1 // SPDX-License-Identifier: GPL-2.0-or-later
5 * Code which implements the kernel side of a minimal userspace
6 * interface to our DLM. This file handles the virtual file system
7 * used for communication with userspace. Credit should go to ramfs,
8 * which was a template for the fs side of this module.
10 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
13 /* Simple VFS hooks based on: */
15 * Resizable simple ram filesystem for Linux.
17 * Copyright (C) 2000 Linus Torvalds.
18 * 2000 Transmeta Corp.
21 #include <linux/module.h>
23 #include <linux/pagemap.h>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/highmem.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/poll.h>
32 #include <linux/uaccess.h>
34 #include "../stackglue.h"
37 #define MLOG_MASK_PREFIX ML_DLMFS
38 #include "../cluster/masklog.h"
41 static const struct super_operations dlmfs_ops;
42 static const struct file_operations dlmfs_file_operations;
43 static const struct inode_operations dlmfs_dir_inode_operations;
44 static const struct inode_operations dlmfs_root_inode_operations;
45 static const struct inode_operations dlmfs_file_inode_operations;
46 static struct kmem_cache *dlmfs_inode_cache;
48 struct workqueue_struct *user_dlm_worker;
53 * These are the ABI capabilities of dlmfs.
55 * Over time, dlmfs has added some features that were not part of the
56 * initial ABI. Unfortunately, some of these features are not detectable
57 * via standard usage. For example, Linux's default poll always returns
58 * EPOLLIN, so there is no way for a caller of poll(2) to know when dlmfs
59 * added poll support. Instead, we provide this list of new capabilities.
61 * Capabilities is a read-only attribute. We do it as a module parameter
62 * so we can discover it whether dlmfs is built in, loaded, or even not
65 * The ABI features are local to this machine's dlmfs mount. This is
66 * distinct from the locking protocol, which is concerned with inter-node
70 * - bast : EPOLLIN against the file descriptor of a held lock
71 * signifies a bast fired on the lock.
73 #define DLMFS_CAPABILITIES "bast stackglue"
74 static int param_set_dlmfs_capabilities(const char *val,
75 const struct kernel_param *kp)
77 printk(KERN_ERR "%s: readonly parameter\n", kp->name);
80 static int param_get_dlmfs_capabilities(char *buffer,
81 const struct kernel_param *kp)
83 return strlcpy(buffer, DLMFS_CAPABILITIES,
84 strlen(DLMFS_CAPABILITIES) + 1);
86 module_param_call(capabilities, param_set_dlmfs_capabilities,
87 param_get_dlmfs_capabilities, NULL, 0444);
88 MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES);
92 * decodes a set of open flags into a valid lock level and a set of flags.
93 * returns < 0 if we have invalid flags
94 * flags which mean something to us:
95 * O_RDONLY -> PRMODE level
96 * O_WRONLY -> EXMODE level
98 * O_NONBLOCK -> NOQUEUE
100 static int dlmfs_decode_open_flags(int open_flags,
104 if (open_flags & (O_WRONLY|O_RDWR))
105 *level = DLM_LOCK_EX;
107 *level = DLM_LOCK_PR;
110 if (open_flags & O_NONBLOCK)
111 *flags |= DLM_LKF_NOQUEUE;
116 static int dlmfs_file_open(struct inode *inode,
119 int status, level, flags;
120 struct dlmfs_filp_private *fp = NULL;
121 struct dlmfs_inode_private *ip;
123 if (S_ISDIR(inode->i_mode))
126 mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
129 status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
133 /* We don't want to honor O_APPEND at read/write time as it
134 * doesn't make sense for LVB writes. */
135 file->f_flags &= ~O_APPEND;
137 fp = kmalloc(sizeof(*fp), GFP_NOFS);
142 fp->fp_lock_level = level;
146 status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
148 /* this is a strange error to return here but I want
149 * to be able userspace to be able to distinguish a
150 * valid lock request from one that simply couldn't be
152 if (flags & DLM_LKF_NOQUEUE && status == -EAGAIN)
158 file->private_data = fp;
163 static int dlmfs_file_release(struct inode *inode,
167 struct dlmfs_inode_private *ip = DLMFS_I(inode);
168 struct dlmfs_filp_private *fp = file->private_data;
170 if (S_ISDIR(inode->i_mode))
173 mlog(0, "close called on inode %lu\n", inode->i_ino);
176 level = fp->fp_lock_level;
177 if (level != DLM_LOCK_IV)
178 user_dlm_cluster_unlock(&ip->ip_lockres, level);
181 file->private_data = NULL;
188 * We do ->setattr() just to override size changes. Our size is the size
189 * of the LVB and nothing else.
191 static int dlmfs_file_setattr(struct user_namespace *mnt_userns,
192 struct dentry *dentry, struct iattr *attr)
195 struct inode *inode = d_inode(dentry);
197 attr->ia_valid &= ~ATTR_SIZE;
198 error = setattr_prepare(&init_user_ns, dentry, attr);
202 setattr_copy(&init_user_ns, inode, attr);
203 mark_inode_dirty(inode);
207 static __poll_t dlmfs_file_poll(struct file *file, poll_table *wait)
210 struct inode *inode = file_inode(file);
211 struct dlmfs_inode_private *ip = DLMFS_I(inode);
213 poll_wait(file, &ip->ip_lockres.l_event, wait);
215 spin_lock(&ip->ip_lockres.l_lock);
216 if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED)
217 event = EPOLLIN | EPOLLRDNORM;
218 spin_unlock(&ip->ip_lockres.l_lock);
223 static ssize_t dlmfs_file_read(struct file *file,
228 char lvb[DLM_LVB_LEN];
230 if (!user_dlm_read_lvb(file_inode(file), lvb))
233 return simple_read_from_buffer(buf, count, ppos, lvb, sizeof(lvb));
236 static ssize_t dlmfs_file_write(struct file *filp,
237 const char __user *buf,
241 char lvb_buf[DLM_LVB_LEN];
243 struct inode *inode = file_inode(filp);
245 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
246 inode->i_ino, count, *ppos);
248 if (*ppos >= DLM_LVB_LEN)
251 /* don't write past the lvb */
252 if (count > DLM_LVB_LEN - *ppos)
253 count = DLM_LVB_LEN - *ppos;
258 bytes_left = copy_from_user(lvb_buf, buf, count);
261 user_dlm_write_lvb(inode, lvb_buf, count);
263 *ppos = *ppos + count;
264 mlog(0, "wrote %zu bytes\n", count);
268 static void dlmfs_init_once(void *foo)
270 struct dlmfs_inode_private *ip =
271 (struct dlmfs_inode_private *) foo;
274 ip->ip_parent = NULL;
276 inode_init_once(&ip->ip_vfs_inode);
279 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
281 struct dlmfs_inode_private *ip;
283 ip = alloc_inode_sb(sb, dlmfs_inode_cache, GFP_NOFS);
287 return &ip->ip_vfs_inode;
290 static void dlmfs_free_inode(struct inode *inode)
292 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
295 static void dlmfs_evict_inode(struct inode *inode)
298 struct dlmfs_inode_private *ip;
302 mlog(0, "inode %lu\n", inode->i_ino);
306 if (S_ISREG(inode->i_mode)) {
307 status = user_dlm_destroy_lock(&ip->ip_lockres);
314 mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn);
315 /* we must be a directory. If required, lets unregister the
316 * dlm context now. */
318 user_dlm_unregister(ip->ip_conn);
320 ip->ip_parent = NULL;
324 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
326 struct inode *inode = new_inode(sb);
327 umode_t mode = S_IFDIR | 0755;
330 inode->i_ino = get_next_ino();
331 inode_init_owner(&init_user_ns, inode, NULL, mode);
332 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
335 inode->i_fop = &simple_dir_operations;
336 inode->i_op = &dlmfs_root_inode_operations;
342 static struct inode *dlmfs_get_inode(struct inode *parent,
343 struct dentry *dentry,
346 struct super_block *sb = parent->i_sb;
347 struct inode * inode = new_inode(sb);
348 struct dlmfs_inode_private *ip;
353 inode->i_ino = get_next_ino();
354 inode_init_owner(&init_user_ns, inode, parent, mode);
355 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
358 ip->ip_conn = DLMFS_I(parent)->ip_conn;
360 switch (mode & S_IFMT) {
362 /* for now we don't support anything other than
363 * directories and regular files. */
367 inode->i_op = &dlmfs_file_inode_operations;
368 inode->i_fop = &dlmfs_file_operations;
370 i_size_write(inode, DLM_LVB_LEN);
372 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
374 /* released at clear_inode time, this insures that we
375 * get to drop the dlm reference on each lock *before*
376 * we call the unregister code for releasing parent
378 ip->ip_parent = igrab(parent);
379 BUG_ON(!ip->ip_parent);
382 inode->i_op = &dlmfs_dir_inode_operations;
383 inode->i_fop = &simple_dir_operations;
385 /* directory inodes start off with i_nlink ==
386 * 2 (for "." entry) */
394 * File creation. Allocate an inode, and we're done..
397 static int dlmfs_mkdir(struct user_namespace * mnt_userns,
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 user_namespace *mnt_userns,
448 struct dentry *dentry,
454 const struct qstr *name = &dentry->d_name;
456 mlog(0, "create %.*s\n", name->len, name->name);
458 /* verify name is valid and doesn't contain any dlm reserved
460 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
461 name->name[0] == '$') {
463 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
468 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
475 d_instantiate(dentry, inode);
476 dget(dentry); /* Extra count - pin the dentry in core */
481 static int dlmfs_unlink(struct inode *dir,
482 struct dentry *dentry)
485 struct inode *inode = d_inode(dentry);
487 mlog(0, "unlink inode %lu\n", inode->i_ino);
489 /* if there are no current holders, or none that are waiting
490 * to acquire a lock, this basically destroys our lockres. */
491 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
493 mlog(ML_ERROR, "unlink %pd, error %d from destroy\n",
497 status = simple_unlink(dir, dentry);
502 static int dlmfs_fill_super(struct super_block * sb,
506 sb->s_maxbytes = MAX_LFS_FILESIZE;
507 sb->s_blocksize = PAGE_SIZE;
508 sb->s_blocksize_bits = PAGE_SHIFT;
509 sb->s_magic = DLMFS_MAGIC;
510 sb->s_op = &dlmfs_ops;
511 sb->s_root = d_make_root(dlmfs_get_root_inode(sb));
517 static const struct file_operations dlmfs_file_operations = {
518 .open = dlmfs_file_open,
519 .release = dlmfs_file_release,
520 .poll = dlmfs_file_poll,
521 .read = dlmfs_file_read,
522 .write = dlmfs_file_write,
523 .llseek = default_llseek,
526 static const struct inode_operations dlmfs_dir_inode_operations = {
527 .create = dlmfs_create,
528 .lookup = simple_lookup,
529 .unlink = dlmfs_unlink,
532 /* this way we can restrict mkdir to only the toplevel of the fs. */
533 static const struct inode_operations dlmfs_root_inode_operations = {
534 .lookup = simple_lookup,
535 .mkdir = dlmfs_mkdir,
536 .rmdir = simple_rmdir,
539 static const struct super_operations dlmfs_ops = {
540 .statfs = simple_statfs,
541 .alloc_inode = dlmfs_alloc_inode,
542 .free_inode = dlmfs_free_inode,
543 .evict_inode = dlmfs_evict_inode,
544 .drop_inode = generic_delete_inode,
547 static const struct inode_operations dlmfs_file_inode_operations = {
548 .getattr = simple_getattr,
549 .setattr = dlmfs_file_setattr,
552 static struct dentry *dlmfs_mount(struct file_system_type *fs_type,
553 int flags, const char *dev_name, void *data)
555 return mount_nodev(fs_type, flags, data, dlmfs_fill_super);
558 static struct file_system_type dlmfs_fs_type = {
559 .owner = THIS_MODULE,
560 .name = "ocfs2_dlmfs",
561 .mount = dlmfs_mount,
562 .kill_sb = kill_litter_super,
564 MODULE_ALIAS_FS("ocfs2_dlmfs");
566 static int __init init_dlmfs_fs(void)
569 int cleanup_inode = 0, cleanup_worker = 0;
571 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
572 sizeof(struct dlmfs_inode_private),
573 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
574 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
576 if (!dlmfs_inode_cache) {
582 user_dlm_worker = alloc_workqueue("user_dlm", WQ_MEM_RECLAIM, 0);
583 if (!user_dlm_worker) {
589 user_dlm_set_locking_protocol();
590 status = register_filesystem(&dlmfs_fs_type);
594 kmem_cache_destroy(dlmfs_inode_cache);
596 destroy_workqueue(user_dlm_worker);
598 printk("OCFS2 User DLM kernel interface loaded\n");
602 static void __exit exit_dlmfs_fs(void)
604 unregister_filesystem(&dlmfs_fs_type);
606 destroy_workqueue(user_dlm_worker);
609 * Make sure all delayed rcu free inodes are flushed before we
613 kmem_cache_destroy(dlmfs_inode_cache);
617 MODULE_AUTHOR("Oracle");
618 MODULE_LICENSE("GPL");
619 MODULE_DESCRIPTION("OCFS2 DLM-Filesystem");
621 module_init(init_dlmfs_fs)
622 module_exit(exit_dlmfs_fs)