are waiting
506 * to acquire a lock, this basically destroys our lockres. */
507 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
509 mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
510 dentry->d_name.len, dentry->d_name.name, status);
513 status = simple_unlink(dir, dentry);
518 static int dlmfs_fill_super(struct super_block * sb,
522 struct inode * inode;
523 struct dentry * root;
525 sb->s_maxbytes = MAX_LFS_FILESIZE;
526 sb->s_blocksize = PAGE_CACHE_SIZE;
527 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
528 sb->s_magic = DLMFS_MAGIC;
529 sb->s_op = &dlmfs_ops;
530 inode = dlmfs_get_root_inode(sb);
534 root = d_alloc_root(inode);
543 static struct file_operations dlmfs_file_operations = {
544 .open = dlmfs_file_open,
545 .release = dlmfs_file_release,
546 .read = dlmfs_file_read,
547 .write = dlmfs_file_write,
550 static struct inode_operations dlmfs_dir_inode_operations = {
551 .create = dlmfs_create,
552 .lookup = simple_lookup,
553 .unlink = dlmfs_unlink,
556 /* this way we can restrict mkdir to only the toplevel of the fs. */
557 static struct inode_operations dlmfs_root_inode_operations = {
558 .lookup = simple_lookup,
559 .mkdir = dlmfs_mkdir,
560 .rmdir = simple_rmdir,
563 static struct super_operations dlmfs_ops = {
564 .statfs = simple_statfs,
565 .alloc_inode = dlmfs_alloc_inode,
566 .destroy_inode = dlmfs_destroy_inode,
567 .clear_inode = dlmfs_clear_inode,
568 .drop_inode = generic_delete_inode,
571 static struct inode_operations dlmfs_file_inode_operations = {
572 .getattr = simple_getattr,
575 static int dlmfs_get_sb(struct file_system_type *fs_type,
576 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
578 return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super, mnt);
581 static struct file_system_type dlmfs_fs_type = {
582 .owner = THIS_MODULE,
583 .name = "ocfs2_dlmfs",
584 .get_sb = dlmfs_get_sb,
585 .kill_sb = kill_litter_super,
588 static int __init init_dlmfs_fs(void)
591 int cleanup_inode = 0, cleanup_worker = 0;
593 dlmfs_print_version();
595 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
596 sizeof(struct dlmfs_inode_private),
597 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
599 dlmfs_init_once, NULL);
600 if (!dlmfs_inode_cache)
604 user_dlm_worker = create_singlethread_workqueue("user_dlm");
605 if (!user_dlm_worker) {
611 status = register_filesystem(&dlmfs_fs_type);
615 kmem_cache_destroy(dlmfs_inode_cache);
617 destroy_workqueue(user_dlm_worker);
619 printk("OCFS2 User DLM kernel interface loaded\n");
623 static void __exit exit_dlmfs_fs(void)
625 unregister_filesystem(&dlmfs_fs_type);
627 flush_workqueue(user_dlm_worker);
628 destroy_workqueue(user_dlm_worker);
630 kmem_cache_destroy(dlmfs_inode_cache);
633 MODULE_AUTHOR("Oracle");
634 MODULE_LICENSE("GPL");
636 module_init(init_dlmfs_fs)
637 module_exit(exit_dlmfs_fs)