aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ocfs2/dlmfs
diff options
context:
space:
mode:
Diffstat (limited to 'fs/ocfs2/dlmfs')
-rw-r--r--fs/ocfs2/dlmfs/Makefile5
-rw-r--r--fs/ocfs2/dlmfs/dlmfs.c715
-rw-r--r--fs/ocfs2/dlmfs/dlmfsver.c42
-rw-r--r--fs/ocfs2/dlmfs/dlmfsver.h31
-rw-r--r--fs/ocfs2/dlmfs/userdlm.c688
-rw-r--r--fs/ocfs2/dlmfs/userdlm.h113
6 files changed, 1594 insertions, 0 deletions
diff --git a/fs/ocfs2/dlmfs/Makefile b/fs/ocfs2/dlmfs/Makefile
new file mode 100644
index 000000000000..df69b4856d0d
--- /dev/null
+++ b/fs/ocfs2/dlmfs/Makefile
@@ -0,0 +1,5 @@
1EXTRA_CFLAGS += -Ifs/ocfs2
2
3obj-$(CONFIG_OCFS2_FS) += ocfs2_dlmfs.o
4
5ocfs2_dlmfs-objs := userdlm.o dlmfs.o dlmfsver.o
diff --git a/fs/ocfs2/dlmfs/dlmfs.c b/fs/ocfs2/dlmfs/dlmfs.c
new file mode 100644
index 000000000000..1b0de157a08c
--- /dev/null
+++ b/fs/ocfs2/dlmfs/dlmfs.c
@@ -0,0 +1,715 @@
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#include <linux/poll.h>
47
48#include <asm/uaccess.h>
49
50#include "stackglue.h"
51#include "userdlm.h"
52#include "dlmfsver.h"
53
54#define MLOG_MASK_PREFIX ML_DLMFS
55#include "cluster/masklog.h"
56
57
58static const struct super_operations dlmfs_ops;
59static const struct file_operations dlmfs_file_operations;
60static 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;
63static struct kmem_cache *dlmfs_inode_cache;
64
65struct workqueue_struct *user_dlm_worker;
66
67
68
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.
85 *
86 * Capabilities:
87 * - bast : POLLIN against the file descriptor of a held lock
88 * signifies a bast fired on the lock.
89 */
90#define DLMFS_CAPABILITIES "bast stackglue"
91extern int param_set_dlmfs_capabilities(const char *val,
92 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
108/*
109 * 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 *
115 * O_NONBLOCK -> LKM_NOQUEUE
116 */
117static int dlmfs_decode_open_flags(int open_flags,
118 int *level,
119 int *flags)
120{
121 if (open_flags & (O_WRONLY|O_RDWR))
122 *level = LKM_EXMODE;
123 else
124 *level = LKM_PRMODE;
125
126 *flags = 0;
127 if (open_flags & O_NONBLOCK)
128 *flags |= LKM_NOQUEUE;
129
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
154 fp = kmalloc(sizeof(*fp), GFP_NOFS);
155 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. */
169 if (flags & LKM_NOQUEUE && status == -EAGAIN)
170 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);
185 struct dlmfs_filp_private *fp =
186 (struct dlmfs_filp_private *) file->private_data;
187
188 if (S_ISDIR(inode->i_mode))
189 BUG();
190
191 mlog(0, "close called on inode %lu\n", inode->i_ino);
192
193 status = 0;
194 if (fp) {
195 level = fp->fp_lock_level;
196 if (level != LKM_IVMODE)
197 user_dlm_cluster_unlock(&ip->ip_lockres, level);
198
199 kfree(fp);
200 file->private_data = NULL;
201 }
202
203 return 0;
204}
205
206/*
207 * We do ->setattr() just to override size changes. Our size is the size
208 * of the LVB and nothing else.
209 */
210static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr)
211{
212 int error;
213 struct inode *inode = dentry->d_inode;
214
215 attr->ia_valid &= ~ATTR_SIZE;
216 error = inode_change_ok(inode, attr);
217 if (!error)
218 error = inode_setattr(inode, attr);
219
220 return error;
221}
222
223static unsigned int dlmfs_file_poll(struct file *file, poll_table *wait)
224{
225 int event = 0;
226 struct inode *inode = file->f_path.dentry->d_inode;
227 struct dlmfs_inode_private *ip = DLMFS_I(inode);
228
229 poll_wait(file, &ip->ip_lockres.l_event, wait);
230
231 spin_lock(&ip->ip_lockres.l_lock);
232 if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED)
233 event = POLLIN | POLLRDNORM;
234 spin_unlock(&ip->ip_lockres.l_lock);
235
236 return event;
237}
238
239static ssize_t dlmfs_file_read(struct file *filp,
240 char __user *buf,
241 size_t count,
242 loff_t *ppos)
243{
244 int bytes_left;
245 ssize_t readlen, got;
246 char *lvb_buf;
247 struct inode *inode = filp->f_path.dentry->d_inode;
248
249 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
250 inode->i_ino, count, *ppos);
251
252 if (*ppos >= i_size_read(inode))
253 return 0;
254
255 if (!count)
256 return 0;
257
258 if (!access_ok(VERIFY_WRITE, buf, count))
259 return -EFAULT;
260
261 /* don't read past the lvb */
262 if ((count + *ppos) > i_size_read(inode))
263 readlen = i_size_read(inode) - *ppos;
264 else
265 readlen = count - *ppos;
266
267 lvb_buf = kmalloc(readlen, GFP_NOFS);
268 if (!lvb_buf)
269 return -ENOMEM;
270
271 got = user_dlm_read_lvb(inode, lvb_buf, readlen);
272 if (got) {
273 BUG_ON(got != readlen);
274 bytes_left = __copy_to_user(buf, lvb_buf, readlen);
275 readlen -= bytes_left;
276 } else
277 readlen = 0;
278
279 kfree(lvb_buf);
280
281 *ppos = *ppos + readlen;
282
283 mlog(0, "read %zd bytes\n", readlen);
284 return readlen;
285}
286
287static ssize_t dlmfs_file_write(struct file *filp,
288 const char __user *buf,
289 size_t count,
290 loff_t *ppos)
291{
292 int bytes_left;
293 ssize_t writelen;
294 char *lvb_buf;
295 struct inode *inode = filp->f_path.dentry->d_inode;
296
297 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
298 inode->i_ino, count, *ppos);
299
300 if (*ppos >= i_size_read(inode))
301 return -ENOSPC;
302
303 if (!count)
304 return 0;
305
306 if (!access_ok(VERIFY_READ, buf, count))
307 return -EFAULT;
308
309 /* don't write past the lvb */
310 if ((count + *ppos) > i_size_read(inode))
311 writelen = i_size_read(inode) - *ppos;
312 else
313 writelen = count - *ppos;
314
315 lvb_buf = kmalloc(writelen, GFP_NOFS);
316 if (!lvb_buf)
317 return -ENOMEM;
318
319 bytes_left = copy_from_user(lvb_buf, buf, writelen);
320 writelen -= bytes_left;
321 if (writelen)
322 user_dlm_write_lvb(inode, lvb_buf, writelen);
323
324 kfree(lvb_buf);
325
326 *ppos = *ppos + writelen;
327 mlog(0, "wrote %zd bytes\n", writelen);
328 return writelen;
329}
330
331static void dlmfs_init_once(void *foo)
332{
333 struct dlmfs_inode_private *ip =
334 (struct dlmfs_inode_private *) foo;
335
336 ip->ip_conn = NULL;
337 ip->ip_parent = NULL;
338
339 inode_init_once(&ip->ip_vfs_inode);
340}
341
342static struct inode *dlmfs_alloc_inode(struct super_block *sb)
343{
344 struct dlmfs_inode_private *ip;
345
346 ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
347 if (!ip)
348 return NULL;
349
350 return &ip->ip_vfs_inode;
351}
352
353static void dlmfs_destroy_inode(struct inode *inode)
354{
355 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
356}
357
358static void dlmfs_clear_inode(struct inode *inode)
359{
360 int status;
361 struct dlmfs_inode_private *ip;
362
363 if (!inode)
364 return;
365
366 mlog(0, "inode %lu\n", inode->i_ino);
367
368 ip = DLMFS_I(inode);
369
370 if (S_ISREG(inode->i_mode)) {
371 status = user_dlm_destroy_lock(&ip->ip_lockres);
372 if (status < 0)
373 mlog_errno(status);
374 iput(ip->ip_parent);
375 goto clear_fields;
376 }
377
378 mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn);
379 /* we must be a directory. If required, lets unregister the
380 * dlm context now. */
381 if (ip->ip_conn)
382 user_dlm_unregister(ip->ip_conn);
383clear_fields:
384 ip->ip_parent = NULL;
385 ip->ip_conn = NULL;
386}
387
388static struct backing_dev_info dlmfs_backing_dev_info = {
389 .name = "ocfs2-dlmfs",
390 .ra_pages = 0, /* No readahead */
391 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
392};
393
394static struct inode *dlmfs_get_root_inode(struct super_block *sb)
395{
396 struct inode *inode = new_inode(sb);
397 int mode = S_IFDIR | 0755;
398 struct dlmfs_inode_private *ip;
399
400 if (inode) {
401 ip = DLMFS_I(inode);
402
403 inode->i_mode = mode;
404 inode->i_uid = current_fsuid();
405 inode->i_gid = current_fsgid();
406 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
407 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
408 inc_nlink(inode);
409
410 inode->i_fop = &simple_dir_operations;
411 inode->i_op = &dlmfs_root_inode_operations;
412 }
413
414 return inode;
415}
416
417static struct inode *dlmfs_get_inode(struct inode *parent,
418 struct dentry *dentry,
419 int mode)
420{
421 struct super_block *sb = parent->i_sb;
422 struct inode * inode = new_inode(sb);
423 struct dlmfs_inode_private *ip;
424
425 if (!inode)
426 return NULL;
427
428 inode->i_mode = mode;
429 inode->i_uid = current_fsuid();
430 inode->i_gid = current_fsgid();
431 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
432 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
433
434 ip = DLMFS_I(inode);
435 ip->ip_conn = DLMFS_I(parent)->ip_conn;
436
437 switch (mode & S_IFMT) {
438 default:
439 /* for now we don't support anything other than
440 * directories and regular files. */
441 BUG();
442 break;
443 case S_IFREG:
444 inode->i_op = &dlmfs_file_inode_operations;
445 inode->i_fop = &dlmfs_file_operations;
446
447 i_size_write(inode, DLM_LVB_LEN);
448
449 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
450
451 /* released at clear_inode time, this insures that we
452 * get to drop the dlm reference on each lock *before*
453 * we call the unregister code for releasing parent
454 * directories. */
455 ip->ip_parent = igrab(parent);
456 BUG_ON(!ip->ip_parent);
457 break;
458 case S_IFDIR:
459 inode->i_op = &dlmfs_dir_inode_operations;
460 inode->i_fop = &simple_dir_operations;
461
462 /* directory inodes start off with i_nlink ==
463 * 2 (for "." entry) */
464 inc_nlink(inode);
465 break;
466 }
467
468 if (parent->i_mode & S_ISGID) {
469 inode->i_gid = parent->i_gid;
470 if (S_ISDIR(mode))
471 inode->i_mode |= S_ISGID;
472 }
473
474 return inode;
475}
476
477/*
478 * File creation. Allocate an inode, and we're done..
479 */
480/* SMP-safe */
481static int dlmfs_mkdir(struct inode * dir,
482 struct dentry * dentry,
483 int mode)
484{
485 int status;
486 struct inode *inode = NULL;
487 struct qstr *domain = &dentry->d_name;
488 struct dlmfs_inode_private *ip;
489 struct ocfs2_cluster_connection *conn;
490
491 mlog(0, "mkdir %.*s\n", domain->len, domain->name);
492
493 /* verify that we have a proper domain */
494 if (domain->len >= GROUP_NAME_MAX) {
495 status = -EINVAL;
496 mlog(ML_ERROR, "invalid domain name for directory.\n");
497 goto bail;
498 }
499
500 inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
501 if (!inode) {
502 status = -ENOMEM;
503 mlog_errno(status);
504 goto bail;
505 }
506
507 ip = DLMFS_I(inode);
508
509 conn = user_dlm_register(domain);
510 if (IS_ERR(conn)) {
511 status = PTR_ERR(conn);
512 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
513 status, domain->len, domain->name);
514 goto bail;
515 }
516 ip->ip_conn = conn;
517
518 inc_nlink(dir);
519 d_instantiate(dentry, inode);
520 dget(dentry); /* Extra count - pin the dentry in core */
521
522 status = 0;
523bail:
524 if (status < 0)
525 iput(inode);
526 return status;
527}
528
529static int dlmfs_create(struct inode *dir,
530 struct dentry *dentry,
531 int mode,
532 struct nameidata *nd)
533{
534 int status = 0;
535 struct inode *inode;
536 struct qstr *name = &dentry->d_name;
537
538 mlog(0, "create %.*s\n", name->len, name->name);
539
540 /* verify name is valid and doesn't contain any dlm reserved
541 * characters */
542 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
543 name->name[0] == '$') {
544 status = -EINVAL;
545 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
546 name->name);
547 goto bail;
548 }
549
550 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
551 if (!inode) {
552 status = -ENOMEM;
553 mlog_errno(status);
554 goto bail;
555 }
556
557 d_instantiate(dentry, inode);
558 dget(dentry); /* Extra count - pin the dentry in core */
559bail:
560 return status;
561}
562
563static int dlmfs_unlink(struct inode *dir,
564 struct dentry *dentry)
565{
566 int status;
567 struct inode *inode = dentry->d_inode;
568
569 mlog(0, "unlink inode %lu\n", inode->i_ino);
570
571 /* if there are no current holders, or none that are waiting
572 * to acquire a lock, this basically destroys our lockres. */
573 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
574 if (status < 0) {
575 mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
576 dentry->d_name.len, dentry->d_name.name, status);
577 goto bail;
578 }
579 status = simple_unlink(dir, dentry);
580bail:
581 return status;
582}
583
584static int dlmfs_fill_super(struct super_block * sb,
585 void * data,
586 int silent)
587{
588 struct inode * inode;
589 struct dentry * root;
590
591 sb->s_maxbytes = MAX_LFS_FILESIZE;
592 sb->s_blocksize = PAGE_CACHE_SIZE;
593 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
594 sb->s_magic = DLMFS_MAGIC;
595 sb->s_op = &dlmfs_ops;
596 inode = dlmfs_get_root_inode(sb);
597 if (!inode)
598 return -ENOMEM;
599
600 root = d_alloc_root(inode);
601 if (!root) {
602 iput(inode);
603 return -ENOMEM;
604 }
605 sb->s_root = root;
606 return 0;
607}
608
609static const struct file_operations dlmfs_file_operations = {
610 .open = dlmfs_file_open,
611 .release = dlmfs_file_release,
612 .poll = dlmfs_file_poll,
613 .read = dlmfs_file_read,
614 .write = dlmfs_file_write,
615};
616
617static const struct inode_operations dlmfs_dir_inode_operations = {
618 .create = dlmfs_create,
619 .lookup = simple_lookup,
620 .unlink = dlmfs_unlink,
621};
622
623/* this way we can restrict mkdir to only the toplevel of the fs. */
624static const struct inode_operations dlmfs_root_inode_operations = {
625 .lookup = simple_lookup,
626 .mkdir = dlmfs_mkdir,
627 .rmdir = simple_rmdir,
628};
629
630static const struct super_operations dlmfs_ops = {
631 .statfs = simple_statfs,
632 .alloc_inode = dlmfs_alloc_inode,
633 .destroy_inode = dlmfs_destroy_inode,
634 .clear_inode = dlmfs_clear_inode,
635 .drop_inode = generic_delete_inode,
636};
637
638static const struct inode_operations dlmfs_file_inode_operations = {
639 .getattr = simple_getattr,
640 .setattr = dlmfs_file_setattr,
641};
642
643static int dlmfs_get_sb(struct file_system_type *fs_type,
644 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
645{
646 return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super, mnt);
647}
648
649static struct file_system_type dlmfs_fs_type = {
650 .owner = THIS_MODULE,
651 .name = "ocfs2_dlmfs",
652 .get_sb = dlmfs_get_sb,
653 .kill_sb = kill_litter_super,
654};
655
656static int __init init_dlmfs_fs(void)
657{
658 int status;
659 int cleanup_inode = 0, cleanup_worker = 0;
660
661 dlmfs_print_version();
662
663 status = bdi_init(&dlmfs_backing_dev_info);
664 if (status)
665 return status;
666
667 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
668 sizeof(struct dlmfs_inode_private),
669 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
670 SLAB_MEM_SPREAD),
671 dlmfs_init_once);
672 if (!dlmfs_inode_cache) {
673 status = -ENOMEM;
674 goto bail;
675 }
676 cleanup_inode = 1;
677
678 user_dlm_worker = create_singlethread_workqueue("user_dlm");
679 if (!user_dlm_worker) {
680 status = -ENOMEM;
681 goto bail;
682 }
683 cleanup_worker = 1;
684
685 user_dlm_set_locking_protocol();
686 status = register_filesystem(&dlmfs_fs_type);
687bail:
688 if (status) {
689 if (cleanup_inode)
690 kmem_cache_destroy(dlmfs_inode_cache);
691 if (cleanup_worker)
692 destroy_workqueue(user_dlm_worker);
693 bdi_destroy(&dlmfs_backing_dev_info);
694 } else
695 printk("OCFS2 User DLM kernel interface loaded\n");
696 return status;
697}
698
699static void __exit exit_dlmfs_fs(void)
700{
701 unregister_filesystem(&dlmfs_fs_type);
702
703 flush_workqueue(user_dlm_worker);
704 destroy_workqueue(user_dlm_worker);
705
706 kmem_cache_destroy(dlmfs_inode_cache);
707
708 bdi_destroy(&dlmfs_backing_dev_info);
709}
710
711MODULE_AUTHOR("Oracle");
712MODULE_LICENSE("GPL");
713
714module_init(init_dlmfs_fs)
715module_exit(exit_dlmfs_fs)
diff --git a/fs/ocfs2/dlmfs/dlmfsver.c b/fs/ocfs2/dlmfs/dlmfsver.c
new file mode 100644
index 000000000000..a733b3321f83
--- /dev/null
+++ b/fs/ocfs2/dlmfs/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 "1.5.0"
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/dlmfs/dlmfsver.h b/fs/ocfs2/dlmfs/dlmfsver.h
new file mode 100644
index 000000000000..f35eadbed25c
--- /dev/null
+++ b/fs/ocfs2/dlmfs/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/dlmfs/userdlm.c b/fs/ocfs2/dlmfs/userdlm.c
new file mode 100644
index 000000000000..0499e3fb7bdb
--- /dev/null
+++ b/fs/ocfs2/dlmfs/userdlm.c
@@ -0,0 +1,688 @@
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 <linux/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#include "ocfs2_lockingver.h"
38#include "stackglue.h"
39#include "userdlm.h"
40
41#define MLOG_MASK_PREFIX ML_DLMFS
42#include "cluster/masklog.h"
43
44
45static inline struct user_lock_res *user_lksb_to_lock_res(struct ocfs2_dlm_lksb *lksb)
46{
47 return container_of(lksb, struct user_lock_res, l_lksb);
48}
49
50static inline int user_check_wait_flag(struct user_lock_res *lockres,
51 int flag)
52{
53 int ret;
54
55 spin_lock(&lockres->l_lock);
56 ret = lockres->l_flags & flag;
57 spin_unlock(&lockres->l_lock);
58
59 return ret;
60}
61
62static inline void user_wait_on_busy_lock(struct user_lock_res *lockres)
63
64{
65 wait_event(lockres->l_event,
66 !user_check_wait_flag(lockres, USER_LOCK_BUSY));
67}
68
69static inline void user_wait_on_blocked_lock(struct user_lock_res *lockres)
70
71{
72 wait_event(lockres->l_event,
73 !user_check_wait_flag(lockres, USER_LOCK_BLOCKED));
74}
75
76/* I heart container_of... */
77static inline struct ocfs2_cluster_connection *
78cluster_connection_from_user_lockres(struct user_lock_res *lockres)
79{
80 struct dlmfs_inode_private *ip;
81
82 ip = container_of(lockres,
83 struct dlmfs_inode_private,
84 ip_lockres);
85 return ip->ip_conn;
86}
87
88static struct inode *
89user_dlm_inode_from_user_lockres(struct user_lock_res *lockres)
90{
91 struct dlmfs_inode_private *ip;
92
93 ip = container_of(lockres,
94 struct dlmfs_inode_private,
95 ip_lockres);
96 return &ip->ip_vfs_inode;
97}
98
99static inline void user_recover_from_dlm_error(struct user_lock_res *lockres)
100{
101 spin_lock(&lockres->l_lock);
102 lockres->l_flags &= ~USER_LOCK_BUSY;
103 spin_unlock(&lockres->l_lock);
104}
105
106#define user_log_dlm_error(_func, _stat, _lockres) do { \
107 mlog(ML_ERROR, "Dlm error %d while calling %s on " \
108 "resource %.*s\n", _stat, _func, \
109 _lockres->l_namelen, _lockres->l_name); \
110} while (0)
111
112/* WARNING: This function lives in a world where the only three lock
113 * levels are EX, PR, and NL. It *will* have to be adjusted when more
114 * lock types are added. */
115static inline int user_highest_compat_lock_level(int level)
116{
117 int new_level = DLM_LOCK_EX;
118
119 if (level == DLM_LOCK_EX)
120 new_level = DLM_LOCK_NL;
121 else if (level == DLM_LOCK_PR)
122 new_level = DLM_LOCK_PR;
123 return new_level;
124}
125
126static void user_ast(struct ocfs2_dlm_lksb *lksb)
127{
128 struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
129 int status;
130
131 mlog(ML_BASTS, "AST fired for lockres %.*s, level %d => %d\n",
132 lockres->l_namelen, lockres->l_name, lockres->l_level,
133 lockres->l_requested);
134
135 spin_lock(&lockres->l_lock);
136
137 status = ocfs2_dlm_lock_status(&lockres->l_lksb);
138 if (status) {
139 mlog(ML_ERROR, "lksb status value of %u on lockres %.*s\n",
140 status, lockres->l_namelen, lockres->l_name);
141 spin_unlock(&lockres->l_lock);
142 return;
143 }
144
145 mlog_bug_on_msg(lockres->l_requested == DLM_LOCK_IV,
146 "Lockres %.*s, requested ivmode. flags 0x%x\n",
147 lockres->l_namelen, lockres->l_name, lockres->l_flags);
148
149 /* we're downconverting. */
150 if (lockres->l_requested < lockres->l_level) {
151 if (lockres->l_requested <=
152 user_highest_compat_lock_level(lockres->l_blocking)) {
153 lockres->l_blocking = DLM_LOCK_NL;
154 lockres->l_flags &= ~USER_LOCK_BLOCKED;
155 }
156 }
157
158 lockres->l_level = lockres->l_requested;
159 lockres->l_requested = DLM_LOCK_IV;
160 lockres->l_flags |= USER_LOCK_ATTACHED;
161 lockres->l_flags &= ~USER_LOCK_BUSY;
162
163 spin_unlock(&lockres->l_lock);
164
165 wake_up(&lockres->l_event);
166}
167
168static inline void user_dlm_grab_inode_ref(struct user_lock_res *lockres)
169{
170 struct inode *inode;
171 inode = user_dlm_inode_from_user_lockres(lockres);
172 if (!igrab(inode))
173 BUG();
174}
175
176static void user_dlm_unblock_lock(struct work_struct *work);
177
178static void __user_dlm_queue_lockres(struct user_lock_res *lockres)
179{
180 if (!(lockres->l_flags & USER_LOCK_QUEUED)) {
181 user_dlm_grab_inode_ref(lockres);
182
183 INIT_WORK(&lockres->l_work, user_dlm_unblock_lock);
184
185 queue_work(user_dlm_worker, &lockres->l_work);
186 lockres->l_flags |= USER_LOCK_QUEUED;
187 }
188}
189
190static void __user_dlm_cond_queue_lockres(struct user_lock_res *lockres)
191{
192 int queue = 0;
193
194 if (!(lockres->l_flags & USER_LOCK_BLOCKED))
195 return;
196
197 switch (lockres->l_blocking) {
198 case DLM_LOCK_EX:
199 if (!lockres->l_ex_holders && !lockres->l_ro_holders)
200 queue = 1;
201 break;
202 case DLM_LOCK_PR:
203 if (!lockres->l_ex_holders)
204 queue = 1;
205 break;
206 default:
207 BUG();
208 }
209
210 if (queue)
211 __user_dlm_queue_lockres(lockres);
212}
213
214static void user_bast(struct ocfs2_dlm_lksb *lksb, int level)
215{
216 struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
217
218 mlog(ML_BASTS, "BAST fired for lockres %.*s, blocking %d, level %d\n",
219 lockres->l_namelen, lockres->l_name, level, lockres->l_level);
220
221 spin_lock(&lockres->l_lock);
222 lockres->l_flags |= USER_LOCK_BLOCKED;
223 if (level > lockres->l_blocking)
224 lockres->l_blocking = level;
225
226 __user_dlm_queue_lockres(lockres);
227 spin_unlock(&lockres->l_lock);
228
229 wake_up(&lockres->l_event);
230}
231
232static void user_unlock_ast(struct ocfs2_dlm_lksb *lksb, int status)
233{
234 struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
235
236 mlog(ML_BASTS, "UNLOCK AST fired for lockres %.*s, flags 0x%x\n",
237 lockres->l_namelen, lockres->l_name, lockres->l_flags);
238
239 if (status)
240 mlog(ML_ERROR, "dlm returns status %d\n", status);
241
242 spin_lock(&lockres->l_lock);
243 /* The teardown flag gets set early during the unlock process,
244 * so test the cancel flag to make sure that this ast isn't
245 * for a concurrent cancel. */
246 if (lockres->l_flags & USER_LOCK_IN_TEARDOWN
247 && !(lockres->l_flags & USER_LOCK_IN_CANCEL)) {
248 lockres->l_level = DLM_LOCK_IV;
249 } else if (status == DLM_CANCELGRANT) {
250 /* We tried to cancel a convert request, but it was
251 * already granted. Don't clear the busy flag - the
252 * ast should've done this already. */
253 BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
254 lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
255 goto out_noclear;
256 } else {
257 BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
258 /* Cancel succeeded, we want to re-queue */
259 lockres->l_requested = DLM_LOCK_IV; /* cancel an
260 * upconvert
261 * request. */
262 lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
263 /* we want the unblock thread to look at it again
264 * now. */
265 if (lockres->l_flags & USER_LOCK_BLOCKED)
266 __user_dlm_queue_lockres(lockres);
267 }
268
269 lockres->l_flags &= ~USER_LOCK_BUSY;
270out_noclear:
271 spin_unlock(&lockres->l_lock);
272
273 wake_up(&lockres->l_event);
274}
275
276/*
277 * This is the userdlmfs locking protocol version.
278 *
279 * See fs/ocfs2/dlmglue.c for more details on locking versions.
280 */
281static struct ocfs2_locking_protocol user_dlm_lproto = {
282 .lp_max_version = {
283 .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
284 .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
285 },
286 .lp_lock_ast = user_ast,
287 .lp_blocking_ast = user_bast,
288 .lp_unlock_ast = user_unlock_ast,
289};
290
291static inline void user_dlm_drop_inode_ref(struct user_lock_res *lockres)
292{
293 struct inode *inode;
294 inode = user_dlm_inode_from_user_lockres(lockres);
295 iput(inode);
296}
297
298static void user_dlm_unblock_lock(struct work_struct *work)
299{
300 int new_level, status;
301 struct user_lock_res *lockres =
302 container_of(work, struct user_lock_res, l_work);
303 struct ocfs2_cluster_connection *conn =
304 cluster_connection_from_user_lockres(lockres);
305
306 mlog(0, "lockres %.*s\n", lockres->l_namelen, lockres->l_name);
307
308 spin_lock(&lockres->l_lock);
309
310 mlog_bug_on_msg(!(lockres->l_flags & USER_LOCK_QUEUED),
311 "Lockres %.*s, flags 0x%x\n",
312 lockres->l_namelen, lockres->l_name, lockres->l_flags);
313
314 /* notice that we don't clear USER_LOCK_BLOCKED here. If it's
315 * set, we want user_ast clear it. */
316 lockres->l_flags &= ~USER_LOCK_QUEUED;
317
318 /* It's valid to get here and no longer be blocked - if we get
319 * several basts in a row, we might be queued by the first
320 * one, the unblock thread might run and clear the queued
321 * flag, and finally we might get another bast which re-queues
322 * us before our ast for the downconvert is called. */
323 if (!(lockres->l_flags & USER_LOCK_BLOCKED)) {
324 mlog(ML_BASTS, "lockres %.*s USER_LOCK_BLOCKED\n",
325 lockres->l_namelen, lockres->l_name);
326 spin_unlock(&lockres->l_lock);
327 goto drop_ref;
328 }
329
330 if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
331 mlog(ML_BASTS, "lockres %.*s USER_LOCK_IN_TEARDOWN\n",
332 lockres->l_namelen, lockres->l_name);
333 spin_unlock(&lockres->l_lock);
334 goto drop_ref;
335 }
336
337 if (lockres->l_flags & USER_LOCK_BUSY) {
338 if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
339 mlog(ML_BASTS, "lockres %.*s USER_LOCK_IN_CANCEL\n",
340 lockres->l_namelen, lockres->l_name);
341 spin_unlock(&lockres->l_lock);
342 goto drop_ref;
343 }
344
345 lockres->l_flags |= USER_LOCK_IN_CANCEL;
346 spin_unlock(&lockres->l_lock);
347
348 status = ocfs2_dlm_unlock(conn, &lockres->l_lksb,
349 DLM_LKF_CANCEL);
350 if (status)
351 user_log_dlm_error("ocfs2_dlm_unlock", status, lockres);
352 goto drop_ref;
353 }
354
355 /* If there are still incompat holders, we can exit safely
356 * without worrying about re-queueing this lock as that will
357 * happen on the last call to user_cluster_unlock. */
358 if ((lockres->l_blocking == DLM_LOCK_EX)
359 && (lockres->l_ex_holders || lockres->l_ro_holders)) {
360 spin_unlock(&lockres->l_lock);
361 mlog(ML_BASTS, "lockres %.*s, EX/PR Holders %u,%u\n",
362 lockres->l_namelen, lockres->l_name,
363 lockres->l_ex_holders, lockres->l_ro_holders);
364 goto drop_ref;
365 }
366
367 if ((lockres->l_blocking == DLM_LOCK_PR)
368 && lockres->l_ex_holders) {
369 spin_unlock(&lockres->l_lock);
370 mlog(ML_BASTS, "lockres %.*s, EX Holders %u\n",
371 lockres->l_namelen, lockres->l_name,
372 lockres->l_ex_holders);
373 goto drop_ref;
374 }
375
376 /* yay, we can downconvert now. */
377 new_level = user_highest_compat_lock_level(lockres->l_blocking);
378 lockres->l_requested = new_level;
379 lockres->l_flags |= USER_LOCK_BUSY;
380 mlog(ML_BASTS, "lockres %.*s, downconvert %d => %d\n",
381 lockres->l_namelen, lockres->l_name, lockres->l_level, new_level);
382 spin_unlock(&lockres->l_lock);
383
384 /* need lock downconvert request now... */
385 status = ocfs2_dlm_lock(conn, new_level, &lockres->l_lksb,
386 DLM_LKF_CONVERT|DLM_LKF_VALBLK,
387 lockres->l_name,
388 lockres->l_namelen);
389 if (status) {
390 user_log_dlm_error("ocfs2_dlm_lock", status, lockres);
391 user_recover_from_dlm_error(lockres);
392 }
393
394drop_ref:
395 user_dlm_drop_inode_ref(lockres);
396}
397
398static inline void user_dlm_inc_holders(struct user_lock_res *lockres,
399 int level)
400{
401 switch(level) {
402 case DLM_LOCK_EX:
403 lockres->l_ex_holders++;
404 break;
405 case DLM_LOCK_PR:
406 lockres->l_ro_holders++;
407 break;
408 default:
409 BUG();
410 }
411}
412
413/* predict what lock level we'll be dropping down to on behalf
414 * of another node, and return true if the currently wanted
415 * level will be compatible with it. */
416static inline int
417user_may_continue_on_blocked_lock(struct user_lock_res *lockres,
418 int wanted)
419{
420 BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));
421
422 return wanted <= user_highest_compat_lock_level(lockres->l_blocking);
423}
424
425int user_dlm_cluster_lock(struct user_lock_res *lockres,
426 int level,
427 int lkm_flags)
428{
429 int status, local_flags;
430 struct ocfs2_cluster_connection *conn =
431 cluster_connection_from_user_lockres(lockres);
432
433 if (level != DLM_LOCK_EX &&
434 level != DLM_LOCK_PR) {
435 mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
436 lockres->l_namelen, lockres->l_name);
437 status = -EINVAL;
438 goto bail;
439 }
440
441 mlog(ML_BASTS, "lockres %.*s, level %d, flags = 0x%x\n",
442 lockres->l_namelen, lockres->l_name, level, lkm_flags);
443
444again:
445 if (signal_pending(current)) {
446 status = -ERESTARTSYS;
447 goto bail;
448 }
449
450 spin_lock(&lockres->l_lock);
451
452 /* We only compare against the currently granted level
453 * here. If the lock is blocked waiting on a downconvert,
454 * we'll get caught below. */
455 if ((lockres->l_flags & USER_LOCK_BUSY) &&
456 (level > lockres->l_level)) {
457 /* is someone sitting in dlm_lock? If so, wait on
458 * them. */
459 spin_unlock(&lockres->l_lock);
460
461 user_wait_on_busy_lock(lockres);
462 goto again;
463 }
464
465 if ((lockres->l_flags & USER_LOCK_BLOCKED) &&
466 (!user_may_continue_on_blocked_lock(lockres, level))) {
467 /* is the lock is currently blocked on behalf of
468 * another node */
469 spin_unlock(&lockres->l_lock);
470
471 user_wait_on_blocked_lock(lockres);
472 goto again;
473 }
474
475 if (level > lockres->l_level) {
476 local_flags = lkm_flags | DLM_LKF_VALBLK;
477 if (lockres->l_level != DLM_LOCK_IV)
478 local_flags |= DLM_LKF_CONVERT;
479
480 lockres->l_requested = level;
481 lockres->l_flags |= USER_LOCK_BUSY;
482 spin_unlock(&lockres->l_lock);
483
484 BUG_ON(level == DLM_LOCK_IV);
485 BUG_ON(level == DLM_LOCK_NL);
486
487 /* call dlm_lock to upgrade lock now */
488 status = ocfs2_dlm_lock(conn, level, &lockres->l_lksb,
489 local_flags, lockres->l_name,
490 lockres->l_namelen);
491 if (status) {
492 if ((lkm_flags & DLM_LKF_NOQUEUE) &&
493 (status != -EAGAIN))
494 user_log_dlm_error("ocfs2_dlm_lock",
495 status, lockres);
496 user_recover_from_dlm_error(lockres);
497 goto bail;
498 }
499
500 user_wait_on_busy_lock(lockres);
501 goto again;
502 }
503
504 user_dlm_inc_holders(lockres, level);
505 spin_unlock(&lockres->l_lock);
506
507 status = 0;
508bail:
509 return status;
510}
511
512static inline void user_dlm_dec_holders(struct user_lock_res *lockres,
513 int level)
514{
515 switch(level) {
516 case DLM_LOCK_EX:
517 BUG_ON(!lockres->l_ex_holders);
518 lockres->l_ex_holders--;
519 break;
520 case DLM_LOCK_PR:
521 BUG_ON(!lockres->l_ro_holders);
522 lockres->l_ro_holders--;
523 break;
524 default:
525 BUG();
526 }
527}
528
529void user_dlm_cluster_unlock(struct user_lock_res *lockres,
530 int level)
531{
532 if (level != DLM_LOCK_EX &&
533 level != DLM_LOCK_PR) {
534 mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
535 lockres->l_namelen, lockres->l_name);
536 return;
537 }
538
539 spin_lock(&lockres->l_lock);
540 user_dlm_dec_holders(lockres, level);
541 __user_dlm_cond_queue_lockres(lockres);
542 spin_unlock(&lockres->l_lock);
543}
544
545void user_dlm_write_lvb(struct inode *inode,
546 const char *val,
547 unsigned int len)
548{
549 struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
550 char *lvb;
551
552 BUG_ON(len > DLM_LVB_LEN);
553
554 spin_lock(&lockres->l_lock);
555
556 BUG_ON(lockres->l_level < DLM_LOCK_EX);
557 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
558 memcpy(lvb, val, len);
559
560 spin_unlock(&lockres->l_lock);
561}
562
563ssize_t user_dlm_read_lvb(struct inode *inode,
564 char *val,
565 unsigned int len)
566{
567 struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
568 char *lvb;
569 ssize_t ret = len;
570
571 BUG_ON(len > DLM_LVB_LEN);
572
573 spin_lock(&lockres->l_lock);
574
575 BUG_ON(lockres->l_level < DLM_LOCK_PR);
576 if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)) {
577 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
578 memcpy(val, lvb, len);
579 } else
580 ret = 0;
581
582 spin_unlock(&lockres->l_lock);
583 return ret;
584}
585
586void user_dlm_lock_res_init(struct user_lock_res *lockres,
587 struct dentry *dentry)
588{
589 memset(lockres, 0, sizeof(*lockres));
590
591 spin_lock_init(&lockres->l_lock);
592 init_waitqueue_head(&lockres->l_event);
593 lockres->l_level = DLM_LOCK_IV;
594 lockres->l_requested = DLM_LOCK_IV;
595 lockres->l_blocking = DLM_LOCK_IV;
596
597 /* should have been checked before getting here. */
598 BUG_ON(dentry->d_name.len >= USER_DLM_LOCK_ID_MAX_LEN);
599
600 memcpy(lockres->l_name,
601 dentry->d_name.name,
602 dentry->d_name.len);
603 lockres->l_namelen = dentry->d_name.len;
604}
605
606int user_dlm_destroy_lock(struct user_lock_res *lockres)
607{
608 int status = -EBUSY;
609 struct ocfs2_cluster_connection *conn =
610 cluster_connection_from_user_lockres(lockres);
611
612 mlog(ML_BASTS, "lockres %.*s\n", lockres->l_namelen, lockres->l_name);
613
614 spin_lock(&lockres->l_lock);
615 if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
616 spin_unlock(&lockres->l_lock);
617 return 0;
618 }
619
620 lockres->l_flags |= USER_LOCK_IN_TEARDOWN;
621
622 while (lockres->l_flags & USER_LOCK_BUSY) {
623 spin_unlock(&lockres->l_lock);
624
625 user_wait_on_busy_lock(lockres);
626
627 spin_lock(&lockres->l_lock);
628 }
629
630 if (lockres->l_ro_holders || lockres->l_ex_holders) {
631 spin_unlock(&lockres->l_lock);
632 goto bail;
633 }
634
635 status = 0;
636 if (!(lockres->l_flags & USER_LOCK_ATTACHED)) {
637 spin_unlock(&lockres->l_lock);
638 goto bail;
639 }
640
641 lockres->l_flags &= ~USER_LOCK_ATTACHED;
642 lockres->l_flags |= USER_LOCK_BUSY;
643 spin_unlock(&lockres->l_lock);
644
645 status = ocfs2_dlm_unlock(conn, &lockres->l_lksb, DLM_LKF_VALBLK);
646 if (status) {
647 user_log_dlm_error("ocfs2_dlm_unlock", status, lockres);
648 goto bail;
649 }
650
651 user_wait_on_busy_lock(lockres);
652
653 status = 0;
654bail:
655 return status;
656}
657
658static void user_dlm_recovery_handler_noop(int node_num,
659 void *recovery_data)
660{
661 /* We ignore recovery events */
662 return;
663}
664
665void user_dlm_set_locking_protocol(void)
666{
667 ocfs2_stack_glue_set_max_proto_version(&user_dlm_lproto.lp_max_version);
668}
669
670struct ocfs2_cluster_connection *user_dlm_register(struct qstr *name)
671{
672 int rc;
673 struct ocfs2_cluster_connection *conn;
674
675 rc = ocfs2_cluster_connect_agnostic(name->name, name->len,
676 &user_dlm_lproto,
677 user_dlm_recovery_handler_noop,
678 NULL, &conn);
679 if (rc)
680 mlog_errno(rc);
681
682 return rc ? ERR_PTR(rc) : conn;
683}
684
685void user_dlm_unregister(struct ocfs2_cluster_connection *conn)
686{
687 ocfs2_cluster_disconnect(conn, 0);
688}
diff --git a/fs/ocfs2/dlmfs/userdlm.h b/fs/ocfs2/dlmfs/userdlm.h
new file mode 100644
index 000000000000..3b42d79531d7
--- /dev/null
+++ b/fs/ocfs2/dlmfs/userdlm.h
@@ -0,0 +1,113 @@
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) /* we have 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_namelen;
57 int l_level;
58 unsigned int l_ro_holders;
59 unsigned int l_ex_holders;
60 struct ocfs2_dlm_lksb l_lksb;
61
62 int l_requested;
63 int l_blocking;
64
65 wait_queue_head_t l_event;
66
67 struct work_struct l_work;
68};
69
70extern struct workqueue_struct *user_dlm_worker;
71
72void user_dlm_lock_res_init(struct user_lock_res *lockres,
73 struct dentry *dentry);
74int user_dlm_destroy_lock(struct user_lock_res *lockres);
75int user_dlm_cluster_lock(struct user_lock_res *lockres,
76 int level,
77 int lkm_flags);
78void user_dlm_cluster_unlock(struct user_lock_res *lockres,
79 int level);
80void user_dlm_write_lvb(struct inode *inode,
81 const char *val,
82 unsigned int len);
83ssize_t user_dlm_read_lvb(struct inode *inode,
84 char *val,
85 unsigned int len);
86struct ocfs2_cluster_connection *user_dlm_register(struct qstr *name);
87void user_dlm_unregister(struct ocfs2_cluster_connection *conn);
88void user_dlm_set_locking_protocol(void);
89
90struct dlmfs_inode_private {
91 struct ocfs2_cluster_connection *ip_conn;
92
93 struct user_lock_res ip_lockres; /* unused for directories. */
94 struct inode *ip_parent;
95
96 struct inode ip_vfs_inode;
97};
98
99static inline struct dlmfs_inode_private *
100DLMFS_I(struct inode *inode)
101{
102 return container_of(inode,
103 struct dlmfs_inode_private,
104 ip_vfs_inode);
105}
106
107struct dlmfs_filp_private {
108 int fp_lock_level;
109};
110
111#define DLMFS_MAGIC 0x76a9f425
112
113#endif /* USERDLM_H */
generated by cgit v1.2.2 (git 2.25.0) at 2025-02-19 02:39:54 -0500