aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ocfs2/dlm
diff options
context:
space:
mode:
authorMark Fasheh <mark.fasheh@oracle.com>2005-12-15 17:31:23 -0500
committerJoel Becker <joel.becker@oracle.com>2006-01-03 14:45:47 -0500
commit8df08c89c668e1bd922a053fdb5ba1fadbecbb38 (patch)
treeab1febb732c01a66c0a9bfe9c8952ba2bb66fba8 /fs/ocfs2/dlm
parent6714d8e86bf443f6f7af50f9d432025649f091f5 (diff)
[PATCH] OCFS2: The Second Oracle Cluster Filesystem
dlmfs: A minimal dlm userspace interface implemented via a virtual file system. Most of the OCFS2 tools make use of this to take cluster locks when doing operations on the file system. Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Diffstat (limited to 'fs/ocfs2/dlm')
-rw-r--r--fs/ocfs2/dlm/Makefile4
-rw-r--r--fs/ocfs2/dlm/dlmfs.c640
-rw-r--r--fs/ocfs2/dlm/dlmfsver.c42
-rw-r--r--fs/ocfs2/dlm/dlmfsver.h31
-rw-r--r--fs/ocfs2/dlm/userdlm.c658
-rw-r--r--fs/ocfs2/dlm/userdlm.h111
6 files changed, 1485 insertions, 1 deletions
diff --git a/fs/ocfs2/dlm/Makefile b/fs/ocfs2/dlm/Makefile
index 2a5274bcc8bb..ce3f7c29d270 100644
--- a/fs/ocfs2/dlm/Makefile
+++ b/fs/ocfs2/dlm/Makefile
@@ -1,6 +1,8 @@
1EXTRA_CFLAGS += -Ifs/ocfs2 1EXTRA_CFLAGS += -Ifs/ocfs2
2 2
3obj-$(CONFIG_OCFS2_FS) += ocfs2_dlm.o 3obj-$(CONFIG_OCFS2_FS) += ocfs2_dlm.o ocfs2_dlmfs.o
4 4
5ocfs2_dlm-objs := dlmdomain.o dlmdebug.o dlmthread.o dlmrecovery.o \ 5ocfs2_dlm-objs := dlmdomain.o dlmdebug.o dlmthread.o dlmrecovery.o \
6 dlmmaster.o dlmast.o dlmconvert.o dlmlock.o dlmunlock.o dlmver.o 6 dlmmaster.o dlmast.o dlmconvert.o dlmlock.o dlmunlock.o dlmver.o
7
8ocfs2_dlmfs-objs := userdlm.o dlmfs.o dlmfsver.o
diff --git a/fs/ocfs2/dlm/dlmfs.c b/fs/ocfs2/dlm/dlmfs.c
new file mode 100644
index 000000000000..dd2d24dc25e0
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmfs.c
@@ -0,0 +1,640 @@
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/smp_lock.h>
46#include <linux/backing-dev.h>
47
48#include <asm/uaccess.h>
49
50
51#include "cluster/nodemanager.h"
52#include "cluster/heartbeat.h"
53#include "cluster/tcp.h"
54
55#include "dlmapi.h"
56
57#include "userdlm.h"
58
59#include "dlmfsver.h"
60
61#define MLOG_MASK_PREFIX ML_DLMFS
62#include "cluster/masklog.h"
63
64static struct super_operations dlmfs_ops;
65static struct file_operations dlmfs_file_operations;
66static struct inode_operations dlmfs_dir_inode_operations;
67static struct inode_operations dlmfs_root_inode_operations;
68static struct inode_operations dlmfs_file_inode_operations;
69static kmem_cache_t *dlmfs_inode_cache;
70
71struct workqueue_struct *user_dlm_worker;
72
73/*
74 * decodes a set of open flags into a valid lock level and a set of flags.
75 * returns < 0 if we have invalid flags
76 * flags which mean something to us:
77 * O_RDONLY -> PRMODE level
78 * O_WRONLY -> EXMODE level
79 *
80 * O_NONBLOCK -> LKM_NOQUEUE
81 */
82static int dlmfs_decode_open_flags(int open_flags,
83 int *level,
84 int *flags)
85{
86 if (open_flags & (O_WRONLY|O_RDWR))
87 *level = LKM_EXMODE;
88 else
89 *level = LKM_PRMODE;
90
91 *flags = 0;
92 if (open_flags & O_NONBLOCK)
93 *flags |= LKM_NOQUEUE;
94
95 return 0;
96}
97
98static int dlmfs_file_open(struct inode *inode,
99 struct file *file)
100{
101 int status, level, flags;
102 struct dlmfs_filp_private *fp = NULL;
103 struct dlmfs_inode_private *ip;
104
105 if (S_ISDIR(inode->i_mode))
106 BUG();
107
108 mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
109 file->f_flags);
110
111 status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
112 if (status < 0)
113 goto bail;
114
115 /* We don't want to honor O_APPEND at read/write time as it
116 * doesn't make sense for LVB writes. */
117 file->f_flags &= ~O_APPEND;
118
119 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
120 if (!fp) {
121 status = -ENOMEM;
122 goto bail;
123 }
124 fp->fp_lock_level = level;
125
126 ip = DLMFS_I(inode);
127
128 status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
129 if (status < 0) {
130 /* this is a strange error to return here but I want
131 * to be able userspace to be able to distinguish a
132 * valid lock request from one that simply couldn't be
133 * granted. */
134 if (flags & LKM_NOQUEUE && status == -EAGAIN)
135 status = -ETXTBSY;
136 kfree(fp);
137 goto bail;
138 }
139
140 file->private_data = fp;
141bail:
142 return status;
143}
144
145static int dlmfs_file_release(struct inode *inode,
146 struct file *file)
147{
148 int level, status;
149 struct dlmfs_inode_private *ip = DLMFS_I(inode);
150 struct dlmfs_filp_private *fp =
151 (struct dlmfs_filp_private *) file->private_data;
152
153 if (S_ISDIR(inode->i_mode))
154 BUG();
155
156 mlog(0, "close called on inode %lu\n", inode->i_ino);
157
158 status = 0;
159 if (fp) {
160 level = fp->fp_lock_level;
161 if (level != LKM_IVMODE)
162 user_dlm_cluster_unlock(&ip->ip_lockres, level);
163
164 kfree(fp);
165 file->private_data = NULL;
166 }
167
168 return 0;
169}
170
171static ssize_t dlmfs_file_read(struct file *filp,
172 char __user *buf,
173 size_t count,
174 loff_t *ppos)
175{
176 int bytes_left;
177 ssize_t readlen;
178 char *lvb_buf;
179 struct inode *inode = filp->f_dentry->d_inode;
180
181 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
182 inode->i_ino, count, *ppos);
183
184 if (*ppos >= i_size_read(inode))
185 return 0;
186
187 if (!count)
188 return 0;
189
190 if (!access_ok(VERIFY_WRITE, buf, count))
191 return -EFAULT;
192
193 /* don't read past the lvb */
194 if ((count + *ppos) > i_size_read(inode))
195 readlen = i_size_read(inode) - *ppos;
196 else
197 readlen = count - *ppos;
198
199 lvb_buf = kmalloc(readlen, GFP_KERNEL);
200 if (!lvb_buf)
201 return -ENOMEM;
202
203 user_dlm_read_lvb(inode, lvb_buf, readlen);
204 bytes_left = __copy_to_user(buf, lvb_buf, readlen);
205 readlen -= bytes_left;
206
207 kfree(lvb_buf);
208
209 *ppos = *ppos + readlen;
210
211 mlog(0, "read %zd bytes\n", readlen);
212 return readlen;
213}
214
215static ssize_t dlmfs_file_write(struct file *filp,
216 const char __user *buf,
217 size_t count,
218 loff_t *ppos)
219{
220 int bytes_left;
221 ssize_t writelen;
222 char *lvb_buf;
223 struct inode *inode = filp->f_dentry->d_inode;
224
225 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
226 inode->i_ino, count, *ppos);
227
228 if (*ppos >= i_size_read(inode))
229 return -ENOSPC;
230
231 if (!count)
232 return 0;
233
234 if (!access_ok(VERIFY_READ, buf, count))
235 return -EFAULT;
236
237 /* don't write past the lvb */
238 if ((count + *ppos) > i_size_read(inode))
239 writelen = i_size_read(inode) - *ppos;
240 else
241 writelen = count - *ppos;
242
243 lvb_buf = kmalloc(writelen, GFP_KERNEL);
244 if (!lvb_buf)
245 return -ENOMEM;
246
247 bytes_left = copy_from_user(lvb_buf, buf, writelen);
248 writelen -= bytes_left;
249 if (writelen)
250 user_dlm_write_lvb(inode, lvb_buf, writelen);
251
252 kfree(lvb_buf);
253
254 *ppos = *ppos + writelen;
255 mlog(0, "wrote %zd bytes\n", writelen);
256 return writelen;
257}
258
259static void dlmfs_init_once(void *foo,
260 kmem_cache_t *cachep,
261 unsigned long flags)
262{
263 struct dlmfs_inode_private *ip =
264 (struct dlmfs_inode_private *) foo;
265
266 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
267 SLAB_CTOR_CONSTRUCTOR) {
268 ip->ip_dlm = NULL;
269 ip->ip_parent = NULL;
270
271 inode_init_once(&ip->ip_vfs_inode);
272 }
273}
274
275static struct inode *dlmfs_alloc_inode(struct super_block *sb)
276{
277 struct dlmfs_inode_private *ip;
278
279 ip = kmem_cache_alloc(dlmfs_inode_cache, SLAB_NOFS);
280 if (!ip)
281 return NULL;
282
283 return &ip->ip_vfs_inode;
284}
285
286static void dlmfs_destroy_inode(struct inode *inode)
287{
288 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
289}
290
291static void dlmfs_clear_inode(struct inode *inode)
292{
293 int status;
294 struct dlmfs_inode_private *ip;
295
296 if (!inode)
297 return;
298
299 mlog(0, "inode %lu\n", inode->i_ino);
300
301 ip = DLMFS_I(inode);
302
303 if (S_ISREG(inode->i_mode)) {
304 status = user_dlm_destroy_lock(&ip->ip_lockres);
305 if (status < 0)
306 mlog_errno(status);
307 iput(ip->ip_parent);
308 goto clear_fields;
309 }
310
311 mlog(0, "we're a directory, ip->ip_dlm = 0x%p\n", ip->ip_dlm);
312 /* we must be a directory. If required, lets unregister the
313 * dlm context now. */
314 if (ip->ip_dlm)
315 user_dlm_unregister_context(ip->ip_dlm);
316clear_fields:
317 ip->ip_parent = NULL;
318 ip->ip_dlm = NULL;
319}
320
321static struct backing_dev_info dlmfs_backing_dev_info = {
322 .ra_pages = 0, /* No readahead */
323 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
324};
325
326static struct inode *dlmfs_get_root_inode(struct super_block *sb)
327{
328 struct inode *inode = new_inode(sb);
329 int mode = S_IFDIR | 0755;
330 struct dlmfs_inode_private *ip;
331
332 if (inode) {
333 ip = DLMFS_I(inode);
334
335 inode->i_mode = mode;
336 inode->i_uid = current->fsuid;
337 inode->i_gid = current->fsgid;
338 inode->i_blksize = PAGE_CACHE_SIZE;
339 inode->i_blocks = 0;
340 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
341 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
342 inode->i_nlink++;
343
344 inode->i_fop = &simple_dir_operations;
345 inode->i_op = &dlmfs_root_inode_operations;
346 }
347
348 return inode;
349}
350
351static struct inode *dlmfs_get_inode(struct inode *parent,
352 struct dentry *dentry,
353 int mode)
354{
355 struct super_block *sb = parent->i_sb;
356 struct inode * inode = new_inode(sb);
357 struct dlmfs_inode_private *ip;
358
359 if (!inode)
360 return NULL;
361
362 inode->i_mode = mode;
363 inode->i_uid = current->fsuid;
364 inode->i_gid = current->fsgid;
365 inode->i_blksize = PAGE_CACHE_SIZE;
366 inode->i_blocks = 0;
367 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
368 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
369
370 ip = DLMFS_I(inode);
371 ip->ip_dlm = DLMFS_I(parent)->ip_dlm;
372
373 switch (mode & S_IFMT) {
374 default:
375 /* for now we don't support anything other than
376 * directories and regular files. */
377 BUG();
378 break;
379 case S_IFREG:
380 inode->i_op = &dlmfs_file_inode_operations;
381 inode->i_fop = &dlmfs_file_operations;
382
383 i_size_write(inode, DLM_LVB_LEN);
384
385 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
386
387 /* released at clear_inode time, this insures that we
388 * get to drop the dlm reference on each lock *before*
389 * we call the unregister code for releasing parent
390 * directories. */
391 ip->ip_parent = igrab(parent);
392 BUG_ON(!ip->ip_parent);
393 break;
394 case S_IFDIR:
395 inode->i_op = &dlmfs_dir_inode_operations;
396 inode->i_fop = &simple_dir_operations;
397
398 /* directory inodes start off with i_nlink ==
399 * 2 (for "." entry) */
400 inode->i_nlink++;
401 break;
402 }
403
404 if (parent->i_mode & S_ISGID) {
405 inode->i_gid = parent->i_gid;
406 if (S_ISDIR(mode))
407 inode->i_mode |= S_ISGID;
408 }
409
410 return inode;
411}
412
413/*
414 * File creation. Allocate an inode, and we're done..
415 */
416/* SMP-safe */
417static int dlmfs_mkdir(struct inode * dir,
418 struct dentry * dentry,
419 int mode)
420{
421 int status;
422 struct inode *inode = NULL;
423 struct qstr *domain = &dentry->d_name;
424 struct dlmfs_inode_private *ip;
425 struct dlm_ctxt *dlm;
426
427 mlog(0, "mkdir %.*s\n", domain->len, domain->name);
428
429 /* verify that we have a proper domain */
430 if (domain->len >= O2NM_MAX_NAME_LEN) {
431 status = -EINVAL;
432 mlog(ML_ERROR, "invalid domain name for directory.\n");
433 goto bail;
434 }
435
436 inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
437 if (!inode) {
438 status = -ENOMEM;
439 mlog_errno(status);
440 goto bail;
441 }
442
443 ip = DLMFS_I(inode);
444
445 dlm = user_dlm_register_context(domain);
446 if (IS_ERR(dlm)) {
447 status = PTR_ERR(dlm);
448 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
449 status, domain->len, domain->name);
450 goto bail;
451 }
452 ip->ip_dlm = dlm;
453
454 dir->i_nlink++;
455 d_instantiate(dentry, inode);
456 dget(dentry); /* Extra count - pin the dentry in core */
457
458 status = 0;
459bail:
460 if (status < 0)
461 iput(inode);
462 return status;
463}
464
465static int dlmfs_create(struct inode *dir,
466 struct dentry *dentry,
467 int mode,
468 struct nameidata *nd)
469{
470 int status = 0;
471 struct inode *inode;
472 struct qstr *name = &dentry->d_name;
473
474 mlog(0, "create %.*s\n", name->len, name->name);
475
476 /* verify name is valid and doesn't contain any dlm reserved
477 * characters */
478 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
479 name->name[0] == '$') {
480 status = -EINVAL;
481 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
482 name->name);
483 goto bail;
484 }
485
486 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
487 if (!inode) {
488 status = -ENOMEM;
489 mlog_errno(status);
490 goto bail;
491 }
492
493 d_instantiate(dentry, inode);
494 dget(dentry); /* Extra count - pin the dentry in core */
495bail:
496 return status;
497}
498
499static int dlmfs_unlink(struct inode *dir,
500 struct dentry *dentry)
501{
502 int status;
503 struct inode *inode = dentry->d_inode;
504
505 mlog(0, "unlink inode %lu\n", inode->i_ino);
506
507 /* if there are no current holders, or none that are waiting
508 * to acquire a lock, this basically destroys our lockres. */
509 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
510 if (status < 0) {
511 mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
512 dentry->d_name.len, dentry->d_name.name, status);
513 goto bail;
514 }
515 status = simple_unlink(dir, dentry);
516bail:
517 return status;
518}
519
520static int dlmfs_fill_super(struct super_block * sb,
521 void * data,
522 int silent)
523{
524 struct inode * inode;
525 struct dentry * root;
526
527 sb->s_maxbytes = MAX_LFS_FILESIZE;
528 sb->s_blocksize = PAGE_CACHE_SIZE;
529 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
530 sb->s_magic = DLMFS_MAGIC;
531 sb->s_op = &dlmfs_ops;
532 inode = dlmfs_get_root_inode(sb);
533 if (!inode)
534 return -ENOMEM;
535
536 root = d_alloc_root(inode);
537 if (!root) {
538 iput(inode);
539 return -ENOMEM;
540 }
541 sb->s_root = root;
542 return 0;
543}
544
545static struct file_operations dlmfs_file_operations = {
546 .open = dlmfs_file_open,
547 .release = dlmfs_file_release,
548 .read = dlmfs_file_read,
549 .write = dlmfs_file_write,
550};
551
552static struct inode_operations dlmfs_dir_inode_operations = {
553 .create = dlmfs_create,
554 .lookup = simple_lookup,
555 .unlink = dlmfs_unlink,
556};
557
558/* this way we can restrict mkdir to only the toplevel of the fs. */
559static struct inode_operations dlmfs_root_inode_operations = {
560 .lookup = simple_lookup,
561 .mkdir = dlmfs_mkdir,
562 .rmdir = simple_rmdir,
563};
564
565static struct super_operations dlmfs_ops = {
566 .statfs = simple_statfs,
567 .alloc_inode = dlmfs_alloc_inode,
568 .destroy_inode = dlmfs_destroy_inode,
569 .clear_inode = dlmfs_clear_inode,
570 .drop_inode = generic_delete_inode,
571};
572
573static struct inode_operations dlmfs_file_inode_operations = {
574 .getattr = simple_getattr,
575};
576
577static struct super_block *dlmfs_get_sb(struct file_system_type *fs_type,
578 int flags, const char *dev_name, void *data)
579{
580 return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super);
581}
582
583static struct file_system_type dlmfs_fs_type = {
584 .owner = THIS_MODULE,
585 .name = "ocfs2_dlmfs",
586 .get_sb = dlmfs_get_sb,
587 .kill_sb = kill_litter_super,
588};
589
590static int __init init_dlmfs_fs(void)
591{
592 int status;
593 int cleanup_inode = 0, cleanup_worker = 0;
594
595 dlmfs_print_version();
596
597 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
598 sizeof(struct dlmfs_inode_private),
599 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
600 dlmfs_init_once, NULL);
601 if (!dlmfs_inode_cache)
602 return -ENOMEM;
603 cleanup_inode = 1;
604
605 user_dlm_worker = create_singlethread_workqueue("user_dlm");
606 if (!user_dlm_worker) {
607 status = -ENOMEM;
608 goto bail;
609 }
610 cleanup_worker = 1;
611
612 status = register_filesystem(&dlmfs_fs_type);
613bail:
614 if (status) {
615 if (cleanup_inode)
616 kmem_cache_destroy(dlmfs_inode_cache);
617 if (cleanup_worker)
618 destroy_workqueue(user_dlm_worker);
619 } else
620 printk("OCFS2 User DLM kernel interface loaded\n");
621 return status;
622}
623
624static void __exit exit_dlmfs_fs(void)
625{
626 unregister_filesystem(&dlmfs_fs_type);
627
628 flush_workqueue(user_dlm_worker);
629 destroy_workqueue(user_dlm_worker);
630
631 if (kmem_cache_destroy(dlmfs_inode_cache))
632 printk(KERN_INFO "dlmfs_inode_cache: not all structures "
633 "were freed\n");
634}
635
636MODULE_AUTHOR("Oracle");
637MODULE_LICENSE("GPL");
638
639module_init(init_dlmfs_fs)
640module_exit(exit_dlmfs_fs)
diff --git a/fs/ocfs2/dlm/dlmfsver.c b/fs/ocfs2/dlm/dlmfsver.c
new file mode 100644
index 000000000000..d2be3ad841f9
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmfsver.c
@@ -0,0 +1,42 @@
1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dlmfsver.c
5 *
6 * version string
7 *
8 * Copyright (C) 2002, 2005 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26#include <linux/module.h>
27#include <linux/kernel.h>
28
29#include "dlmfsver.h"
30
31#define DLM_BUILD_VERSION "1.3.3"
32
33#define VERSION_STR "OCFS2 DLMFS " DLM_BUILD_VERSION
34
35void dlmfs_print_version(void)
36{
37 printk(KERN_INFO "%s\n", VERSION_STR);
38}
39
40MODULE_DESCRIPTION(VERSION_STR);
41
42MODULE_VERSION(DLM_BUILD_VERSION);
diff --git a/fs/ocfs2/dlm/dlmfsver.h b/fs/ocfs2/dlm/dlmfsver.h
new file mode 100644
index 000000000000..f35eadbed25c
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmfsver.h
@@ -0,0 +1,31 @@
1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dlmver.h
5 *
6 * Function prototypes
7 *
8 * Copyright (C) 2005 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26#ifndef DLMFS_VER_H
27#define DLMFS_VER_H
28
29void dlmfs_print_version(void);
30
31#endif /* DLMFS_VER_H */
diff --git a/fs/ocfs2/dlm/userdlm.c b/fs/ocfs2/dlm/userdlm.c
new file mode 100644
index 000000000000..e1fdd288796e
--- /dev/null
+++ b/fs/ocfs2/dlm/userdlm.c
@@ -0,0 +1,658 @@
1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * userdlm.c
5 *
6 * Code which implements the kernel side of a minimal userspace
7 * interface to our DLM.
8 *
9 * Many of the functions here are pared down versions of dlmglue.c
10 * functions.
11 *
12 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
13 *
14 * This program is free software; you can redistribute it and/or
15 * modify it under the terms of the GNU General Public
16 * License as published by the Free Software Foundation; either
17 * version 2 of the License, or (at your option) any later version.
18 *
19 * This program is distributed in the hope that it will be useful,
20 * but WITHOUT ANY WARRANTY; without even the implied warranty of
21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22 * General Public License for more details.
23 *
24 * You should have received a copy of the GNU General Public
25 * License along with this program; if not, write to the
26 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
27 * Boston, MA 021110-1307, USA.
28 */
29
30#include <asm/signal.h>
31
32#include <linux/module.h>
33#include <linux/fs.h>
34#include <linux/types.h>
35#include <linux/crc32.h>
36
37
38#include "cluster/nodemanager.h"
39#include "cluster/heartbeat.h"
40#include "cluster/tcp.h"
41
42#include "dlmapi.h"
43
44#include "userdlm.h"
45
46#define MLOG_MASK_PREFIX ML_DLMFS
47#include "cluster/masklog.h"
48
49static inline int user_check_wait_flag(struct user_lock_res *lockres,
50 int flag)
51{
52 int ret;
53
54 spin_lock(&lockres->l_lock);
55 ret = lockres->l_flags & flag;
56 spin_unlock(&lockres->l_lock);
57
58 return ret;
59}
60
61static inline void user_wait_on_busy_lock(struct user_lock_res *lockres)
62
63{
64 wait_event(lockres->l_event,
65 !user_check_wait_flag(lockres, USER_LOCK_BUSY));
66}
67
68static inline void user_wait_on_blocked_lock(struct user_lock_res *lockres)
69
70{
71 wait_event(lockres->l_event,
72 !user_check_wait_flag(lockres, USER_LOCK_BLOCKED));
73}
74
75/* I heart container_of... */
76static inline struct dlm_ctxt *
77dlm_ctxt_from_user_lockres(struct user_lock_res *lockres)
78{
79 struct dlmfs_inode_private *ip;
80
81 ip = container_of(lockres,
82 struct dlmfs_inode_private,
83 ip_lockres);
84 return ip->ip_dlm;
85}
86
87static struct inode *
88user_dlm_inode_from_user_lockres(struct user_lock_res *lockres)
89{
90 struct dlmfs_inode_private *ip;
91
92 ip = container_of(lockres,
93 struct dlmfs_inode_private,
94 ip_lockres);
95 return &ip->ip_vfs_inode;
96}
97
98static inline void user_recover_from_dlm_error(struct user_lock_res *lockres)
99{
100 spin_lock(&lockres->l_lock);
101 lockres->l_flags &= ~USER_LOCK_BUSY;
102 spin_unlock(&lockres->l_lock);
103}
104
105#define user_log_dlm_error(_func, _stat, _lockres) do { \
106 mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on " \
107 "resource %s: %s\n", dlm_errname(_stat), _func, \
108 _lockres->l_name, dlm_errmsg(_stat)); \
109} while (0)
110
111/* WARNING: This function lives in a world where the only three lock
112 * levels are EX, PR, and NL. It *will* have to be adjusted when more
113 * lock types are added. */
114static inline int user_highest_compat_lock_level(int level)
115{
116 int new_level = LKM_EXMODE;
117
118 if (level == LKM_EXMODE)
119 new_level = LKM_NLMODE;
120 else if (level == LKM_PRMODE)
121 new_level = LKM_PRMODE;
122 return new_level;
123}
124
125static void user_ast(void *opaque)
126{
127 struct user_lock_res *lockres = opaque;
128 struct dlm_lockstatus *lksb;
129
130 mlog(0, "AST fired for lockres %s\n", lockres->l_name);
131
132 spin_lock(&lockres->l_lock);
133
134 lksb = &(lockres->l_lksb);
135 if (lksb->status != DLM_NORMAL) {
136 mlog(ML_ERROR, "lksb status value of %u on lockres %s\n",
137 lksb->status, lockres->l_name);
138 spin_unlock(&lockres->l_lock);
139 return;
140 }
141
142 /* we're downconverting. */
143 if (lockres->l_requested < lockres->l_level) {
144 if (lockres->l_requested <=
145 user_highest_compat_lock_level(lockres->l_blocking)) {
146 lockres->l_blocking = LKM_NLMODE;
147 lockres->l_flags &= ~USER_LOCK_BLOCKED;
148 }
149 }
150
151 lockres->l_level = lockres->l_requested;
152 lockres->l_requested = LKM_IVMODE;
153 lockres->l_flags |= USER_LOCK_ATTACHED;
154 lockres->l_flags &= ~USER_LOCK_BUSY;
155
156 spin_unlock(&lockres->l_lock);
157
158 wake_up(&lockres->l_event);
159}
160
161static inline void user_dlm_grab_inode_ref(struct user_lock_res *lockres)
162{
163 struct inode *inode;
164 inode = user_dlm_inode_from_user_lockres(lockres);
165 if (!igrab(inode))
166 BUG();
167}
168
169static void user_dlm_unblock_lock(void *opaque);
170
171static void __user_dlm_queue_lockres(struct user_lock_res *lockres)
172{
173 if (!(lockres->l_flags & USER_LOCK_QUEUED)) {
174 user_dlm_grab_inode_ref(lockres);
175
176 INIT_WORK(&lockres->l_work, user_dlm_unblock_lock,
177 lockres);
178
179 queue_work(user_dlm_worker, &lockres->l_work);
180 lockres->l_flags |= USER_LOCK_QUEUED;
181 }
182}
183
184static void __user_dlm_cond_queue_lockres(struct user_lock_res *lockres)
185{
186 int queue = 0;
187
188 if (!(lockres->l_flags & USER_LOCK_BLOCKED))
189 return;
190
191 switch (lockres->l_blocking) {
192 case LKM_EXMODE:
193 if (!lockres->l_ex_holders && !lockres->l_ro_holders)
194 queue = 1;
195 break;
196 case LKM_PRMODE:
197 if (!lockres->l_ex_holders)
198 queue = 1;
199 break;
200 default:
201 BUG();
202 }
203
204 if (queue)
205 __user_dlm_queue_lockres(lockres);
206}
207
208static void user_bast(void *opaque, int level)
209{
210 struct user_lock_res *lockres = opaque;
211
212 mlog(0, "Blocking AST fired for lockres %s. Blocking level %d\n",
213 lockres->l_name, level);
214
215 spin_lock(&lockres->l_lock);
216 lockres->l_flags |= USER_LOCK_BLOCKED;
217 if (level > lockres->l_blocking)
218 lockres->l_blocking = level;
219
220 __user_dlm_queue_lockres(lockres);
221 spin_unlock(&lockres->l_lock);
222
223 wake_up(&lockres->l_event);
224}
225
226static void user_unlock_ast(void *opaque, enum dlm_status status)
227{
228 struct user_lock_res *lockres = opaque;
229
230 mlog(0, "UNLOCK AST called on lock %s\n", lockres->l_name);
231
232 if (status != DLM_NORMAL)
233 mlog(ML_ERROR, "Dlm returns status %d\n", status);
234
235 spin_lock(&lockres->l_lock);
236 if (lockres->l_flags & USER_LOCK_IN_TEARDOWN)
237 lockres->l_level = LKM_IVMODE;
238 else {
239 lockres->l_requested = LKM_IVMODE; /* cancel an
240 * upconvert
241 * request. */
242 lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
243 /* we want the unblock thread to look at it again
244 * now. */
245 __user_dlm_queue_lockres(lockres);
246 }
247
248 lockres->l_flags &= ~USER_LOCK_BUSY;
249 spin_unlock(&lockres->l_lock);
250
251 wake_up(&lockres->l_event);
252}
253
254static inline void user_dlm_drop_inode_ref(struct user_lock_res *lockres)
255{
256 struct inode *inode;
257 inode = user_dlm_inode_from_user_lockres(lockres);
258 iput(inode);
259}
260
261static void user_dlm_unblock_lock(void *opaque)
262{
263 int new_level, status;
264 struct user_lock_res *lockres = (struct user_lock_res *) opaque;
265 struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
266
267 mlog(0, "processing lockres %s\n", lockres->l_name);
268
269 spin_lock(&lockres->l_lock);
270
271 BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));
272 BUG_ON(!(lockres->l_flags & USER_LOCK_QUEUED));
273
274 /* notice that we don't clear USER_LOCK_BLOCKED here. That's
275 * for user_ast to do. */
276 lockres->l_flags &= ~USER_LOCK_QUEUED;
277
278 if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
279 mlog(0, "lock is in teardown so we do nothing\n");
280 spin_unlock(&lockres->l_lock);
281 goto drop_ref;
282 }
283
284 if (lockres->l_flags & USER_LOCK_BUSY) {
285 mlog(0, "BUSY flag detected...\n");
286 if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
287 spin_unlock(&lockres->l_lock);
288 goto drop_ref;
289 }
290
291 lockres->l_flags |= USER_LOCK_IN_CANCEL;
292 spin_unlock(&lockres->l_lock);
293
294 status = dlmunlock(dlm,
295 &lockres->l_lksb,
296 LKM_CANCEL,
297 user_unlock_ast,
298 lockres);
299 if (status == DLM_CANCELGRANT) {
300 /* If we got this, then the ast was fired
301 * before we could cancel. We cleanup our
302 * state, and restart the function. */
303 spin_lock(&lockres->l_lock);
304 lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
305 spin_unlock(&lockres->l_lock);
306 } else if (status != DLM_NORMAL)
307 user_log_dlm_error("dlmunlock", status, lockres);
308 goto drop_ref;
309 }
310
311 /* If there are still incompat holders, we can exit safely
312 * without worrying about re-queueing this lock as that will
313 * happen on the last call to user_cluster_unlock. */
314 if ((lockres->l_blocking == LKM_EXMODE)
315 && (lockres->l_ex_holders || lockres->l_ro_holders)) {
316 spin_unlock(&lockres->l_lock);
317 mlog(0, "can't downconvert for ex: ro = %u, ex = %u\n",
318 lockres->l_ro_holders, lockres->l_ex_holders);
319 goto drop_ref;
320 }
321
322 if ((lockres->l_blocking == LKM_PRMODE)
323 && lockres->l_ex_holders) {
324 spin_unlock(&lockres->l_lock);
325 mlog(0, "can't downconvert for pr: ex = %u\n",
326 lockres->l_ex_holders);
327 goto drop_ref;
328 }
329
330 /* yay, we can downconvert now. */
331 new_level = user_highest_compat_lock_level(lockres->l_blocking);
332 lockres->l_requested = new_level;
333 lockres->l_flags |= USER_LOCK_BUSY;
334 mlog(0, "Downconvert lock from %d to %d\n",
335 lockres->l_level, new_level);
336 spin_unlock(&lockres->l_lock);
337
338 /* need lock downconvert request now... */
339 status = dlmlock(dlm,
340 new_level,
341 &lockres->l_lksb,
342 LKM_CONVERT|LKM_VALBLK,
343 lockres->l_name,
344 user_ast,
345 lockres,
346 user_bast);
347 if (status != DLM_NORMAL) {
348 user_log_dlm_error("dlmlock", status, lockres);
349 user_recover_from_dlm_error(lockres);
350 }
351
352drop_ref:
353 user_dlm_drop_inode_ref(lockres);
354}
355
356static inline void user_dlm_inc_holders(struct user_lock_res *lockres,
357 int level)
358{
359 switch(level) {
360 case LKM_EXMODE:
361 lockres->l_ex_holders++;
362 break;
363 case LKM_PRMODE:
364 lockres->l_ro_holders++;
365 break;
366 default:
367 BUG();
368 }
369}
370
371/* predict what lock level we'll be dropping down to on behalf
372 * of another node, and return true if the currently wanted
373 * level will be compatible with it. */
374static inline int
375user_may_continue_on_blocked_lock(struct user_lock_res *lockres,
376 int wanted)
377{
378 BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));
379
380 return wanted <= user_highest_compat_lock_level(lockres->l_blocking);
381}
382
383int user_dlm_cluster_lock(struct user_lock_res *lockres,
384 int level,
385 int lkm_flags)
386{
387 int status, local_flags;
388 struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
389
390 if (level != LKM_EXMODE &&
391 level != LKM_PRMODE) {
392 mlog(ML_ERROR, "lockres %s: invalid request!\n",
393 lockres->l_name);
394 status = -EINVAL;
395 goto bail;
396 }
397
398 mlog(0, "lockres %s: asking for %s lock, passed flags = 0x%x\n",
399 lockres->l_name,
400 (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE",
401 lkm_flags);
402
403again:
404 if (signal_pending(current)) {
405 status = -ERESTARTSYS;
406 goto bail;
407 }
408
409 spin_lock(&lockres->l_lock);
410
411 /* We only compare against the currently granted level
412 * here. If the lock is blocked waiting on a downconvert,
413 * we'll get caught below. */
414 if ((lockres->l_flags & USER_LOCK_BUSY) &&
415 (level > lockres->l_level)) {
416 /* is someone sitting in dlm_lock? If so, wait on
417 * them. */
418 spin_unlock(&lockres->l_lock);
419
420 user_wait_on_busy_lock(lockres);
421 goto again;
422 }
423
424 if ((lockres->l_flags & USER_LOCK_BLOCKED) &&
425 (!user_may_continue_on_blocked_lock(lockres, level))) {
426 /* is the lock is currently blocked on behalf of
427 * another node */
428 spin_unlock(&lockres->l_lock);
429
430 user_wait_on_blocked_lock(lockres);
431 goto again;
432 }
433
434 if (level > lockres->l_level) {
435 local_flags = lkm_flags | LKM_VALBLK;
436 if (lockres->l_level != LKM_IVMODE)
437 local_flags |= LKM_CONVERT;
438
439 lockres->l_requested = level;
440 lockres->l_flags |= USER_LOCK_BUSY;
441 spin_unlock(&lockres->l_lock);
442
443 BUG_ON(level == LKM_IVMODE);
444 BUG_ON(level == LKM_NLMODE);
445
446 mlog(0, "lock %s, get lock from %d to level = %d\n",
447 lockres->l_name, lockres->l_level, level);
448
449 /* call dlm_lock to upgrade lock now */
450 status = dlmlock(dlm,
451 level,
452 &lockres->l_lksb,
453 local_flags,
454 lockres->l_name,
455 user_ast,
456 lockres,
457 user_bast);
458 if (status != DLM_NORMAL) {
459 if ((lkm_flags & LKM_NOQUEUE) &&
460 (status == DLM_NOTQUEUED))
461 status = -EAGAIN;
462 else {
463 user_log_dlm_error("dlmlock", status, lockres);
464 status = -EINVAL;
465 }
466 user_recover_from_dlm_error(lockres);
467 goto bail;
468 }
469
470 mlog(0, "lock %s, successfull return from dlmlock\n",
471 lockres->l_name);
472
473 user_wait_on_busy_lock(lockres);
474 goto again;
475 }
476
477 user_dlm_inc_holders(lockres, level);
478 spin_unlock(&lockres->l_lock);
479
480 mlog(0, "lockres %s: Got %s lock!\n", lockres->l_name,
481 (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE");
482
483 status = 0;
484bail:
485 return status;
486}
487
488static inline void user_dlm_dec_holders(struct user_lock_res *lockres,
489 int level)
490{
491 switch(level) {
492 case LKM_EXMODE:
493 BUG_ON(!lockres->l_ex_holders);
494 lockres->l_ex_holders--;
495 break;
496 case LKM_PRMODE:
497 BUG_ON(!lockres->l_ro_holders);
498 lockres->l_ro_holders--;
499 break;
500 default:
501 BUG();
502 }
503}
504
505void user_dlm_cluster_unlock(struct user_lock_res *lockres,
506 int level)
507{
508 if (level != LKM_EXMODE &&
509 level != LKM_PRMODE) {
510 mlog(ML_ERROR, "lockres %s: invalid request!\n", lockres->l_name);
511 return;
512 }
513
514 mlog(0, "lockres %s: dropping %s lock\n", lockres->l_name,
515 (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE");
516
517 spin_lock(&lockres->l_lock);
518 user_dlm_dec_holders(lockres, level);
519 __user_dlm_cond_queue_lockres(lockres);
520 spin_unlock(&lockres->l_lock);
521}
522
523void user_dlm_write_lvb(struct inode *inode,
524 const char *val,
525 unsigned int len)
526{
527 struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
528 char *lvb = lockres->l_lksb.lvb;
529
530 BUG_ON(len > DLM_LVB_LEN);
531
532 spin_lock(&lockres->l_lock);
533
534 BUG_ON(lockres->l_level < LKM_EXMODE);
535 memcpy(lvb, val, len);
536
537 spin_unlock(&lockres->l_lock);
538}
539
540void user_dlm_read_lvb(struct inode *inode,
541 char *val,
542 unsigned int len)
543{
544 struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
545 char *lvb = lockres->l_lksb.lvb;
546
547 BUG_ON(len > DLM_LVB_LEN);
548
549 spin_lock(&lockres->l_lock);
550
551 BUG_ON(lockres->l_level < LKM_PRMODE);
552 memcpy(val, lvb, len);
553
554 spin_unlock(&lockres->l_lock);
555}
556
557void user_dlm_lock_res_init(struct user_lock_res *lockres,
558 struct dentry *dentry)
559{
560 memset(lockres, 0, sizeof(*lockres));
561
562 spin_lock_init(&lockres->l_lock);
563 init_waitqueue_head(&lockres->l_event);
564 lockres->l_level = LKM_IVMODE;
565 lockres->l_requested = LKM_IVMODE;
566 lockres->l_blocking = LKM_IVMODE;
567
568 /* should have been checked before getting here. */
569 BUG_ON(dentry->d_name.len >= USER_DLM_LOCK_ID_MAX_LEN);
570
571 memcpy(lockres->l_name,
572 dentry->d_name.name,
573 dentry->d_name.len);
574}
575
576int user_dlm_destroy_lock(struct user_lock_res *lockres)
577{
578 int status = -EBUSY;
579 struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
580
581 mlog(0, "asked to destroy %s\n", lockres->l_name);
582
583 spin_lock(&lockres->l_lock);
584 while (lockres->l_flags & USER_LOCK_BUSY) {
585 spin_unlock(&lockres->l_lock);
586
587 mlog(0, "lock %s is busy\n", lockres->l_name);
588
589 user_wait_on_busy_lock(lockres);
590
591 spin_lock(&lockres->l_lock);
592 }
593
594 if (lockres->l_ro_holders || lockres->l_ex_holders) {
595 spin_unlock(&lockres->l_lock);
596 mlog(0, "lock %s has holders\n", lockres->l_name);
597 goto bail;
598 }
599
600 status = 0;
601 if (!(lockres->l_flags & USER_LOCK_ATTACHED)) {
602 spin_unlock(&lockres->l_lock);
603 mlog(0, "lock %s is not attached\n", lockres->l_name);
604 goto bail;
605 }
606
607 lockres->l_flags &= ~USER_LOCK_ATTACHED;
608 lockres->l_flags |= USER_LOCK_BUSY;
609 lockres->l_flags |= USER_LOCK_IN_TEARDOWN;
610 spin_unlock(&lockres->l_lock);
611
612 mlog(0, "unlocking lockres %s\n", lockres->l_name);
613 status = dlmunlock(dlm,
614 &lockres->l_lksb,
615 LKM_VALBLK,
616 user_unlock_ast,
617 lockres);
618 if (status != DLM_NORMAL) {
619 user_log_dlm_error("dlmunlock", status, lockres);
620 status = -EINVAL;
621 goto bail;
622 }
623
624 user_wait_on_busy_lock(lockres);
625
626 status = 0;
627bail:
628 return status;
629}
630
631struct dlm_ctxt *user_dlm_register_context(struct qstr *name)
632{
633 struct dlm_ctxt *dlm;
634 u32 dlm_key;
635 char *domain;
636
637 domain = kmalloc(name->len + 1, GFP_KERNEL);
638 if (!domain) {
639 mlog_errno(-ENOMEM);
640 return ERR_PTR(-ENOMEM);
641 }
642
643 dlm_key = crc32_le(0, name->name, name->len);
644
645 snprintf(domain, name->len + 1, "%.*s", name->len, name->name);
646
647 dlm = dlm_register_domain(domain, dlm_key);
648 if (IS_ERR(dlm))
649 mlog_errno(PTR_ERR(dlm));
650
651 kfree(domain);
652 return dlm;
653}
654
655void user_dlm_unregister_context(struct dlm_ctxt *dlm)
656{
657 dlm_unregister_domain(dlm);
658}
diff --git a/fs/ocfs2/dlm/userdlm.h b/fs/ocfs2/dlm/userdlm.h
new file mode 100644
index 000000000000..04178bc40b76
--- /dev/null
+++ b/fs/ocfs2/dlm/userdlm.h
@@ -0,0 +1,111 @@
1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * userdlm.h
5 *
6 * Userspace dlm defines
7 *
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26
27#ifndef USERDLM_H
28#define USERDLM_H
29
30#include <linux/module.h>
31#include <linux/fs.h>
32#include <linux/types.h>
33#include <linux/workqueue.h>
34
35/* user_lock_res->l_flags flags. */
36#define USER_LOCK_ATTACHED (0x00000001) /* have we initialized
37 * the lvb */
38#define USER_LOCK_BUSY (0x00000002) /* we are currently in
39 * dlm_lock */
40#define USER_LOCK_BLOCKED (0x00000004) /* blocked waiting to
41 * downconvert*/
42#define USER_LOCK_IN_TEARDOWN (0x00000008) /* we're currently
43 * destroying this
44 * lock. */
45#define USER_LOCK_QUEUED (0x00000010) /* lock is on the
46 * workqueue */
47#define USER_LOCK_IN_CANCEL (0x00000020)
48
49struct user_lock_res {
50 spinlock_t l_lock;
51
52 int l_flags;
53
54#define USER_DLM_LOCK_ID_MAX_LEN 32
55 char l_name[USER_DLM_LOCK_ID_MAX_LEN];
56 int l_level;
57 unsigned int l_ro_holders;
58 unsigned int l_ex_holders;
59 struct dlm_lockstatus l_lksb;
60
61 int l_requested;
62 int l_blocking;
63
64 wait_queue_head_t l_event;
65
66 struct work_struct l_work;
67};
68
69extern struct workqueue_struct *user_dlm_worker;
70
71void user_dlm_lock_res_init(struct user_lock_res *lockres,
72 struct dentry *dentry);
73int user_dlm_destroy_lock(struct user_lock_res *lockres);
74int user_dlm_cluster_lock(struct user_lock_res *lockres,
75 int level,
76 int lkm_flags);
77void user_dlm_cluster_unlock(struct user_lock_res *lockres,
78 int level);
79void user_dlm_write_lvb(struct inode *inode,
80 const char *val,
81 unsigned int len);
82void user_dlm_read_lvb(struct inode *inode,
83 char *val,
84 unsigned int len);
85struct dlm_ctxt *user_dlm_register_context(struct qstr *name);
86void user_dlm_unregister_context(struct dlm_ctxt *dlm);
87
88struct dlmfs_inode_private {
89 struct dlm_ctxt *ip_dlm;
90
91 struct user_lock_res ip_lockres; /* unused for directories. */
92 struct inode *ip_parent;
93
94 struct inode ip_vfs_inode;
95};
96
97static inline struct dlmfs_inode_private *
98DLMFS_I(struct inode *inode)
99{
100 return container_of(inode,
101 struct dlmfs_inode_private,
102 ip_vfs_inode);
103}
104
105struct dlmfs_filp_private {
106 int fp_lock_level;
107};
108
109#define DLMFS_MAGIC 0x76a9f425
110
111#endif /* USERDLM_H */
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-17 03:34:33 -0500