1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dlmfs.c
5 *
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.
10 *
11 * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
12 *
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.
17 *
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.
22 *
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.
27 */
28
29/* Simple VFS hooks based on: */
30/*
31 * Resizable simple ram filesystem for Linux.
32 *
33 * Copyright (C) 2000 Linus Torvalds.
34 *               2000 Transmeta Corp.
35 */
36
37#include <linux/module.h>
38#include <linux/fs.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
47#include <asm/uaccess.h>
48
49
50#include "cluster/nodemanager.h"
51#include "cluster/heartbeat.h"
52#include "cluster/tcp.h"
53
54#include "dlmapi.h"
55
56#include "userdlm.h"
57
58#include "dlmfsver.h"
59
60#define MLOG_MASK_PREFIX ML_DLMFS
61#include "cluster/masklog.h"
62
63static const struct super_operations dlmfs_ops;
64static const struct file_operations dlmfs_file_operations;
65static const struct inode_operations dlmfs_dir_inode_operations;
66static const struct inode_operations dlmfs_root_inode_operations;
67static const struct inode_operations dlmfs_file_inode_operations;
68static struct kmem_cache *dlmfs_inode_cache;
69
70struct workqueue_struct *user_dlm_worker;
71
72/*
73 * decodes a set of open flags into a valid lock level and a set of flags.
74 * returns < 0 if we have invalid flags
75 * flags which mean something to us:
76 * O_RDONLY -> PRMODE level
77 * O_WRONLY -> EXMODE level
78 *
79 * O_NONBLOCK -> LKM_NOQUEUE
80 */
81static int dlmfs_decode_open_flags(int open_flags,
82				   int *level,
83				   int *flags)
84{
85	if (open_flags & (O_WRONLY|O_RDWR))
86		*level = LKM_EXMODE;
87	else
88		*level = LKM_PRMODE;
89
90	*flags = 0;
91	if (open_flags & O_NONBLOCK)
92		*flags |= LKM_NOQUEUE;
93
94	return 0;
95}
96
97static int dlmfs_file_open(struct inode *inode,
98			   struct file *file)
99{
100	int status, level, flags;
101	struct dlmfs_filp_private *fp = NULL;
102	struct dlmfs_inode_private *ip;
103
104	if (S_ISDIR(inode->i_mode))
105		BUG();
106
107	mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
108		file->f_flags);
109
110	status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
111	if (status < 0)
112		goto bail;
113
114	/* We don't want to honor O_APPEND at read/write time as it
115	 * doesn't make sense for LVB writes. */
116	file->f_flags &= ~O_APPEND;
117
118	fp = kmalloc(sizeof(*fp), GFP_NOFS);
119	if (!fp) {
120		status = -ENOMEM;
121		goto bail;
122	}
123	fp->fp_lock_level = level;
124
125	ip = DLMFS_I(inode);
126
127	status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
128	if (status < 0) {
129		/* this is a strange error to return here but I want
130		 * to be able userspace to be able to distinguish a
131		 * valid lock request from one that simply couldn't be
132		 * granted. */
133		if (flags & LKM_NOQUEUE && status == -EAGAIN)
134			status = -ETXTBSY;
135		kfree(fp);
136		goto bail;
137	}
138
139	file->private_data = fp;
140bail:
141	return status;
142}
143
144static int dlmfs_file_release(struct inode *inode,
145			      struct file *file)
146{
147	int level, status;
148	struct dlmfs_inode_private *ip = DLMFS_I(inode);
149	struct dlmfs_filp_private *fp =
150		(struct dlmfs_filp_private *) file->private_data;
151
152	if (S_ISDIR(inode->i_mode))
153		BUG();
154
155	mlog(0, "close called on inode %lu\n", inode->i_ino);
156
157	status = 0;
158	if (fp) {
159		level = fp->fp_lock_level;
160		if (level != LKM_IVMODE)
161			user_dlm_cluster_unlock(&ip->ip_lockres, level);
162
163		kfree(fp);
164		file->private_data = NULL;
165	}
166
167	return 0;
168}
169
170static ssize_t dlmfs_file_read(struct file *filp,
171			       char __user *buf,
172			       size_t count,
173			       loff_t *ppos)
174{
175	int bytes_left;
176	ssize_t readlen;
177	char *lvb_buf;
178	struct inode *inode = filp->f_path.dentry->d_inode;
179
180	mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
181		inode->i_ino, count, *ppos);
182
183	if (*ppos >= i_size_read(inode))
184		return 0;
185
186	if (!count)
187		return 0;
188
189	if (!access_ok(VERIFY_WRITE, buf, count))
190		return -EFAULT;
191
192	/* don't read past the lvb */
193	if ((count + *ppos) > i_size_read(inode))
194		readlen = i_size_read(inode) - *ppos;
195	else
196		readlen = count - *ppos;
197
198	lvb_buf = kmalloc(readlen, GFP_NOFS);
199	if (!lvb_buf)
200		return -ENOMEM;
201
202	user_dlm_read_lvb(inode, lvb_buf, readlen);
203	bytes_left = __copy_to_user(buf, lvb_buf, readlen);
204	readlen -= bytes_left;
205
206	kfree(lvb_buf);
207
208	*ppos = *ppos + readlen;
209
210	mlog(0, "read %zd bytes\n", readlen);
211	return readlen;
212}
213
214static ssize_t dlmfs_file_write(struct file *filp,
215				const char __user *buf,
216				size_t count,
217				loff_t *ppos)
218{
219	int bytes_left;
220	ssize_t writelen;
221	char *lvb_buf;
222	struct inode *inode = filp->f_path.dentry->d_inode;
223
224	mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
225		inode->i_ino, count, *ppos);
226
227	if (*ppos >= i_size_read(inode))
228		return -ENOSPC;
229
230	if (!count)
231		return 0;
232
233	if (!access_ok(VERIFY_READ, buf, count))
234		return -EFAULT;
235
236	/* don't write past the lvb */
237	if ((count + *ppos) > i_size_read(inode))
238		writelen = i_size_read(inode) - *ppos;
239	else
240		writelen = count - *ppos;
241
242	lvb_buf = kmalloc(writelen, GFP_NOFS);
243	if (!lvb_buf)
244		return -ENOMEM;
245
246	bytes_left = copy_from_user(lvb_buf, buf, writelen);
247	writelen -= bytes_left;
248	if (writelen)
249		user_dlm_write_lvb(inode, lvb_buf, writelen);
250
251	kfree(lvb_buf);
252
253	*ppos = *ppos + writelen;
254	mlog(0, "wrote %zd bytes\n", writelen);
255	return writelen;
256}
257
258static void dlmfs_init_once(void *foo,
259			    struct kmem_cache *cachep,
260			    unsigned long flags)
261{
262	struct dlmfs_inode_private *ip =
263		(struct dlmfs_inode_private *) foo;
264
265	ip->ip_dlm = NULL;
266	ip->ip_parent = NULL;
267
268	inode_init_once(&ip->ip_vfs_inode);
269}
270
271static struct inode *dlmfs_alloc_inode(struct super_block *sb)
272{
273	struct dlmfs_inode_private *ip;
274
275	ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
276	if (!ip)
277		return NULL;
278
279	return &ip->ip_vfs_inode;
280}
281
282static void dlmfs_destroy_inode(struct inode *inode)
283{
284	kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
285}
286
287static void dlmfs_clear_inode(struct inode *inode)
288{
289	int status;
290	struct dlmfs_inode_private *ip;
291
292	if (!inode)
293		return;
294
295	mlog(0, "inode %lu\n", inode->i_ino);
296
297	ip = DLMFS_I(inode);
298
299	if (S_ISREG(inode->i_mode)) {
300		status = user_dlm_destroy_lock(&ip->ip_lockres);
301		if (status < 0)
302			mlog_errno(status);
303		iput(ip->ip_parent);
304		goto clear_fields;
305	}
306
307	mlog(0, "we're a directory, ip->ip_dlm = 0x%p\n", ip->ip_dlm);
308	/* we must be a directory. If required, lets unregister the
309	 * dlm context now. */
310	if (ip->ip_dlm)
311		user_dlm_unregister_context(ip->ip_dlm);
312clear_fields:
313	ip->ip_parent = NULL;
314	ip->ip_dlm = NULL;
315}
316
317static struct backing_dev_info dlmfs_backing_dev_info = {
318	.ra_pages	= 0,	/* No readahead */
319	.capabilities	= BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
320};
321
322static struct inode *dlmfs_get_root_inode(struct super_block *sb)
323{
324	struct inode *inode = new_inode(sb);
325	int mode = S_IFDIR | 0755;
326	struct dlmfs_inode_private *ip;
327
328	if (inode) {
329		ip = DLMFS_I(inode);
330
331		inode->i_mode = mode;
332		inode->i_uid = current->fsuid;
333		inode->i_gid = current->fsgid;
334		inode->i_blocks = 0;
335		inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
336		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
337		inc_nlink(inode);
338
339		inode->i_fop = &simple_dir_operations;
340		inode->i_op = &dlmfs_root_inode_operations;
341	}
342
343	return inode;
344}
345
346static struct inode *dlmfs_get_inode(struct inode *parent,
347				     struct dentry *dentry,
348				     int mode)
349{
350	struct super_block *sb = parent->i_sb;
351	struct inode * inode = new_inode(sb);
352	struct dlmfs_inode_private *ip;
353
354	if (!inode)
355		return NULL;
356
357	inode->i_mode = mode;
358	inode->i_uid = current->fsuid;
359	inode->i_gid = current->fsgid;
360	inode->i_blocks = 0;
361	inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
362	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
363
364	ip = DLMFS_I(inode);
365	ip->ip_dlm = DLMFS_I(parent)->ip_dlm;
366
367	switch (mode & S_IFMT) {
368	default:
369		/* for now we don't support anything other than
370		 * directories and regular files. */
371		BUG();
372		break;
373	case S_IFREG:
374		inode->i_op = &dlmfs_file_inode_operations;
375		inode->i_fop = &dlmfs_file_operations;
376
377		i_size_write(inode,  DLM_LVB_LEN);
378
379		user_dlm_lock_res_init(&ip->ip_lockres, dentry);
380
381		/* released at clear_inode time, this insures that we
382		 * get to drop the dlm reference on each lock *before*
383		 * we call the unregister code for releasing parent
384		 * directories. */
385		ip->ip_parent = igrab(parent);
386		BUG_ON(!ip->ip_parent);
387		break;
388	case S_IFDIR:
389		inode->i_op = &dlmfs_dir_inode_operations;
390		inode->i_fop = &simple_dir_operations;
391
392		/* directory inodes start off with i_nlink ==
393		 * 2 (for "." entry) */
394		inc_nlink(inode);
395		break;
396	}
397
398	if (parent->i_mode & S_ISGID) {
399		inode->i_gid = parent->i_gid;
400		if (S_ISDIR(mode))
401			inode->i_mode |= S_ISGID;
402	}
403
404	return inode;
405}
406
407/*
408 * File creation. Allocate an inode, and we're done..
409 */
410/* SMP-safe */
411static int dlmfs_mkdir(struct inode * dir,
412		       struct dentry * dentry,
413		       int mode)
414{
415	int status;
416	struct inode *inode = NULL;
417	struct qstr *domain = &dentry->d_name;
418	struct dlmfs_inode_private *ip;
419	struct dlm_ctxt *dlm;
420
421	mlog(0, "mkdir %.*s\n", domain->len, domain->name);
422
423	/* verify that we have a proper domain */
424	if (domain->len >= O2NM_MAX_NAME_LEN) {
425		status = -EINVAL;
426		mlog(ML_ERROR, "invalid domain name for directory.\n");
427		goto bail;
428	}
429
430	inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
431	if (!inode) {
432		status = -ENOMEM;
433		mlog_errno(status);
434		goto bail;
435	}
436
437	ip = DLMFS_I(inode);
438
439	dlm = user_dlm_register_context(domain);
440	if (IS_ERR(dlm)) {
441		status = PTR_ERR(dlm);
442		mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
443		     status, domain->len, domain->name);
444		goto bail;
445	}
446	ip->ip_dlm = dlm;
447
448	inc_nlink(dir);
449	d_instantiate(dentry, inode);
450	dget(dentry);	/* Extra count - pin the dentry in core */
451
452	status = 0;
453bail:
454	if (status < 0)
455		iput(inode);
456	return status;
457}
458
459static int dlmfs_create(struct inode *dir,
460			struct dentry *dentry,
461			int mode,
462			struct nameidata *nd)
463{
464	int status = 0;
465	struct inode *inode;
466	struct qstr *name = &dentry->d_name;
467
468	mlog(0, "create %.*s\n", name->len, name->name);
469
470	/* verify name is valid and doesn't contain any dlm reserved
471	 * characters */
472	if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
473	    name->name[0] == '$') {
474		status = -EINVAL;
475		mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
476		     name->name);
477		goto bail;
478	}
479
480	inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
481	if (!inode) {
482		status = -ENOMEM;
483		mlog_errno(status);
484		goto bail;
485	}
486
487	d_instantiate(dentry, inode);
488	dget(dentry);	/* Extra count - pin the dentry in core */
489bail:
490	return status;
491}
492
493static int dlmfs_unlink(struct inode *dir,
494			struct dentry *dentry)
495{
496	int status;
497	struct inode *inode = dentry->d_inode;
498
499	mlog(0, "unlink inode %lu\n", inode->i_ino);
500
501	/* if there are no current holders, or none that are waiting
502	 * to acquire a lock, this basically destroys our lockres. */
503	status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
504	if (status < 0) {
505		mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
506		     dentry->d_name.len, dentry->d_name.name, status);
507		goto bail;
508	}
509	status = simple_unlink(dir, dentry);
510bail:
511	return status;
512}
513
514static int dlmfs_fill_super(struct super_block * sb,
515			    void * data,
516			    int silent)
517{
518	struct inode * inode;
519	struct dentry * root;
520
521	sb->s_maxbytes = MAX_LFS_FILESIZE;
522	sb->s_blocksize = PAGE_CACHE_SIZE;
523	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
524	sb->s_magic = DLMFS_MAGIC;
525	sb->s_op = &dlmfs_ops;
526	inode = dlmfs_get_root_inode(sb);
527	if (!inode)
528		return -ENOMEM;
529
530	root = d_alloc_root(inode);
531	if (!root) {
532		iput(inode);
533		return -ENOMEM;
534	}
535	sb->s_root = root;
536	return 0;
537}
538
539static const struct file_operations dlmfs_file_operations = {
540	.open		= dlmfs_file_open,
541	.release	= dlmfs_file_release,
542	.read		= dlmfs_file_read,
543	.write		= dlmfs_file_write,
544};
545
546static const struct inode_operations dlmfs_dir_inode_operations = {
547	.create		= dlmfs_create,
548	.lookup		= simple_lookup,
549	.unlink		= dlmfs_unlink,
550};
551
552/* this way we can restrict mkdir to only the toplevel of the fs. */
553static const struct inode_operations dlmfs_root_inode_operations = {
554	.lookup		= simple_lookup,
555	.mkdir		= dlmfs_mkdir,
556	.rmdir		= simple_rmdir,
557};
558
559static const struct super_operations dlmfs_ops = {
560	.statfs		= simple_statfs,
561	.alloc_inode	= dlmfs_alloc_inode,
562	.destroy_inode	= dlmfs_destroy_inode,
563	.clear_inode	= dlmfs_clear_inode,
564	.drop_inode	= generic_delete_inode,
565};
566
567static const struct inode_operations dlmfs_file_inode_operations = {
568	.getattr	= simple_getattr,
569};
570
571static int dlmfs_get_sb(struct file_system_type *fs_type,
572	int flags, const char *dev_name, void *data, struct vfsmount *mnt)
573{
574	return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super, mnt);
575}
576
577static struct file_system_type dlmfs_fs_type = {
578	.owner		= THIS_MODULE,
579	.name		= "ocfs2_dlmfs",
580	.get_sb		= dlmfs_get_sb,
581	.kill_sb	= kill_litter_super,
582};
583
584static int __init init_dlmfs_fs(void)
585{
586	int status;
587	int cleanup_inode = 0, cleanup_worker = 0;
588
589	dlmfs_print_version();
590
591	dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
592				sizeof(struct dlmfs_inode_private),
593				0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
594					SLAB_MEM_SPREAD),
595				dlmfs_init_once, NULL);
596	if (!dlmfs_inode_cache)
597		return -ENOMEM;
598	cleanup_inode = 1;
599
600	user_dlm_worker = create_singlethread_workqueue("user_dlm");
601	if (!user_dlm_worker) {
602		status = -ENOMEM;
603		goto bail;
604	}
605	cleanup_worker = 1;
606
607	status = register_filesystem(&dlmfs_fs_type);
608bail:
609	if (status) {
610		if (cleanup_inode)
611			kmem_cache_destroy(dlmfs_inode_cache);
612		if (cleanup_worker)
613			destroy_workqueue(user_dlm_worker);
614	} else
615		printk("OCFS2 User DLM kernel interface loaded\n");
616	return status;
617}
618
619static void __exit exit_dlmfs_fs(void)
620{
621	unregister_filesystem(&dlmfs_fs_type);
622
623	flush_workqueue(user_dlm_worker);
624	destroy_workqueue(user_dlm_worker);
625
626	kmem_cache_destroy(dlmfs_inode_cache);
627}
628
629MODULE_AUTHOR("Oracle");
630MODULE_LICENSE("GPL");
631
632module_init(init_dlmfs_fs)
633module_exit(exit_dlmfs_fs)
634