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>
47 #include <linux/vs_tag.h>
49 #include <asm/uaccess.h>
52 #include "cluster/nodemanager.h"
53 #include "cluster/heartbeat.h"
54 #include "cluster/tcp.h"
62 #define MLOG_MASK_PREFIX ML_DLMFS
63 #include "cluster/masklog.h"
65 static struct super_operations dlmfs_ops;
66 static struct file_operations dlmfs_file_operations;
67 static struct inode_operations dlmfs_dir_inode_operations;
68 static struct inode_operations dlmfs_root_inode_operations;
69 static struct inode_operations dlmfs_file_inode_operations;
70 static struct kmem_cache *dlmfs_inode_cache;
72 struct workqueue_struct *user_dlm_worker;
75 * decodes a set of open flags into a valid lock level and a set of flags.
76 * returns < 0 if we have invalid flags
77 * flags which mean something to us:
78 * O_RDONLY -> PRMODE level
79 * O_WRONLY -> EXMODE level
81 * O_NONBLOCK -> LKM_NOQUEUE
83 static int dlmfs_decode_open_flags(int open_flags,
87 if (open_flags & (O_WRONLY|O_RDWR))
93 if (open_flags & O_NONBLOCK)
94 *flags |= LKM_NOQUEUE;
99 static int dlmfs_file_open(struct inode *inode,
102 int status, level, flags;
103 struct dlmfs_filp_private *fp = NULL;
104 struct dlmfs_inode_private *ip;
106 if (S_ISDIR(inode->i_mode))
109 mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
112 status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
116 /* We don't want to honor O_APPEND at read/write time as it
117 * doesn't make sense for LVB writes. */
118 file->f_flags &= ~O_APPEND;
120 fp = kmalloc(sizeof(*fp), GFP_NOFS);
125 fp->fp_lock_level = level;
129 status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
131 /* this is a strange error to return here but I want
132 * to be able userspace to be able to distinguish a
133 * valid lock request from one that simply couldn't be
135 if (flags & LKM_NOQUEUE && status == -EAGAIN)
141 file->private_data = fp;
146 static int dlmfs_file_release(struct inode *inode,
150 struct dlmfs_inode_private *ip = DLMFS_I(inode);
151 struct dlmfs_filp_private *fp =
152 (struct dlmfs_filp_private *) file->private_data;
154 if (S_ISDIR(inode->i_mode))
157 mlog(0, "close called on inode %lu\n", inode->i_ino);
161 level = fp->fp_lock_level;
162 if (level != LKM_IVMODE)
163 user_dlm_cluster_unlock(&ip->ip_lockres, level);
166 file->private_data = NULL;
172 static ssize_t dlmfs_file_read(struct file *filp,
180 struct inode *inode = filp->f_path.dentry->d_inode;
182 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
183 inode->i_ino, count, *ppos);
185 if (*ppos >= i_size_read(inode))
191 if (!access_ok(VERIFY_WRITE, buf, count))
194 /* don't read past the lvb */
195 if ((count + *ppos) > i_size_read(inode))
196 readlen = i_size_read(inode) - *ppos;
198 readlen = count - *ppos;
200 lvb_buf = kmalloc(readlen, GFP_NOFS);
204 user_dlm_read_lvb(inode, lvb_buf, readlen);
205 bytes_left = __copy_to_user(buf, lvb_buf, readlen);
206 readlen -= bytes_left;
210 *ppos = *ppos + readlen;
212 mlog(0, "read %zd bytes\n", readlen);
216 static ssize_t dlmfs_file_write(struct file *filp,
217 const char __user *buf,
224 struct inode *inode = filp->f_path.dentry->d_inode;
226 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
227 inode->i_ino, count, *ppos);
229 if (*ppos >= i_size_read(inode))
235 if (!access_ok(VERIFY_READ, buf, count))
238 /* don't write past the lvb */
239 if ((count + *ppos) > i_size_read(inode))
240 writelen = i_size_read(inode) - *ppos;
242 writelen = count - *ppos;
244 lvb_buf = kmalloc(writelen, GFP_NOFS);
248 bytes_left = copy_from_user(lvb_buf, buf, writelen);
249 writelen -= bytes_left;
251 user_dlm_write_lvb(inode, lvb_buf, writelen);
255 *ppos = *ppos + writelen;
256 mlog(0, "wrote %zd bytes\n", writelen);
260 static void dlmfs_init_once(void *foo,
261 struct kmem_cache *cachep,
264 struct dlmfs_inode_private *ip =
265 (struct dlmfs_inode_private *) foo;
267 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
268 SLAB_CTOR_CONSTRUCTOR) {
270 ip->ip_parent = NULL;
272 inode_init_once(&ip->ip_vfs_inode);
276 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
278 struct dlmfs_inode_private *ip;
280 ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
284 return &ip->ip_vfs_inode;
287 static void dlmfs_destroy_inode(struct inode *inode)
289 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
292 static void dlmfs_clear_inode(struct inode *inode)
295 struct dlmfs_inode_private *ip;
300 mlog(0, "inode %lu\n", inode->i_ino);
304 if (S_ISREG(inode->i_mode)) {
305 status = user_dlm_destroy_lock(&ip->ip_lockres);
312 mlog(0, "we're a directory, ip->ip_dlm = 0x%p\n", ip->ip_dlm);
313 /* we must be a directory. If required, lets unregister the
314 * dlm context now. */
316 user_dlm_unregister_context(ip->ip_dlm);
318 ip->ip_parent = NULL;
322 static struct backing_dev_info dlmfs_backing_dev_info = {
323 .ra_pages = 0, /* No readahead */
324 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
327 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
329 struct inode *inode = new_inode(sb);
330 int mode = S_IFDIR | 0755;
331 struct dlmfs_inode_private *ip;
336 inode->i_mode = mode;
337 inode->i_uid = current->fsuid;
338 inode->i_gid = current->fsgid;
339 inode->i_tag = dx_current_fstag(sb);
341 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
342 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
345 inode->i_fop = &simple_dir_operations;
346 inode->i_op = &dlmfs_root_inode_operations;
352 static struct inode *dlmfs_get_inode(struct inode *parent,
353 struct dentry *dentry,
356 struct super_block *sb = parent->i_sb;
357 struct inode * inode = new_inode(sb);
358 struct dlmfs_inode_private *ip;
363 inode->i_mode = mode;
364 inode->i_uid = current->fsuid;
365 inode->i_gid = current->fsgid;
366 inode->i_tag = dx_current_fstag(sb);
368 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
369 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
372 ip->ip_dlm = DLMFS_I(parent)->ip_dlm;
374 switch (mode & S_IFMT) {
376 /* for now we don't support anything other than
377 * directories and regular files. */
381 inode->i_op = &dlmfs_file_inode_operations;
382 inode->i_fop = &dlmfs_file_operations;
384 i_size_write(inode, DLM_LVB_LEN);
386 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
388 /* released at clear_inode time, this insures that we
389 * get to drop the dlm reference on each lock *before*
390 * we call the unregister code for releasing parent
392 ip->ip_parent = igrab(parent);
393 BUG_ON(!ip->ip_parent);
396 inode->i_op = &dlmfs_dir_inode_operations;
397 inode->i_fop = &simple_dir_operations;
399 /* directory inodes start off with i_nlink ==
400 * 2 (for "." entry) */
405 if (parent->i_mode & S_ISGID) {
406 inode->i_gid = parent->i_gid;
408 inode->i_mode |= S_ISGID;
415 * File creation. Allocate an inode, and we're done..
418 static int dlmfs_mkdir(struct inode * dir,
419 struct dentry * dentry,
423 struct inode *inode = NULL;
424 struct qstr *domain = &dentry->d_name;
425 struct dlmfs_inode_private *ip;
426 struct dlm_ctxt *dlm;
428 mlog(0, "mkdir %.*s\n", domain->len, domain->name);
430 /* verify that we have a proper domain */
431 if (domain->len >= O2NM_MAX_NAME_LEN) {
433 mlog(ML_ERROR, "invalid domain name for directory.\n");
437 inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
446 dlm = user_dlm_register_context(domain);
448 status = PTR_ERR(dlm);
449 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
450 status, domain->len, domain->name);
456 d_instantiate(dentry, inode);
457 dget(dentry); /* Extra count - pin the dentry in core */
466 static int dlmfs_create(struct inode *dir,
467 struct dentry *dentry,
469 struct nameidata *nd)
473 struct qstr *name = &dentry->d_name;
475 mlog(0, "create %.*s\n", name->len, name->name);
477 /* verify name is valid and doesn't contain any dlm reserved
479 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
480 name->name[0] == '$') {
482 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
487 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
494 d_instantiate(dentry, inode);
495 dget(dentry); /* Extra count - pin the dentry in core */
500 static int dlmfs_unlink(struct inode *dir,
501 struct dentry *dentry)
504 struct inode *inode = dentry->d_inode;
506 mlog(0, "unlink inode %lu\n", inode->i_ino);
508 /* if there are no current holders, or none that are waiting
509 * to acquire a lock, this basically destroys our lockres. */
510 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
512 mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
513 dentry->d_name.len, dentry->d_name.name, status);
516 status = simple_unlink(dir, dentry);
521 static int dlmfs_fill_super(struct super_block * sb,
525 struct inode * inode;
526 struct dentry * root;
528 sb->s_maxbytes = MAX_LFS_FILESIZE;
529 sb->s_blocksize = PAGE_CACHE_SIZE;
530 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
531 sb->s_magic = DLMFS_MAGIC;
532 sb->s_op = &dlmfs_ops;
533 inode = dlmfs_get_root_inode(sb);
537 root = d_alloc_root(inode);
546 static struct file_operations dlmfs_file_operations = {
547 .open = dlmfs_file_open,
548 .release = dlmfs_file_release,
549 .read = dlmfs_file_read,
550 .write = dlmfs_file_write,
553 static struct inode_operations dlmfs_dir_inode_operations = {
554 .create = dlmfs_create,
555 .lookup = simple_lookup,
556 .unlink = dlmfs_unlink,
559 /* this way we can restrict mkdir to only the toplevel of the fs. */
560 static struct inode_operations dlmfs_root_inode_operations = {
561 .lookup = simple_lookup,
562 .mkdir = dlmfs_mkdir,
563 .rmdir = simple_rmdir,
566 static struct super_operations dlmfs_ops = {
567 .statfs = simple_statfs,
568 .alloc_inode = dlmfs_alloc_inode,
569 .destroy_inode = dlmfs_destroy_inode,
570 .clear_inode = dlmfs_clear_inode,
571 .drop_inode = generic_delete_inode,
574 static struct inode_operations dlmfs_file_inode_operations = {
575 .getattr = simple_getattr,
578 static int dlmfs_get_sb(struct file_system_type *fs_type,
579 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
581 return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super, mnt);
584 static struct file_system_type dlmfs_fs_type = {
585 .owner = THIS_MODULE,
586 .name = "ocfs2_dlmfs",
587 .get_sb = dlmfs_get_sb,
588 .kill_sb = kill_litter_super,
591 static int __init init_dlmfs_fs(void)
594 int cleanup_inode = 0, cleanup_worker = 0;
596 dlmfs_print_version();
598 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
599 sizeof(struct dlmfs_inode_private),
600 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
602 dlmfs_init_once, NULL);
603 if (!dlmfs_inode_cache)
607 user_dlm_worker = create_singlethread_workqueue("user_dlm");
608 if (!user_dlm_worker) {
614 status = register_filesystem(&dlmfs_fs_type);
618 kmem_cache_destroy(dlmfs_inode_cache);
620 destroy_workqueue(user_dlm_worker);
622 printk("OCFS2 User DLM kernel interface loaded\n");
626 static void __exit exit_dlmfs_fs(void)
628 unregister_filesystem(&dlmfs_fs_type);
630 flush_workqueue(user_dlm_worker);
631 destroy_workqueue(user_dlm_worker);
633 kmem_cache_destroy(dlmfs_inode_cache);
636 MODULE_AUTHOR("Oracle");
637 MODULE_LICENSE("GPL");
639 module_init(init_dlmfs_fs)
640 module_exit(exit_dlmfs_fs)