blob: 12bafb7265ceb8d2a13e8a1fb3aa86a73149a43e [file] [log] [blame]
Mark Fasheh8df08c82005-12-15 14:31:23 -08001/* -*- 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>
Mark Fasheh8df08c82005-12-15 14:31:23 -080045#include <linux/backing-dev.h>
Joel Becker65b6f342010-01-26 22:14:20 -080046#include <linux/poll.h>
Mark Fasheh8df08c82005-12-15 14:31:23 -080047
48#include <asm/uaccess.h>
49
Joel Becker0016eed2010-01-30 04:33:50 -080050#include "stackglue.h"
Mark Fasheh8df08c82005-12-15 14:31:23 -080051#include "userdlm.h"
Mark Fasheh8df08c82005-12-15 14:31:23 -080052#include "dlmfsver.h"
53
54#define MLOG_MASK_PREFIX ML_DLMFS
55#include "cluster/masklog.h"
56
Joel Beckerd24fbcd2008-01-25 17:02:21 -080057
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -080058static const struct super_operations dlmfs_ops;
Arjan van de Ven00977a52007-02-12 00:55:34 -080059static const struct file_operations dlmfs_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080060static const struct inode_operations dlmfs_dir_inode_operations;
61static const struct inode_operations dlmfs_root_inode_operations;
62static const struct inode_operations dlmfs_file_inode_operations;
Christoph Lametere18b8902006-12-06 20:33:20 -080063static struct kmem_cache *dlmfs_inode_cache;
Mark Fasheh8df08c82005-12-15 14:31:23 -080064
65struct workqueue_struct *user_dlm_worker;
66
Joel Beckerd24fbcd2008-01-25 17:02:21 -080067
Joel Becker14a437c2010-02-04 15:37:27 -080068
69/*
70 * These are the ABI capabilities of dlmfs.
71 *
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.
77 *
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
80 * loaded.
81 *
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
84 * interaction.
Joel Becker65b6f342010-01-26 22:14:20 -080085 *
86 * Capabilities:
87 * - bast : POLLIN against the file descriptor of a held lock
88 * signifies a bast fired on the lock.
Joel Becker14a437c2010-02-04 15:37:27 -080089 */
Joel Beckercbe0e332010-01-30 06:02:10 -080090#define DLMFS_CAPABILITIES "bast stackglue"
Robin Dong98ba0732011-05-06 16:01:40 +080091static int param_set_dlmfs_capabilities(const char *val,
Joel Becker14a437c2010-02-04 15:37:27 -080092 struct kernel_param *kp)
93{
94 printk(KERN_ERR "%s: readonly parameter\n", kp->name);
95 return -EINVAL;
96}
97static int param_get_dlmfs_capabilities(char *buffer,
98 struct kernel_param *kp)
99{
100 return strlcpy(buffer, DLMFS_CAPABILITIES,
101 strlen(DLMFS_CAPABILITIES) + 1);
102}
103module_param_call(capabilities, param_set_dlmfs_capabilities,
104 param_get_dlmfs_capabilities, NULL, 0444);
105MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES);
106
107
Joel Beckerd24fbcd2008-01-25 17:02:21 -0800108/*
Mark Fasheh8df08c82005-12-15 14:31:23 -0800109 * 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
114 *
Tao Maefd647f2010-03-30 13:21:31 +0800115 * O_NONBLOCK -> NOQUEUE
Mark Fasheh8df08c82005-12-15 14:31:23 -0800116 */
117static int dlmfs_decode_open_flags(int open_flags,
118 int *level,
119 int *flags)
120{
121 if (open_flags & (O_WRONLY|O_RDWR))
Tao Maefd647f2010-03-30 13:21:31 +0800122 *level = DLM_LOCK_EX;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800123 else
Tao Maefd647f2010-03-30 13:21:31 +0800124 *level = DLM_LOCK_PR;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800125
126 *flags = 0;
127 if (open_flags & O_NONBLOCK)
Tao Maefd647f2010-03-30 13:21:31 +0800128 *flags |= DLM_LKF_NOQUEUE;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800129
130 return 0;
131}
132
133static int dlmfs_file_open(struct inode *inode,
134 struct file *file)
135{
136 int status, level, flags;
137 struct dlmfs_filp_private *fp = NULL;
138 struct dlmfs_inode_private *ip;
139
140 if (S_ISDIR(inode->i_mode))
141 BUG();
142
143 mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
144 file->f_flags);
145
146 status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
147 if (status < 0)
148 goto bail;
149
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;
153
Kurt Hackelad8100e2006-05-01 14:25:21 -0700154 fp = kmalloc(sizeof(*fp), GFP_NOFS);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800155 if (!fp) {
156 status = -ENOMEM;
157 goto bail;
158 }
159 fp->fp_lock_level = level;
160
161 ip = DLMFS_I(inode);
162
163 status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
164 if (status < 0) {
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
168 * granted. */
Tao Maefd647f2010-03-30 13:21:31 +0800169 if (flags & DLM_LKF_NOQUEUE && status == -EAGAIN)
Mark Fasheh8df08c82005-12-15 14:31:23 -0800170 status = -ETXTBSY;
171 kfree(fp);
172 goto bail;
173 }
174
175 file->private_data = fp;
176bail:
177 return status;
178}
179
180static int dlmfs_file_release(struct inode *inode,
181 struct file *file)
182{
183 int level, status;
184 struct dlmfs_inode_private *ip = DLMFS_I(inode);
Joe Perches33fa1d92010-07-12 13:50:19 -0700185 struct dlmfs_filp_private *fp = file->private_data;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800186
187 if (S_ISDIR(inode->i_mode))
188 BUG();
189
190 mlog(0, "close called on inode %lu\n", inode->i_ino);
191
192 status = 0;
193 if (fp) {
194 level = fp->fp_lock_level;
Tao Maefd647f2010-03-30 13:21:31 +0800195 if (level != DLM_LOCK_IV)
Mark Fasheh8df08c82005-12-15 14:31:23 -0800196 user_dlm_cluster_unlock(&ip->ip_lockres, level);
197
198 kfree(fp);
199 file->private_data = NULL;
200 }
201
202 return 0;
203}
204
Joel Beckere8fce482010-02-09 17:52:13 -0800205/*
206 * We do ->setattr() just to override size changes. Our size is the size
207 * of the LVB and nothing else.
208 */
209static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr)
210{
211 int error;
212 struct inode *inode = dentry->d_inode;
213
214 attr->ia_valid &= ~ATTR_SIZE;
215 error = inode_change_ok(inode, attr);
Christoph Hellwig10257742010-06-04 11:30:02 +0200216 if (error)
217 return error;
Joel Beckere8fce482010-02-09 17:52:13 -0800218
Christoph Hellwig10257742010-06-04 11:30:02 +0200219 setattr_copy(inode, attr);
220 mark_inode_dirty(inode);
221 return 0;
Joel Beckere8fce482010-02-09 17:52:13 -0800222}
223
Joel Becker65b6f342010-01-26 22:14:20 -0800224static unsigned int dlmfs_file_poll(struct file *file, poll_table *wait)
225{
226 int event = 0;
Al Viro496ad9a2013-01-23 17:07:38 -0500227 struct inode *inode = file_inode(file);
Joel Becker65b6f342010-01-26 22:14:20 -0800228 struct dlmfs_inode_private *ip = DLMFS_I(inode);
229
230 poll_wait(file, &ip->ip_lockres.l_event, wait);
231
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);
236
237 return event;
238}
239
Mark Fasheh8df08c82005-12-15 14:31:23 -0800240static ssize_t dlmfs_file_read(struct file *filp,
241 char __user *buf,
242 size_t count,
243 loff_t *ppos)
244{
245 int bytes_left;
Joel Becker0016eed2010-01-30 04:33:50 -0800246 ssize_t readlen, got;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800247 char *lvb_buf;
Al Viro496ad9a2013-01-23 17:07:38 -0500248 struct inode *inode = file_inode(filp);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800249
250 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
251 inode->i_ino, count, *ppos);
252
253 if (*ppos >= i_size_read(inode))
254 return 0;
255
256 if (!count)
257 return 0;
258
259 if (!access_ok(VERIFY_WRITE, buf, count))
260 return -EFAULT;
261
262 /* don't read past the lvb */
263 if ((count + *ppos) > i_size_read(inode))
264 readlen = i_size_read(inode) - *ppos;
265 else
Joel Beckera36d5152010-04-23 15:24:59 -0700266 readlen = count;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800267
Kurt Hackelad8100e2006-05-01 14:25:21 -0700268 lvb_buf = kmalloc(readlen, GFP_NOFS);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800269 if (!lvb_buf)
270 return -ENOMEM;
271
Joel Becker0016eed2010-01-30 04:33:50 -0800272 got = user_dlm_read_lvb(inode, lvb_buf, readlen);
273 if (got) {
274 BUG_ON(got != readlen);
275 bytes_left = __copy_to_user(buf, lvb_buf, readlen);
276 readlen -= bytes_left;
277 } else
278 readlen = 0;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800279
280 kfree(lvb_buf);
281
282 *ppos = *ppos + readlen;
283
284 mlog(0, "read %zd bytes\n", readlen);
285 return readlen;
286}
287
288static ssize_t dlmfs_file_write(struct file *filp,
289 const char __user *buf,
290 size_t count,
291 loff_t *ppos)
292{
293 int bytes_left;
294 ssize_t writelen;
295 char *lvb_buf;
Al Viro496ad9a2013-01-23 17:07:38 -0500296 struct inode *inode = file_inode(filp);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800297
298 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
299 inode->i_ino, count, *ppos);
300
301 if (*ppos >= i_size_read(inode))
302 return -ENOSPC;
303
304 if (!count)
305 return 0;
306
307 if (!access_ok(VERIFY_READ, buf, count))
308 return -EFAULT;
309
310 /* don't write past the lvb */
311 if ((count + *ppos) > i_size_read(inode))
312 writelen = i_size_read(inode) - *ppos;
313 else
314 writelen = count - *ppos;
315
Kurt Hackelad8100e2006-05-01 14:25:21 -0700316 lvb_buf = kmalloc(writelen, GFP_NOFS);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800317 if (!lvb_buf)
318 return -ENOMEM;
319
320 bytes_left = copy_from_user(lvb_buf, buf, writelen);
321 writelen -= bytes_left;
322 if (writelen)
323 user_dlm_write_lvb(inode, lvb_buf, writelen);
324
325 kfree(lvb_buf);
326
327 *ppos = *ppos + writelen;
328 mlog(0, "wrote %zd bytes\n", writelen);
329 return writelen;
330}
331
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700332static void dlmfs_init_once(void *foo)
Mark Fasheh8df08c82005-12-15 14:31:23 -0800333{
334 struct dlmfs_inode_private *ip =
335 (struct dlmfs_inode_private *) foo;
336
Joel Becker0016eed2010-01-30 04:33:50 -0800337 ip->ip_conn = NULL;
Christoph Lametera35afb82007-05-16 22:10:57 -0700338 ip->ip_parent = NULL;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800339
Christoph Lametera35afb82007-05-16 22:10:57 -0700340 inode_init_once(&ip->ip_vfs_inode);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800341}
342
343static struct inode *dlmfs_alloc_inode(struct super_block *sb)
344{
345 struct dlmfs_inode_private *ip;
346
Christoph Lametere6b4f8d2006-12-06 20:33:14 -0800347 ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800348 if (!ip)
349 return NULL;
350
351 return &ip->ip_vfs_inode;
352}
353
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100354static void dlmfs_i_callback(struct rcu_head *head)
355{
356 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100357 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
358}
359
Mark Fasheh8df08c82005-12-15 14:31:23 -0800360static void dlmfs_destroy_inode(struct inode *inode)
361{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100362 call_rcu(&inode->i_rcu, dlmfs_i_callback);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800363}
364
Al Virob57922d2010-06-07 14:34:48 -0400365static void dlmfs_evict_inode(struct inode *inode)
Mark Fasheh8df08c82005-12-15 14:31:23 -0800366{
367 int status;
368 struct dlmfs_inode_private *ip;
369
Jan Karadbd57682012-05-03 14:48:02 +0200370 clear_inode(inode);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800371
372 mlog(0, "inode %lu\n", inode->i_ino);
373
374 ip = DLMFS_I(inode);
375
376 if (S_ISREG(inode->i_mode)) {
377 status = user_dlm_destroy_lock(&ip->ip_lockres);
378 if (status < 0)
379 mlog_errno(status);
380 iput(ip->ip_parent);
381 goto clear_fields;
382 }
383
Joel Becker0016eed2010-01-30 04:33:50 -0800384 mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800385 /* we must be a directory. If required, lets unregister the
386 * dlm context now. */
Joel Becker0016eed2010-01-30 04:33:50 -0800387 if (ip->ip_conn)
388 user_dlm_unregister(ip->ip_conn);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800389clear_fields:
390 ip->ip_parent = NULL;
Joel Becker0016eed2010-01-30 04:33:50 -0800391 ip->ip_conn = NULL;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800392}
393
394static struct backing_dev_info dlmfs_backing_dev_info = {
Jens Axboed9938312009-06-12 14:45:52 +0200395 .name = "ocfs2-dlmfs",
Mark Fasheh8df08c82005-12-15 14:31:23 -0800396 .ra_pages = 0, /* No readahead */
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700397 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Mark Fasheh8df08c82005-12-15 14:31:23 -0800398};
399
400static struct inode *dlmfs_get_root_inode(struct super_block *sb)
401{
402 struct inode *inode = new_inode(sb);
Al Viro67697cb2011-07-26 02:55:32 -0400403 umode_t mode = S_IFDIR | 0755;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800404 struct dlmfs_inode_private *ip;
405
406 if (inode) {
407 ip = DLMFS_I(inode);
408
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400409 inode->i_ino = get_next_ino();
Al Viro2b15ad02011-07-24 13:43:19 -0400410 inode_init_owner(inode, NULL, mode);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800411 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
412 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -0700413 inc_nlink(inode);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800414
415 inode->i_fop = &simple_dir_operations;
416 inode->i_op = &dlmfs_root_inode_operations;
417 }
418
419 return inode;
420}
421
422static struct inode *dlmfs_get_inode(struct inode *parent,
423 struct dentry *dentry,
Al Viro67697cb2011-07-26 02:55:32 -0400424 umode_t mode)
Mark Fasheh8df08c82005-12-15 14:31:23 -0800425{
426 struct super_block *sb = parent->i_sb;
427 struct inode * inode = new_inode(sb);
428 struct dlmfs_inode_private *ip;
429
430 if (!inode)
431 return NULL;
432
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400433 inode->i_ino = get_next_ino();
Al Viro2b15ad02011-07-24 13:43:19 -0400434 inode_init_owner(inode, parent, mode);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800435 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
436 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
437
438 ip = DLMFS_I(inode);
Joel Becker0016eed2010-01-30 04:33:50 -0800439 ip->ip_conn = DLMFS_I(parent)->ip_conn;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800440
441 switch (mode & S_IFMT) {
442 default:
443 /* for now we don't support anything other than
444 * directories and regular files. */
445 BUG();
446 break;
447 case S_IFREG:
448 inode->i_op = &dlmfs_file_inode_operations;
449 inode->i_fop = &dlmfs_file_operations;
450
451 i_size_write(inode, DLM_LVB_LEN);
452
453 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
454
455 /* released at clear_inode time, this insures that we
456 * get to drop the dlm reference on each lock *before*
457 * we call the unregister code for releasing parent
458 * directories. */
459 ip->ip_parent = igrab(parent);
460 BUG_ON(!ip->ip_parent);
461 break;
462 case S_IFDIR:
463 inode->i_op = &dlmfs_dir_inode_operations;
464 inode->i_fop = &simple_dir_operations;
465
466 /* directory inodes start off with i_nlink ==
467 * 2 (for "." entry) */
Dave Hansend8c76e62006-09-30 23:29:04 -0700468 inc_nlink(inode);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800469 break;
470 }
Mark Fasheh8df08c82005-12-15 14:31:23 -0800471 return inode;
472}
473
474/*
475 * File creation. Allocate an inode, and we're done..
476 */
477/* SMP-safe */
478static int dlmfs_mkdir(struct inode * dir,
479 struct dentry * dentry,
Al Viro18bb1db2011-07-26 01:41:39 -0400480 umode_t mode)
Mark Fasheh8df08c82005-12-15 14:31:23 -0800481{
482 int status;
483 struct inode *inode = NULL;
484 struct qstr *domain = &dentry->d_name;
485 struct dlmfs_inode_private *ip;
Joel Becker0016eed2010-01-30 04:33:50 -0800486 struct ocfs2_cluster_connection *conn;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800487
488 mlog(0, "mkdir %.*s\n", domain->len, domain->name);
489
490 /* verify that we have a proper domain */
Joel Becker0016eed2010-01-30 04:33:50 -0800491 if (domain->len >= GROUP_NAME_MAX) {
Mark Fasheh8df08c82005-12-15 14:31:23 -0800492 status = -EINVAL;
493 mlog(ML_ERROR, "invalid domain name for directory.\n");
494 goto bail;
495 }
496
497 inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
498 if (!inode) {
499 status = -ENOMEM;
500 mlog_errno(status);
501 goto bail;
502 }
503
504 ip = DLMFS_I(inode);
505
Joel Becker0016eed2010-01-30 04:33:50 -0800506 conn = user_dlm_register(domain);
507 if (IS_ERR(conn)) {
508 status = PTR_ERR(conn);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800509 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
510 status, domain->len, domain->name);
511 goto bail;
512 }
Joel Becker0016eed2010-01-30 04:33:50 -0800513 ip->ip_conn = conn;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800514
Dave Hansend8c76e62006-09-30 23:29:04 -0700515 inc_nlink(dir);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800516 d_instantiate(dentry, inode);
517 dget(dentry); /* Extra count - pin the dentry in core */
518
519 status = 0;
520bail:
521 if (status < 0)
522 iput(inode);
523 return status;
524}
525
526static int dlmfs_create(struct inode *dir,
527 struct dentry *dentry,
Al Viro4acdaf22011-07-26 01:42:34 -0400528 umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -0400529 bool excl)
Mark Fasheh8df08c82005-12-15 14:31:23 -0800530{
531 int status = 0;
532 struct inode *inode;
533 struct qstr *name = &dentry->d_name;
534
535 mlog(0, "create %.*s\n", name->len, name->name);
536
537 /* verify name is valid and doesn't contain any dlm reserved
538 * characters */
539 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
540 name->name[0] == '$') {
541 status = -EINVAL;
542 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
543 name->name);
544 goto bail;
545 }
546
547 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
548 if (!inode) {
549 status = -ENOMEM;
550 mlog_errno(status);
551 goto bail;
552 }
553
554 d_instantiate(dentry, inode);
555 dget(dentry); /* Extra count - pin the dentry in core */
556bail:
557 return status;
558}
559
560static int dlmfs_unlink(struct inode *dir,
561 struct dentry *dentry)
562{
563 int status;
564 struct inode *inode = dentry->d_inode;
565
566 mlog(0, "unlink inode %lu\n", inode->i_ino);
567
568 /* if there are no current holders, or none that are waiting
569 * to acquire a lock, this basically destroys our lockres. */
570 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
571 if (status < 0) {
572 mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
573 dentry->d_name.len, dentry->d_name.name, status);
574 goto bail;
575 }
576 status = simple_unlink(dir, dentry);
577bail:
578 return status;
579}
580
581static int dlmfs_fill_super(struct super_block * sb,
582 void * data,
583 int silent)
584{
Mark Fasheh8df08c82005-12-15 14:31:23 -0800585 sb->s_maxbytes = MAX_LFS_FILESIZE;
586 sb->s_blocksize = PAGE_CACHE_SIZE;
587 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
588 sb->s_magic = DLMFS_MAGIC;
589 sb->s_op = &dlmfs_ops;
Al Viro48fde702012-01-08 22:15:13 -0500590 sb->s_root = d_make_root(dlmfs_get_root_inode(sb));
591 if (!sb->s_root)
Mark Fasheh8df08c82005-12-15 14:31:23 -0800592 return -ENOMEM;
Mark Fasheh8df08c82005-12-15 14:31:23 -0800593 return 0;
594}
595
Arjan van de Ven00977a52007-02-12 00:55:34 -0800596static const struct file_operations dlmfs_file_operations = {
Mark Fasheh8df08c82005-12-15 14:31:23 -0800597 .open = dlmfs_file_open,
598 .release = dlmfs_file_release,
Joel Becker65b6f342010-01-26 22:14:20 -0800599 .poll = dlmfs_file_poll,
Mark Fasheh8df08c82005-12-15 14:31:23 -0800600 .read = dlmfs_file_read,
601 .write = dlmfs_file_write,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200602 .llseek = default_llseek,
Mark Fasheh8df08c82005-12-15 14:31:23 -0800603};
604
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800605static const struct inode_operations dlmfs_dir_inode_operations = {
Mark Fasheh8df08c82005-12-15 14:31:23 -0800606 .create = dlmfs_create,
607 .lookup = simple_lookup,
608 .unlink = dlmfs_unlink,
609};
610
611/* this way we can restrict mkdir to only the toplevel of the fs. */
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800612static const struct inode_operations dlmfs_root_inode_operations = {
Mark Fasheh8df08c82005-12-15 14:31:23 -0800613 .lookup = simple_lookup,
614 .mkdir = dlmfs_mkdir,
615 .rmdir = simple_rmdir,
616};
617
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800618static const struct super_operations dlmfs_ops = {
Mark Fasheh8df08c82005-12-15 14:31:23 -0800619 .statfs = simple_statfs,
620 .alloc_inode = dlmfs_alloc_inode,
621 .destroy_inode = dlmfs_destroy_inode,
Al Virob57922d2010-06-07 14:34:48 -0400622 .evict_inode = dlmfs_evict_inode,
Mark Fasheh8df08c82005-12-15 14:31:23 -0800623 .drop_inode = generic_delete_inode,
624};
625
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800626static const struct inode_operations dlmfs_file_inode_operations = {
Mark Fasheh8df08c82005-12-15 14:31:23 -0800627 .getattr = simple_getattr,
Joel Beckere8fce482010-02-09 17:52:13 -0800628 .setattr = dlmfs_file_setattr,
Mark Fasheh8df08c82005-12-15 14:31:23 -0800629};
630
Al Viro3c26ff62010-07-25 11:46:36 +0400631static struct dentry *dlmfs_mount(struct file_system_type *fs_type,
632 int flags, const char *dev_name, void *data)
Mark Fasheh8df08c82005-12-15 14:31:23 -0800633{
Al Viro3c26ff62010-07-25 11:46:36 +0400634 return mount_nodev(fs_type, flags, data, dlmfs_fill_super);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800635}
636
637static struct file_system_type dlmfs_fs_type = {
638 .owner = THIS_MODULE,
639 .name = "ocfs2_dlmfs",
Al Viro3c26ff62010-07-25 11:46:36 +0400640 .mount = dlmfs_mount,
Mark Fasheh8df08c82005-12-15 14:31:23 -0800641 .kill_sb = kill_litter_super,
642};
Eric W. Biederman7f78e032013-03-02 19:39:14 -0800643MODULE_ALIAS_FS("ocfs2_dlmfs");
Mark Fasheh8df08c82005-12-15 14:31:23 -0800644
645static int __init init_dlmfs_fs(void)
646{
647 int status;
648 int cleanup_inode = 0, cleanup_worker = 0;
649
650 dlmfs_print_version();
651
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700652 status = bdi_init(&dlmfs_backing_dev_info);
653 if (status)
654 return status;
655
Mark Fasheh8df08c82005-12-15 14:31:23 -0800656 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
657 sizeof(struct dlmfs_inode_private),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800658 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
659 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +0900660 dlmfs_init_once);
Coly Li07d9a392008-11-17 12:38:22 +0800661 if (!dlmfs_inode_cache) {
662 status = -ENOMEM;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700663 goto bail;
Coly Li07d9a392008-11-17 12:38:22 +0800664 }
Mark Fasheh8df08c82005-12-15 14:31:23 -0800665 cleanup_inode = 1;
666
667 user_dlm_worker = create_singlethread_workqueue("user_dlm");
668 if (!user_dlm_worker) {
669 status = -ENOMEM;
670 goto bail;
671 }
672 cleanup_worker = 1;
673
Joel Becker0016eed2010-01-30 04:33:50 -0800674 user_dlm_set_locking_protocol();
Mark Fasheh8df08c82005-12-15 14:31:23 -0800675 status = register_filesystem(&dlmfs_fs_type);
676bail:
677 if (status) {
678 if (cleanup_inode)
679 kmem_cache_destroy(dlmfs_inode_cache);
680 if (cleanup_worker)
681 destroy_workqueue(user_dlm_worker);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700682 bdi_destroy(&dlmfs_backing_dev_info);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800683 } else
684 printk("OCFS2 User DLM kernel interface loaded\n");
685 return status;
686}
687
688static void __exit exit_dlmfs_fs(void)
689{
690 unregister_filesystem(&dlmfs_fs_type);
691
692 flush_workqueue(user_dlm_worker);
693 destroy_workqueue(user_dlm_worker);
694
Kirill A. Shutemov8c0a8532012-09-26 11:33:07 +1000695 /*
696 * Make sure all delayed rcu free inodes are flushed before we
697 * destroy cache.
698 */
699 rcu_barrier();
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -0700700 kmem_cache_destroy(dlmfs_inode_cache);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700701
702 bdi_destroy(&dlmfs_backing_dev_info);
Mark Fasheh8df08c82005-12-15 14:31:23 -0800703}
704
705MODULE_AUTHOR("Oracle");
706MODULE_LICENSE("GPL");
707
708module_init(init_dlmfs_fs)
709module_exit(exit_dlmfs_fs)