xref: /openbmc/linux/fs/ocfs2/dlmglue.c (revision 22246614)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * dlmglue.c
5  *
6  * Code which implements an OCFS2 specific interface to our DLM.
7  *
8  * Copyright (C) 2003, 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 #include <linux/types.h>
27 #include <linux/slab.h>
28 #include <linux/highmem.h>
29 #include <linux/mm.h>
30 #include <linux/kthread.h>
31 #include <linux/pagemap.h>
32 #include <linux/debugfs.h>
33 #include <linux/seq_file.h>
34 
35 #define MLOG_MASK_PREFIX ML_DLM_GLUE
36 #include <cluster/masklog.h>
37 
38 #include "ocfs2.h"
39 #include "ocfs2_lockingver.h"
40 
41 #include "alloc.h"
42 #include "dcache.h"
43 #include "dlmglue.h"
44 #include "extent_map.h"
45 #include "file.h"
46 #include "heartbeat.h"
47 #include "inode.h"
48 #include "journal.h"
49 #include "stackglue.h"
50 #include "slot_map.h"
51 #include "super.h"
52 #include "uptodate.h"
53 
54 #include "buffer_head_io.h"
55 
56 struct ocfs2_mask_waiter {
57 	struct list_head	mw_item;
58 	int			mw_status;
59 	struct completion	mw_complete;
60 	unsigned long		mw_mask;
61 	unsigned long		mw_goal;
62 };
63 
64 static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres);
65 static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres);
66 static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres);
67 
68 /*
69  * Return value from ->downconvert_worker functions.
70  *
71  * These control the precise actions of ocfs2_unblock_lock()
72  * and ocfs2_process_blocked_lock()
73  *
74  */
75 enum ocfs2_unblock_action {
76 	UNBLOCK_CONTINUE	= 0, /* Continue downconvert */
77 	UNBLOCK_CONTINUE_POST	= 1, /* Continue downconvert, fire
78 				      * ->post_unlock callback */
79 	UNBLOCK_STOP_POST	= 2, /* Do not downconvert, fire
80 				      * ->post_unlock() callback. */
81 };
82 
83 struct ocfs2_unblock_ctl {
84 	int requeue;
85 	enum ocfs2_unblock_action unblock_action;
86 };
87 
88 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
89 					int new_level);
90 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);
91 
92 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
93 				     int blocking);
94 
95 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
96 				       int blocking);
97 
98 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
99 				     struct ocfs2_lock_res *lockres);
100 
101 
102 #define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
103 
104 /* This aids in debugging situations where a bad LVB might be involved. */
105 static void ocfs2_dump_meta_lvb_info(u64 level,
106 				     const char *function,
107 				     unsigned int line,
108 				     struct ocfs2_lock_res *lockres)
109 {
110 	struct ocfs2_meta_lvb *lvb =
111 		(struct ocfs2_meta_lvb *)ocfs2_dlm_lvb(&lockres->l_lksb);
112 
113 	mlog(level, "LVB information for %s (called from %s:%u):\n",
114 	     lockres->l_name, function, line);
115 	mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
116 	     lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
117 	     be32_to_cpu(lvb->lvb_igeneration));
118 	mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
119 	     (unsigned long long)be64_to_cpu(lvb->lvb_isize),
120 	     be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
121 	     be16_to_cpu(lvb->lvb_imode));
122 	mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
123 	     "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
124 	     (long long)be64_to_cpu(lvb->lvb_iatime_packed),
125 	     (long long)be64_to_cpu(lvb->lvb_ictime_packed),
126 	     (long long)be64_to_cpu(lvb->lvb_imtime_packed),
127 	     be32_to_cpu(lvb->lvb_iattr));
128 }
129 
130 
131 /*
132  * OCFS2 Lock Resource Operations
133  *
134  * These fine tune the behavior of the generic dlmglue locking infrastructure.
135  *
136  * The most basic of lock types can point ->l_priv to their respective
137  * struct ocfs2_super and allow the default actions to manage things.
138  *
139  * Right now, each lock type also needs to implement an init function,
140  * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
141  * should be called when the lock is no longer needed (i.e., object
142  * destruction time).
143  */
144 struct ocfs2_lock_res_ops {
145 	/*
146 	 * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
147 	 * this callback if ->l_priv is not an ocfs2_super pointer
148 	 */
149 	struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
150 
151 	/*
152 	 * Optionally called in the downconvert thread after a
153 	 * successful downconvert. The lockres will not be referenced
154 	 * after this callback is called, so it is safe to free
155 	 * memory, etc.
156 	 *
157 	 * The exact semantics of when this is called are controlled
158 	 * by ->downconvert_worker()
159 	 */
160 	void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
161 
162 	/*
163 	 * Allow a lock type to add checks to determine whether it is
164 	 * safe to downconvert a lock. Return 0 to re-queue the
165 	 * downconvert at a later time, nonzero to continue.
166 	 *
167 	 * For most locks, the default checks that there are no
168 	 * incompatible holders are sufficient.
169 	 *
170 	 * Called with the lockres spinlock held.
171 	 */
172 	int (*check_downconvert)(struct ocfs2_lock_res *, int);
173 
174 	/*
175 	 * Allows a lock type to populate the lock value block. This
176 	 * is called on downconvert, and when we drop a lock.
177 	 *
178 	 * Locks that want to use this should set LOCK_TYPE_USES_LVB
179 	 * in the flags field.
180 	 *
181 	 * Called with the lockres spinlock held.
182 	 */
183 	void (*set_lvb)(struct ocfs2_lock_res *);
184 
185 	/*
186 	 * Called from the downconvert thread when it is determined
187 	 * that a lock will be downconverted. This is called without
188 	 * any locks held so the function can do work that might
189 	 * schedule (syncing out data, etc).
190 	 *
191 	 * This should return any one of the ocfs2_unblock_action
192 	 * values, depending on what it wants the thread to do.
193 	 */
194 	int (*downconvert_worker)(struct ocfs2_lock_res *, int);
195 
196 	/*
197 	 * LOCK_TYPE_* flags which describe the specific requirements
198 	 * of a lock type. Descriptions of each individual flag follow.
199 	 */
200 	int flags;
201 };
202 
203 /*
204  * Some locks want to "refresh" potentially stale data when a
205  * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
206  * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
207  * individual lockres l_flags member from the ast function. It is
208  * expected that the locking wrapper will clear the
209  * OCFS2_LOCK_NEEDS_REFRESH flag when done.
210  */
211 #define LOCK_TYPE_REQUIRES_REFRESH 0x1
212 
213 /*
214  * Indicate that a lock type makes use of the lock value block. The
215  * ->set_lvb lock type callback must be defined.
216  */
217 #define LOCK_TYPE_USES_LVB		0x2
218 
219 static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
220 	.get_osb	= ocfs2_get_inode_osb,
221 	.flags		= 0,
222 };
223 
224 static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = {
225 	.get_osb	= ocfs2_get_inode_osb,
226 	.check_downconvert = ocfs2_check_meta_downconvert,
227 	.set_lvb	= ocfs2_set_meta_lvb,
228 	.downconvert_worker = ocfs2_data_convert_worker,
229 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
230 };
231 
232 static struct ocfs2_lock_res_ops ocfs2_super_lops = {
233 	.flags		= LOCK_TYPE_REQUIRES_REFRESH,
234 };
235 
236 static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
237 	.flags		= 0,
238 };
239 
240 static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
241 	.get_osb	= ocfs2_get_dentry_osb,
242 	.post_unlock	= ocfs2_dentry_post_unlock,
243 	.downconvert_worker = ocfs2_dentry_convert_worker,
244 	.flags		= 0,
245 };
246 
247 static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
248 	.get_osb	= ocfs2_get_inode_osb,
249 	.flags		= 0,
250 };
251 
252 static struct ocfs2_lock_res_ops ocfs2_flock_lops = {
253 	.get_osb	= ocfs2_get_file_osb,
254 	.flags		= 0,
255 };
256 
257 static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
258 {
259 	return lockres->l_type == OCFS2_LOCK_TYPE_META ||
260 		lockres->l_type == OCFS2_LOCK_TYPE_RW ||
261 		lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
262 }
263 
264 static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
265 {
266 	BUG_ON(!ocfs2_is_inode_lock(lockres));
267 
268 	return (struct inode *) lockres->l_priv;
269 }
270 
271 static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
272 {
273 	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
274 
275 	return (struct ocfs2_dentry_lock *)lockres->l_priv;
276 }
277 
278 static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
279 {
280 	if (lockres->l_ops->get_osb)
281 		return lockres->l_ops->get_osb(lockres);
282 
283 	return (struct ocfs2_super *)lockres->l_priv;
284 }
285 
286 static int ocfs2_lock_create(struct ocfs2_super *osb,
287 			     struct ocfs2_lock_res *lockres,
288 			     int level,
289 			     u32 dlm_flags);
290 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
291 						     int wanted);
292 static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
293 				 struct ocfs2_lock_res *lockres,
294 				 int level);
295 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
296 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
297 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
298 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
299 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
300 					struct ocfs2_lock_res *lockres);
301 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
302 						int convert);
303 #define ocfs2_log_dlm_error(_func, _err, _lockres) do {			\
304 	mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n", \
305 	     _err, _func, _lockres->l_name);				\
306 } while (0)
307 static int ocfs2_downconvert_thread(void *arg);
308 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
309 					struct ocfs2_lock_res *lockres);
310 static int ocfs2_inode_lock_update(struct inode *inode,
311 				  struct buffer_head **bh);
312 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
313 static inline int ocfs2_highest_compat_lock_level(int level);
314 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
315 					      int new_level);
316 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
317 				  struct ocfs2_lock_res *lockres,
318 				  int new_level,
319 				  int lvb,
320 				  unsigned int generation);
321 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
322 				        struct ocfs2_lock_res *lockres);
323 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
324 				struct ocfs2_lock_res *lockres);
325 
326 
327 static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
328 				  u64 blkno,
329 				  u32 generation,
330 				  char *name)
331 {
332 	int len;
333 
334 	mlog_entry_void();
335 
336 	BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
337 
338 	len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
339 		       ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
340 		       (long long)blkno, generation);
341 
342 	BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
343 
344 	mlog(0, "built lock resource with name: %s\n", name);
345 
346 	mlog_exit_void();
347 }
348 
349 static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
350 
351 static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
352 				       struct ocfs2_dlm_debug *dlm_debug)
353 {
354 	mlog(0, "Add tracking for lockres %s\n", res->l_name);
355 
356 	spin_lock(&ocfs2_dlm_tracking_lock);
357 	list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
358 	spin_unlock(&ocfs2_dlm_tracking_lock);
359 }
360 
361 static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
362 {
363 	spin_lock(&ocfs2_dlm_tracking_lock);
364 	if (!list_empty(&res->l_debug_list))
365 		list_del_init(&res->l_debug_list);
366 	spin_unlock(&ocfs2_dlm_tracking_lock);
367 }
368 
369 static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
370 				       struct ocfs2_lock_res *res,
371 				       enum ocfs2_lock_type type,
372 				       struct ocfs2_lock_res_ops *ops,
373 				       void *priv)
374 {
375 	res->l_type          = type;
376 	res->l_ops           = ops;
377 	res->l_priv          = priv;
378 
379 	res->l_level         = DLM_LOCK_IV;
380 	res->l_requested     = DLM_LOCK_IV;
381 	res->l_blocking      = DLM_LOCK_IV;
382 	res->l_action        = OCFS2_AST_INVALID;
383 	res->l_unlock_action = OCFS2_UNLOCK_INVALID;
384 
385 	res->l_flags         = OCFS2_LOCK_INITIALIZED;
386 
387 	ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
388 }
389 
390 void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
391 {
392 	/* This also clears out the lock status block */
393 	memset(res, 0, sizeof(struct ocfs2_lock_res));
394 	spin_lock_init(&res->l_lock);
395 	init_waitqueue_head(&res->l_event);
396 	INIT_LIST_HEAD(&res->l_blocked_list);
397 	INIT_LIST_HEAD(&res->l_mask_waiters);
398 }
399 
400 void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
401 			       enum ocfs2_lock_type type,
402 			       unsigned int generation,
403 			       struct inode *inode)
404 {
405 	struct ocfs2_lock_res_ops *ops;
406 
407 	switch(type) {
408 		case OCFS2_LOCK_TYPE_RW:
409 			ops = &ocfs2_inode_rw_lops;
410 			break;
411 		case OCFS2_LOCK_TYPE_META:
412 			ops = &ocfs2_inode_inode_lops;
413 			break;
414 		case OCFS2_LOCK_TYPE_OPEN:
415 			ops = &ocfs2_inode_open_lops;
416 			break;
417 		default:
418 			mlog_bug_on_msg(1, "type: %d\n", type);
419 			ops = NULL; /* thanks, gcc */
420 			break;
421 	};
422 
423 	ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
424 			      generation, res->l_name);
425 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
426 }
427 
428 static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
429 {
430 	struct inode *inode = ocfs2_lock_res_inode(lockres);
431 
432 	return OCFS2_SB(inode->i_sb);
433 }
434 
435 static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres)
436 {
437 	struct ocfs2_file_private *fp = lockres->l_priv;
438 
439 	return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb);
440 }
441 
442 static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
443 {
444 	__be64 inode_blkno_be;
445 
446 	memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
447 	       sizeof(__be64));
448 
449 	return be64_to_cpu(inode_blkno_be);
450 }
451 
452 static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
453 {
454 	struct ocfs2_dentry_lock *dl = lockres->l_priv;
455 
456 	return OCFS2_SB(dl->dl_inode->i_sb);
457 }
458 
459 void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
460 				u64 parent, struct inode *inode)
461 {
462 	int len;
463 	u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
464 	__be64 inode_blkno_be = cpu_to_be64(inode_blkno);
465 	struct ocfs2_lock_res *lockres = &dl->dl_lockres;
466 
467 	ocfs2_lock_res_init_once(lockres);
468 
469 	/*
470 	 * Unfortunately, the standard lock naming scheme won't work
471 	 * here because we have two 16 byte values to use. Instead,
472 	 * we'll stuff the inode number as a binary value. We still
473 	 * want error prints to show something without garbling the
474 	 * display, so drop a null byte in there before the inode
475 	 * number. A future version of OCFS2 will likely use all
476 	 * binary lock names. The stringified names have been a
477 	 * tremendous aid in debugging, but now that the debugfs
478 	 * interface exists, we can mangle things there if need be.
479 	 *
480 	 * NOTE: We also drop the standard "pad" value (the total lock
481 	 * name size stays the same though - the last part is all
482 	 * zeros due to the memset in ocfs2_lock_res_init_once()
483 	 */
484 	len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
485 		       "%c%016llx",
486 		       ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
487 		       (long long)parent);
488 
489 	BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
490 
491 	memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
492 	       sizeof(__be64));
493 
494 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
495 				   OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
496 				   dl);
497 }
498 
499 static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
500 				      struct ocfs2_super *osb)
501 {
502 	/* Superblock lockres doesn't come from a slab so we call init
503 	 * once on it manually.  */
504 	ocfs2_lock_res_init_once(res);
505 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
506 			      0, res->l_name);
507 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
508 				   &ocfs2_super_lops, osb);
509 }
510 
511 static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
512 				       struct ocfs2_super *osb)
513 {
514 	/* Rename lockres doesn't come from a slab so we call init
515 	 * once on it manually.  */
516 	ocfs2_lock_res_init_once(res);
517 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
518 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
519 				   &ocfs2_rename_lops, osb);
520 }
521 
522 void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
523 			      struct ocfs2_file_private *fp)
524 {
525 	struct inode *inode = fp->fp_file->f_mapping->host;
526 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
527 
528 	ocfs2_lock_res_init_once(lockres);
529 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
530 			      inode->i_generation, lockres->l_name);
531 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
532 				   OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
533 				   fp);
534 	lockres->l_flags |= OCFS2_LOCK_NOCACHE;
535 }
536 
537 void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
538 {
539 	mlog_entry_void();
540 
541 	if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
542 		return;
543 
544 	ocfs2_remove_lockres_tracking(res);
545 
546 	mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
547 			"Lockres %s is on the blocked list\n",
548 			res->l_name);
549 	mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
550 			"Lockres %s has mask waiters pending\n",
551 			res->l_name);
552 	mlog_bug_on_msg(spin_is_locked(&res->l_lock),
553 			"Lockres %s is locked\n",
554 			res->l_name);
555 	mlog_bug_on_msg(res->l_ro_holders,
556 			"Lockres %s has %u ro holders\n",
557 			res->l_name, res->l_ro_holders);
558 	mlog_bug_on_msg(res->l_ex_holders,
559 			"Lockres %s has %u ex holders\n",
560 			res->l_name, res->l_ex_holders);
561 
562 	/* Need to clear out the lock status block for the dlm */
563 	memset(&res->l_lksb, 0, sizeof(res->l_lksb));
564 
565 	res->l_flags = 0UL;
566 	mlog_exit_void();
567 }
568 
569 static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
570 				     int level)
571 {
572 	mlog_entry_void();
573 
574 	BUG_ON(!lockres);
575 
576 	switch(level) {
577 	case DLM_LOCK_EX:
578 		lockres->l_ex_holders++;
579 		break;
580 	case DLM_LOCK_PR:
581 		lockres->l_ro_holders++;
582 		break;
583 	default:
584 		BUG();
585 	}
586 
587 	mlog_exit_void();
588 }
589 
590 static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
591 				     int level)
592 {
593 	mlog_entry_void();
594 
595 	BUG_ON(!lockres);
596 
597 	switch(level) {
598 	case DLM_LOCK_EX:
599 		BUG_ON(!lockres->l_ex_holders);
600 		lockres->l_ex_holders--;
601 		break;
602 	case DLM_LOCK_PR:
603 		BUG_ON(!lockres->l_ro_holders);
604 		lockres->l_ro_holders--;
605 		break;
606 	default:
607 		BUG();
608 	}
609 	mlog_exit_void();
610 }
611 
612 /* WARNING: This function lives in a world where the only three lock
613  * levels are EX, PR, and NL. It *will* have to be adjusted when more
614  * lock types are added. */
615 static inline int ocfs2_highest_compat_lock_level(int level)
616 {
617 	int new_level = DLM_LOCK_EX;
618 
619 	if (level == DLM_LOCK_EX)
620 		new_level = DLM_LOCK_NL;
621 	else if (level == DLM_LOCK_PR)
622 		new_level = DLM_LOCK_PR;
623 	return new_level;
624 }
625 
626 static void lockres_set_flags(struct ocfs2_lock_res *lockres,
627 			      unsigned long newflags)
628 {
629 	struct ocfs2_mask_waiter *mw, *tmp;
630 
631  	assert_spin_locked(&lockres->l_lock);
632 
633 	lockres->l_flags = newflags;
634 
635 	list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
636 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
637 			continue;
638 
639 		list_del_init(&mw->mw_item);
640 		mw->mw_status = 0;
641 		complete(&mw->mw_complete);
642 	}
643 }
644 static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
645 {
646 	lockres_set_flags(lockres, lockres->l_flags | or);
647 }
648 static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
649 				unsigned long clear)
650 {
651 	lockres_set_flags(lockres, lockres->l_flags & ~clear);
652 }
653 
654 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
655 {
656 	mlog_entry_void();
657 
658 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
659 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
660 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
661 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
662 
663 	lockres->l_level = lockres->l_requested;
664 	if (lockres->l_level <=
665 	    ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
666 		lockres->l_blocking = DLM_LOCK_NL;
667 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
668 	}
669 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
670 
671 	mlog_exit_void();
672 }
673 
674 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
675 {
676 	mlog_entry_void();
677 
678 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
679 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
680 
681 	/* Convert from RO to EX doesn't really need anything as our
682 	 * information is already up to data. Convert from NL to
683 	 * *anything* however should mark ourselves as needing an
684 	 * update */
685 	if (lockres->l_level == DLM_LOCK_NL &&
686 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
687 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
688 
689 	lockres->l_level = lockres->l_requested;
690 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
691 
692 	mlog_exit_void();
693 }
694 
695 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
696 {
697 	mlog_entry_void();
698 
699 	BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
700 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
701 
702 	if (lockres->l_requested > DLM_LOCK_NL &&
703 	    !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
704 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
705 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
706 
707 	lockres->l_level = lockres->l_requested;
708 	lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
709 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
710 
711 	mlog_exit_void();
712 }
713 
714 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
715 				     int level)
716 {
717 	int needs_downconvert = 0;
718 	mlog_entry_void();
719 
720 	assert_spin_locked(&lockres->l_lock);
721 
722 	lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
723 
724 	if (level > lockres->l_blocking) {
725 		/* only schedule a downconvert if we haven't already scheduled
726 		 * one that goes low enough to satisfy the level we're
727 		 * blocking.  this also catches the case where we get
728 		 * duplicate BASTs */
729 		if (ocfs2_highest_compat_lock_level(level) <
730 		    ocfs2_highest_compat_lock_level(lockres->l_blocking))
731 			needs_downconvert = 1;
732 
733 		lockres->l_blocking = level;
734 	}
735 
736 	mlog_exit(needs_downconvert);
737 	return needs_downconvert;
738 }
739 
740 /*
741  * OCFS2_LOCK_PENDING and l_pending_gen.
742  *
743  * Why does OCFS2_LOCK_PENDING exist?  To close a race between setting
744  * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock().  See ocfs2_unblock_lock()
745  * for more details on the race.
746  *
747  * OCFS2_LOCK_PENDING closes the race quite nicely.  However, it introduces
748  * a race on itself.  In o2dlm, we can get the ast before ocfs2_dlm_lock()
749  * returns.  The ast clears OCFS2_LOCK_BUSY, and must therefore clear
750  * OCFS2_LOCK_PENDING at the same time.  When ocfs2_dlm_lock() returns,
751  * the caller is going to try to clear PENDING again.  If nothing else is
752  * happening, __lockres_clear_pending() sees PENDING is unset and does
753  * nothing.
754  *
755  * But what if another path (eg downconvert thread) has just started a
756  * new locking action?  The other path has re-set PENDING.  Our path
757  * cannot clear PENDING, because that will re-open the original race
758  * window.
759  *
760  * [Example]
761  *
762  * ocfs2_meta_lock()
763  *  ocfs2_cluster_lock()
764  *   set BUSY
765  *   set PENDING
766  *   drop l_lock
767  *   ocfs2_dlm_lock()
768  *    ocfs2_locking_ast()		ocfs2_downconvert_thread()
769  *     clear PENDING			 ocfs2_unblock_lock()
770  *					  take_l_lock
771  *					  !BUSY
772  *					  ocfs2_prepare_downconvert()
773  *					   set BUSY
774  *					   set PENDING
775  *					  drop l_lock
776  *   take l_lock
777  *   clear PENDING
778  *   drop l_lock
779  *			<window>
780  *					  ocfs2_dlm_lock()
781  *
782  * So as you can see, we now have a window where l_lock is not held,
783  * PENDING is not set, and ocfs2_dlm_lock() has not been called.
784  *
785  * The core problem is that ocfs2_cluster_lock() has cleared the PENDING
786  * set by ocfs2_prepare_downconvert().  That wasn't nice.
787  *
788  * To solve this we introduce l_pending_gen.  A call to
789  * lockres_clear_pending() will only do so when it is passed a generation
790  * number that matches the lockres.  lockres_set_pending() will return the
791  * current generation number.  When ocfs2_cluster_lock() goes to clear
792  * PENDING, it passes the generation it got from set_pending().  In our
793  * example above, the generation numbers will *not* match.  Thus,
794  * ocfs2_cluster_lock() will not clear the PENDING set by
795  * ocfs2_prepare_downconvert().
796  */
797 
798 /* Unlocked version for ocfs2_locking_ast() */
799 static void __lockres_clear_pending(struct ocfs2_lock_res *lockres,
800 				    unsigned int generation,
801 				    struct ocfs2_super *osb)
802 {
803 	assert_spin_locked(&lockres->l_lock);
804 
805 	/*
806 	 * The ast and locking functions can race us here.  The winner
807 	 * will clear pending, the loser will not.
808 	 */
809 	if (!(lockres->l_flags & OCFS2_LOCK_PENDING) ||
810 	    (lockres->l_pending_gen != generation))
811 		return;
812 
813 	lockres_clear_flags(lockres, OCFS2_LOCK_PENDING);
814 	lockres->l_pending_gen++;
815 
816 	/*
817 	 * The downconvert thread may have skipped us because we
818 	 * were PENDING.  Wake it up.
819 	 */
820 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
821 		ocfs2_wake_downconvert_thread(osb);
822 }
823 
824 /* Locked version for callers of ocfs2_dlm_lock() */
825 static void lockres_clear_pending(struct ocfs2_lock_res *lockres,
826 				  unsigned int generation,
827 				  struct ocfs2_super *osb)
828 {
829 	unsigned long flags;
830 
831 	spin_lock_irqsave(&lockres->l_lock, flags);
832 	__lockres_clear_pending(lockres, generation, osb);
833 	spin_unlock_irqrestore(&lockres->l_lock, flags);
834 }
835 
836 static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres)
837 {
838 	assert_spin_locked(&lockres->l_lock);
839 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
840 
841 	lockres_or_flags(lockres, OCFS2_LOCK_PENDING);
842 
843 	return lockres->l_pending_gen;
844 }
845 
846 
847 static void ocfs2_blocking_ast(void *opaque, int level)
848 {
849 	struct ocfs2_lock_res *lockres = opaque;
850 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
851 	int needs_downconvert;
852 	unsigned long flags;
853 
854 	BUG_ON(level <= DLM_LOCK_NL);
855 
856 	mlog(0, "BAST fired for lockres %s, blocking %d, level %d type %s\n",
857 	     lockres->l_name, level, lockres->l_level,
858 	     ocfs2_lock_type_string(lockres->l_type));
859 
860 	/*
861 	 * We can skip the bast for locks which don't enable caching -
862 	 * they'll be dropped at the earliest possible time anyway.
863 	 */
864 	if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
865 		return;
866 
867 	spin_lock_irqsave(&lockres->l_lock, flags);
868 	needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
869 	if (needs_downconvert)
870 		ocfs2_schedule_blocked_lock(osb, lockres);
871 	spin_unlock_irqrestore(&lockres->l_lock, flags);
872 
873 	wake_up(&lockres->l_event);
874 
875 	ocfs2_wake_downconvert_thread(osb);
876 }
877 
878 static void ocfs2_locking_ast(void *opaque)
879 {
880 	struct ocfs2_lock_res *lockres = opaque;
881 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
882 	unsigned long flags;
883 	int status;
884 
885 	spin_lock_irqsave(&lockres->l_lock, flags);
886 
887 	status = ocfs2_dlm_lock_status(&lockres->l_lksb);
888 
889 	if (status == -EAGAIN) {
890 		lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
891 		goto out;
892 	}
893 
894 	if (status) {
895 		mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n",
896 		     lockres->l_name, status);
897 		spin_unlock_irqrestore(&lockres->l_lock, flags);
898 		return;
899 	}
900 
901 	switch(lockres->l_action) {
902 	case OCFS2_AST_ATTACH:
903 		ocfs2_generic_handle_attach_action(lockres);
904 		lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
905 		break;
906 	case OCFS2_AST_CONVERT:
907 		ocfs2_generic_handle_convert_action(lockres);
908 		break;
909 	case OCFS2_AST_DOWNCONVERT:
910 		ocfs2_generic_handle_downconvert_action(lockres);
911 		break;
912 	default:
913 		mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u "
914 		     "lockres flags = 0x%lx, unlock action: %u\n",
915 		     lockres->l_name, lockres->l_action, lockres->l_flags,
916 		     lockres->l_unlock_action);
917 		BUG();
918 	}
919 out:
920 	/* set it to something invalid so if we get called again we
921 	 * can catch it. */
922 	lockres->l_action = OCFS2_AST_INVALID;
923 
924 	/* Did we try to cancel this lock?  Clear that state */
925 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT)
926 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
927 
928 	/*
929 	 * We may have beaten the locking functions here.  We certainly
930 	 * know that dlm_lock() has been called :-)
931 	 * Because we can't have two lock calls in flight at once, we
932 	 * can use lockres->l_pending_gen.
933 	 */
934 	__lockres_clear_pending(lockres, lockres->l_pending_gen,  osb);
935 
936 	wake_up(&lockres->l_event);
937 	spin_unlock_irqrestore(&lockres->l_lock, flags);
938 }
939 
940 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
941 						int convert)
942 {
943 	unsigned long flags;
944 
945 	mlog_entry_void();
946 	spin_lock_irqsave(&lockres->l_lock, flags);
947 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
948 	if (convert)
949 		lockres->l_action = OCFS2_AST_INVALID;
950 	else
951 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
952 	spin_unlock_irqrestore(&lockres->l_lock, flags);
953 
954 	wake_up(&lockres->l_event);
955 	mlog_exit_void();
956 }
957 
958 /* Note: If we detect another process working on the lock (i.e.,
959  * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
960  * to do the right thing in that case.
961  */
962 static int ocfs2_lock_create(struct ocfs2_super *osb,
963 			     struct ocfs2_lock_res *lockres,
964 			     int level,
965 			     u32 dlm_flags)
966 {
967 	int ret = 0;
968 	unsigned long flags;
969 	unsigned int gen;
970 
971 	mlog_entry_void();
972 
973 	mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
974 	     dlm_flags);
975 
976 	spin_lock_irqsave(&lockres->l_lock, flags);
977 	if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
978 	    (lockres->l_flags & OCFS2_LOCK_BUSY)) {
979 		spin_unlock_irqrestore(&lockres->l_lock, flags);
980 		goto bail;
981 	}
982 
983 	lockres->l_action = OCFS2_AST_ATTACH;
984 	lockres->l_requested = level;
985 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
986 	gen = lockres_set_pending(lockres);
987 	spin_unlock_irqrestore(&lockres->l_lock, flags);
988 
989 	ret = ocfs2_dlm_lock(osb->cconn,
990 			     level,
991 			     &lockres->l_lksb,
992 			     dlm_flags,
993 			     lockres->l_name,
994 			     OCFS2_LOCK_ID_MAX_LEN - 1,
995 			     lockres);
996 	lockres_clear_pending(lockres, gen, osb);
997 	if (ret) {
998 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
999 		ocfs2_recover_from_dlm_error(lockres, 1);
1000 	}
1001 
1002 	mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
1003 
1004 bail:
1005 	mlog_exit(ret);
1006 	return ret;
1007 }
1008 
1009 static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
1010 					int flag)
1011 {
1012 	unsigned long flags;
1013 	int ret;
1014 
1015 	spin_lock_irqsave(&lockres->l_lock, flags);
1016 	ret = lockres->l_flags & flag;
1017 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1018 
1019 	return ret;
1020 }
1021 
1022 static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
1023 
1024 {
1025 	wait_event(lockres->l_event,
1026 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
1027 }
1028 
1029 static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
1030 
1031 {
1032 	wait_event(lockres->l_event,
1033 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
1034 }
1035 
1036 /* predict what lock level we'll be dropping down to on behalf
1037  * of another node, and return true if the currently wanted
1038  * level will be compatible with it. */
1039 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
1040 						     int wanted)
1041 {
1042 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
1043 
1044 	return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
1045 }
1046 
1047 static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
1048 {
1049 	INIT_LIST_HEAD(&mw->mw_item);
1050 	init_completion(&mw->mw_complete);
1051 }
1052 
1053 static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
1054 {
1055 	wait_for_completion(&mw->mw_complete);
1056 	/* Re-arm the completion in case we want to wait on it again */
1057 	INIT_COMPLETION(mw->mw_complete);
1058 	return mw->mw_status;
1059 }
1060 
1061 static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
1062 				    struct ocfs2_mask_waiter *mw,
1063 				    unsigned long mask,
1064 				    unsigned long goal)
1065 {
1066 	BUG_ON(!list_empty(&mw->mw_item));
1067 
1068 	assert_spin_locked(&lockres->l_lock);
1069 
1070 	list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
1071 	mw->mw_mask = mask;
1072 	mw->mw_goal = goal;
1073 }
1074 
1075 /* returns 0 if the mw that was removed was already satisfied, -EBUSY
1076  * if the mask still hadn't reached its goal */
1077 static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1078 				      struct ocfs2_mask_waiter *mw)
1079 {
1080 	unsigned long flags;
1081 	int ret = 0;
1082 
1083 	spin_lock_irqsave(&lockres->l_lock, flags);
1084 	if (!list_empty(&mw->mw_item)) {
1085 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
1086 			ret = -EBUSY;
1087 
1088 		list_del_init(&mw->mw_item);
1089 		init_completion(&mw->mw_complete);
1090 	}
1091 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1092 
1093 	return ret;
1094 
1095 }
1096 
1097 static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
1098 					     struct ocfs2_lock_res *lockres)
1099 {
1100 	int ret;
1101 
1102 	ret = wait_for_completion_interruptible(&mw->mw_complete);
1103 	if (ret)
1104 		lockres_remove_mask_waiter(lockres, mw);
1105 	else
1106 		ret = mw->mw_status;
1107 	/* Re-arm the completion in case we want to wait on it again */
1108 	INIT_COMPLETION(mw->mw_complete);
1109 	return ret;
1110 }
1111 
1112 static int ocfs2_cluster_lock(struct ocfs2_super *osb,
1113 			      struct ocfs2_lock_res *lockres,
1114 			      int level,
1115 			      u32 lkm_flags,
1116 			      int arg_flags)
1117 {
1118 	struct ocfs2_mask_waiter mw;
1119 	int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
1120 	int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1121 	unsigned long flags;
1122 	unsigned int gen;
1123 	int noqueue_attempted = 0;
1124 
1125 	mlog_entry_void();
1126 
1127 	ocfs2_init_mask_waiter(&mw);
1128 
1129 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
1130 		lkm_flags |= DLM_LKF_VALBLK;
1131 
1132 again:
1133 	wait = 0;
1134 
1135 	if (catch_signals && signal_pending(current)) {
1136 		ret = -ERESTARTSYS;
1137 		goto out;
1138 	}
1139 
1140 	spin_lock_irqsave(&lockres->l_lock, flags);
1141 
1142 	mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
1143 			"Cluster lock called on freeing lockres %s! flags "
1144 			"0x%lx\n", lockres->l_name, lockres->l_flags);
1145 
1146 	/* We only compare against the currently granted level
1147 	 * here. If the lock is blocked waiting on a downconvert,
1148 	 * we'll get caught below. */
1149 	if (lockres->l_flags & OCFS2_LOCK_BUSY &&
1150 	    level > lockres->l_level) {
1151 		/* is someone sitting in dlm_lock? If so, wait on
1152 		 * them. */
1153 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1154 		wait = 1;
1155 		goto unlock;
1156 	}
1157 
1158 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
1159 	    !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
1160 		/* is the lock is currently blocked on behalf of
1161 		 * another node */
1162 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
1163 		wait = 1;
1164 		goto unlock;
1165 	}
1166 
1167 	if (level > lockres->l_level) {
1168 		if (noqueue_attempted > 0) {
1169 			ret = -EAGAIN;
1170 			goto unlock;
1171 		}
1172 		if (lkm_flags & DLM_LKF_NOQUEUE)
1173 			noqueue_attempted = 1;
1174 
1175 		if (lockres->l_action != OCFS2_AST_INVALID)
1176 			mlog(ML_ERROR, "lockres %s has action %u pending\n",
1177 			     lockres->l_name, lockres->l_action);
1178 
1179 		if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1180 			lockres->l_action = OCFS2_AST_ATTACH;
1181 			lkm_flags &= ~DLM_LKF_CONVERT;
1182 		} else {
1183 			lockres->l_action = OCFS2_AST_CONVERT;
1184 			lkm_flags |= DLM_LKF_CONVERT;
1185 		}
1186 
1187 		lockres->l_requested = level;
1188 		lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1189 		gen = lockres_set_pending(lockres);
1190 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1191 
1192 		BUG_ON(level == DLM_LOCK_IV);
1193 		BUG_ON(level == DLM_LOCK_NL);
1194 
1195 		mlog(0, "lock %s, convert from %d to level = %d\n",
1196 		     lockres->l_name, lockres->l_level, level);
1197 
1198 		/* call dlm_lock to upgrade lock now */
1199 		ret = ocfs2_dlm_lock(osb->cconn,
1200 				     level,
1201 				     &lockres->l_lksb,
1202 				     lkm_flags,
1203 				     lockres->l_name,
1204 				     OCFS2_LOCK_ID_MAX_LEN - 1,
1205 				     lockres);
1206 		lockres_clear_pending(lockres, gen, osb);
1207 		if (ret) {
1208 			if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
1209 			    (ret != -EAGAIN)) {
1210 				ocfs2_log_dlm_error("ocfs2_dlm_lock",
1211 						    ret, lockres);
1212 			}
1213 			ocfs2_recover_from_dlm_error(lockres, 1);
1214 			goto out;
1215 		}
1216 
1217 		mlog(0, "lock %s, successfull return from ocfs2_dlm_lock\n",
1218 		     lockres->l_name);
1219 
1220 		/* At this point we've gone inside the dlm and need to
1221 		 * complete our work regardless. */
1222 		catch_signals = 0;
1223 
1224 		/* wait for busy to clear and carry on */
1225 		goto again;
1226 	}
1227 
1228 	/* Ok, if we get here then we're good to go. */
1229 	ocfs2_inc_holders(lockres, level);
1230 
1231 	ret = 0;
1232 unlock:
1233 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1234 out:
1235 	/*
1236 	 * This is helping work around a lock inversion between the page lock
1237 	 * and dlm locks.  One path holds the page lock while calling aops
1238 	 * which block acquiring dlm locks.  The voting thread holds dlm
1239 	 * locks while acquiring page locks while down converting data locks.
1240 	 * This block is helping an aop path notice the inversion and back
1241 	 * off to unlock its page lock before trying the dlm lock again.
1242 	 */
1243 	if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1244 	    mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1245 		wait = 0;
1246 		if (lockres_remove_mask_waiter(lockres, &mw))
1247 			ret = -EAGAIN;
1248 		else
1249 			goto again;
1250 	}
1251 	if (wait) {
1252 		ret = ocfs2_wait_for_mask(&mw);
1253 		if (ret == 0)
1254 			goto again;
1255 		mlog_errno(ret);
1256 	}
1257 
1258 	mlog_exit(ret);
1259 	return ret;
1260 }
1261 
1262 static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
1263 				 struct ocfs2_lock_res *lockres,
1264 				 int level)
1265 {
1266 	unsigned long flags;
1267 
1268 	mlog_entry_void();
1269 	spin_lock_irqsave(&lockres->l_lock, flags);
1270 	ocfs2_dec_holders(lockres, level);
1271 	ocfs2_downconvert_on_unlock(osb, lockres);
1272 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1273 	mlog_exit_void();
1274 }
1275 
1276 static int ocfs2_create_new_lock(struct ocfs2_super *osb,
1277 				 struct ocfs2_lock_res *lockres,
1278 				 int ex,
1279 				 int local)
1280 {
1281 	int level =  ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1282 	unsigned long flags;
1283 	u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
1284 
1285 	spin_lock_irqsave(&lockres->l_lock, flags);
1286 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1287 	lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1288 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1289 
1290 	return ocfs2_lock_create(osb, lockres, level, lkm_flags);
1291 }
1292 
1293 /* Grants us an EX lock on the data and metadata resources, skipping
1294  * the normal cluster directory lookup. Use this ONLY on newly created
1295  * inodes which other nodes can't possibly see, and which haven't been
1296  * hashed in the inode hash yet. This can give us a good performance
1297  * increase as it'll skip the network broadcast normally associated
1298  * with creating a new lock resource. */
1299 int ocfs2_create_new_inode_locks(struct inode *inode)
1300 {
1301 	int ret;
1302 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1303 
1304 	BUG_ON(!inode);
1305 	BUG_ON(!ocfs2_inode_is_new(inode));
1306 
1307 	mlog_entry_void();
1308 
1309 	mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
1310 
1311 	/* NOTE: That we don't increment any of the holder counts, nor
1312 	 * do we add anything to a journal handle. Since this is
1313 	 * supposed to be a new inode which the cluster doesn't know
1314 	 * about yet, there is no need to.  As far as the LVB handling
1315 	 * is concerned, this is basically like acquiring an EX lock
1316 	 * on a resource which has an invalid one -- we'll set it
1317 	 * valid when we release the EX. */
1318 
1319 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
1320 	if (ret) {
1321 		mlog_errno(ret);
1322 		goto bail;
1323 	}
1324 
1325 	/*
1326 	 * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
1327 	 * don't use a generation in their lock names.
1328 	 */
1329 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
1330 	if (ret) {
1331 		mlog_errno(ret);
1332 		goto bail;
1333 	}
1334 
1335 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
1336 	if (ret) {
1337 		mlog_errno(ret);
1338 		goto bail;
1339 	}
1340 
1341 bail:
1342 	mlog_exit(ret);
1343 	return ret;
1344 }
1345 
1346 int ocfs2_rw_lock(struct inode *inode, int write)
1347 {
1348 	int status, level;
1349 	struct ocfs2_lock_res *lockres;
1350 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1351 
1352 	BUG_ON(!inode);
1353 
1354 	mlog_entry_void();
1355 
1356 	mlog(0, "inode %llu take %s RW lock\n",
1357 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1358 	     write ? "EXMODE" : "PRMODE");
1359 
1360 	if (ocfs2_mount_local(osb))
1361 		return 0;
1362 
1363 	lockres = &OCFS2_I(inode)->ip_rw_lockres;
1364 
1365 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1366 
1367 	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
1368 				    0);
1369 	if (status < 0)
1370 		mlog_errno(status);
1371 
1372 	mlog_exit(status);
1373 	return status;
1374 }
1375 
1376 void ocfs2_rw_unlock(struct inode *inode, int write)
1377 {
1378 	int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1379 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
1380 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1381 
1382 	mlog_entry_void();
1383 
1384 	mlog(0, "inode %llu drop %s RW lock\n",
1385 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1386 	     write ? "EXMODE" : "PRMODE");
1387 
1388 	if (!ocfs2_mount_local(osb))
1389 		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1390 
1391 	mlog_exit_void();
1392 }
1393 
1394 /*
1395  * ocfs2_open_lock always get PR mode lock.
1396  */
1397 int ocfs2_open_lock(struct inode *inode)
1398 {
1399 	int status = 0;
1400 	struct ocfs2_lock_res *lockres;
1401 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1402 
1403 	BUG_ON(!inode);
1404 
1405 	mlog_entry_void();
1406 
1407 	mlog(0, "inode %llu take PRMODE open lock\n",
1408 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1409 
1410 	if (ocfs2_mount_local(osb))
1411 		goto out;
1412 
1413 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1414 
1415 	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
1416 				    DLM_LOCK_PR, 0, 0);
1417 	if (status < 0)
1418 		mlog_errno(status);
1419 
1420 out:
1421 	mlog_exit(status);
1422 	return status;
1423 }
1424 
1425 int ocfs2_try_open_lock(struct inode *inode, int write)
1426 {
1427 	int status = 0, level;
1428 	struct ocfs2_lock_res *lockres;
1429 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1430 
1431 	BUG_ON(!inode);
1432 
1433 	mlog_entry_void();
1434 
1435 	mlog(0, "inode %llu try to take %s open lock\n",
1436 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1437 	     write ? "EXMODE" : "PRMODE");
1438 
1439 	if (ocfs2_mount_local(osb))
1440 		goto out;
1441 
1442 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1443 
1444 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1445 
1446 	/*
1447 	 * The file system may already holding a PRMODE/EXMODE open lock.
1448 	 * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
1449 	 * other nodes and the -EAGAIN will indicate to the caller that
1450 	 * this inode is still in use.
1451 	 */
1452 	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
1453 				    level, DLM_LKF_NOQUEUE, 0);
1454 
1455 out:
1456 	mlog_exit(status);
1457 	return status;
1458 }
1459 
1460 /*
1461  * ocfs2_open_unlock unlock PR and EX mode open locks.
1462  */
1463 void ocfs2_open_unlock(struct inode *inode)
1464 {
1465 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
1466 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1467 
1468 	mlog_entry_void();
1469 
1470 	mlog(0, "inode %llu drop open lock\n",
1471 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1472 
1473 	if (ocfs2_mount_local(osb))
1474 		goto out;
1475 
1476 	if(lockres->l_ro_holders)
1477 		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
1478 				     DLM_LOCK_PR);
1479 	if(lockres->l_ex_holders)
1480 		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
1481 				     DLM_LOCK_EX);
1482 
1483 out:
1484 	mlog_exit_void();
1485 }
1486 
1487 static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
1488 				     int level)
1489 {
1490 	int ret;
1491 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1492 	unsigned long flags;
1493 	struct ocfs2_mask_waiter mw;
1494 
1495 	ocfs2_init_mask_waiter(&mw);
1496 
1497 retry_cancel:
1498 	spin_lock_irqsave(&lockres->l_lock, flags);
1499 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
1500 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
1501 		if (ret) {
1502 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1503 			ret = ocfs2_cancel_convert(osb, lockres);
1504 			if (ret < 0) {
1505 				mlog_errno(ret);
1506 				goto out;
1507 			}
1508 			goto retry_cancel;
1509 		}
1510 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1511 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1512 
1513 		ocfs2_wait_for_mask(&mw);
1514 		goto retry_cancel;
1515 	}
1516 
1517 	ret = -ERESTARTSYS;
1518 	/*
1519 	 * We may still have gotten the lock, in which case there's no
1520 	 * point to restarting the syscall.
1521 	 */
1522 	if (lockres->l_level == level)
1523 		ret = 0;
1524 
1525 	mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
1526 	     lockres->l_flags, lockres->l_level, lockres->l_action);
1527 
1528 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1529 
1530 out:
1531 	return ret;
1532 }
1533 
1534 /*
1535  * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
1536  * flock() calls. The locking approach this requires is sufficiently
1537  * different from all other cluster lock types that we implement a
1538  * seperate path to the "low-level" dlm calls. In particular:
1539  *
1540  * - No optimization of lock levels is done - we take at exactly
1541  *   what's been requested.
1542  *
1543  * - No lock caching is employed. We immediately downconvert to
1544  *   no-lock at unlock time. This also means flock locks never go on
1545  *   the blocking list).
1546  *
1547  * - Since userspace can trivially deadlock itself with flock, we make
1548  *   sure to allow cancellation of a misbehaving applications flock()
1549  *   request.
1550  *
1551  * - Access to any flock lockres doesn't require concurrency, so we
1552  *   can simplify the code by requiring the caller to guarantee
1553  *   serialization of dlmglue flock calls.
1554  */
1555 int ocfs2_file_lock(struct file *file, int ex, int trylock)
1556 {
1557 	int ret, level = ex ? LKM_EXMODE : LKM_PRMODE;
1558 	unsigned int lkm_flags = trylock ? LKM_NOQUEUE : 0;
1559 	unsigned long flags;
1560 	struct ocfs2_file_private *fp = file->private_data;
1561 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
1562 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1563 	struct ocfs2_mask_waiter mw;
1564 
1565 	ocfs2_init_mask_waiter(&mw);
1566 
1567 	if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
1568 	    (lockres->l_level > DLM_LOCK_NL)) {
1569 		mlog(ML_ERROR,
1570 		     "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
1571 		     "level: %u\n", lockres->l_name, lockres->l_flags,
1572 		     lockres->l_level);
1573 		return -EINVAL;
1574 	}
1575 
1576 	spin_lock_irqsave(&lockres->l_lock, flags);
1577 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1578 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1579 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1580 
1581 		/*
1582 		 * Get the lock at NLMODE to start - that way we
1583 		 * can cancel the upconvert request if need be.
1584 		 */
1585 		ret = ocfs2_lock_create(osb, lockres, LKM_NLMODE, 0);
1586 		if (ret < 0) {
1587 			mlog_errno(ret);
1588 			goto out;
1589 		}
1590 
1591 		ret = ocfs2_wait_for_mask(&mw);
1592 		if (ret) {
1593 			mlog_errno(ret);
1594 			goto out;
1595 		}
1596 		spin_lock_irqsave(&lockres->l_lock, flags);
1597 	}
1598 
1599 	lockres->l_action = OCFS2_AST_CONVERT;
1600 	lkm_flags |= LKM_CONVERT;
1601 	lockres->l_requested = level;
1602 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1603 
1604 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1605 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1606 
1607 	ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags,
1608 			     lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1,
1609 			     lockres);
1610 	if (ret) {
1611 		if (!trylock || (ret != -EAGAIN)) {
1612 			ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
1613 			ret = -EINVAL;
1614 		}
1615 
1616 		ocfs2_recover_from_dlm_error(lockres, 1);
1617 		lockres_remove_mask_waiter(lockres, &mw);
1618 		goto out;
1619 	}
1620 
1621 	ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
1622 	if (ret == -ERESTARTSYS) {
1623 		/*
1624 		 * Userspace can cause deadlock itself with
1625 		 * flock(). Current behavior locally is to allow the
1626 		 * deadlock, but abort the system call if a signal is
1627 		 * received. We follow this example, otherwise a
1628 		 * poorly written program could sit in kernel until
1629 		 * reboot.
1630 		 *
1631 		 * Handling this is a bit more complicated for Ocfs2
1632 		 * though. We can't exit this function with an
1633 		 * outstanding lock request, so a cancel convert is
1634 		 * required. We intentionally overwrite 'ret' - if the
1635 		 * cancel fails and the lock was granted, it's easier
1636 		 * to just bubble sucess back up to the user.
1637 		 */
1638 		ret = ocfs2_flock_handle_signal(lockres, level);
1639 	} else if (!ret && (level > lockres->l_level)) {
1640 		/* Trylock failed asynchronously */
1641 		BUG_ON(!trylock);
1642 		ret = -EAGAIN;
1643 	}
1644 
1645 out:
1646 
1647 	mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
1648 	     lockres->l_name, ex, trylock, ret);
1649 	return ret;
1650 }
1651 
1652 void ocfs2_file_unlock(struct file *file)
1653 {
1654 	int ret;
1655 	unsigned int gen;
1656 	unsigned long flags;
1657 	struct ocfs2_file_private *fp = file->private_data;
1658 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
1659 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1660 	struct ocfs2_mask_waiter mw;
1661 
1662 	ocfs2_init_mask_waiter(&mw);
1663 
1664 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
1665 		return;
1666 
1667 	if (lockres->l_level == LKM_NLMODE)
1668 		return;
1669 
1670 	mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
1671 	     lockres->l_name, lockres->l_flags, lockres->l_level,
1672 	     lockres->l_action);
1673 
1674 	spin_lock_irqsave(&lockres->l_lock, flags);
1675 	/*
1676 	 * Fake a blocking ast for the downconvert code.
1677 	 */
1678 	lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
1679 	lockres->l_blocking = DLM_LOCK_EX;
1680 
1681 	gen = ocfs2_prepare_downconvert(lockres, LKM_NLMODE);
1682 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1683 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1684 
1685 	ret = ocfs2_downconvert_lock(osb, lockres, LKM_NLMODE, 0, gen);
1686 	if (ret) {
1687 		mlog_errno(ret);
1688 		return;
1689 	}
1690 
1691 	ret = ocfs2_wait_for_mask(&mw);
1692 	if (ret)
1693 		mlog_errno(ret);
1694 }
1695 
1696 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
1697 					struct ocfs2_lock_res *lockres)
1698 {
1699 	int kick = 0;
1700 
1701 	mlog_entry_void();
1702 
1703 	/* If we know that another node is waiting on our lock, kick
1704 	 * the downconvert thread * pre-emptively when we reach a release
1705 	 * condition. */
1706 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
1707 		switch(lockres->l_blocking) {
1708 		case DLM_LOCK_EX:
1709 			if (!lockres->l_ex_holders && !lockres->l_ro_holders)
1710 				kick = 1;
1711 			break;
1712 		case DLM_LOCK_PR:
1713 			if (!lockres->l_ex_holders)
1714 				kick = 1;
1715 			break;
1716 		default:
1717 			BUG();
1718 		}
1719 	}
1720 
1721 	if (kick)
1722 		ocfs2_wake_downconvert_thread(osb);
1723 
1724 	mlog_exit_void();
1725 }
1726 
1727 #define OCFS2_SEC_BITS   34
1728 #define OCFS2_SEC_SHIFT  (64 - 34)
1729 #define OCFS2_NSEC_MASK  ((1ULL << OCFS2_SEC_SHIFT) - 1)
1730 
1731 /* LVB only has room for 64 bits of time here so we pack it for
1732  * now. */
1733 static u64 ocfs2_pack_timespec(struct timespec *spec)
1734 {
1735 	u64 res;
1736 	u64 sec = spec->tv_sec;
1737 	u32 nsec = spec->tv_nsec;
1738 
1739 	res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
1740 
1741 	return res;
1742 }
1743 
1744 /* Call this with the lockres locked. I am reasonably sure we don't
1745  * need ip_lock in this function as anyone who would be changing those
1746  * values is supposed to be blocked in ocfs2_inode_lock right now. */
1747 static void __ocfs2_stuff_meta_lvb(struct inode *inode)
1748 {
1749 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1750 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
1751 	struct ocfs2_meta_lvb *lvb;
1752 
1753 	mlog_entry_void();
1754 
1755 	lvb = (struct ocfs2_meta_lvb *)ocfs2_dlm_lvb(&lockres->l_lksb);
1756 
1757 	/*
1758 	 * Invalidate the LVB of a deleted inode - this way other
1759 	 * nodes are forced to go to disk and discover the new inode
1760 	 * status.
1761 	 */
1762 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
1763 		lvb->lvb_version = 0;
1764 		goto out;
1765 	}
1766 
1767 	lvb->lvb_version   = OCFS2_LVB_VERSION;
1768 	lvb->lvb_isize	   = cpu_to_be64(i_size_read(inode));
1769 	lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
1770 	lvb->lvb_iuid      = cpu_to_be32(inode->i_uid);
1771 	lvb->lvb_igid      = cpu_to_be32(inode->i_gid);
1772 	lvb->lvb_imode     = cpu_to_be16(inode->i_mode);
1773 	lvb->lvb_inlink    = cpu_to_be16(inode->i_nlink);
1774 	lvb->lvb_iatime_packed  =
1775 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
1776 	lvb->lvb_ictime_packed =
1777 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
1778 	lvb->lvb_imtime_packed =
1779 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
1780 	lvb->lvb_iattr    = cpu_to_be32(oi->ip_attr);
1781 	lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
1782 	lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
1783 
1784 out:
1785 	mlog_meta_lvb(0, lockres);
1786 
1787 	mlog_exit_void();
1788 }
1789 
1790 static void ocfs2_unpack_timespec(struct timespec *spec,
1791 				  u64 packed_time)
1792 {
1793 	spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
1794 	spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
1795 }
1796 
1797 static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
1798 {
1799 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1800 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
1801 	struct ocfs2_meta_lvb *lvb;
1802 
1803 	mlog_entry_void();
1804 
1805 	mlog_meta_lvb(0, lockres);
1806 
1807 	lvb = (struct ocfs2_meta_lvb *)ocfs2_dlm_lvb(&lockres->l_lksb);
1808 
1809 	/* We're safe here without the lockres lock... */
1810 	spin_lock(&oi->ip_lock);
1811 	oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
1812 	i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
1813 
1814 	oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
1815 	oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
1816 	ocfs2_set_inode_flags(inode);
1817 
1818 	/* fast-symlinks are a special case */
1819 	if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
1820 		inode->i_blocks = 0;
1821 	else
1822 		inode->i_blocks = ocfs2_inode_sector_count(inode);
1823 
1824 	inode->i_uid     = be32_to_cpu(lvb->lvb_iuid);
1825 	inode->i_gid     = be32_to_cpu(lvb->lvb_igid);
1826 	inode->i_mode    = be16_to_cpu(lvb->lvb_imode);
1827 	inode->i_nlink   = be16_to_cpu(lvb->lvb_inlink);
1828 	ocfs2_unpack_timespec(&inode->i_atime,
1829 			      be64_to_cpu(lvb->lvb_iatime_packed));
1830 	ocfs2_unpack_timespec(&inode->i_mtime,
1831 			      be64_to_cpu(lvb->lvb_imtime_packed));
1832 	ocfs2_unpack_timespec(&inode->i_ctime,
1833 			      be64_to_cpu(lvb->lvb_ictime_packed));
1834 	spin_unlock(&oi->ip_lock);
1835 
1836 	mlog_exit_void();
1837 }
1838 
1839 static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
1840 					      struct ocfs2_lock_res *lockres)
1841 {
1842 	struct ocfs2_meta_lvb *lvb =
1843 		(struct ocfs2_meta_lvb *)ocfs2_dlm_lvb(&lockres->l_lksb);
1844 
1845 	if (lvb->lvb_version == OCFS2_LVB_VERSION
1846 	    && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
1847 		return 1;
1848 	return 0;
1849 }
1850 
1851 /* Determine whether a lock resource needs to be refreshed, and
1852  * arbitrate who gets to refresh it.
1853  *
1854  *   0 means no refresh needed.
1855  *
1856  *   > 0 means you need to refresh this and you MUST call
1857  *   ocfs2_complete_lock_res_refresh afterwards. */
1858 static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
1859 {
1860 	unsigned long flags;
1861 	int status = 0;
1862 
1863 	mlog_entry_void();
1864 
1865 refresh_check:
1866 	spin_lock_irqsave(&lockres->l_lock, flags);
1867 	if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
1868 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1869 		goto bail;
1870 	}
1871 
1872 	if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
1873 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1874 
1875 		ocfs2_wait_on_refreshing_lock(lockres);
1876 		goto refresh_check;
1877 	}
1878 
1879 	/* Ok, I'll be the one to refresh this lock. */
1880 	lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
1881 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1882 
1883 	status = 1;
1884 bail:
1885 	mlog_exit(status);
1886 	return status;
1887 }
1888 
1889 /* If status is non zero, I'll mark it as not being in refresh
1890  * anymroe, but i won't clear the needs refresh flag. */
1891 static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
1892 						   int status)
1893 {
1894 	unsigned long flags;
1895 	mlog_entry_void();
1896 
1897 	spin_lock_irqsave(&lockres->l_lock, flags);
1898 	lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
1899 	if (!status)
1900 		lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
1901 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1902 
1903 	wake_up(&lockres->l_event);
1904 
1905 	mlog_exit_void();
1906 }
1907 
1908 /* may or may not return a bh if it went to disk. */
1909 static int ocfs2_inode_lock_update(struct inode *inode,
1910 				  struct buffer_head **bh)
1911 {
1912 	int status = 0;
1913 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1914 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
1915 	struct ocfs2_dinode *fe;
1916 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1917 
1918 	mlog_entry_void();
1919 
1920 	if (ocfs2_mount_local(osb))
1921 		goto bail;
1922 
1923 	spin_lock(&oi->ip_lock);
1924 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
1925 		mlog(0, "Orphaned inode %llu was deleted while we "
1926 		     "were waiting on a lock. ip_flags = 0x%x\n",
1927 		     (unsigned long long)oi->ip_blkno, oi->ip_flags);
1928 		spin_unlock(&oi->ip_lock);
1929 		status = -ENOENT;
1930 		goto bail;
1931 	}
1932 	spin_unlock(&oi->ip_lock);
1933 
1934 	if (!ocfs2_should_refresh_lock_res(lockres))
1935 		goto bail;
1936 
1937 	/* This will discard any caching information we might have had
1938 	 * for the inode metadata. */
1939 	ocfs2_metadata_cache_purge(inode);
1940 
1941 	ocfs2_extent_map_trunc(inode, 0);
1942 
1943 	if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
1944 		mlog(0, "Trusting LVB on inode %llu\n",
1945 		     (unsigned long long)oi->ip_blkno);
1946 		ocfs2_refresh_inode_from_lvb(inode);
1947 	} else {
1948 		/* Boo, we have to go to disk. */
1949 		/* read bh, cast, ocfs2_refresh_inode */
1950 		status = ocfs2_read_block(OCFS2_SB(inode->i_sb), oi->ip_blkno,
1951 					  bh, OCFS2_BH_CACHED, inode);
1952 		if (status < 0) {
1953 			mlog_errno(status);
1954 			goto bail_refresh;
1955 		}
1956 		fe = (struct ocfs2_dinode *) (*bh)->b_data;
1957 
1958 		/* This is a good chance to make sure we're not
1959 		 * locking an invalid object.
1960 		 *
1961 		 * We bug on a stale inode here because we checked
1962 		 * above whether it was wiped from disk. The wiping
1963 		 * node provides a guarantee that we receive that
1964 		 * message and can mark the inode before dropping any
1965 		 * locks associated with it. */
1966 		if (!OCFS2_IS_VALID_DINODE(fe)) {
1967 			OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
1968 			status = -EIO;
1969 			goto bail_refresh;
1970 		}
1971 		mlog_bug_on_msg(inode->i_generation !=
1972 				le32_to_cpu(fe->i_generation),
1973 				"Invalid dinode %llu disk generation: %u "
1974 				"inode->i_generation: %u\n",
1975 				(unsigned long long)oi->ip_blkno,
1976 				le32_to_cpu(fe->i_generation),
1977 				inode->i_generation);
1978 		mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
1979 				!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
1980 				"Stale dinode %llu dtime: %llu flags: 0x%x\n",
1981 				(unsigned long long)oi->ip_blkno,
1982 				(unsigned long long)le64_to_cpu(fe->i_dtime),
1983 				le32_to_cpu(fe->i_flags));
1984 
1985 		ocfs2_refresh_inode(inode, fe);
1986 	}
1987 
1988 	status = 0;
1989 bail_refresh:
1990 	ocfs2_complete_lock_res_refresh(lockres, status);
1991 bail:
1992 	mlog_exit(status);
1993 	return status;
1994 }
1995 
1996 static int ocfs2_assign_bh(struct inode *inode,
1997 			   struct buffer_head **ret_bh,
1998 			   struct buffer_head *passed_bh)
1999 {
2000 	int status;
2001 
2002 	if (passed_bh) {
2003 		/* Ok, the update went to disk for us, use the
2004 		 * returned bh. */
2005 		*ret_bh = passed_bh;
2006 		get_bh(*ret_bh);
2007 
2008 		return 0;
2009 	}
2010 
2011 	status = ocfs2_read_block(OCFS2_SB(inode->i_sb),
2012 				  OCFS2_I(inode)->ip_blkno,
2013 				  ret_bh,
2014 				  OCFS2_BH_CACHED,
2015 				  inode);
2016 	if (status < 0)
2017 		mlog_errno(status);
2018 
2019 	return status;
2020 }
2021 
2022 /*
2023  * returns < 0 error if the callback will never be called, otherwise
2024  * the result of the lock will be communicated via the callback.
2025  */
2026 int ocfs2_inode_lock_full(struct inode *inode,
2027 			 struct buffer_head **ret_bh,
2028 			 int ex,
2029 			 int arg_flags)
2030 {
2031 	int status, level, acquired;
2032 	u32 dlm_flags;
2033 	struct ocfs2_lock_res *lockres = NULL;
2034 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2035 	struct buffer_head *local_bh = NULL;
2036 
2037 	BUG_ON(!inode);
2038 
2039 	mlog_entry_void();
2040 
2041 	mlog(0, "inode %llu, take %s META lock\n",
2042 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2043 	     ex ? "EXMODE" : "PRMODE");
2044 
2045 	status = 0;
2046 	acquired = 0;
2047 	/* We'll allow faking a readonly metadata lock for
2048 	 * rodevices. */
2049 	if (ocfs2_is_hard_readonly(osb)) {
2050 		if (ex)
2051 			status = -EROFS;
2052 		goto bail;
2053 	}
2054 
2055 	if (ocfs2_mount_local(osb))
2056 		goto local;
2057 
2058 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2059 		ocfs2_wait_for_recovery(osb);
2060 
2061 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2062 	level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2063 	dlm_flags = 0;
2064 	if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
2065 		dlm_flags |= DLM_LKF_NOQUEUE;
2066 
2067 	status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags);
2068 	if (status < 0) {
2069 		if (status != -EAGAIN && status != -EIOCBRETRY)
2070 			mlog_errno(status);
2071 		goto bail;
2072 	}
2073 
2074 	/* Notify the error cleanup path to drop the cluster lock. */
2075 	acquired = 1;
2076 
2077 	/* We wait twice because a node may have died while we were in
2078 	 * the lower dlm layers. The second time though, we've
2079 	 * committed to owning this lock so we don't allow signals to
2080 	 * abort the operation. */
2081 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2082 		ocfs2_wait_for_recovery(osb);
2083 
2084 local:
2085 	/*
2086 	 * We only see this flag if we're being called from
2087 	 * ocfs2_read_locked_inode(). It means we're locking an inode
2088 	 * which hasn't been populated yet, so clear the refresh flag
2089 	 * and let the caller handle it.
2090 	 */
2091 	if (inode->i_state & I_NEW) {
2092 		status = 0;
2093 		if (lockres)
2094 			ocfs2_complete_lock_res_refresh(lockres, 0);
2095 		goto bail;
2096 	}
2097 
2098 	/* This is fun. The caller may want a bh back, or it may
2099 	 * not. ocfs2_inode_lock_update definitely wants one in, but
2100 	 * may or may not read one, depending on what's in the
2101 	 * LVB. The result of all of this is that we've *only* gone to
2102 	 * disk if we have to, so the complexity is worthwhile. */
2103 	status = ocfs2_inode_lock_update(inode, &local_bh);
2104 	if (status < 0) {
2105 		if (status != -ENOENT)
2106 			mlog_errno(status);
2107 		goto bail;
2108 	}
2109 
2110 	if (ret_bh) {
2111 		status = ocfs2_assign_bh(inode, ret_bh, local_bh);
2112 		if (status < 0) {
2113 			mlog_errno(status);
2114 			goto bail;
2115 		}
2116 	}
2117 
2118 bail:
2119 	if (status < 0) {
2120 		if (ret_bh && (*ret_bh)) {
2121 			brelse(*ret_bh);
2122 			*ret_bh = NULL;
2123 		}
2124 		if (acquired)
2125 			ocfs2_inode_unlock(inode, ex);
2126 	}
2127 
2128 	if (local_bh)
2129 		brelse(local_bh);
2130 
2131 	mlog_exit(status);
2132 	return status;
2133 }
2134 
2135 /*
2136  * This is working around a lock inversion between tasks acquiring DLM
2137  * locks while holding a page lock and the downconvert thread which
2138  * blocks dlm lock acquiry while acquiring page locks.
2139  *
2140  * ** These _with_page variantes are only intended to be called from aop
2141  * methods that hold page locks and return a very specific *positive* error
2142  * code that aop methods pass up to the VFS -- test for errors with != 0. **
2143  *
2144  * The DLM is called such that it returns -EAGAIN if it would have
2145  * blocked waiting for the downconvert thread.  In that case we unlock
2146  * our page so the downconvert thread can make progress.  Once we've
2147  * done this we have to return AOP_TRUNCATED_PAGE so the aop method
2148  * that called us can bubble that back up into the VFS who will then
2149  * immediately retry the aop call.
2150  *
2151  * We do a blocking lock and immediate unlock before returning, though, so that
2152  * the lock has a great chance of being cached on this node by the time the VFS
2153  * calls back to retry the aop.    This has a potential to livelock as nodes
2154  * ping locks back and forth, but that's a risk we're willing to take to avoid
2155  * the lock inversion simply.
2156  */
2157 int ocfs2_inode_lock_with_page(struct inode *inode,
2158 			      struct buffer_head **ret_bh,
2159 			      int ex,
2160 			      struct page *page)
2161 {
2162 	int ret;
2163 
2164 	ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
2165 	if (ret == -EAGAIN) {
2166 		unlock_page(page);
2167 		if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
2168 			ocfs2_inode_unlock(inode, ex);
2169 		ret = AOP_TRUNCATED_PAGE;
2170 	}
2171 
2172 	return ret;
2173 }
2174 
2175 int ocfs2_inode_lock_atime(struct inode *inode,
2176 			  struct vfsmount *vfsmnt,
2177 			  int *level)
2178 {
2179 	int ret;
2180 
2181 	mlog_entry_void();
2182 	ret = ocfs2_inode_lock(inode, NULL, 0);
2183 	if (ret < 0) {
2184 		mlog_errno(ret);
2185 		return ret;
2186 	}
2187 
2188 	/*
2189 	 * If we should update atime, we will get EX lock,
2190 	 * otherwise we just get PR lock.
2191 	 */
2192 	if (ocfs2_should_update_atime(inode, vfsmnt)) {
2193 		struct buffer_head *bh = NULL;
2194 
2195 		ocfs2_inode_unlock(inode, 0);
2196 		ret = ocfs2_inode_lock(inode, &bh, 1);
2197 		if (ret < 0) {
2198 			mlog_errno(ret);
2199 			return ret;
2200 		}
2201 		*level = 1;
2202 		if (ocfs2_should_update_atime(inode, vfsmnt))
2203 			ocfs2_update_inode_atime(inode, bh);
2204 		if (bh)
2205 			brelse(bh);
2206 	} else
2207 		*level = 0;
2208 
2209 	mlog_exit(ret);
2210 	return ret;
2211 }
2212 
2213 void ocfs2_inode_unlock(struct inode *inode,
2214 		       int ex)
2215 {
2216 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2217 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
2218 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2219 
2220 	mlog_entry_void();
2221 
2222 	mlog(0, "inode %llu drop %s META lock\n",
2223 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2224 	     ex ? "EXMODE" : "PRMODE");
2225 
2226 	if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) &&
2227 	    !ocfs2_mount_local(osb))
2228 		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
2229 
2230 	mlog_exit_void();
2231 }
2232 
2233 int ocfs2_super_lock(struct ocfs2_super *osb,
2234 		     int ex)
2235 {
2236 	int status = 0;
2237 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2238 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2239 
2240 	mlog_entry_void();
2241 
2242 	if (ocfs2_is_hard_readonly(osb))
2243 		return -EROFS;
2244 
2245 	if (ocfs2_mount_local(osb))
2246 		goto bail;
2247 
2248 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
2249 	if (status < 0) {
2250 		mlog_errno(status);
2251 		goto bail;
2252 	}
2253 
2254 	/* The super block lock path is really in the best position to
2255 	 * know when resources covered by the lock need to be
2256 	 * refreshed, so we do it here. Of course, making sense of
2257 	 * everything is up to the caller :) */
2258 	status = ocfs2_should_refresh_lock_res(lockres);
2259 	if (status < 0) {
2260 		mlog_errno(status);
2261 		goto bail;
2262 	}
2263 	if (status) {
2264 		status = ocfs2_refresh_slot_info(osb);
2265 
2266 		ocfs2_complete_lock_res_refresh(lockres, status);
2267 
2268 		if (status < 0)
2269 			mlog_errno(status);
2270 	}
2271 bail:
2272 	mlog_exit(status);
2273 	return status;
2274 }
2275 
2276 void ocfs2_super_unlock(struct ocfs2_super *osb,
2277 			int ex)
2278 {
2279 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2280 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2281 
2282 	if (!ocfs2_mount_local(osb))
2283 		ocfs2_cluster_unlock(osb, lockres, level);
2284 }
2285 
2286 int ocfs2_rename_lock(struct ocfs2_super *osb)
2287 {
2288 	int status;
2289 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2290 
2291 	if (ocfs2_is_hard_readonly(osb))
2292 		return -EROFS;
2293 
2294 	if (ocfs2_mount_local(osb))
2295 		return 0;
2296 
2297 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2298 	if (status < 0)
2299 		mlog_errno(status);
2300 
2301 	return status;
2302 }
2303 
2304 void ocfs2_rename_unlock(struct ocfs2_super *osb)
2305 {
2306 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2307 
2308 	if (!ocfs2_mount_local(osb))
2309 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2310 }
2311 
2312 int ocfs2_dentry_lock(struct dentry *dentry, int ex)
2313 {
2314 	int ret;
2315 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2316 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2317 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2318 
2319 	BUG_ON(!dl);
2320 
2321 	if (ocfs2_is_hard_readonly(osb))
2322 		return -EROFS;
2323 
2324 	if (ocfs2_mount_local(osb))
2325 		return 0;
2326 
2327 	ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2328 	if (ret < 0)
2329 		mlog_errno(ret);
2330 
2331 	return ret;
2332 }
2333 
2334 void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2335 {
2336 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2337 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2338 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2339 
2340 	if (!ocfs2_mount_local(osb))
2341 		ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
2342 }
2343 
2344 /* Reference counting of the dlm debug structure. We want this because
2345  * open references on the debug inodes can live on after a mount, so
2346  * we can't rely on the ocfs2_super to always exist. */
2347 static void ocfs2_dlm_debug_free(struct kref *kref)
2348 {
2349 	struct ocfs2_dlm_debug *dlm_debug;
2350 
2351 	dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
2352 
2353 	kfree(dlm_debug);
2354 }
2355 
2356 void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
2357 {
2358 	if (dlm_debug)
2359 		kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
2360 }
2361 
2362 static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
2363 {
2364 	kref_get(&debug->d_refcnt);
2365 }
2366 
2367 struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
2368 {
2369 	struct ocfs2_dlm_debug *dlm_debug;
2370 
2371 	dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
2372 	if (!dlm_debug) {
2373 		mlog_errno(-ENOMEM);
2374 		goto out;
2375 	}
2376 
2377 	kref_init(&dlm_debug->d_refcnt);
2378 	INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
2379 	dlm_debug->d_locking_state = NULL;
2380 out:
2381 	return dlm_debug;
2382 }
2383 
2384 /* Access to this is arbitrated for us via seq_file->sem. */
2385 struct ocfs2_dlm_seq_priv {
2386 	struct ocfs2_dlm_debug *p_dlm_debug;
2387 	struct ocfs2_lock_res p_iter_res;
2388 	struct ocfs2_lock_res p_tmp_res;
2389 };
2390 
2391 static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
2392 						 struct ocfs2_dlm_seq_priv *priv)
2393 {
2394 	struct ocfs2_lock_res *iter, *ret = NULL;
2395 	struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
2396 
2397 	assert_spin_locked(&ocfs2_dlm_tracking_lock);
2398 
2399 	list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
2400 		/* discover the head of the list */
2401 		if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
2402 			mlog(0, "End of list found, %p\n", ret);
2403 			break;
2404 		}
2405 
2406 		/* We track our "dummy" iteration lockres' by a NULL
2407 		 * l_ops field. */
2408 		if (iter->l_ops != NULL) {
2409 			ret = iter;
2410 			break;
2411 		}
2412 	}
2413 
2414 	return ret;
2415 }
2416 
2417 static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
2418 {
2419 	struct ocfs2_dlm_seq_priv *priv = m->private;
2420 	struct ocfs2_lock_res *iter;
2421 
2422 	spin_lock(&ocfs2_dlm_tracking_lock);
2423 	iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
2424 	if (iter) {
2425 		/* Since lockres' have the lifetime of their container
2426 		 * (which can be inodes, ocfs2_supers, etc) we want to
2427 		 * copy this out to a temporary lockres while still
2428 		 * under the spinlock. Obviously after this we can't
2429 		 * trust any pointers on the copy returned, but that's
2430 		 * ok as the information we want isn't typically held
2431 		 * in them. */
2432 		priv->p_tmp_res = *iter;
2433 		iter = &priv->p_tmp_res;
2434 	}
2435 	spin_unlock(&ocfs2_dlm_tracking_lock);
2436 
2437 	return iter;
2438 }
2439 
2440 static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
2441 {
2442 }
2443 
2444 static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
2445 {
2446 	struct ocfs2_dlm_seq_priv *priv = m->private;
2447 	struct ocfs2_lock_res *iter = v;
2448 	struct ocfs2_lock_res *dummy = &priv->p_iter_res;
2449 
2450 	spin_lock(&ocfs2_dlm_tracking_lock);
2451 	iter = ocfs2_dlm_next_res(iter, priv);
2452 	list_del_init(&dummy->l_debug_list);
2453 	if (iter) {
2454 		list_add(&dummy->l_debug_list, &iter->l_debug_list);
2455 		priv->p_tmp_res = *iter;
2456 		iter = &priv->p_tmp_res;
2457 	}
2458 	spin_unlock(&ocfs2_dlm_tracking_lock);
2459 
2460 	return iter;
2461 }
2462 
2463 /* So that debugfs.ocfs2 can determine which format is being used */
2464 #define OCFS2_DLM_DEBUG_STR_VERSION 1
2465 static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
2466 {
2467 	int i;
2468 	char *lvb;
2469 	struct ocfs2_lock_res *lockres = v;
2470 
2471 	if (!lockres)
2472 		return -EINVAL;
2473 
2474 	seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
2475 
2476 	if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
2477 		seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
2478 			   lockres->l_name,
2479 			   (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
2480 	else
2481 		seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
2482 
2483 	seq_printf(m, "%d\t"
2484 		   "0x%lx\t"
2485 		   "0x%x\t"
2486 		   "0x%x\t"
2487 		   "%u\t"
2488 		   "%u\t"
2489 		   "%d\t"
2490 		   "%d\t",
2491 		   lockres->l_level,
2492 		   lockres->l_flags,
2493 		   lockres->l_action,
2494 		   lockres->l_unlock_action,
2495 		   lockres->l_ro_holders,
2496 		   lockres->l_ex_holders,
2497 		   lockres->l_requested,
2498 		   lockres->l_blocking);
2499 
2500 	/* Dump the raw LVB */
2501 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2502 	for(i = 0; i < DLM_LVB_LEN; i++)
2503 		seq_printf(m, "0x%x\t", lvb[i]);
2504 
2505 	/* End the line */
2506 	seq_printf(m, "\n");
2507 	return 0;
2508 }
2509 
2510 static const struct seq_operations ocfs2_dlm_seq_ops = {
2511 	.start =	ocfs2_dlm_seq_start,
2512 	.stop =		ocfs2_dlm_seq_stop,
2513 	.next =		ocfs2_dlm_seq_next,
2514 	.show =		ocfs2_dlm_seq_show,
2515 };
2516 
2517 static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
2518 {
2519 	struct seq_file *seq = (struct seq_file *) file->private_data;
2520 	struct ocfs2_dlm_seq_priv *priv = seq->private;
2521 	struct ocfs2_lock_res *res = &priv->p_iter_res;
2522 
2523 	ocfs2_remove_lockres_tracking(res);
2524 	ocfs2_put_dlm_debug(priv->p_dlm_debug);
2525 	return seq_release_private(inode, file);
2526 }
2527 
2528 static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
2529 {
2530 	int ret;
2531 	struct ocfs2_dlm_seq_priv *priv;
2532 	struct seq_file *seq;
2533 	struct ocfs2_super *osb;
2534 
2535 	priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
2536 	if (!priv) {
2537 		ret = -ENOMEM;
2538 		mlog_errno(ret);
2539 		goto out;
2540 	}
2541 	osb = inode->i_private;
2542 	ocfs2_get_dlm_debug(osb->osb_dlm_debug);
2543 	priv->p_dlm_debug = osb->osb_dlm_debug;
2544 	INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
2545 
2546 	ret = seq_open(file, &ocfs2_dlm_seq_ops);
2547 	if (ret) {
2548 		kfree(priv);
2549 		mlog_errno(ret);
2550 		goto out;
2551 	}
2552 
2553 	seq = (struct seq_file *) file->private_data;
2554 	seq->private = priv;
2555 
2556 	ocfs2_add_lockres_tracking(&priv->p_iter_res,
2557 				   priv->p_dlm_debug);
2558 
2559 out:
2560 	return ret;
2561 }
2562 
2563 static const struct file_operations ocfs2_dlm_debug_fops = {
2564 	.open =		ocfs2_dlm_debug_open,
2565 	.release =	ocfs2_dlm_debug_release,
2566 	.read =		seq_read,
2567 	.llseek =	seq_lseek,
2568 };
2569 
2570 static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
2571 {
2572 	int ret = 0;
2573 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2574 
2575 	dlm_debug->d_locking_state = debugfs_create_file("locking_state",
2576 							 S_IFREG|S_IRUSR,
2577 							 osb->osb_debug_root,
2578 							 osb,
2579 							 &ocfs2_dlm_debug_fops);
2580 	if (!dlm_debug->d_locking_state) {
2581 		ret = -EINVAL;
2582 		mlog(ML_ERROR,
2583 		     "Unable to create locking state debugfs file.\n");
2584 		goto out;
2585 	}
2586 
2587 	ocfs2_get_dlm_debug(dlm_debug);
2588 out:
2589 	return ret;
2590 }
2591 
2592 static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
2593 {
2594 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2595 
2596 	if (dlm_debug) {
2597 		debugfs_remove(dlm_debug->d_locking_state);
2598 		ocfs2_put_dlm_debug(dlm_debug);
2599 	}
2600 }
2601 
2602 int ocfs2_dlm_init(struct ocfs2_super *osb)
2603 {
2604 	int status = 0;
2605 	struct ocfs2_cluster_connection *conn = NULL;
2606 
2607 	mlog_entry_void();
2608 
2609 	if (ocfs2_mount_local(osb)) {
2610 		osb->node_num = 0;
2611 		goto local;
2612 	}
2613 
2614 	status = ocfs2_dlm_init_debug(osb);
2615 	if (status < 0) {
2616 		mlog_errno(status);
2617 		goto bail;
2618 	}
2619 
2620 	/* launch downconvert thread */
2621 	osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc");
2622 	if (IS_ERR(osb->dc_task)) {
2623 		status = PTR_ERR(osb->dc_task);
2624 		osb->dc_task = NULL;
2625 		mlog_errno(status);
2626 		goto bail;
2627 	}
2628 
2629 	/* for now, uuid == domain */
2630 	status = ocfs2_cluster_connect(osb->osb_cluster_stack,
2631 				       osb->uuid_str,
2632 				       strlen(osb->uuid_str),
2633 				       ocfs2_do_node_down, osb,
2634 				       &conn);
2635 	if (status) {
2636 		mlog_errno(status);
2637 		goto bail;
2638 	}
2639 
2640 	status = ocfs2_cluster_this_node(&osb->node_num);
2641 	if (status < 0) {
2642 		mlog_errno(status);
2643 		mlog(ML_ERROR,
2644 		     "could not find this host's node number\n");
2645 		ocfs2_cluster_disconnect(conn, 0);
2646 		goto bail;
2647 	}
2648 
2649 local:
2650 	ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
2651 	ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
2652 
2653 	osb->cconn = conn;
2654 
2655 	status = 0;
2656 bail:
2657 	if (status < 0) {
2658 		ocfs2_dlm_shutdown_debug(osb);
2659 		if (osb->dc_task)
2660 			kthread_stop(osb->dc_task);
2661 	}
2662 
2663 	mlog_exit(status);
2664 	return status;
2665 }
2666 
2667 void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
2668 			int hangup_pending)
2669 {
2670 	mlog_entry_void();
2671 
2672 	ocfs2_drop_osb_locks(osb);
2673 
2674 	/*
2675 	 * Now that we have dropped all locks and ocfs2_dismount_volume()
2676 	 * has disabled recovery, the DLM won't be talking to us.  It's
2677 	 * safe to tear things down before disconnecting the cluster.
2678 	 */
2679 
2680 	if (osb->dc_task) {
2681 		kthread_stop(osb->dc_task);
2682 		osb->dc_task = NULL;
2683 	}
2684 
2685 	ocfs2_lock_res_free(&osb->osb_super_lockres);
2686 	ocfs2_lock_res_free(&osb->osb_rename_lockres);
2687 
2688 	ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
2689 	osb->cconn = NULL;
2690 
2691 	ocfs2_dlm_shutdown_debug(osb);
2692 
2693 	mlog_exit_void();
2694 }
2695 
2696 static void ocfs2_unlock_ast(void *opaque, int error)
2697 {
2698 	struct ocfs2_lock_res *lockres = opaque;
2699 	unsigned long flags;
2700 
2701 	mlog_entry_void();
2702 
2703 	mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name,
2704 	     lockres->l_unlock_action);
2705 
2706 	spin_lock_irqsave(&lockres->l_lock, flags);
2707 	if (error) {
2708 		mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
2709 		     "unlock_action %d\n", error, lockres->l_name,
2710 		     lockres->l_unlock_action);
2711 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2712 		return;
2713 	}
2714 
2715 	switch(lockres->l_unlock_action) {
2716 	case OCFS2_UNLOCK_CANCEL_CONVERT:
2717 		mlog(0, "Cancel convert success for %s\n", lockres->l_name);
2718 		lockres->l_action = OCFS2_AST_INVALID;
2719 		break;
2720 	case OCFS2_UNLOCK_DROP_LOCK:
2721 		lockres->l_level = DLM_LOCK_IV;
2722 		break;
2723 	default:
2724 		BUG();
2725 	}
2726 
2727 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
2728 	lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
2729 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2730 
2731 	wake_up(&lockres->l_event);
2732 
2733 	mlog_exit_void();
2734 }
2735 
2736 static int ocfs2_drop_lock(struct ocfs2_super *osb,
2737 			   struct ocfs2_lock_res *lockres)
2738 {
2739 	int ret;
2740 	unsigned long flags;
2741 	u32 lkm_flags = 0;
2742 
2743 	/* We didn't get anywhere near actually using this lockres. */
2744 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
2745 		goto out;
2746 
2747 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
2748 		lkm_flags |= DLM_LKF_VALBLK;
2749 
2750 	spin_lock_irqsave(&lockres->l_lock, flags);
2751 
2752 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
2753 			"lockres %s, flags 0x%lx\n",
2754 			lockres->l_name, lockres->l_flags);
2755 
2756 	while (lockres->l_flags & OCFS2_LOCK_BUSY) {
2757 		mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
2758 		     "%u, unlock_action = %u\n",
2759 		     lockres->l_name, lockres->l_flags, lockres->l_action,
2760 		     lockres->l_unlock_action);
2761 
2762 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2763 
2764 		/* XXX: Today we just wait on any busy
2765 		 * locks... Perhaps we need to cancel converts in the
2766 		 * future? */
2767 		ocfs2_wait_on_busy_lock(lockres);
2768 
2769 		spin_lock_irqsave(&lockres->l_lock, flags);
2770 	}
2771 
2772 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
2773 		if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
2774 		    lockres->l_level == DLM_LOCK_EX &&
2775 		    !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
2776 			lockres->l_ops->set_lvb(lockres);
2777 	}
2778 
2779 	if (lockres->l_flags & OCFS2_LOCK_BUSY)
2780 		mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
2781 		     lockres->l_name);
2782 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
2783 		mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
2784 
2785 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
2786 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2787 		goto out;
2788 	}
2789 
2790 	lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
2791 
2792 	/* make sure we never get here while waiting for an ast to
2793 	 * fire. */
2794 	BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
2795 
2796 	/* is this necessary? */
2797 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2798 	lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
2799 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2800 
2801 	mlog(0, "lock %s\n", lockres->l_name);
2802 
2803 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags,
2804 			       lockres);
2805 	if (ret) {
2806 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
2807 		mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
2808 		ocfs2_dlm_dump_lksb(&lockres->l_lksb);
2809 		BUG();
2810 	}
2811 	mlog(0, "lock %s, successfull return from ocfs2_dlm_unlock\n",
2812 	     lockres->l_name);
2813 
2814 	ocfs2_wait_on_busy_lock(lockres);
2815 out:
2816 	mlog_exit(0);
2817 	return 0;
2818 }
2819 
2820 /* Mark the lockres as being dropped. It will no longer be
2821  * queued if blocking, but we still may have to wait on it
2822  * being dequeued from the downconvert thread before we can consider
2823  * it safe to drop.
2824  *
2825  * You can *not* attempt to call cluster_lock on this lockres anymore. */
2826 void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
2827 {
2828 	int status;
2829 	struct ocfs2_mask_waiter mw;
2830 	unsigned long flags;
2831 
2832 	ocfs2_init_mask_waiter(&mw);
2833 
2834 	spin_lock_irqsave(&lockres->l_lock, flags);
2835 	lockres->l_flags |= OCFS2_LOCK_FREEING;
2836 	while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
2837 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
2838 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2839 
2840 		mlog(0, "Waiting on lockres %s\n", lockres->l_name);
2841 
2842 		status = ocfs2_wait_for_mask(&mw);
2843 		if (status)
2844 			mlog_errno(status);
2845 
2846 		spin_lock_irqsave(&lockres->l_lock, flags);
2847 	}
2848 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2849 }
2850 
2851 void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
2852 			       struct ocfs2_lock_res *lockres)
2853 {
2854 	int ret;
2855 
2856 	ocfs2_mark_lockres_freeing(lockres);
2857 	ret = ocfs2_drop_lock(osb, lockres);
2858 	if (ret)
2859 		mlog_errno(ret);
2860 }
2861 
2862 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
2863 {
2864 	ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
2865 	ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
2866 }
2867 
2868 int ocfs2_drop_inode_locks(struct inode *inode)
2869 {
2870 	int status, err;
2871 
2872 	mlog_entry_void();
2873 
2874 	/* No need to call ocfs2_mark_lockres_freeing here -
2875 	 * ocfs2_clear_inode has done it for us. */
2876 
2877 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2878 			      &OCFS2_I(inode)->ip_open_lockres);
2879 	if (err < 0)
2880 		mlog_errno(err);
2881 
2882 	status = err;
2883 
2884 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2885 			      &OCFS2_I(inode)->ip_inode_lockres);
2886 	if (err < 0)
2887 		mlog_errno(err);
2888 	if (err < 0 && !status)
2889 		status = err;
2890 
2891 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2892 			      &OCFS2_I(inode)->ip_rw_lockres);
2893 	if (err < 0)
2894 		mlog_errno(err);
2895 	if (err < 0 && !status)
2896 		status = err;
2897 
2898 	mlog_exit(status);
2899 	return status;
2900 }
2901 
2902 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
2903 					      int new_level)
2904 {
2905 	assert_spin_locked(&lockres->l_lock);
2906 
2907 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
2908 
2909 	if (lockres->l_level <= new_level) {
2910 		mlog(ML_ERROR, "lockres->l_level (%d) <= new_level (%d)\n",
2911 		     lockres->l_level, new_level);
2912 		BUG();
2913 	}
2914 
2915 	mlog(0, "lock %s, new_level = %d, l_blocking = %d\n",
2916 	     lockres->l_name, new_level, lockres->l_blocking);
2917 
2918 	lockres->l_action = OCFS2_AST_DOWNCONVERT;
2919 	lockres->l_requested = new_level;
2920 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2921 	return lockres_set_pending(lockres);
2922 }
2923 
2924 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
2925 				  struct ocfs2_lock_res *lockres,
2926 				  int new_level,
2927 				  int lvb,
2928 				  unsigned int generation)
2929 {
2930 	int ret;
2931 	u32 dlm_flags = DLM_LKF_CONVERT;
2932 
2933 	mlog_entry_void();
2934 
2935 	if (lvb)
2936 		dlm_flags |= DLM_LKF_VALBLK;
2937 
2938 	ret = ocfs2_dlm_lock(osb->cconn,
2939 			     new_level,
2940 			     &lockres->l_lksb,
2941 			     dlm_flags,
2942 			     lockres->l_name,
2943 			     OCFS2_LOCK_ID_MAX_LEN - 1,
2944 			     lockres);
2945 	lockres_clear_pending(lockres, generation, osb);
2946 	if (ret) {
2947 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
2948 		ocfs2_recover_from_dlm_error(lockres, 1);
2949 		goto bail;
2950 	}
2951 
2952 	ret = 0;
2953 bail:
2954 	mlog_exit(ret);
2955 	return ret;
2956 }
2957 
2958 /* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
2959 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
2960 				        struct ocfs2_lock_res *lockres)
2961 {
2962 	assert_spin_locked(&lockres->l_lock);
2963 
2964 	mlog_entry_void();
2965 	mlog(0, "lock %s\n", lockres->l_name);
2966 
2967 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
2968 		/* If we're already trying to cancel a lock conversion
2969 		 * then just drop the spinlock and allow the caller to
2970 		 * requeue this lock. */
2971 
2972 		mlog(0, "Lockres %s, skip convert\n", lockres->l_name);
2973 		return 0;
2974 	}
2975 
2976 	/* were we in a convert when we got the bast fire? */
2977 	BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
2978 	       lockres->l_action != OCFS2_AST_DOWNCONVERT);
2979 	/* set things up for the unlockast to know to just
2980 	 * clear out the ast_action and unset busy, etc. */
2981 	lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
2982 
2983 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
2984 			"lock %s, invalid flags: 0x%lx\n",
2985 			lockres->l_name, lockres->l_flags);
2986 
2987 	return 1;
2988 }
2989 
2990 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
2991 				struct ocfs2_lock_res *lockres)
2992 {
2993 	int ret;
2994 
2995 	mlog_entry_void();
2996 	mlog(0, "lock %s\n", lockres->l_name);
2997 
2998 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
2999 			       DLM_LKF_CANCEL, lockres);
3000 	if (ret) {
3001 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3002 		ocfs2_recover_from_dlm_error(lockres, 0);
3003 	}
3004 
3005 	mlog(0, "lock %s return from ocfs2_dlm_unlock\n", lockres->l_name);
3006 
3007 	mlog_exit(ret);
3008 	return ret;
3009 }
3010 
3011 static int ocfs2_unblock_lock(struct ocfs2_super *osb,
3012 			      struct ocfs2_lock_res *lockres,
3013 			      struct ocfs2_unblock_ctl *ctl)
3014 {
3015 	unsigned long flags;
3016 	int blocking;
3017 	int new_level;
3018 	int ret = 0;
3019 	int set_lvb = 0;
3020 	unsigned int gen;
3021 
3022 	mlog_entry_void();
3023 
3024 	spin_lock_irqsave(&lockres->l_lock, flags);
3025 
3026 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
3027 
3028 recheck:
3029 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
3030 		/* XXX
3031 		 * This is a *big* race.  The OCFS2_LOCK_PENDING flag
3032 		 * exists entirely for one reason - another thread has set
3033 		 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
3034 		 *
3035 		 * If we do ocfs2_cancel_convert() before the other thread
3036 		 * calls dlm_lock(), our cancel will do nothing.  We will
3037 		 * get no ast, and we will have no way of knowing the
3038 		 * cancel failed.  Meanwhile, the other thread will call
3039 		 * into dlm_lock() and wait...forever.
3040 		 *
3041 		 * Why forever?  Because another node has asked for the
3042 		 * lock first; that's why we're here in unblock_lock().
3043 		 *
3044 		 * The solution is OCFS2_LOCK_PENDING.  When PENDING is
3045 		 * set, we just requeue the unblock.  Only when the other
3046 		 * thread has called dlm_lock() and cleared PENDING will
3047 		 * we then cancel their request.
3048 		 *
3049 		 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
3050 		 * at the same time they set OCFS2_DLM_BUSY.  They must
3051 		 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
3052 		 */
3053 		if (lockres->l_flags & OCFS2_LOCK_PENDING)
3054 			goto leave_requeue;
3055 
3056 		ctl->requeue = 1;
3057 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
3058 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3059 		if (ret) {
3060 			ret = ocfs2_cancel_convert(osb, lockres);
3061 			if (ret < 0)
3062 				mlog_errno(ret);
3063 		}
3064 		goto leave;
3065 	}
3066 
3067 	/* if we're blocking an exclusive and we have *any* holders,
3068 	 * then requeue. */
3069 	if ((lockres->l_blocking == DLM_LOCK_EX)
3070 	    && (lockres->l_ex_holders || lockres->l_ro_holders))
3071 		goto leave_requeue;
3072 
3073 	/* If it's a PR we're blocking, then only
3074 	 * requeue if we've got any EX holders */
3075 	if (lockres->l_blocking == DLM_LOCK_PR &&
3076 	    lockres->l_ex_holders)
3077 		goto leave_requeue;
3078 
3079 	/*
3080 	 * Can we get a lock in this state if the holder counts are
3081 	 * zero? The meta data unblock code used to check this.
3082 	 */
3083 	if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
3084 	    && (lockres->l_flags & OCFS2_LOCK_REFRESHING))
3085 		goto leave_requeue;
3086 
3087 	new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
3088 
3089 	if (lockres->l_ops->check_downconvert
3090 	    && !lockres->l_ops->check_downconvert(lockres, new_level))
3091 		goto leave_requeue;
3092 
3093 	/* If we get here, then we know that there are no more
3094 	 * incompatible holders (and anyone asking for an incompatible
3095 	 * lock is blocked). We can now downconvert the lock */
3096 	if (!lockres->l_ops->downconvert_worker)
3097 		goto downconvert;
3098 
3099 	/* Some lockres types want to do a bit of work before
3100 	 * downconverting a lock. Allow that here. The worker function
3101 	 * may sleep, so we save off a copy of what we're blocking as
3102 	 * it may change while we're not holding the spin lock. */
3103 	blocking = lockres->l_blocking;
3104 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3105 
3106 	ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
3107 
3108 	if (ctl->unblock_action == UNBLOCK_STOP_POST)
3109 		goto leave;
3110 
3111 	spin_lock_irqsave(&lockres->l_lock, flags);
3112 	if (blocking != lockres->l_blocking) {
3113 		/* If this changed underneath us, then we can't drop
3114 		 * it just yet. */
3115 		goto recheck;
3116 	}
3117 
3118 downconvert:
3119 	ctl->requeue = 0;
3120 
3121 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3122 		if (lockres->l_level == DLM_LOCK_EX)
3123 			set_lvb = 1;
3124 
3125 		/*
3126 		 * We only set the lvb if the lock has been fully
3127 		 * refreshed - otherwise we risk setting stale
3128 		 * data. Otherwise, there's no need to actually clear
3129 		 * out the lvb here as it's value is still valid.
3130 		 */
3131 		if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3132 			lockres->l_ops->set_lvb(lockres);
3133 	}
3134 
3135 	gen = ocfs2_prepare_downconvert(lockres, new_level);
3136 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3137 	ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
3138 				     gen);
3139 
3140 leave:
3141 	mlog_exit(ret);
3142 	return ret;
3143 
3144 leave_requeue:
3145 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3146 	ctl->requeue = 1;
3147 
3148 	mlog_exit(0);
3149 	return 0;
3150 }
3151 
3152 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
3153 				     int blocking)
3154 {
3155 	struct inode *inode;
3156 	struct address_space *mapping;
3157 
3158        	inode = ocfs2_lock_res_inode(lockres);
3159 	mapping = inode->i_mapping;
3160 
3161 	if (!S_ISREG(inode->i_mode))
3162 		goto out;
3163 
3164 	/*
3165 	 * We need this before the filemap_fdatawrite() so that it can
3166 	 * transfer the dirty bit from the PTE to the
3167 	 * page. Unfortunately this means that even for EX->PR
3168 	 * downconverts, we'll lose our mappings and have to build
3169 	 * them up again.
3170 	 */
3171 	unmap_mapping_range(mapping, 0, 0, 0);
3172 
3173 	if (filemap_fdatawrite(mapping)) {
3174 		mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
3175 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
3176 	}
3177 	sync_mapping_buffers(mapping);
3178 	if (blocking == DLM_LOCK_EX) {
3179 		truncate_inode_pages(mapping, 0);
3180 	} else {
3181 		/* We only need to wait on the I/O if we're not also
3182 		 * truncating pages because truncate_inode_pages waits
3183 		 * for us above. We don't truncate pages if we're
3184 		 * blocking anything < EXMODE because we want to keep
3185 		 * them around in that case. */
3186 		filemap_fdatawait(mapping);
3187 	}
3188 
3189 out:
3190 	return UNBLOCK_CONTINUE;
3191 }
3192 
3193 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
3194 					int new_level)
3195 {
3196 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3197 	int checkpointed = ocfs2_inode_fully_checkpointed(inode);
3198 
3199 	BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
3200 	BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
3201 
3202 	if (checkpointed)
3203 		return 1;
3204 
3205 	ocfs2_start_checkpoint(OCFS2_SB(inode->i_sb));
3206 	return 0;
3207 }
3208 
3209 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
3210 {
3211 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3212 
3213 	__ocfs2_stuff_meta_lvb(inode);
3214 }
3215 
3216 /*
3217  * Does the final reference drop on our dentry lock. Right now this
3218  * happens in the downconvert thread, but we could choose to simplify the
3219  * dlmglue API and push these off to the ocfs2_wq in the future.
3220  */
3221 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
3222 				     struct ocfs2_lock_res *lockres)
3223 {
3224 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3225 	ocfs2_dentry_lock_put(osb, dl);
3226 }
3227 
3228 /*
3229  * d_delete() matching dentries before the lock downconvert.
3230  *
3231  * At this point, any process waiting to destroy the
3232  * dentry_lock due to last ref count is stopped by the
3233  * OCFS2_LOCK_QUEUED flag.
3234  *
3235  * We have two potential problems
3236  *
3237  * 1) If we do the last reference drop on our dentry_lock (via dput)
3238  *    we'll wind up in ocfs2_release_dentry_lock(), waiting on
3239  *    the downconvert to finish. Instead we take an elevated
3240  *    reference and push the drop until after we've completed our
3241  *    unblock processing.
3242  *
3243  * 2) There might be another process with a final reference,
3244  *    waiting on us to finish processing. If this is the case, we
3245  *    detect it and exit out - there's no more dentries anyway.
3246  */
3247 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
3248 				       int blocking)
3249 {
3250 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3251 	struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
3252 	struct dentry *dentry;
3253 	unsigned long flags;
3254 	int extra_ref = 0;
3255 
3256 	/*
3257 	 * This node is blocking another node from getting a read
3258 	 * lock. This happens when we've renamed within a
3259 	 * directory. We've forced the other nodes to d_delete(), but
3260 	 * we never actually dropped our lock because it's still
3261 	 * valid. The downconvert code will retain a PR for this node,
3262 	 * so there's no further work to do.
3263 	 */
3264 	if (blocking == DLM_LOCK_PR)
3265 		return UNBLOCK_CONTINUE;
3266 
3267 	/*
3268 	 * Mark this inode as potentially orphaned. The code in
3269 	 * ocfs2_delete_inode() will figure out whether it actually
3270 	 * needs to be freed or not.
3271 	 */
3272 	spin_lock(&oi->ip_lock);
3273 	oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
3274 	spin_unlock(&oi->ip_lock);
3275 
3276 	/*
3277 	 * Yuck. We need to make sure however that the check of
3278 	 * OCFS2_LOCK_FREEING and the extra reference are atomic with
3279 	 * respect to a reference decrement or the setting of that
3280 	 * flag.
3281 	 */
3282 	spin_lock_irqsave(&lockres->l_lock, flags);
3283 	spin_lock(&dentry_attach_lock);
3284 	if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
3285 	    && dl->dl_count) {
3286 		dl->dl_count++;
3287 		extra_ref = 1;
3288 	}
3289 	spin_unlock(&dentry_attach_lock);
3290 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3291 
3292 	mlog(0, "extra_ref = %d\n", extra_ref);
3293 
3294 	/*
3295 	 * We have a process waiting on us in ocfs2_dentry_iput(),
3296 	 * which means we can't have any more outstanding
3297 	 * aliases. There's no need to do any more work.
3298 	 */
3299 	if (!extra_ref)
3300 		return UNBLOCK_CONTINUE;
3301 
3302 	spin_lock(&dentry_attach_lock);
3303 	while (1) {
3304 		dentry = ocfs2_find_local_alias(dl->dl_inode,
3305 						dl->dl_parent_blkno, 1);
3306 		if (!dentry)
3307 			break;
3308 		spin_unlock(&dentry_attach_lock);
3309 
3310 		mlog(0, "d_delete(%.*s);\n", dentry->d_name.len,
3311 		     dentry->d_name.name);
3312 
3313 		/*
3314 		 * The following dcache calls may do an
3315 		 * iput(). Normally we don't want that from the
3316 		 * downconverting thread, but in this case it's ok
3317 		 * because the requesting node already has an
3318 		 * exclusive lock on the inode, so it can't be queued
3319 		 * for a downconvert.
3320 		 */
3321 		d_delete(dentry);
3322 		dput(dentry);
3323 
3324 		spin_lock(&dentry_attach_lock);
3325 	}
3326 	spin_unlock(&dentry_attach_lock);
3327 
3328 	/*
3329 	 * If we are the last holder of this dentry lock, there is no
3330 	 * reason to downconvert so skip straight to the unlock.
3331 	 */
3332 	if (dl->dl_count == 1)
3333 		return UNBLOCK_STOP_POST;
3334 
3335 	return UNBLOCK_CONTINUE_POST;
3336 }
3337 
3338 /*
3339  * This is the filesystem locking protocol.  It provides the lock handling
3340  * hooks for the underlying DLM.  It has a maximum version number.
3341  * The version number allows interoperability with systems running at
3342  * the same major number and an equal or smaller minor number.
3343  *
3344  * Whenever the filesystem does new things with locks (adds or removes a
3345  * lock, orders them differently, does different things underneath a lock),
3346  * the version must be changed.  The protocol is negotiated when joining
3347  * the dlm domain.  A node may join the domain if its major version is
3348  * identical to all other nodes and its minor version is greater than
3349  * or equal to all other nodes.  When its minor version is greater than
3350  * the other nodes, it will run at the minor version specified by the
3351  * other nodes.
3352  *
3353  * If a locking change is made that will not be compatible with older
3354  * versions, the major number must be increased and the minor version set
3355  * to zero.  If a change merely adds a behavior that can be disabled when
3356  * speaking to older versions, the minor version must be increased.  If a
3357  * change adds a fully backwards compatible change (eg, LVB changes that
3358  * are just ignored by older versions), the version does not need to be
3359  * updated.
3360  */
3361 static struct ocfs2_locking_protocol lproto = {
3362 	.lp_max_version = {
3363 		.pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
3364 		.pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
3365 	},
3366 	.lp_lock_ast		= ocfs2_locking_ast,
3367 	.lp_blocking_ast	= ocfs2_blocking_ast,
3368 	.lp_unlock_ast		= ocfs2_unlock_ast,
3369 };
3370 
3371 void ocfs2_set_locking_protocol(void)
3372 {
3373 	ocfs2_stack_glue_set_locking_protocol(&lproto);
3374 }
3375 
3376 
3377 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3378 				       struct ocfs2_lock_res *lockres)
3379 {
3380 	int status;
3381 	struct ocfs2_unblock_ctl ctl = {0, 0,};
3382 	unsigned long flags;
3383 
3384 	/* Our reference to the lockres in this function can be
3385 	 * considered valid until we remove the OCFS2_LOCK_QUEUED
3386 	 * flag. */
3387 
3388 	mlog_entry_void();
3389 
3390 	BUG_ON(!lockres);
3391 	BUG_ON(!lockres->l_ops);
3392 
3393 	mlog(0, "lockres %s blocked.\n", lockres->l_name);
3394 
3395 	/* Detect whether a lock has been marked as going away while
3396 	 * the downconvert thread was processing other things. A lock can
3397 	 * still be marked with OCFS2_LOCK_FREEING after this check,
3398 	 * but short circuiting here will still save us some
3399 	 * performance. */
3400 	spin_lock_irqsave(&lockres->l_lock, flags);
3401 	if (lockres->l_flags & OCFS2_LOCK_FREEING)
3402 		goto unqueue;
3403 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3404 
3405 	status = ocfs2_unblock_lock(osb, lockres, &ctl);
3406 	if (status < 0)
3407 		mlog_errno(status);
3408 
3409 	spin_lock_irqsave(&lockres->l_lock, flags);
3410 unqueue:
3411 	if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
3412 		lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
3413 	} else
3414 		ocfs2_schedule_blocked_lock(osb, lockres);
3415 
3416 	mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name,
3417 	     ctl.requeue ? "yes" : "no");
3418 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3419 
3420 	if (ctl.unblock_action != UNBLOCK_CONTINUE
3421 	    && lockres->l_ops->post_unlock)
3422 		lockres->l_ops->post_unlock(osb, lockres);
3423 
3424 	mlog_exit_void();
3425 }
3426 
3427 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
3428 					struct ocfs2_lock_res *lockres)
3429 {
3430 	mlog_entry_void();
3431 
3432 	assert_spin_locked(&lockres->l_lock);
3433 
3434 	if (lockres->l_flags & OCFS2_LOCK_FREEING) {
3435 		/* Do not schedule a lock for downconvert when it's on
3436 		 * the way to destruction - any nodes wanting access
3437 		 * to the resource will get it soon. */
3438 		mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n",
3439 		     lockres->l_name, lockres->l_flags);
3440 		return;
3441 	}
3442 
3443 	lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
3444 
3445 	spin_lock(&osb->dc_task_lock);
3446 	if (list_empty(&lockres->l_blocked_list)) {
3447 		list_add_tail(&lockres->l_blocked_list,
3448 			      &osb->blocked_lock_list);
3449 		osb->blocked_lock_count++;
3450 	}
3451 	spin_unlock(&osb->dc_task_lock);
3452 
3453 	mlog_exit_void();
3454 }
3455 
3456 static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
3457 {
3458 	unsigned long processed;
3459 	struct ocfs2_lock_res *lockres;
3460 
3461 	mlog_entry_void();
3462 
3463 	spin_lock(&osb->dc_task_lock);
3464 	/* grab this early so we know to try again if a state change and
3465 	 * wake happens part-way through our work  */
3466 	osb->dc_work_sequence = osb->dc_wake_sequence;
3467 
3468 	processed = osb->blocked_lock_count;
3469 	while (processed) {
3470 		BUG_ON(list_empty(&osb->blocked_lock_list));
3471 
3472 		lockres = list_entry(osb->blocked_lock_list.next,
3473 				     struct ocfs2_lock_res, l_blocked_list);
3474 		list_del_init(&lockres->l_blocked_list);
3475 		osb->blocked_lock_count--;
3476 		spin_unlock(&osb->dc_task_lock);
3477 
3478 		BUG_ON(!processed);
3479 		processed--;
3480 
3481 		ocfs2_process_blocked_lock(osb, lockres);
3482 
3483 		spin_lock(&osb->dc_task_lock);
3484 	}
3485 	spin_unlock(&osb->dc_task_lock);
3486 
3487 	mlog_exit_void();
3488 }
3489 
3490 static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
3491 {
3492 	int empty = 0;
3493 
3494 	spin_lock(&osb->dc_task_lock);
3495 	if (list_empty(&osb->blocked_lock_list))
3496 		empty = 1;
3497 
3498 	spin_unlock(&osb->dc_task_lock);
3499 	return empty;
3500 }
3501 
3502 static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
3503 {
3504 	int should_wake = 0;
3505 
3506 	spin_lock(&osb->dc_task_lock);
3507 	if (osb->dc_work_sequence != osb->dc_wake_sequence)
3508 		should_wake = 1;
3509 	spin_unlock(&osb->dc_task_lock);
3510 
3511 	return should_wake;
3512 }
3513 
3514 static int ocfs2_downconvert_thread(void *arg)
3515 {
3516 	int status = 0;
3517 	struct ocfs2_super *osb = arg;
3518 
3519 	/* only quit once we've been asked to stop and there is no more
3520 	 * work available */
3521 	while (!(kthread_should_stop() &&
3522 		ocfs2_downconvert_thread_lists_empty(osb))) {
3523 
3524 		wait_event_interruptible(osb->dc_event,
3525 					 ocfs2_downconvert_thread_should_wake(osb) ||
3526 					 kthread_should_stop());
3527 
3528 		mlog(0, "downconvert_thread: awoken\n");
3529 
3530 		ocfs2_downconvert_thread_do_work(osb);
3531 	}
3532 
3533 	osb->dc_task = NULL;
3534 	return status;
3535 }
3536 
3537 void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
3538 {
3539 	spin_lock(&osb->dc_task_lock);
3540 	/* make sure the voting thread gets a swipe at whatever changes
3541 	 * the caller may have made to the voting state */
3542 	osb->dc_wake_sequence++;
3543 	spin_unlock(&osb->dc_task_lock);
3544 	wake_up(&osb->dc_event);
3545 }
3546