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>
51 #include "cluster/nodemanager.h"
52 #include "cluster/heartbeat.h"
53 #include "cluster/tcp.h"
55 #include "dlm/dlmapi.h"
61 #define MLOG_MASK_PREFIX ML_DLMFS
62 #include "cluster/masklog.h"
64 #include "ocfs2_lockingver.h"
66 static const struct super_operations dlmfs_ops;
67 static const struct file_operations dlmfs_file_operations;
68 static const struct inode_operations dlmfs_dir_inode_operations;
69 static const struct inode_operations dlmfs_root_inode_operations;
70 static const struct inode_operations dlmfs_file_inode_operations;
71 static struct kmem_cache *dlmfs_inode_cache;
73 struct workqueue_struct *user_dlm_worker;
76 * This is the userdlmfs locking protocol version.
78 * See fs/ocfs2/dlmglue.c for more details on locking versions.
80 static const struct dlm_protocol_version user_locking_protocol = {
81 .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
82 .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
87 * These are the ABI capabilities of dlmfs.
89 * Over time, dlmfs has added some features that were not part of the
90 * initial ABI. Unfortunately, some of these features are not detectable
91 * via standard usage. For example, Linux's default poll always returns
92 * POLLIN, so there is no way for a caller of poll(2) to know when dlmfs
93 * added poll support. Instead, we provide this list of new capabilities.
95 * Capabilities is a read-only attribute. We do it as a module parameter
96 * so we can discover it whether dlmfs is built in, loaded, or even not
99 * The ABI features are local to this machine's dlmfs mount. This is
100 * distinct from the locking protocol, which is concerned with inter-node
104 * - bast : POLLIN against the file descriptor of a held lock
105 * signifies a bast fired on the lock.
107 #define DLMFS_CAPABILITIES "bast"
108 extern int param_set_dlmfs_capabilities(const char *val,
109 struct kernel_param *kp)
111 printk(KERN_ERR "%s: readonly parameter\n", kp->name);
114 static int param_get_dlmfs_capabilities(char *buffer,
115 struct kernel_param *kp)
117 return strlcpy(buffer, DLMFS_CAPABILITIES,
118 strlen(DLMFS_CAPABILITIES) + 1);
120 module_param_call(capabilities, param_set_dlmfs_capabilities,
121 param_get_dlmfs_capabilities, NULL, 0444);
122 MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES);
126 * decodes a set of open flags into a valid lock level and a set of flags.
127 * returns < 0 if we have invalid flags
128 * flags which mean something to us:
129 * O_RDONLY -> PRMODE level
130 * O_WRONLY -> EXMODE level
132 * O_NONBLOCK -> LKM_NOQUEUE
134 static int dlmfs_decode_open_flags(int open_flags,
138 if (open_flags & (O_WRONLY|O_RDWR))
144 if (open_flags & O_NONBLOCK)
145 *flags |= LKM_NOQUEUE;
150 static int dlmfs_file_open(struct inode *inode,
153 int status, level, flags;
154 struct dlmfs_filp_private *fp = NULL;
155 struct dlmfs_inode_private *ip;
157 if (S_ISDIR(inode->i_mode))
160 mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
163 status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
167 /* We don't want to honor O_APPEND at read/write time as it
168 * doesn't make sense for LVB writes. */
169 file->f_flags &= ~O_APPEND;
171 fp = kmalloc(sizeof(*fp), GFP_NOFS);
176 fp->fp_lock_level = level;
180 status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
182 /* this is a strange error to return here but I want
183 * to be able userspace to be able to distinguish a
184 * valid lock request from one that simply couldn't be
186 if (flags & LKM_NOQUEUE && status == -EAGAIN)
192 file->private_data = fp;
197 static int dlmfs_file_release(struct inode *inode,
201 struct dlmfs_inode_private *ip = DLMFS_I(inode);
202 struct dlmfs_filp_private *fp =
203 (struct dlmfs_filp_private *) file->private_data;
205 if (S_ISDIR(inode->i_mode))
208 mlog(0, "close called on inode %lu\n", inode->i_ino);
212 level = fp->fp_lock_level;
213 if (level != LKM_IVMODE)
214 user_dlm_cluster_unlock(&ip->ip_lockres, level);
217 file->private_data = NULL;
224 * We do ->setattr() just to override size changes. Our size is the size
225 * of the LVB and nothing else.
227 static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr)
230 struct inode *inode = dentry->d_inode;
232 attr->ia_valid &= ~ATTR_SIZE;
233 error = inode_change_ok(inode, attr);
235 error = inode_setattr(inode, attr);
240 static unsigned int dlmfs_file_poll(struct file *file, poll_table *wait)
243 struct inode *inode = file->f_path.dentry->d_inode;
244 struct dlmfs_inode_private *ip = DLMFS_I(inode);
246 poll_wait(file, &ip->ip_lockres.l_event, wait);
248 spin_lock(&ip->ip_lockres.l_lock);
249 if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED)
250 event = POLLIN | POLLRDNORM;
251 spin_unlock(&ip->ip_lockres.l_lock);
256 static ssize_t dlmfs_file_read(struct file *filp,
264 struct inode *inode = filp->f_path.dentry->d_inode;
266 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
267 inode->i_ino, count, *ppos);
269 if (*ppos >= i_size_read(inode))
275 if (!access_ok(VERIFY_WRITE, buf, count))
278 /* don't read past the lvb */
279 if ((count + *ppos) > i_size_read(inode))
280 readlen = i_size_read(inode) - *ppos;
282 readlen = count - *ppos;
284 lvb_buf = kmalloc(readlen, GFP_NOFS);
288 user_dlm_read_lvb(inode, lvb_buf, readlen);
289 bytes_left = __copy_to_user(buf, lvb_buf, readlen);
290 readlen -= bytes_left;
294 *ppos = *ppos + readlen;
296 mlog(0, "read %zd bytes\n", readlen);
300 static ssize_t dlmfs_file_write(struct file *filp,
301 const char __user *buf,
308 struct inode *inode = filp->f_path.dentry->d_inode;
310 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
311 inode->i_ino, count, *ppos);
313 if (*ppos >= i_size_read(inode))
319 if (!access_ok(VERIFY_READ, buf, count))
322 /* don't write past the lvb */
323 if ((count + *ppos) > i_size_read(inode))
324 writelen = i_size_read(inode) - *ppos;
326 writelen = count - *ppos;
328 lvb_buf = kmalloc(writelen, GFP_NOFS);
332 bytes_left = copy_from_user(lvb_buf, buf, writelen);
333 writelen -= bytes_left;
335 user_dlm_write_lvb(inode, lvb_buf, writelen);
339 *ppos = *ppos + writelen;
340 mlog(0, "wrote %zd bytes\n", writelen);
344 static void dlmfs_init_once(void *foo)
346 struct dlmfs_inode_private *ip =
347 (struct dlmfs_inode_private *) foo;
350 ip->ip_parent = NULL;
352 inode_init_once(&ip->ip_vfs_inode);
355 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
357 struct dlmfs_inode_private *ip;
359 ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
363 return &ip->ip_vfs_inode;
366 static void dlmfs_destroy_inode(struct inode *inode)
368 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
371 static void dlmfs_clear_inode(struct inode *inode)
374 struct dlmfs_inode_private *ip;
379 mlog(0, "inode %lu\n", inode->i_ino);
383 if (S_ISREG(inode->i_mode)) {
384 status = user_dlm_destroy_lock(&ip->ip_lockres);
391 mlog(0, "we're a directory, ip->ip_dlm = 0x%p\n", ip->ip_dlm);
392 /* we must be a directory. If required, lets unregister the
393 * dlm context now. */
395 user_dlm_unregister_context(ip->ip_dlm);
397 ip->ip_parent = NULL;
401 static struct backing_dev_info dlmfs_backing_dev_info = {
402 .name = "ocfs2-dlmfs",
403 .ra_pages = 0, /* No readahead */
404 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
407 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
409 struct inode *inode = new_inode(sb);
410 int mode = S_IFDIR | 0755;
411 struct dlmfs_inode_private *ip;
416 inode->i_mode = mode;
417 inode->i_uid = current_fsuid();
418 inode->i_gid = current_fsgid();
419 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
420 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
423 inode->i_fop = &simple_dir_operations;
424 inode->i_op = &dlmfs_root_inode_operations;
430 static struct inode *dlmfs_get_inode(struct inode *parent,
431 struct dentry *dentry,
434 struct super_block *sb = parent->i_sb;
435 struct inode * inode = new_inode(sb);
436 struct dlmfs_inode_private *ip;
441 inode->i_mode = mode;
442 inode->i_uid = current_fsuid();
443 inode->i_gid = current_fsgid();
444 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
445 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
448 ip->ip_dlm = DLMFS_I(parent)->ip_dlm;
450 switch (mode & S_IFMT) {
452 /* for now we don't support anything other than
453 * directories and regular files. */
457 inode->i_op = &dlmfs_file_inode_operations;
458 inode->i_fop = &dlmfs_file_operations;
460 i_size_write(inode, DLM_LVB_LEN);
462 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
464 /* released at clear_inode time, this insures that we
465 * get to drop the dlm reference on each lock *before*
466 * we call the unregister code for releasing parent
468 ip->ip_parent = igrab(parent);
469 BUG_ON(!ip->ip_parent);
472 inode->i_op = &dlmfs_dir_inode_operations;
473 inode->i_fop = &simple_dir_operations;
475 /* directory inodes start off with i_nlink ==
476 * 2 (for "." entry) */
481 if (parent->i_mode & S_ISGID) {
482 inode->i_gid = parent->i_gid;
484 inode->i_mode |= S_ISGID;
491 * File creation. Allocate an inode, and we're done..
494 static int dlmfs_mkdir(struct inode * dir,
495 struct dentry * dentry,
499 struct inode *inode = NULL;
500 struct qstr *domain = &dentry->d_name;
501 struct dlmfs_inode_private *ip;
502 struct dlm_ctxt *dlm;
503 struct dlm_protocol_version proto = user_locking_protocol;
505 mlog(0, "mkdir %.*s\n", domain->len, domain->name);
507 /* verify that we have a proper domain */
508 if (domain->len >= O2NM_MAX_NAME_LEN) {
510 mlog(ML_ERROR, "invalid domain name for directory.\n");
514 inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
523 dlm = user_dlm_register_context(domain, &proto);
525 status = PTR_ERR(dlm);
526 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
527 status, domain->len, domain->name);
533 d_instantiate(dentry, inode);
534 dget(dentry); /* Extra count - pin the dentry in core */
543 static int dlmfs_create(struct inode *dir,
544 struct dentry *dentry,
546 struct nameidata *nd)
550 struct qstr *name = &dentry->d_name;
552 mlog(0, "create %.*s\n", name->len, name->name);
554 /* verify name is valid and doesn't contain any dlm reserved
556 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
557 name->name[0] == '$') {
559 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
564 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
571 d_instantiate(dentry, inode);
572 dget(dentry); /* Extra count - pin the dentry in core */
577 static int dlmfs_unlink(struct inode *dir,
578 struct dentry *dentry)
581 struct inode *inode = dentry->d_inode;
583 mlog(0, "unlink inode %lu\n", inode->i_ino);
585 /* if there are no current holders, or none that are waiting
586 * to acquire a lock, this basically destroys our lockres. */
587 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
589 mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
590 dentry->d_name.len, dentry->d_name.name, status);
593 status = simple_unlink(dir, dentry);
598 static int dlmfs_fill_super(struct super_block * sb,
602 struct inode * inode;
603 struct dentry * root;
605 sb->s_maxbytes = MAX_LFS_FILESIZE;
606 sb->s_blocksize = PAGE_CACHE_SIZE;
607 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
608 sb->s_magic = DLMFS_MAGIC;
609 sb->s_op = &dlmfs_ops;
610 inode = dlmfs_get_root_inode(sb);
614 root = d_alloc_root(inode);
623 static const struct file_operations dlmfs_file_operations = {
624 .open = dlmfs_file_open,
625 .release = dlmfs_file_release,
626 .poll = dlmfs_file_poll,
627 .read = dlmfs_file_read,
628 .write = dlmfs_file_write,
631 static const struct inode_operations dlmfs_dir_inode_operations = {
632 .create = dlmfs_create,
633 .lookup = simple_lookup,
634 .unlink = dlmfs_unlink,
637 /* this way we can restrict mkdir to only the toplevel of the fs. */
638 static const struct inode_operations dlmfs_root_inode_operations = {
639 .lookup = simple_lookup,
640 .mkdir = dlmfs_mkdir,
641 .rmdir = simple_rmdir,
644 static const struct super_operations dlmfs_ops = {
645 .statfs = simple_statfs,
646 .alloc_inode = dlmfs_alloc_inode,
647 .destroy_inode = dlmfs_destroy_inode,
648 .clear_inode = dlmfs_clear_inode,
649 .drop_inode = generic_delete_inode,
652 static const struct inode_operations dlmfs_file_inode_operations = {
653 .getattr = simple_getattr,
654 .setattr = dlmfs_file_setattr,
657 static int dlmfs_get_sb(struct file_system_type *fs_type,
658 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
660 return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super, mnt);
663 static struct file_system_type dlmfs_fs_type = {
664 .owner = THIS_MODULE,
665 .name = "ocfs2_dlmfs",
666 .get_sb = dlmfs_get_sb,
667 .kill_sb = kill_litter_super,
670 static int __init init_dlmfs_fs(void)
673 int cleanup_inode = 0, cleanup_worker = 0;
675 dlmfs_print_version();
677 status = bdi_init(&dlmfs_backing_dev_info);
681 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
682 sizeof(struct dlmfs_inode_private),
683 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
686 if (!dlmfs_inode_cache) {
692 user_dlm_worker = create_singlethread_workqueue("user_dlm");
693 if (!user_dlm_worker) {
699 status = register_filesystem(&dlmfs_fs_type);
703 kmem_cache_destroy(dlmfs_inode_cache);
705 destroy_workqueue(user_dlm_worker);
706 bdi_destroy(&dlmfs_backing_dev_info);
708 printk("OCFS2 User DLM kernel interface loaded\n");
712 static void __exit exit_dlmfs_fs(void)
714 unregister_filesystem(&dlmfs_fs_type);
716 flush_workqueue(user_dlm_worker);
717 destroy_workqueue(user_dlm_worker);
719 kmem_cache_destroy(dlmfs_inode_cache);
721 bdi_destroy(&dlmfs_backing_dev_info);
724 MODULE_AUTHOR("Oracle");
725 MODULE_LICENSE("GPL");
727 module_init(init_dlmfs_fs)
728 module_exit(exit_dlmfs_fs)