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 <asm/uaccess.h>
50 #include "stackglue.h"
54 #define MLOG_MASK_PREFIX ML_DLMFS
55 #include "cluster/masklog.h"
58 static const struct super_operations dlmfs_ops;
59 static const struct file_operations dlmfs_file_operations;
60 static const struct inode_operations dlmfs_dir_inode_operations;
61 static const struct inode_operations dlmfs_root_inode_operations;
62 static const struct inode_operations dlmfs_file_inode_operations;
63 static struct kmem_cache *dlmfs_inode_cache;
65 struct workqueue_struct *user_dlm_worker;
70 * These are the ABI capabilities of dlmfs.
72 * Over time, dlmfs has added some features that were not part of the
73 * initial ABI. Unfortunately, some of these features are not detectable
74 * via standard usage. For example, Linux's default poll always returns
75 * POLLIN, so there is no way for a caller of poll(2) to know when dlmfs
76 * added poll support. Instead, we provide this list of new capabilities.
78 * Capabilities is a read-only attribute. We do it as a module parameter
79 * so we can discover it whether dlmfs is built in, loaded, or even not
82 * The ABI features are local to this machine's dlmfs mount. This is
83 * distinct from the locking protocol, which is concerned with inter-node
87 * - bast : POLLIN against the file descriptor of a held lock
88 * signifies a bast fired on the lock.
90 #define DLMFS_CAPABILITIES "bast stackglue"
91 extern int param_set_dlmfs_capabilities(const char *val,
92 struct kernel_param *kp)
94 printk(KERN_ERR "%s: readonly parameter\n", kp->name);
97 static int param_get_dlmfs_capabilities(char *buffer,
98 struct kernel_param *kp)
100 return strlcpy(buffer, DLMFS_CAPABILITIES,
101 strlen(DLMFS_CAPABILITIES) + 1);
103 module_param_call(capabilities, param_set_dlmfs_capabilities,
104 param_get_dlmfs_capabilities, NULL, 0444);
105 MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES);
109 * decodes a set of open flags into a valid lock level and a set of flags.
110 * returns < 0 if we have invalid flags
111 * flags which mean something to us:
112 * O_RDONLY -> PRMODE level
113 * O_WRONLY -> EXMODE level
115 * O_NONBLOCK -> NOQUEUE
117 static int dlmfs_decode_open_flags(int open_flags,
121 if (open_flags & (O_WRONLY|O_RDWR))
122 *level = DLM_LOCK_EX;
124 *level = DLM_LOCK_PR;
127 if (open_flags & O_NONBLOCK)
128 *flags |= DLM_LKF_NOQUEUE;
133 static int dlmfs_file_open(struct inode *inode,
136 int status, level, flags;
137 struct dlmfs_filp_private *fp = NULL;
138 struct dlmfs_inode_private *ip;
140 if (S_ISDIR(inode->i_mode))
143 mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
146 status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
150 /* We don't want to honor O_APPEND at read/write time as it
151 * doesn't make sense for LVB writes. */
152 file->f_flags &= ~O_APPEND;
154 fp = kmalloc(sizeof(*fp), GFP_NOFS);
159 fp->fp_lock_level = level;
163 status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
165 /* this is a strange error to return here but I want
166 * to be able userspace to be able to distinguish a
167 * valid lock request from one that simply couldn't be
169 if (flags & DLM_LKF_NOQUEUE && status == -EAGAIN)
175 file->private_data = fp;
180 static int dlmfs_file_release(struct inode *inode,
184 struct dlmfs_inode_private *ip = DLMFS_I(inode);
185 struct dlmfs_filp_private *fp = file->private_data;
187 if (S_ISDIR(inode->i_mode))
190 mlog(0, "close called on inode %lu\n", inode->i_ino);
194 level = fp->fp_lock_level;
195 if (level != DLM_LOCK_IV)
196 user_dlm_cluster_unlock(&ip->ip_lockres, level);
199 file->private_data = NULL;
206 * We do ->setattr() just to override size changes. Our size is the size
207 * of the LVB and nothing else.
209 static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr)
212 struct inode *inode = dentry->d_inode;
214 attr->ia_valid &= ~ATTR_SIZE;
215 error = inode_change_ok(inode, attr);
219 setattr_copy(inode, attr);
220 mark_inode_dirty(inode);
224 static unsigned int dlmfs_file_poll(struct file *file, poll_table *wait)
227 struct inode *inode = file->f_path.dentry->d_inode;
228 struct dlmfs_inode_private *ip = DLMFS_I(inode);
230 poll_wait(file, &ip->ip_lockres.l_event, wait);
232 spin_lock(&ip->ip_lockres.l_lock);
233 if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED)
234 event = POLLIN | POLLRDNORM;
235 spin_unlock(&ip->ip_lockres.l_lock);
240 static ssize_t dlmfs_file_read(struct file *filp,
246 ssize_t readlen, got;
248 struct inode *inode = filp->f_path.dentry->d_inode;
250 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
251 inode->i_ino, count, *ppos);
253 if (*ppos >= i_size_read(inode))
259 if (!access_ok(VERIFY_WRITE, buf, count))
262 /* don't read past the lvb */
263 if ((count + *ppos) > i_size_read(inode))
264 readlen = i_size_read(inode) - *ppos;
268 lvb_buf = kmalloc(readlen, GFP_NOFS);
272 got = user_dlm_read_lvb(inode, lvb_buf, readlen);
274 BUG_ON(got != readlen);
275 bytes_left = __copy_to_user(buf, lvb_buf, readlen);
276 readlen -= bytes_left;
282 *ppos = *ppos + readlen;
284 mlog(0, "read %zd bytes\n", readlen);
288 static ssize_t dlmfs_file_write(struct file *filp,
289 const char __user *buf,
296 struct inode *inode = filp->f_path.dentry->d_inode;
298 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
299 inode->i_ino, count, *ppos);
301 if (*ppos >= i_size_read(inode))
307 if (!access_ok(VERIFY_READ, buf, count))
310 /* don't write past the lvb */
311 if ((count + *ppos) > i_size_read(inode))
312 writelen = i_size_read(inode) - *ppos;
314 writelen = count - *ppos;
316 lvb_buf = kmalloc(writelen, GFP_NOFS);
320 bytes_left = copy_from_user(lvb_buf, buf, writelen);
321 writelen -= bytes_left;
323 user_dlm_write_lvb(inode, lvb_buf, writelen);
327 *ppos = *ppos + writelen;
328 mlog(0, "wrote %zd bytes\n", writelen);
332 static void dlmfs_init_once(void *foo)
334 struct dlmfs_inode_private *ip =
335 (struct dlmfs_inode_private *) foo;
338 ip->ip_parent = NULL;
340 inode_init_once(&ip->ip_vfs_inode);
343 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
345 struct dlmfs_inode_private *ip;
347 ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
351 return &ip->ip_vfs_inode;
354 static void dlmfs_destroy_inode(struct inode *inode)
356 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
359 static void dlmfs_evict_inode(struct inode *inode)
362 struct dlmfs_inode_private *ip;
364 end_writeback(inode);
366 mlog(0, "inode %lu\n", inode->i_ino);
370 if (S_ISREG(inode->i_mode)) {
371 status = user_dlm_destroy_lock(&ip->ip_lockres);
378 mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn);
379 /* we must be a directory. If required, lets unregister the
380 * dlm context now. */
382 user_dlm_unregister(ip->ip_conn);
384 ip->ip_parent = NULL;
388 static struct backing_dev_info dlmfs_backing_dev_info = {
389 .name = "ocfs2-dlmfs",
390 .ra_pages = 0, /* No readahead */
391 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
394 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
396 struct inode *inode = new_inode(sb);
397 int mode = S_IFDIR | 0755;
398 struct dlmfs_inode_private *ip;
403 inode->i_ino = get_next_ino();
404 inode->i_mode = mode;
405 inode->i_uid = current_fsuid();
406 inode->i_gid = current_fsgid();
407 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
408 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
411 inode->i_fop = &simple_dir_operations;
412 inode->i_op = &dlmfs_root_inode_operations;
418 static struct inode *dlmfs_get_inode(struct inode *parent,
419 struct dentry *dentry,
422 struct super_block *sb = parent->i_sb;
423 struct inode * inode = new_inode(sb);
424 struct dlmfs_inode_private *ip;
429 inode->i_ino = get_next_ino();
430 inode->i_mode = mode;
431 inode->i_uid = current_fsuid();
432 inode->i_gid = current_fsgid();
433 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
434 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
437 ip->ip_conn = DLMFS_I(parent)->ip_conn;
439 switch (mode & S_IFMT) {
441 /* for now we don't support anything other than
442 * directories and regular files. */
446 inode->i_op = &dlmfs_file_inode_operations;
447 inode->i_fop = &dlmfs_file_operations;
449 i_size_write(inode, DLM_LVB_LEN);
451 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
453 /* released at clear_inode time, this insures that we
454 * get to drop the dlm reference on each lock *before*
455 * we call the unregister code for releasing parent
457 ip->ip_parent = igrab(parent);
458 BUG_ON(!ip->ip_parent);
461 inode->i_op = &dlmfs_dir_inode_operations;
462 inode->i_fop = &simple_dir_operations;
464 /* directory inodes start off with i_nlink ==
465 * 2 (for "." entry) */
470 if (parent->i_mode & S_ISGID) {
471 inode->i_gid = parent->i_gid;
473 inode->i_mode |= S_ISGID;
480 * File creation. Allocate an inode, and we're done..
483 static int dlmfs_mkdir(struct inode * dir,
484 struct dentry * dentry,
488 struct inode *inode = NULL;
489 struct qstr *domain = &dentry->d_name;
490 struct dlmfs_inode_private *ip;
491 struct ocfs2_cluster_connection *conn;
493 mlog(0, "mkdir %.*s\n", domain->len, domain->name);
495 /* verify that we have a proper domain */
496 if (domain->len >= GROUP_NAME_MAX) {
498 mlog(ML_ERROR, "invalid domain name for directory.\n");
502 inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
511 conn = user_dlm_register(domain);
513 status = PTR_ERR(conn);
514 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
515 status, domain->len, domain->name);
521 d_instantiate(dentry, inode);
522 dget(dentry); /* Extra count - pin the dentry in core */
531 static int dlmfs_create(struct inode *dir,
532 struct dentry *dentry,
534 struct nameidata *nd)
538 struct qstr *name = &dentry->d_name;
540 mlog(0, "create %.*s\n", name->len, name->name);
542 /* verify name is valid and doesn't contain any dlm reserved
544 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
545 name->name[0] == '$') {
547 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
552 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
559 d_instantiate(dentry, inode);
560 dget(dentry); /* Extra count - pin the dentry in core */
565 static int dlmfs_unlink(struct inode *dir,
566 struct dentry *dentry)
569 struct inode *inode = dentry->d_inode;
571 mlog(0, "unlink inode %lu\n", inode->i_ino);
573 /* if there are no current holders, or none that are waiting
574 * to acquire a lock, this basically destroys our lockres. */
575 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
577 mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
578 dentry->d_name.len, dentry->d_name.name, status);
581 status = simple_unlink(dir, dentry);
586 static int dlmfs_fill_super(struct super_block * sb,
590 struct inode * inode;
591 struct dentry * root;
593 sb->s_maxbytes = MAX_LFS_FILESIZE;
594 sb->s_blocksize = PAGE_CACHE_SIZE;
595 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
596 sb->s_magic = DLMFS_MAGIC;
597 sb->s_op = &dlmfs_ops;
598 inode = dlmfs_get_root_inode(sb);
602 root = d_alloc_root(inode);
611 static const struct file_operations dlmfs_file_operations = {
612 .open = dlmfs_file_open,
613 .release = dlmfs_file_release,
614 .poll = dlmfs_file_poll,
615 .read = dlmfs_file_read,
616 .write = dlmfs_file_write,
617 .llseek = default_llseek,
620 static const struct inode_operations dlmfs_dir_inode_operations = {
621 .create = dlmfs_create,
622 .lookup = simple_lookup,
623 .unlink = dlmfs_unlink,
626 /* this way we can restrict mkdir to only the toplevel of the fs. */
627 static const struct inode_operations dlmfs_root_inode_operations = {
628 .lookup = simple_lookup,
629 .mkdir = dlmfs_mkdir,
630 .rmdir = simple_rmdir,
633 static const struct super_operations dlmfs_ops = {
634 .statfs = simple_statfs,
635 .alloc_inode = dlmfs_alloc_inode,
636 .destroy_inode = dlmfs_destroy_inode,
637 .evict_inode = dlmfs_evict_inode,
638 .drop_inode = generic_delete_inode,
641 static const struct inode_operations dlmfs_file_inode_operations = {
642 .getattr = simple_getattr,
643 .setattr = dlmfs_file_setattr,
646 static struct dentry *dlmfs_mount(struct file_system_type *fs_type,
647 int flags, const char *dev_name, void *data)
649 return mount_nodev(fs_type, flags, data, dlmfs_fill_super);
652 static struct file_system_type dlmfs_fs_type = {
653 .owner = THIS_MODULE,
654 .name = "ocfs2_dlmfs",
655 .mount = dlmfs_mount,
656 .kill_sb = kill_litter_super,
659 static int __init init_dlmfs_fs(void)
662 int cleanup_inode = 0, cleanup_worker = 0;
664 dlmfs_print_version();
666 status = bdi_init(&dlmfs_backing_dev_info);
670 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
671 sizeof(struct dlmfs_inode_private),
672 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
675 if (!dlmfs_inode_cache) {
681 user_dlm_worker = create_singlethread_workqueue("user_dlm");
682 if (!user_dlm_worker) {
688 user_dlm_set_locking_protocol();
689 status = register_filesystem(&dlmfs_fs_type);
693 kmem_cache_destroy(dlmfs_inode_cache);
695 destroy_workqueue(user_dlm_worker);
696 bdi_destroy(&dlmfs_backing_dev_info);
698 printk("OCFS2 User DLM kernel interface loaded\n");
702 static void __exit exit_dlmfs_fs(void)
704 unregister_filesystem(&dlmfs_fs_type);
706 flush_workqueue(user_dlm_worker);
707 destroy_workqueue(user_dlm_worker);
709 kmem_cache_destroy(dlmfs_inode_cache);
711 bdi_destroy(&dlmfs_backing_dev_info);
714 MODULE_AUTHOR("Oracle");
715 MODULE_LICENSE("GPL");
717 module_init(init_dlmfs_fs)
718 module_exit(exit_dlmfs_fs)