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 #include "ocfs2_lockingver.h"
65 static const struct super_operations dlmfs_ops;
66 static const struct file_operations dlmfs_file_operations;
67 static const struct inode_operations dlmfs_dir_inode_operations;
68 static const struct inode_operations dlmfs_root_inode_operations;
69 static const struct inode_operations dlmfs_file_inode_operations;
70 static struct kmem_cache *dlmfs_inode_cache;
72 struct workqueue_struct *user_dlm_worker;
75 * This is the userdlmfs locking protocol version.
77 * See fs/ocfs2/dlmglue.c for more details on locking versions.
79 static const struct dlm_protocol_version user_locking_protocol = {
80 .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
81 .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
85 * decodes a set of open flags into a valid lock level and a set of flags.
86 * returns < 0 if we have invalid flags
87 * flags which mean something to us:
88 * O_RDONLY -> PRMODE level
89 * O_WRONLY -> EXMODE level
91 * O_NONBLOCK -> LKM_NOQUEUE
93 static int dlmfs_decode_open_flags(int open_flags,
97 if (open_flags & (O_WRONLY|O_RDWR))
103 if (open_flags & O_NONBLOCK)
104 *flags |= LKM_NOQUEUE;
109 static int dlmfs_file_open(struct inode *inode,
112 int status, level, flags;
113 struct dlmfs_filp_private *fp = NULL;
114 struct dlmfs_inode_private *ip;
116 if (S_ISDIR(inode->i_mode))
119 mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
122 status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
126 /* We don't want to honor O_APPEND at read/write time as it
127 * doesn't make sense for LVB writes. */
128 file->f_flags &= ~O_APPEND;
130 fp = kmalloc(sizeof(*fp), GFP_NOFS);
135 fp->fp_lock_level = level;
139 status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
141 /* this is a strange error to return here but I want
142 * to be able userspace to be able to distinguish a
143 * valid lock request from one that simply couldn't be
145 if (flags & LKM_NOQUEUE && status == -EAGAIN)
151 file->private_data = fp;
156 static int dlmfs_file_release(struct inode *inode,
160 struct dlmfs_inode_private *ip = DLMFS_I(inode);
161 struct dlmfs_filp_private *fp =
162 (struct dlmfs_filp_private *) file->private_data;
164 if (S_ISDIR(inode->i_mode))
167 mlog(0, "close called on inode %lu\n", inode->i_ino);
171 level = fp->fp_lock_level;
172 if (level != LKM_IVMODE)
173 user_dlm_cluster_unlock(&ip->ip_lockres, level);
176 file->private_data = NULL;
182 static ssize_t dlmfs_file_read(struct file *filp,
190 struct inode *inode = filp->f_path.dentry->d_inode;
192 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
193 inode->i_ino, count, *ppos);
195 if (*ppos >= i_size_read(inode))
201 if (!access_ok(VERIFY_WRITE, buf, count))
204 /* don't read past the lvb */
205 if ((count + *ppos) > i_size_read(inode))
206 readlen = i_size_read(inode) - *ppos;
208 readlen = count - *ppos;
210 lvb_buf = kmalloc(readlen, GFP_NOFS);
214 user_dlm_read_lvb(inode, lvb_buf, readlen);
215 bytes_left = __copy_to_user(buf, lvb_buf, readlen);
216 readlen -= bytes_left;
220 *ppos = *ppos + readlen;
222 mlog(0, "read %zd bytes\n", readlen);
226 static ssize_t dlmfs_file_write(struct file *filp,
227 const char __user *buf,
234 struct inode *inode = filp->f_path.dentry->d_inode;
236 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
237 inode->i_ino, count, *ppos);
239 if (*ppos >= i_size_read(inode))
245 if (!access_ok(VERIFY_READ, buf, count))
248 /* don't write past the lvb */
249 if ((count + *ppos) > i_size_read(inode))
250 writelen = i_size_read(inode) - *ppos;
252 writelen = count - *ppos;
254 lvb_buf = kmalloc(writelen, GFP_NOFS);
258 bytes_left = copy_from_user(lvb_buf, buf, writelen);
259 writelen -= bytes_left;
261 user_dlm_write_lvb(inode, lvb_buf, writelen);
265 *ppos = *ppos + writelen;
266 mlog(0, "wrote %zd bytes\n", writelen);
270 static void dlmfs_init_once(void *foo)
272 struct dlmfs_inode_private *ip =
273 (struct dlmfs_inode_private *) foo;
276 ip->ip_parent = NULL;
278 inode_init_once(&ip->ip_vfs_inode);
281 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
283 struct dlmfs_inode_private *ip;
285 ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
289 return &ip->ip_vfs_inode;
292 static void dlmfs_destroy_inode(struct inode *inode)
294 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
297 static void dlmfs_clear_inode(struct inode *inode)
300 struct dlmfs_inode_private *ip;
305 mlog(0, "inode %lu\n", inode->i_ino);
309 if (S_ISREG(inode->i_mode)) {
310 status = user_dlm_destroy_lock(&ip->ip_lockres);
317 mlog(0, "we're a directory, ip->ip_dlm = 0x%p\n", ip->ip_dlm);
318 /* we must be a directory. If required, lets unregister the
319 * dlm context now. */
321 user_dlm_unregister_context(ip->ip_dlm);
323 ip->ip_parent = NULL;
327 static struct backing_dev_info dlmfs_backing_dev_info = {
328 .ra_pages = 0, /* No readahead */
329 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
332 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
334 struct inode *inode = new_inode(sb);
335 int mode = S_IFDIR | 0755;
336 struct dlmfs_inode_private *ip;
341 inode->i_mode = mode;
342 inode->i_uid = current_fsuid();
343 inode->i_gid = current_fsgid();
344 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
345 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
348 inode->i_fop = &simple_dir_operations;
349 inode->i_op = &dlmfs_root_inode_operations;
355 static struct inode *dlmfs_get_inode(struct inode *parent,
356 struct dentry *dentry,
359 struct super_block *sb = parent->i_sb;
360 struct inode * inode = new_inode(sb);
361 struct dlmfs_inode_private *ip;
366 inode->i_mode = mode;
367 inode->i_uid = current_fsuid();
368 inode->i_gid = current_fsgid();
369 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
370 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
373 ip->ip_dlm = DLMFS_I(parent)->ip_dlm;
375 switch (mode & S_IFMT) {
377 /* for now we don't support anything other than
378 * directories and regular files. */
382 inode->i_op = &dlmfs_file_inode_operations;
383 inode->i_fop = &dlmfs_file_operations;
385 i_size_write(inode, DLM_LVB_LEN);
387 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
389 /* released at clear_inode time, this insures that we
390 * get to drop the dlm reference on each lock *before*
391 * we call the unregister code for releasing parent
393 ip->ip_parent = igrab(parent);
394 BUG_ON(!ip->ip_parent);
397 inode->i_op = &dlmfs_dir_inode_operations;
398 inode->i_fop = &simple_dir_operations;
400 /* directory inodes start off with i_nlink ==
401 * 2 (for "." entry) */
406 if (parent->i_mode & S_ISGID) {
407 inode->i_gid = parent->i_gid;
409 inode->i_mode |= S_ISGID;
416 * File creation. Allocate an inode, and we're done..
419 static int dlmfs_mkdir(struct inode * dir,
420 struct dentry * dentry,
424 struct inode *inode = NULL;
425 struct qstr *domain = &dentry->d_name;
426 struct dlmfs_inode_private *ip;
427 struct dlm_ctxt *dlm;
428 struct dlm_protocol_version proto = user_locking_protocol;
430 mlog(0, "mkdir %.*s\n", domain->len, domain->name);
432 /* verify that we have a proper domain */
433 if (domain->len >= O2NM_MAX_NAME_LEN) {
435 mlog(ML_ERROR, "invalid domain name for directory.\n");
439 inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
448 dlm = user_dlm_register_context(domain, &proto);
450 status = PTR_ERR(dlm);
451 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
452 status, domain->len, domain->name);
458 d_instantiate(dentry, inode);
459 dget(dentry); /* Extra count - pin the dentry in core */
468 static int dlmfs_create(struct inode *dir,
469 struct dentry *dentry,
471 struct nameidata *nd)
475 struct qstr *name = &dentry->d_name;
477 mlog(0, "create %.*s\n", name->len, name->name);
479 /* verify name is valid and doesn't contain any dlm reserved
481 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
482 name->name[0] == '$') {
484 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
489 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
496 d_instantiate(dentry, inode);
497 dget(dentry); /* Extra count - pin the dentry in core */
502 static int dlmfs_unlink(struct inode *dir,
503 struct dentry *dentry)
506 struct inode *inode = dentry->d_inode;
508 mlog(0, "unlink inode %lu\n", inode->i_ino);
510 /* if there are no current holders, or none that are waiting
511 * to acquire a lock, this basically destroys our lockres. */
512 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
514 mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
515 dentry->d_name.len, dentry->d_name.name, status);
518 status = simple_unlink(dir, dentry);
523 static int dlmfs_fill_super(struct super_block * sb,
527 struct inode * inode;
528 struct dentry * root;
530 sb->s_maxbytes = MAX_LFS_FILESIZE;
531 sb->s_blocksize = PAGE_CACHE_SIZE;
532 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
533 sb->s_magic = DLMFS_MAGIC;
534 sb->s_op = &dlmfs_ops;
535 inode = dlmfs_get_root_inode(sb);
539 root = d_alloc_root(inode);
548 static const struct file_operations dlmfs_file_operations = {
549 .open = dlmfs_file_open,
550 .release = dlmfs_file_release,
551 .read = dlmfs_file_read,
552 .write = dlmfs_file_write,
555 static const struct inode_operations dlmfs_dir_inode_operations = {
556 .create = dlmfs_create,
557 .lookup = simple_lookup,
558 .unlink = dlmfs_unlink,
561 /* this way we can restrict mkdir to only the toplevel of the fs. */
562 static const struct inode_operations dlmfs_root_inode_operations = {
563 .lookup = simple_lookup,
564 .mkdir = dlmfs_mkdir,
565 .rmdir = simple_rmdir,
568 static const struct super_operations dlmfs_ops = {
569 .statfs = simple_statfs,
570 .alloc_inode = dlmfs_alloc_inode,
571 .destroy_inode = dlmfs_destroy_inode,
572 .clear_inode = dlmfs_clear_inode,
573 .drop_inode = generic_delete_inode,
576 static const struct inode_operations dlmfs_file_inode_operations = {
577 .getattr = simple_getattr,
580 static int dlmfs_get_sb(struct file_system_type *fs_type,
581 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
583 return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super, mnt);
586 static struct file_system_type dlmfs_fs_type = {
587 .owner = THIS_MODULE,
588 .name = "ocfs2_dlmfs",
589 .get_sb = dlmfs_get_sb,
590 .kill_sb = kill_litter_super,
593 static int __init init_dlmfs_fs(void)
596 int cleanup_inode = 0, cleanup_worker = 0;
598 dlmfs_print_version();
600 status = bdi_init(&dlmfs_backing_dev_info);
604 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
605 sizeof(struct dlmfs_inode_private),
606 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
609 if (!dlmfs_inode_cache) {
615 user_dlm_worker = create_singlethread_workqueue("user_dlm");
616 if (!user_dlm_worker) {
622 status = register_filesystem(&dlmfs_fs_type);
626 kmem_cache_destroy(dlmfs_inode_cache);
628 destroy_workqueue(user_dlm_worker);
629 bdi_destroy(&dlmfs_backing_dev_info);
631 printk("OCFS2 User DLM kernel interface loaded\n");
635 static void __exit exit_dlmfs_fs(void)
637 unregister_filesystem(&dlmfs_fs_type);
639 flush_workqueue(user_dlm_worker);
640 destroy_workqueue(user_dlm_worker);
642 kmem_cache_destroy(dlmfs_inode_cache);
644 bdi_destroy(&dlmfs_backing_dev_info);
647 MODULE_AUTHOR("Oracle");
648 MODULE_LICENSE("GPL");
650 module_init(init_dlmfs_fs)
651 module_exit(exit_dlmfs_fs)