1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
7 * Code which implements the kernel side of a minimal userspace
8 * interface to our DLM. This file handles the virtual file system
9 * used for communication with userspace. Credit should go to ramfs,
10 * which was a template for the fs side of this module.
12 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
15 /* Simple VFS hooks based on: */
17 * Resizable simple ram filesystem for Linux.
19 * Copyright (C) 2000 Linus Torvalds.
20 * 2000 Transmeta Corp.
23 #include <linux/module.h>
25 #include <linux/pagemap.h>
26 #include <linux/types.h>
27 #include <linux/slab.h>
28 #include <linux/highmem.h>
29 #include <linux/init.h>
30 #include <linux/string.h>
31 #include <linux/backing-dev.h>
32 #include <linux/poll.h>
34 #include <linux/uaccess.h>
36 #include "../stackglue.h"
39 #define MLOG_MASK_PREFIX ML_DLMFS
40 #include "../cluster/masklog.h"
43 static const struct super_operations dlmfs_ops;
44 static const struct file_operations dlmfs_file_operations;
45 static const struct inode_operations dlmfs_dir_inode_operations;
46 static const struct inode_operations dlmfs_root_inode_operations;
47 static const struct inode_operations dlmfs_file_inode_operations;
48 static struct kmem_cache *dlmfs_inode_cache;
50 struct workqueue_struct *user_dlm_worker;
55 * These are the ABI capabilities of dlmfs.
57 * Over time, dlmfs has added some features that were not part of the
58 * initial ABI. Unfortunately, some of these features are not detectable
59 * via standard usage. For example, Linux's default poll always returns
60 * EPOLLIN, so there is no way for a caller of poll(2) to know when dlmfs
61 * added poll support. Instead, we provide this list of new capabilities.
63 * Capabilities is a read-only attribute. We do it as a module parameter
64 * so we can discover it whether dlmfs is built in, loaded, or even not
67 * The ABI features are local to this machine's dlmfs mount. This is
68 * distinct from the locking protocol, which is concerned with inter-node
72 * - bast : EPOLLIN against the file descriptor of a held lock
73 * signifies a bast fired on the lock.
75 #define DLMFS_CAPABILITIES "bast stackglue"
76 static int param_set_dlmfs_capabilities(const char *val,
77 const struct kernel_param *kp)
79 printk(KERN_ERR "%s: readonly parameter\n", kp->name);
82 static int param_get_dlmfs_capabilities(char *buffer,
83 const struct kernel_param *kp)
85 return strlcpy(buffer, DLMFS_CAPABILITIES,
86 strlen(DLMFS_CAPABILITIES) + 1);
88 module_param_call(capabilities, param_set_dlmfs_capabilities,
89 param_get_dlmfs_capabilities, NULL, 0444);
90 MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES);
94 * decodes a set of open flags into a valid lock level and a set of flags.
95 * returns < 0 if we have invalid flags
96 * flags which mean something to us:
97 * O_RDONLY -> PRMODE level
98 * O_WRONLY -> EXMODE level
100 * O_NONBLOCK -> NOQUEUE
102 static int dlmfs_decode_open_flags(int open_flags,
106 if (open_flags & (O_WRONLY|O_RDWR))
107 *level = DLM_LOCK_EX;
109 *level = DLM_LOCK_PR;
112 if (open_flags & O_NONBLOCK)
113 *flags |= DLM_LKF_NOQUEUE;
118 static int dlmfs_file_open(struct inode *inode,
121 int status, level, flags;
122 struct dlmfs_filp_private *fp = NULL;
123 struct dlmfs_inode_private *ip;
125 if (S_ISDIR(inode->i_mode))
128 mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
131 status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
135 /* We don't want to honor O_APPEND at read/write time as it
136 * doesn't make sense for LVB writes. */
137 file->f_flags &= ~O_APPEND;
139 fp = kmalloc(sizeof(*fp), GFP_NOFS);
144 fp->fp_lock_level = level;
148 status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
150 /* this is a strange error to return here but I want
151 * to be able userspace to be able to distinguish a
152 * valid lock request from one that simply couldn't be
154 if (flags & DLM_LKF_NOQUEUE && status == -EAGAIN)
160 file->private_data = fp;
165 static int dlmfs_file_release(struct inode *inode,
169 struct dlmfs_inode_private *ip = DLMFS_I(inode);
170 struct dlmfs_filp_private *fp = file->private_data;
172 if (S_ISDIR(inode->i_mode))
175 mlog(0, "close called on inode %lu\n", inode->i_ino);
178 level = fp->fp_lock_level;
179 if (level != DLM_LOCK_IV)
180 user_dlm_cluster_unlock(&ip->ip_lockres, level);
183 file->private_data = NULL;
190 * We do ->setattr() just to override size changes. Our size is the size
191 * of the LVB and nothing else.
193 static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr)
196 struct inode *inode = d_inode(dentry);
198 attr->ia_valid &= ~ATTR_SIZE;
199 error = setattr_prepare(dentry, attr);
203 setattr_copy(inode, attr);
204 mark_inode_dirty(inode);
208 static __poll_t dlmfs_file_poll(struct file *file, poll_table *wait)
211 struct inode *inode = file_inode(file);
212 struct dlmfs_inode_private *ip = DLMFS_I(inode);
214 poll_wait(file, &ip->ip_lockres.l_event, wait);
216 spin_lock(&ip->ip_lockres.l_lock);
217 if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED)
218 event = EPOLLIN | EPOLLRDNORM;
219 spin_unlock(&ip->ip_lockres.l_lock);
224 static ssize_t dlmfs_file_read(struct file *file,
229 char lvb[DLM_LVB_LEN];
231 if (!user_dlm_read_lvb(file_inode(file), lvb))
234 return simple_read_from_buffer(buf, count, ppos, lvb, sizeof(lvb));
237 static ssize_t dlmfs_file_write(struct file *filp,
238 const char __user *buf,
242 char lvb_buf[DLM_LVB_LEN];
244 struct inode *inode = file_inode(filp);
246 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
247 inode->i_ino, count, *ppos);
249 if (*ppos >= DLM_LVB_LEN)
252 /* don't write past the lvb */
253 if (count > DLM_LVB_LEN - *ppos)
254 count = DLM_LVB_LEN - *ppos;
259 bytes_left = copy_from_user(lvb_buf, buf, count);
262 user_dlm_write_lvb(inode, lvb_buf, count);
264 *ppos = *ppos + count;
265 mlog(0, "wrote %zu bytes\n", count);
269 static void dlmfs_init_once(void *foo)
271 struct dlmfs_inode_private *ip =
272 (struct dlmfs_inode_private *) foo;
275 ip->ip_parent = NULL;
277 inode_init_once(&ip->ip_vfs_inode);
280 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
282 struct dlmfs_inode_private *ip;
284 ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
288 return &ip->ip_vfs_inode;
291 static void dlmfs_free_inode(struct inode *inode)
293 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
296 static void dlmfs_evict_inode(struct inode *inode)
299 struct dlmfs_inode_private *ip;
303 mlog(0, "inode %lu\n", inode->i_ino);
307 if (S_ISREG(inode->i_mode)) {
308 status = user_dlm_destroy_lock(&ip->ip_lockres);
315 mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn);
316 /* we must be a directory. If required, lets unregister the
317 * dlm context now. */
319 user_dlm_unregister(ip->ip_conn);
321 ip->ip_parent = NULL;
325 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
327 struct inode *inode = new_inode(sb);
328 umode_t mode = S_IFDIR | 0755;
331 inode->i_ino = get_next_ino();
332 inode_init_owner(inode, NULL, mode);
333 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
336 inode->i_fop = &simple_dir_operations;
337 inode->i_op = &dlmfs_root_inode_operations;
343 static struct inode *dlmfs_get_inode(struct inode *parent,
344 struct dentry *dentry,
347 struct super_block *sb = parent->i_sb;
348 struct inode * inode = new_inode(sb);
349 struct dlmfs_inode_private *ip;
354 inode->i_ino = get_next_ino();
355 inode_init_owner(inode, parent, mode);
356 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
359 ip->ip_conn = DLMFS_I(parent)->ip_conn;
361 switch (mode & S_IFMT) {
363 /* for now we don't support anything other than
364 * directories and regular files. */
368 inode->i_op = &dlmfs_file_inode_operations;
369 inode->i_fop = &dlmfs_file_operations;
371 i_size_write(inode, DLM_LVB_LEN);
373 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
375 /* released at clear_inode time, this insures that we
376 * get to drop the dlm reference on each lock *before*
377 * we call the unregister code for releasing parent
379 ip->ip_parent = igrab(parent);
380 BUG_ON(!ip->ip_parent);
383 inode->i_op = &dlmfs_dir_inode_operations;
384 inode->i_fop = &simple_dir_operations;
386 /* directory inodes start off with i_nlink ==
387 * 2 (for "." entry) */
395 * File creation. Allocate an inode, and we're done..
398 static int dlmfs_mkdir(struct inode * dir,
399 struct dentry * dentry,
403 struct inode *inode = NULL;
404 const struct qstr *domain = &dentry->d_name;
405 struct dlmfs_inode_private *ip;
406 struct ocfs2_cluster_connection *conn;
408 mlog(0, "mkdir %.*s\n", domain->len, domain->name);
410 /* verify that we have a proper domain */
411 if (domain->len >= GROUP_NAME_MAX) {
413 mlog(ML_ERROR, "invalid domain name for directory.\n");
417 inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
426 conn = user_dlm_register(domain);
428 status = PTR_ERR(conn);
429 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
430 status, domain->len, domain->name);
436 d_instantiate(dentry, inode);
437 dget(dentry); /* Extra count - pin the dentry in core */
446 static int dlmfs_create(struct inode *dir,
447 struct dentry *dentry,
453 const struct qstr *name = &dentry->d_name;
455 mlog(0, "create %.*s\n", name->len, name->name);
457 /* verify name is valid and doesn't contain any dlm reserved
459 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
460 name->name[0] == '$') {
462 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
467 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
474 d_instantiate(dentry, inode);
475 dget(dentry); /* Extra count - pin the dentry in core */
480 static int dlmfs_unlink(struct inode *dir,
481 struct dentry *dentry)
484 struct inode *inode = d_inode(dentry);
486 mlog(0, "unlink inode %lu\n", inode->i_ino);
488 /* if there are no current holders, or none that are waiting
489 * to acquire a lock, this basically destroys our lockres. */
490 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
492 mlog(ML_ERROR, "unlink %pd, error %d from destroy\n",
496 status = simple_unlink(dir, dentry);
501 static int dlmfs_fill_super(struct super_block * sb,
505 sb->s_maxbytes = MAX_LFS_FILESIZE;
506 sb->s_blocksize = PAGE_SIZE;
507 sb->s_blocksize_bits = PAGE_SHIFT;
508 sb->s_magic = DLMFS_MAGIC;
509 sb->s_op = &dlmfs_ops;
510 sb->s_root = d_make_root(dlmfs_get_root_inode(sb));
516 static const struct file_operations dlmfs_file_operations = {
517 .open = dlmfs_file_open,
518 .release = dlmfs_file_release,
519 .poll = dlmfs_file_poll,
520 .read = dlmfs_file_read,
521 .write = dlmfs_file_write,
522 .llseek = default_llseek,
525 static const struct inode_operations dlmfs_dir_inode_operations = {
526 .create = dlmfs_create,
527 .lookup = simple_lookup,
528 .unlink = dlmfs_unlink,
531 /* this way we can restrict mkdir to only the toplevel of the fs. */
532 static const struct inode_operations dlmfs_root_inode_operations = {
533 .lookup = simple_lookup,
534 .mkdir = dlmfs_mkdir,
535 .rmdir = simple_rmdir,
538 static const struct super_operations dlmfs_ops = {
539 .statfs = simple_statfs,
540 .alloc_inode = dlmfs_alloc_inode,
541 .free_inode = dlmfs_free_inode,
542 .evict_inode = dlmfs_evict_inode,
543 .drop_inode = generic_delete_inode,
546 static const struct inode_operations dlmfs_file_inode_operations = {
547 .getattr = simple_getattr,
548 .setattr = dlmfs_file_setattr,
551 static struct dentry *dlmfs_mount(struct file_system_type *fs_type,
552 int flags, const char *dev_name, void *data)
554 return mount_nodev(fs_type, flags, data, dlmfs_fill_super);
557 static struct file_system_type dlmfs_fs_type = {
558 .owner = THIS_MODULE,
559 .name = "ocfs2_dlmfs",
560 .mount = dlmfs_mount,
561 .kill_sb = kill_litter_super,
563 MODULE_ALIAS_FS("ocfs2_dlmfs");
565 static int __init init_dlmfs_fs(void)
568 int cleanup_inode = 0, cleanup_worker = 0;
570 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
571 sizeof(struct dlmfs_inode_private),
572 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
573 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
575 if (!dlmfs_inode_cache) {
581 user_dlm_worker = alloc_workqueue("user_dlm", WQ_MEM_RECLAIM, 0);
582 if (!user_dlm_worker) {
588 user_dlm_set_locking_protocol();
589 status = register_filesystem(&dlmfs_fs_type);
593 kmem_cache_destroy(dlmfs_inode_cache);
595 destroy_workqueue(user_dlm_worker);
597 printk("OCFS2 User DLM kernel interface loaded\n");
601 static void __exit exit_dlmfs_fs(void)
603 unregister_filesystem(&dlmfs_fs_type);
605 destroy_workqueue(user_dlm_worker);
608 * Make sure all delayed rcu free inodes are flushed before we
612 kmem_cache_destroy(dlmfs_inode_cache);
616 MODULE_AUTHOR("Oracle");
617 MODULE_LICENSE("GPL");
618 MODULE_DESCRIPTION("OCFS2 DLM-Filesystem");
620 module_init(init_dlmfs_fs)
621 module_exit(exit_dlmfs_fs)