xref: /openbmc/linux/fs/ocfs2/dlmglue.c (revision b04b4f78)
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 #include <linux/time.h>
35 #include <linux/quotaops.h>
36 
37 #define MLOG_MASK_PREFIX ML_DLM_GLUE
38 #include <cluster/masklog.h>
39 
40 #include "ocfs2.h"
41 #include "ocfs2_lockingver.h"
42 
43 #include "alloc.h"
44 #include "dcache.h"
45 #include "dlmglue.h"
46 #include "extent_map.h"
47 #include "file.h"
48 #include "heartbeat.h"
49 #include "inode.h"
50 #include "journal.h"
51 #include "stackglue.h"
52 #include "slot_map.h"
53 #include "super.h"
54 #include "uptodate.h"
55 #include "quota.h"
56 
57 #include "buffer_head_io.h"
58 
59 struct ocfs2_mask_waiter {
60 	struct list_head	mw_item;
61 	int			mw_status;
62 	struct completion	mw_complete;
63 	unsigned long		mw_mask;
64 	unsigned long		mw_goal;
65 #ifdef CONFIG_OCFS2_FS_STATS
66 	unsigned long long 	mw_lock_start;
67 #endif
68 };
69 
70 static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres);
71 static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres);
72 static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres);
73 static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres);
74 
75 /*
76  * Return value from ->downconvert_worker functions.
77  *
78  * These control the precise actions of ocfs2_unblock_lock()
79  * and ocfs2_process_blocked_lock()
80  *
81  */
82 enum ocfs2_unblock_action {
83 	UNBLOCK_CONTINUE	= 0, /* Continue downconvert */
84 	UNBLOCK_CONTINUE_POST	= 1, /* Continue downconvert, fire
85 				      * ->post_unlock callback */
86 	UNBLOCK_STOP_POST	= 2, /* Do not downconvert, fire
87 				      * ->post_unlock() callback. */
88 };
89 
90 struct ocfs2_unblock_ctl {
91 	int requeue;
92 	enum ocfs2_unblock_action unblock_action;
93 };
94 
95 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
96 					int new_level);
97 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);
98 
99 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
100 				     int blocking);
101 
102 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
103 				       int blocking);
104 
105 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
106 				     struct ocfs2_lock_res *lockres);
107 
108 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres);
109 
110 #define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
111 
112 /* This aids in debugging situations where a bad LVB might be involved. */
113 static void ocfs2_dump_meta_lvb_info(u64 level,
114 				     const char *function,
115 				     unsigned int line,
116 				     struct ocfs2_lock_res *lockres)
117 {
118 	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
119 
120 	mlog(level, "LVB information for %s (called from %s:%u):\n",
121 	     lockres->l_name, function, line);
122 	mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
123 	     lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
124 	     be32_to_cpu(lvb->lvb_igeneration));
125 	mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
126 	     (unsigned long long)be64_to_cpu(lvb->lvb_isize),
127 	     be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
128 	     be16_to_cpu(lvb->lvb_imode));
129 	mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
130 	     "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
131 	     (long long)be64_to_cpu(lvb->lvb_iatime_packed),
132 	     (long long)be64_to_cpu(lvb->lvb_ictime_packed),
133 	     (long long)be64_to_cpu(lvb->lvb_imtime_packed),
134 	     be32_to_cpu(lvb->lvb_iattr));
135 }
136 
137 
138 /*
139  * OCFS2 Lock Resource Operations
140  *
141  * These fine tune the behavior of the generic dlmglue locking infrastructure.
142  *
143  * The most basic of lock types can point ->l_priv to their respective
144  * struct ocfs2_super and allow the default actions to manage things.
145  *
146  * Right now, each lock type also needs to implement an init function,
147  * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
148  * should be called when the lock is no longer needed (i.e., object
149  * destruction time).
150  */
151 struct ocfs2_lock_res_ops {
152 	/*
153 	 * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
154 	 * this callback if ->l_priv is not an ocfs2_super pointer
155 	 */
156 	struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
157 
158 	/*
159 	 * Optionally called in the downconvert thread after a
160 	 * successful downconvert. The lockres will not be referenced
161 	 * after this callback is called, so it is safe to free
162 	 * memory, etc.
163 	 *
164 	 * The exact semantics of when this is called are controlled
165 	 * by ->downconvert_worker()
166 	 */
167 	void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
168 
169 	/*
170 	 * Allow a lock type to add checks to determine whether it is
171 	 * safe to downconvert a lock. Return 0 to re-queue the
172 	 * downconvert at a later time, nonzero to continue.
173 	 *
174 	 * For most locks, the default checks that there are no
175 	 * incompatible holders are sufficient.
176 	 *
177 	 * Called with the lockres spinlock held.
178 	 */
179 	int (*check_downconvert)(struct ocfs2_lock_res *, int);
180 
181 	/*
182 	 * Allows a lock type to populate the lock value block. This
183 	 * is called on downconvert, and when we drop a lock.
184 	 *
185 	 * Locks that want to use this should set LOCK_TYPE_USES_LVB
186 	 * in the flags field.
187 	 *
188 	 * Called with the lockres spinlock held.
189 	 */
190 	void (*set_lvb)(struct ocfs2_lock_res *);
191 
192 	/*
193 	 * Called from the downconvert thread when it is determined
194 	 * that a lock will be downconverted. This is called without
195 	 * any locks held so the function can do work that might
196 	 * schedule (syncing out data, etc).
197 	 *
198 	 * This should return any one of the ocfs2_unblock_action
199 	 * values, depending on what it wants the thread to do.
200 	 */
201 	int (*downconvert_worker)(struct ocfs2_lock_res *, int);
202 
203 	/*
204 	 * LOCK_TYPE_* flags which describe the specific requirements
205 	 * of a lock type. Descriptions of each individual flag follow.
206 	 */
207 	int flags;
208 };
209 
210 /*
211  * Some locks want to "refresh" potentially stale data when a
212  * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
213  * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
214  * individual lockres l_flags member from the ast function. It is
215  * expected that the locking wrapper will clear the
216  * OCFS2_LOCK_NEEDS_REFRESH flag when done.
217  */
218 #define LOCK_TYPE_REQUIRES_REFRESH 0x1
219 
220 /*
221  * Indicate that a lock type makes use of the lock value block. The
222  * ->set_lvb lock type callback must be defined.
223  */
224 #define LOCK_TYPE_USES_LVB		0x2
225 
226 static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
227 	.get_osb	= ocfs2_get_inode_osb,
228 	.flags		= 0,
229 };
230 
231 static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = {
232 	.get_osb	= ocfs2_get_inode_osb,
233 	.check_downconvert = ocfs2_check_meta_downconvert,
234 	.set_lvb	= ocfs2_set_meta_lvb,
235 	.downconvert_worker = ocfs2_data_convert_worker,
236 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
237 };
238 
239 static struct ocfs2_lock_res_ops ocfs2_super_lops = {
240 	.flags		= LOCK_TYPE_REQUIRES_REFRESH,
241 };
242 
243 static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
244 	.flags		= 0,
245 };
246 
247 static struct ocfs2_lock_res_ops ocfs2_nfs_sync_lops = {
248 	.flags		= 0,
249 };
250 
251 static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
252 	.get_osb	= ocfs2_get_dentry_osb,
253 	.post_unlock	= ocfs2_dentry_post_unlock,
254 	.downconvert_worker = ocfs2_dentry_convert_worker,
255 	.flags		= 0,
256 };
257 
258 static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
259 	.get_osb	= ocfs2_get_inode_osb,
260 	.flags		= 0,
261 };
262 
263 static struct ocfs2_lock_res_ops ocfs2_flock_lops = {
264 	.get_osb	= ocfs2_get_file_osb,
265 	.flags		= 0,
266 };
267 
268 static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = {
269 	.set_lvb	= ocfs2_set_qinfo_lvb,
270 	.get_osb	= ocfs2_get_qinfo_osb,
271 	.flags		= LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB,
272 };
273 
274 static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
275 {
276 	return lockres->l_type == OCFS2_LOCK_TYPE_META ||
277 		lockres->l_type == OCFS2_LOCK_TYPE_RW ||
278 		lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
279 }
280 
281 static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
282 {
283 	BUG_ON(!ocfs2_is_inode_lock(lockres));
284 
285 	return (struct inode *) lockres->l_priv;
286 }
287 
288 static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
289 {
290 	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
291 
292 	return (struct ocfs2_dentry_lock *)lockres->l_priv;
293 }
294 
295 static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres)
296 {
297 	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO);
298 
299 	return (struct ocfs2_mem_dqinfo *)lockres->l_priv;
300 }
301 
302 static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
303 {
304 	if (lockres->l_ops->get_osb)
305 		return lockres->l_ops->get_osb(lockres);
306 
307 	return (struct ocfs2_super *)lockres->l_priv;
308 }
309 
310 static int ocfs2_lock_create(struct ocfs2_super *osb,
311 			     struct ocfs2_lock_res *lockres,
312 			     int level,
313 			     u32 dlm_flags);
314 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
315 						     int wanted);
316 static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
317 				 struct ocfs2_lock_res *lockres,
318 				 int level);
319 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
320 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
321 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
322 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
323 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
324 					struct ocfs2_lock_res *lockres);
325 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
326 						int convert);
327 #define ocfs2_log_dlm_error(_func, _err, _lockres) do {					\
328 	if ((_lockres)->l_type != OCFS2_LOCK_TYPE_DENTRY)				\
329 		mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n",	\
330 		     _err, _func, _lockres->l_name);					\
331 	else										\
332 		mlog(ML_ERROR, "DLM error %d while calling %s on resource %.*s%08x\n",	\
333 		     _err, _func, OCFS2_DENTRY_LOCK_INO_START - 1, (_lockres)->l_name,	\
334 		     (unsigned int)ocfs2_get_dentry_lock_ino(_lockres));		\
335 } while (0)
336 static int ocfs2_downconvert_thread(void *arg);
337 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
338 					struct ocfs2_lock_res *lockres);
339 static int ocfs2_inode_lock_update(struct inode *inode,
340 				  struct buffer_head **bh);
341 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
342 static inline int ocfs2_highest_compat_lock_level(int level);
343 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
344 					      int new_level);
345 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
346 				  struct ocfs2_lock_res *lockres,
347 				  int new_level,
348 				  int lvb,
349 				  unsigned int generation);
350 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
351 				        struct ocfs2_lock_res *lockres);
352 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
353 				struct ocfs2_lock_res *lockres);
354 
355 
356 static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
357 				  u64 blkno,
358 				  u32 generation,
359 				  char *name)
360 {
361 	int len;
362 
363 	mlog_entry_void();
364 
365 	BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
366 
367 	len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
368 		       ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
369 		       (long long)blkno, generation);
370 
371 	BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
372 
373 	mlog(0, "built lock resource with name: %s\n", name);
374 
375 	mlog_exit_void();
376 }
377 
378 static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
379 
380 static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
381 				       struct ocfs2_dlm_debug *dlm_debug)
382 {
383 	mlog(0, "Add tracking for lockres %s\n", res->l_name);
384 
385 	spin_lock(&ocfs2_dlm_tracking_lock);
386 	list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
387 	spin_unlock(&ocfs2_dlm_tracking_lock);
388 }
389 
390 static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
391 {
392 	spin_lock(&ocfs2_dlm_tracking_lock);
393 	if (!list_empty(&res->l_debug_list))
394 		list_del_init(&res->l_debug_list);
395 	spin_unlock(&ocfs2_dlm_tracking_lock);
396 }
397 
398 #ifdef CONFIG_OCFS2_FS_STATS
399 static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
400 {
401 	res->l_lock_num_prmode = 0;
402 	res->l_lock_num_prmode_failed = 0;
403 	res->l_lock_total_prmode = 0;
404 	res->l_lock_max_prmode = 0;
405 	res->l_lock_num_exmode = 0;
406 	res->l_lock_num_exmode_failed = 0;
407 	res->l_lock_total_exmode = 0;
408 	res->l_lock_max_exmode = 0;
409 	res->l_lock_refresh = 0;
410 }
411 
412 static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level,
413 				    struct ocfs2_mask_waiter *mw, int ret)
414 {
415 	unsigned long long *num, *sum;
416 	unsigned int *max, *failed;
417 	struct timespec ts = current_kernel_time();
418 	unsigned long long time = timespec_to_ns(&ts) - mw->mw_lock_start;
419 
420 	if (level == LKM_PRMODE) {
421 		num = &res->l_lock_num_prmode;
422 		sum = &res->l_lock_total_prmode;
423 		max = &res->l_lock_max_prmode;
424 		failed = &res->l_lock_num_prmode_failed;
425 	} else if (level == LKM_EXMODE) {
426 		num = &res->l_lock_num_exmode;
427 		sum = &res->l_lock_total_exmode;
428 		max = &res->l_lock_max_exmode;
429 		failed = &res->l_lock_num_exmode_failed;
430 	} else
431 		return;
432 
433 	(*num)++;
434 	(*sum) += time;
435 	if (time > *max)
436 		*max = time;
437 	if (ret)
438 		(*failed)++;
439 }
440 
441 static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
442 {
443 	lockres->l_lock_refresh++;
444 }
445 
446 static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
447 {
448 	struct timespec ts = current_kernel_time();
449 	mw->mw_lock_start = timespec_to_ns(&ts);
450 }
451 #else
452 static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
453 {
454 }
455 static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res,
456 			   int level, struct ocfs2_mask_waiter *mw, int ret)
457 {
458 }
459 static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
460 {
461 }
462 static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
463 {
464 }
465 #endif
466 
467 static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
468 				       struct ocfs2_lock_res *res,
469 				       enum ocfs2_lock_type type,
470 				       struct ocfs2_lock_res_ops *ops,
471 				       void *priv)
472 {
473 	res->l_type          = type;
474 	res->l_ops           = ops;
475 	res->l_priv          = priv;
476 
477 	res->l_level         = DLM_LOCK_IV;
478 	res->l_requested     = DLM_LOCK_IV;
479 	res->l_blocking      = DLM_LOCK_IV;
480 	res->l_action        = OCFS2_AST_INVALID;
481 	res->l_unlock_action = OCFS2_UNLOCK_INVALID;
482 
483 	res->l_flags         = OCFS2_LOCK_INITIALIZED;
484 
485 	ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
486 
487 	ocfs2_init_lock_stats(res);
488 }
489 
490 void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
491 {
492 	/* This also clears out the lock status block */
493 	memset(res, 0, sizeof(struct ocfs2_lock_res));
494 	spin_lock_init(&res->l_lock);
495 	init_waitqueue_head(&res->l_event);
496 	INIT_LIST_HEAD(&res->l_blocked_list);
497 	INIT_LIST_HEAD(&res->l_mask_waiters);
498 }
499 
500 void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
501 			       enum ocfs2_lock_type type,
502 			       unsigned int generation,
503 			       struct inode *inode)
504 {
505 	struct ocfs2_lock_res_ops *ops;
506 
507 	switch(type) {
508 		case OCFS2_LOCK_TYPE_RW:
509 			ops = &ocfs2_inode_rw_lops;
510 			break;
511 		case OCFS2_LOCK_TYPE_META:
512 			ops = &ocfs2_inode_inode_lops;
513 			break;
514 		case OCFS2_LOCK_TYPE_OPEN:
515 			ops = &ocfs2_inode_open_lops;
516 			break;
517 		default:
518 			mlog_bug_on_msg(1, "type: %d\n", type);
519 			ops = NULL; /* thanks, gcc */
520 			break;
521 	};
522 
523 	ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
524 			      generation, res->l_name);
525 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
526 }
527 
528 static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
529 {
530 	struct inode *inode = ocfs2_lock_res_inode(lockres);
531 
532 	return OCFS2_SB(inode->i_sb);
533 }
534 
535 static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres)
536 {
537 	struct ocfs2_mem_dqinfo *info = lockres->l_priv;
538 
539 	return OCFS2_SB(info->dqi_gi.dqi_sb);
540 }
541 
542 static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres)
543 {
544 	struct ocfs2_file_private *fp = lockres->l_priv;
545 
546 	return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb);
547 }
548 
549 static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
550 {
551 	__be64 inode_blkno_be;
552 
553 	memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
554 	       sizeof(__be64));
555 
556 	return be64_to_cpu(inode_blkno_be);
557 }
558 
559 static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
560 {
561 	struct ocfs2_dentry_lock *dl = lockres->l_priv;
562 
563 	return OCFS2_SB(dl->dl_inode->i_sb);
564 }
565 
566 void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
567 				u64 parent, struct inode *inode)
568 {
569 	int len;
570 	u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
571 	__be64 inode_blkno_be = cpu_to_be64(inode_blkno);
572 	struct ocfs2_lock_res *lockres = &dl->dl_lockres;
573 
574 	ocfs2_lock_res_init_once(lockres);
575 
576 	/*
577 	 * Unfortunately, the standard lock naming scheme won't work
578 	 * here because we have two 16 byte values to use. Instead,
579 	 * we'll stuff the inode number as a binary value. We still
580 	 * want error prints to show something without garbling the
581 	 * display, so drop a null byte in there before the inode
582 	 * number. A future version of OCFS2 will likely use all
583 	 * binary lock names. The stringified names have been a
584 	 * tremendous aid in debugging, but now that the debugfs
585 	 * interface exists, we can mangle things there if need be.
586 	 *
587 	 * NOTE: We also drop the standard "pad" value (the total lock
588 	 * name size stays the same though - the last part is all
589 	 * zeros due to the memset in ocfs2_lock_res_init_once()
590 	 */
591 	len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
592 		       "%c%016llx",
593 		       ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
594 		       (long long)parent);
595 
596 	BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
597 
598 	memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
599 	       sizeof(__be64));
600 
601 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
602 				   OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
603 				   dl);
604 }
605 
606 static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
607 				      struct ocfs2_super *osb)
608 {
609 	/* Superblock lockres doesn't come from a slab so we call init
610 	 * once on it manually.  */
611 	ocfs2_lock_res_init_once(res);
612 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
613 			      0, res->l_name);
614 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
615 				   &ocfs2_super_lops, osb);
616 }
617 
618 static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
619 				       struct ocfs2_super *osb)
620 {
621 	/* Rename lockres doesn't come from a slab so we call init
622 	 * once on it manually.  */
623 	ocfs2_lock_res_init_once(res);
624 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
625 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
626 				   &ocfs2_rename_lops, osb);
627 }
628 
629 static void ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res *res,
630 					 struct ocfs2_super *osb)
631 {
632 	/* nfs_sync lockres doesn't come from a slab so we call init
633 	 * once on it manually.  */
634 	ocfs2_lock_res_init_once(res);
635 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_NFS_SYNC, 0, 0, res->l_name);
636 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_NFS_SYNC,
637 				   &ocfs2_nfs_sync_lops, osb);
638 }
639 
640 void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
641 			      struct ocfs2_file_private *fp)
642 {
643 	struct inode *inode = fp->fp_file->f_mapping->host;
644 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
645 
646 	ocfs2_lock_res_init_once(lockres);
647 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
648 			      inode->i_generation, lockres->l_name);
649 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
650 				   OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
651 				   fp);
652 	lockres->l_flags |= OCFS2_LOCK_NOCACHE;
653 }
654 
655 void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
656 			       struct ocfs2_mem_dqinfo *info)
657 {
658 	ocfs2_lock_res_init_once(lockres);
659 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type,
660 			      0, lockres->l_name);
661 	ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres,
662 				   OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops,
663 				   info);
664 }
665 
666 void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
667 {
668 	mlog_entry_void();
669 
670 	if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
671 		return;
672 
673 	ocfs2_remove_lockres_tracking(res);
674 
675 	mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
676 			"Lockres %s is on the blocked list\n",
677 			res->l_name);
678 	mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
679 			"Lockres %s has mask waiters pending\n",
680 			res->l_name);
681 	mlog_bug_on_msg(spin_is_locked(&res->l_lock),
682 			"Lockres %s is locked\n",
683 			res->l_name);
684 	mlog_bug_on_msg(res->l_ro_holders,
685 			"Lockres %s has %u ro holders\n",
686 			res->l_name, res->l_ro_holders);
687 	mlog_bug_on_msg(res->l_ex_holders,
688 			"Lockres %s has %u ex holders\n",
689 			res->l_name, res->l_ex_holders);
690 
691 	/* Need to clear out the lock status block for the dlm */
692 	memset(&res->l_lksb, 0, sizeof(res->l_lksb));
693 
694 	res->l_flags = 0UL;
695 	mlog_exit_void();
696 }
697 
698 static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
699 				     int level)
700 {
701 	mlog_entry_void();
702 
703 	BUG_ON(!lockres);
704 
705 	switch(level) {
706 	case DLM_LOCK_EX:
707 		lockres->l_ex_holders++;
708 		break;
709 	case DLM_LOCK_PR:
710 		lockres->l_ro_holders++;
711 		break;
712 	default:
713 		BUG();
714 	}
715 
716 	mlog_exit_void();
717 }
718 
719 static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
720 				     int level)
721 {
722 	mlog_entry_void();
723 
724 	BUG_ON(!lockres);
725 
726 	switch(level) {
727 	case DLM_LOCK_EX:
728 		BUG_ON(!lockres->l_ex_holders);
729 		lockres->l_ex_holders--;
730 		break;
731 	case DLM_LOCK_PR:
732 		BUG_ON(!lockres->l_ro_holders);
733 		lockres->l_ro_holders--;
734 		break;
735 	default:
736 		BUG();
737 	}
738 	mlog_exit_void();
739 }
740 
741 /* WARNING: This function lives in a world where the only three lock
742  * levels are EX, PR, and NL. It *will* have to be adjusted when more
743  * lock types are added. */
744 static inline int ocfs2_highest_compat_lock_level(int level)
745 {
746 	int new_level = DLM_LOCK_EX;
747 
748 	if (level == DLM_LOCK_EX)
749 		new_level = DLM_LOCK_NL;
750 	else if (level == DLM_LOCK_PR)
751 		new_level = DLM_LOCK_PR;
752 	return new_level;
753 }
754 
755 static void lockres_set_flags(struct ocfs2_lock_res *lockres,
756 			      unsigned long newflags)
757 {
758 	struct ocfs2_mask_waiter *mw, *tmp;
759 
760  	assert_spin_locked(&lockres->l_lock);
761 
762 	lockres->l_flags = newflags;
763 
764 	list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
765 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
766 			continue;
767 
768 		list_del_init(&mw->mw_item);
769 		mw->mw_status = 0;
770 		complete(&mw->mw_complete);
771 	}
772 }
773 static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
774 {
775 	lockres_set_flags(lockres, lockres->l_flags | or);
776 }
777 static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
778 				unsigned long clear)
779 {
780 	lockres_set_flags(lockres, lockres->l_flags & ~clear);
781 }
782 
783 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
784 {
785 	mlog_entry_void();
786 
787 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
788 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
789 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
790 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
791 
792 	lockres->l_level = lockres->l_requested;
793 	if (lockres->l_level <=
794 	    ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
795 		lockres->l_blocking = DLM_LOCK_NL;
796 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
797 	}
798 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
799 
800 	mlog_exit_void();
801 }
802 
803 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
804 {
805 	mlog_entry_void();
806 
807 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
808 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
809 
810 	/* Convert from RO to EX doesn't really need anything as our
811 	 * information is already up to data. Convert from NL to
812 	 * *anything* however should mark ourselves as needing an
813 	 * update */
814 	if (lockres->l_level == DLM_LOCK_NL &&
815 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
816 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
817 
818 	lockres->l_level = lockres->l_requested;
819 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
820 
821 	mlog_exit_void();
822 }
823 
824 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
825 {
826 	mlog_entry_void();
827 
828 	BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
829 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
830 
831 	if (lockres->l_requested > DLM_LOCK_NL &&
832 	    !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
833 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
834 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
835 
836 	lockres->l_level = lockres->l_requested;
837 	lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
838 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
839 
840 	mlog_exit_void();
841 }
842 
843 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
844 				     int level)
845 {
846 	int needs_downconvert = 0;
847 	mlog_entry_void();
848 
849 	assert_spin_locked(&lockres->l_lock);
850 
851 	lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
852 
853 	if (level > lockres->l_blocking) {
854 		/* only schedule a downconvert if we haven't already scheduled
855 		 * one that goes low enough to satisfy the level we're
856 		 * blocking.  this also catches the case where we get
857 		 * duplicate BASTs */
858 		if (ocfs2_highest_compat_lock_level(level) <
859 		    ocfs2_highest_compat_lock_level(lockres->l_blocking))
860 			needs_downconvert = 1;
861 
862 		lockres->l_blocking = level;
863 	}
864 
865 	mlog_exit(needs_downconvert);
866 	return needs_downconvert;
867 }
868 
869 /*
870  * OCFS2_LOCK_PENDING and l_pending_gen.
871  *
872  * Why does OCFS2_LOCK_PENDING exist?  To close a race between setting
873  * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock().  See ocfs2_unblock_lock()
874  * for more details on the race.
875  *
876  * OCFS2_LOCK_PENDING closes the race quite nicely.  However, it introduces
877  * a race on itself.  In o2dlm, we can get the ast before ocfs2_dlm_lock()
878  * returns.  The ast clears OCFS2_LOCK_BUSY, and must therefore clear
879  * OCFS2_LOCK_PENDING at the same time.  When ocfs2_dlm_lock() returns,
880  * the caller is going to try to clear PENDING again.  If nothing else is
881  * happening, __lockres_clear_pending() sees PENDING is unset and does
882  * nothing.
883  *
884  * But what if another path (eg downconvert thread) has just started a
885  * new locking action?  The other path has re-set PENDING.  Our path
886  * cannot clear PENDING, because that will re-open the original race
887  * window.
888  *
889  * [Example]
890  *
891  * ocfs2_meta_lock()
892  *  ocfs2_cluster_lock()
893  *   set BUSY
894  *   set PENDING
895  *   drop l_lock
896  *   ocfs2_dlm_lock()
897  *    ocfs2_locking_ast()		ocfs2_downconvert_thread()
898  *     clear PENDING			 ocfs2_unblock_lock()
899  *					  take_l_lock
900  *					  !BUSY
901  *					  ocfs2_prepare_downconvert()
902  *					   set BUSY
903  *					   set PENDING
904  *					  drop l_lock
905  *   take l_lock
906  *   clear PENDING
907  *   drop l_lock
908  *			<window>
909  *					  ocfs2_dlm_lock()
910  *
911  * So as you can see, we now have a window where l_lock is not held,
912  * PENDING is not set, and ocfs2_dlm_lock() has not been called.
913  *
914  * The core problem is that ocfs2_cluster_lock() has cleared the PENDING
915  * set by ocfs2_prepare_downconvert().  That wasn't nice.
916  *
917  * To solve this we introduce l_pending_gen.  A call to
918  * lockres_clear_pending() will only do so when it is passed a generation
919  * number that matches the lockres.  lockres_set_pending() will return the
920  * current generation number.  When ocfs2_cluster_lock() goes to clear
921  * PENDING, it passes the generation it got from set_pending().  In our
922  * example above, the generation numbers will *not* match.  Thus,
923  * ocfs2_cluster_lock() will not clear the PENDING set by
924  * ocfs2_prepare_downconvert().
925  */
926 
927 /* Unlocked version for ocfs2_locking_ast() */
928 static void __lockres_clear_pending(struct ocfs2_lock_res *lockres,
929 				    unsigned int generation,
930 				    struct ocfs2_super *osb)
931 {
932 	assert_spin_locked(&lockres->l_lock);
933 
934 	/*
935 	 * The ast and locking functions can race us here.  The winner
936 	 * will clear pending, the loser will not.
937 	 */
938 	if (!(lockres->l_flags & OCFS2_LOCK_PENDING) ||
939 	    (lockres->l_pending_gen != generation))
940 		return;
941 
942 	lockres_clear_flags(lockres, OCFS2_LOCK_PENDING);
943 	lockres->l_pending_gen++;
944 
945 	/*
946 	 * The downconvert thread may have skipped us because we
947 	 * were PENDING.  Wake it up.
948 	 */
949 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
950 		ocfs2_wake_downconvert_thread(osb);
951 }
952 
953 /* Locked version for callers of ocfs2_dlm_lock() */
954 static void lockres_clear_pending(struct ocfs2_lock_res *lockres,
955 				  unsigned int generation,
956 				  struct ocfs2_super *osb)
957 {
958 	unsigned long flags;
959 
960 	spin_lock_irqsave(&lockres->l_lock, flags);
961 	__lockres_clear_pending(lockres, generation, osb);
962 	spin_unlock_irqrestore(&lockres->l_lock, flags);
963 }
964 
965 static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres)
966 {
967 	assert_spin_locked(&lockres->l_lock);
968 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
969 
970 	lockres_or_flags(lockres, OCFS2_LOCK_PENDING);
971 
972 	return lockres->l_pending_gen;
973 }
974 
975 
976 static void ocfs2_blocking_ast(void *opaque, int level)
977 {
978 	struct ocfs2_lock_res *lockres = opaque;
979 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
980 	int needs_downconvert;
981 	unsigned long flags;
982 
983 	BUG_ON(level <= DLM_LOCK_NL);
984 
985 	mlog(0, "BAST fired for lockres %s, blocking %d, level %d type %s\n",
986 	     lockres->l_name, level, lockres->l_level,
987 	     ocfs2_lock_type_string(lockres->l_type));
988 
989 	/*
990 	 * We can skip the bast for locks which don't enable caching -
991 	 * they'll be dropped at the earliest possible time anyway.
992 	 */
993 	if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
994 		return;
995 
996 	spin_lock_irqsave(&lockres->l_lock, flags);
997 	needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
998 	if (needs_downconvert)
999 		ocfs2_schedule_blocked_lock(osb, lockres);
1000 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1001 
1002 	wake_up(&lockres->l_event);
1003 
1004 	ocfs2_wake_downconvert_thread(osb);
1005 }
1006 
1007 static void ocfs2_locking_ast(void *opaque)
1008 {
1009 	struct ocfs2_lock_res *lockres = opaque;
1010 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1011 	unsigned long flags;
1012 	int status;
1013 
1014 	spin_lock_irqsave(&lockres->l_lock, flags);
1015 
1016 	status = ocfs2_dlm_lock_status(&lockres->l_lksb);
1017 
1018 	if (status == -EAGAIN) {
1019 		lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1020 		goto out;
1021 	}
1022 
1023 	if (status) {
1024 		mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n",
1025 		     lockres->l_name, status);
1026 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1027 		return;
1028 	}
1029 
1030 	switch(lockres->l_action) {
1031 	case OCFS2_AST_ATTACH:
1032 		ocfs2_generic_handle_attach_action(lockres);
1033 		lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
1034 		break;
1035 	case OCFS2_AST_CONVERT:
1036 		ocfs2_generic_handle_convert_action(lockres);
1037 		break;
1038 	case OCFS2_AST_DOWNCONVERT:
1039 		ocfs2_generic_handle_downconvert_action(lockres);
1040 		break;
1041 	default:
1042 		mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u "
1043 		     "lockres flags = 0x%lx, unlock action: %u\n",
1044 		     lockres->l_name, lockres->l_action, lockres->l_flags,
1045 		     lockres->l_unlock_action);
1046 		BUG();
1047 	}
1048 out:
1049 	/* set it to something invalid so if we get called again we
1050 	 * can catch it. */
1051 	lockres->l_action = OCFS2_AST_INVALID;
1052 
1053 	/* Did we try to cancel this lock?  Clear that state */
1054 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT)
1055 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1056 
1057 	/*
1058 	 * We may have beaten the locking functions here.  We certainly
1059 	 * know that dlm_lock() has been called :-)
1060 	 * Because we can't have two lock calls in flight at once, we
1061 	 * can use lockres->l_pending_gen.
1062 	 */
1063 	__lockres_clear_pending(lockres, lockres->l_pending_gen,  osb);
1064 
1065 	wake_up(&lockres->l_event);
1066 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1067 }
1068 
1069 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
1070 						int convert)
1071 {
1072 	unsigned long flags;
1073 
1074 	mlog_entry_void();
1075 	spin_lock_irqsave(&lockres->l_lock, flags);
1076 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1077 	if (convert)
1078 		lockres->l_action = OCFS2_AST_INVALID;
1079 	else
1080 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1081 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1082 
1083 	wake_up(&lockres->l_event);
1084 	mlog_exit_void();
1085 }
1086 
1087 /* Note: If we detect another process working on the lock (i.e.,
1088  * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
1089  * to do the right thing in that case.
1090  */
1091 static int ocfs2_lock_create(struct ocfs2_super *osb,
1092 			     struct ocfs2_lock_res *lockres,
1093 			     int level,
1094 			     u32 dlm_flags)
1095 {
1096 	int ret = 0;
1097 	unsigned long flags;
1098 	unsigned int gen;
1099 
1100 	mlog_entry_void();
1101 
1102 	mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
1103 	     dlm_flags);
1104 
1105 	spin_lock_irqsave(&lockres->l_lock, flags);
1106 	if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
1107 	    (lockres->l_flags & OCFS2_LOCK_BUSY)) {
1108 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1109 		goto bail;
1110 	}
1111 
1112 	lockres->l_action = OCFS2_AST_ATTACH;
1113 	lockres->l_requested = level;
1114 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1115 	gen = lockres_set_pending(lockres);
1116 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1117 
1118 	ret = ocfs2_dlm_lock(osb->cconn,
1119 			     level,
1120 			     &lockres->l_lksb,
1121 			     dlm_flags,
1122 			     lockres->l_name,
1123 			     OCFS2_LOCK_ID_MAX_LEN - 1,
1124 			     lockres);
1125 	lockres_clear_pending(lockres, gen, osb);
1126 	if (ret) {
1127 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
1128 		ocfs2_recover_from_dlm_error(lockres, 1);
1129 	}
1130 
1131 	mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
1132 
1133 bail:
1134 	mlog_exit(ret);
1135 	return ret;
1136 }
1137 
1138 static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
1139 					int flag)
1140 {
1141 	unsigned long flags;
1142 	int ret;
1143 
1144 	spin_lock_irqsave(&lockres->l_lock, flags);
1145 	ret = lockres->l_flags & flag;
1146 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1147 
1148 	return ret;
1149 }
1150 
1151 static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
1152 
1153 {
1154 	wait_event(lockres->l_event,
1155 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
1156 }
1157 
1158 static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
1159 
1160 {
1161 	wait_event(lockres->l_event,
1162 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
1163 }
1164 
1165 /* predict what lock level we'll be dropping down to on behalf
1166  * of another node, and return true if the currently wanted
1167  * level will be compatible with it. */
1168 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
1169 						     int wanted)
1170 {
1171 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
1172 
1173 	return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
1174 }
1175 
1176 static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
1177 {
1178 	INIT_LIST_HEAD(&mw->mw_item);
1179 	init_completion(&mw->mw_complete);
1180 	ocfs2_init_start_time(mw);
1181 }
1182 
1183 static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
1184 {
1185 	wait_for_completion(&mw->mw_complete);
1186 	/* Re-arm the completion in case we want to wait on it again */
1187 	INIT_COMPLETION(mw->mw_complete);
1188 	return mw->mw_status;
1189 }
1190 
1191 static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
1192 				    struct ocfs2_mask_waiter *mw,
1193 				    unsigned long mask,
1194 				    unsigned long goal)
1195 {
1196 	BUG_ON(!list_empty(&mw->mw_item));
1197 
1198 	assert_spin_locked(&lockres->l_lock);
1199 
1200 	list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
1201 	mw->mw_mask = mask;
1202 	mw->mw_goal = goal;
1203 }
1204 
1205 /* returns 0 if the mw that was removed was already satisfied, -EBUSY
1206  * if the mask still hadn't reached its goal */
1207 static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1208 				      struct ocfs2_mask_waiter *mw)
1209 {
1210 	unsigned long flags;
1211 	int ret = 0;
1212 
1213 	spin_lock_irqsave(&lockres->l_lock, flags);
1214 	if (!list_empty(&mw->mw_item)) {
1215 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
1216 			ret = -EBUSY;
1217 
1218 		list_del_init(&mw->mw_item);
1219 		init_completion(&mw->mw_complete);
1220 	}
1221 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1222 
1223 	return ret;
1224 
1225 }
1226 
1227 static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
1228 					     struct ocfs2_lock_res *lockres)
1229 {
1230 	int ret;
1231 
1232 	ret = wait_for_completion_interruptible(&mw->mw_complete);
1233 	if (ret)
1234 		lockres_remove_mask_waiter(lockres, mw);
1235 	else
1236 		ret = mw->mw_status;
1237 	/* Re-arm the completion in case we want to wait on it again */
1238 	INIT_COMPLETION(mw->mw_complete);
1239 	return ret;
1240 }
1241 
1242 static int ocfs2_cluster_lock(struct ocfs2_super *osb,
1243 			      struct ocfs2_lock_res *lockres,
1244 			      int level,
1245 			      u32 lkm_flags,
1246 			      int arg_flags)
1247 {
1248 	struct ocfs2_mask_waiter mw;
1249 	int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
1250 	int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1251 	unsigned long flags;
1252 	unsigned int gen;
1253 	int noqueue_attempted = 0;
1254 
1255 	mlog_entry_void();
1256 
1257 	ocfs2_init_mask_waiter(&mw);
1258 
1259 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
1260 		lkm_flags |= DLM_LKF_VALBLK;
1261 
1262 again:
1263 	wait = 0;
1264 
1265 	if (catch_signals && signal_pending(current)) {
1266 		ret = -ERESTARTSYS;
1267 		goto out;
1268 	}
1269 
1270 	spin_lock_irqsave(&lockres->l_lock, flags);
1271 
1272 	mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
1273 			"Cluster lock called on freeing lockres %s! flags "
1274 			"0x%lx\n", lockres->l_name, lockres->l_flags);
1275 
1276 	/* We only compare against the currently granted level
1277 	 * here. If the lock is blocked waiting on a downconvert,
1278 	 * we'll get caught below. */
1279 	if (lockres->l_flags & OCFS2_LOCK_BUSY &&
1280 	    level > lockres->l_level) {
1281 		/* is someone sitting in dlm_lock? If so, wait on
1282 		 * them. */
1283 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1284 		wait = 1;
1285 		goto unlock;
1286 	}
1287 
1288 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
1289 	    !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
1290 		/* is the lock is currently blocked on behalf of
1291 		 * another node */
1292 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
1293 		wait = 1;
1294 		goto unlock;
1295 	}
1296 
1297 	if (level > lockres->l_level) {
1298 		if (noqueue_attempted > 0) {
1299 			ret = -EAGAIN;
1300 			goto unlock;
1301 		}
1302 		if (lkm_flags & DLM_LKF_NOQUEUE)
1303 			noqueue_attempted = 1;
1304 
1305 		if (lockres->l_action != OCFS2_AST_INVALID)
1306 			mlog(ML_ERROR, "lockres %s has action %u pending\n",
1307 			     lockres->l_name, lockres->l_action);
1308 
1309 		if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1310 			lockres->l_action = OCFS2_AST_ATTACH;
1311 			lkm_flags &= ~DLM_LKF_CONVERT;
1312 		} else {
1313 			lockres->l_action = OCFS2_AST_CONVERT;
1314 			lkm_flags |= DLM_LKF_CONVERT;
1315 		}
1316 
1317 		lockres->l_requested = level;
1318 		lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1319 		gen = lockres_set_pending(lockres);
1320 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1321 
1322 		BUG_ON(level == DLM_LOCK_IV);
1323 		BUG_ON(level == DLM_LOCK_NL);
1324 
1325 		mlog(0, "lock %s, convert from %d to level = %d\n",
1326 		     lockres->l_name, lockres->l_level, level);
1327 
1328 		/* call dlm_lock to upgrade lock now */
1329 		ret = ocfs2_dlm_lock(osb->cconn,
1330 				     level,
1331 				     &lockres->l_lksb,
1332 				     lkm_flags,
1333 				     lockres->l_name,
1334 				     OCFS2_LOCK_ID_MAX_LEN - 1,
1335 				     lockres);
1336 		lockres_clear_pending(lockres, gen, osb);
1337 		if (ret) {
1338 			if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
1339 			    (ret != -EAGAIN)) {
1340 				ocfs2_log_dlm_error("ocfs2_dlm_lock",
1341 						    ret, lockres);
1342 			}
1343 			ocfs2_recover_from_dlm_error(lockres, 1);
1344 			goto out;
1345 		}
1346 
1347 		mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n",
1348 		     lockres->l_name);
1349 
1350 		/* At this point we've gone inside the dlm and need to
1351 		 * complete our work regardless. */
1352 		catch_signals = 0;
1353 
1354 		/* wait for busy to clear and carry on */
1355 		goto again;
1356 	}
1357 
1358 	/* Ok, if we get here then we're good to go. */
1359 	ocfs2_inc_holders(lockres, level);
1360 
1361 	ret = 0;
1362 unlock:
1363 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1364 out:
1365 	/*
1366 	 * This is helping work around a lock inversion between the page lock
1367 	 * and dlm locks.  One path holds the page lock while calling aops
1368 	 * which block acquiring dlm locks.  The voting thread holds dlm
1369 	 * locks while acquiring page locks while down converting data locks.
1370 	 * This block is helping an aop path notice the inversion and back
1371 	 * off to unlock its page lock before trying the dlm lock again.
1372 	 */
1373 	if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1374 	    mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1375 		wait = 0;
1376 		if (lockres_remove_mask_waiter(lockres, &mw))
1377 			ret = -EAGAIN;
1378 		else
1379 			goto again;
1380 	}
1381 	if (wait) {
1382 		ret = ocfs2_wait_for_mask(&mw);
1383 		if (ret == 0)
1384 			goto again;
1385 		mlog_errno(ret);
1386 	}
1387 	ocfs2_update_lock_stats(lockres, level, &mw, ret);
1388 
1389 	mlog_exit(ret);
1390 	return ret;
1391 }
1392 
1393 static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
1394 				 struct ocfs2_lock_res *lockres,
1395 				 int level)
1396 {
1397 	unsigned long flags;
1398 
1399 	mlog_entry_void();
1400 	spin_lock_irqsave(&lockres->l_lock, flags);
1401 	ocfs2_dec_holders(lockres, level);
1402 	ocfs2_downconvert_on_unlock(osb, lockres);
1403 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1404 	mlog_exit_void();
1405 }
1406 
1407 static int ocfs2_create_new_lock(struct ocfs2_super *osb,
1408 				 struct ocfs2_lock_res *lockres,
1409 				 int ex,
1410 				 int local)
1411 {
1412 	int level =  ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1413 	unsigned long flags;
1414 	u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
1415 
1416 	spin_lock_irqsave(&lockres->l_lock, flags);
1417 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1418 	lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1419 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1420 
1421 	return ocfs2_lock_create(osb, lockres, level, lkm_flags);
1422 }
1423 
1424 /* Grants us an EX lock on the data and metadata resources, skipping
1425  * the normal cluster directory lookup. Use this ONLY on newly created
1426  * inodes which other nodes can't possibly see, and which haven't been
1427  * hashed in the inode hash yet. This can give us a good performance
1428  * increase as it'll skip the network broadcast normally associated
1429  * with creating a new lock resource. */
1430 int ocfs2_create_new_inode_locks(struct inode *inode)
1431 {
1432 	int ret;
1433 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1434 
1435 	BUG_ON(!inode);
1436 	BUG_ON(!ocfs2_inode_is_new(inode));
1437 
1438 	mlog_entry_void();
1439 
1440 	mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
1441 
1442 	/* NOTE: That we don't increment any of the holder counts, nor
1443 	 * do we add anything to a journal handle. Since this is
1444 	 * supposed to be a new inode which the cluster doesn't know
1445 	 * about yet, there is no need to.  As far as the LVB handling
1446 	 * is concerned, this is basically like acquiring an EX lock
1447 	 * on a resource which has an invalid one -- we'll set it
1448 	 * valid when we release the EX. */
1449 
1450 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
1451 	if (ret) {
1452 		mlog_errno(ret);
1453 		goto bail;
1454 	}
1455 
1456 	/*
1457 	 * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
1458 	 * don't use a generation in their lock names.
1459 	 */
1460 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
1461 	if (ret) {
1462 		mlog_errno(ret);
1463 		goto bail;
1464 	}
1465 
1466 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
1467 	if (ret) {
1468 		mlog_errno(ret);
1469 		goto bail;
1470 	}
1471 
1472 bail:
1473 	mlog_exit(ret);
1474 	return ret;
1475 }
1476 
1477 int ocfs2_rw_lock(struct inode *inode, int write)
1478 {
1479 	int status, level;
1480 	struct ocfs2_lock_res *lockres;
1481 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1482 
1483 	BUG_ON(!inode);
1484 
1485 	mlog_entry_void();
1486 
1487 	mlog(0, "inode %llu take %s RW lock\n",
1488 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1489 	     write ? "EXMODE" : "PRMODE");
1490 
1491 	if (ocfs2_mount_local(osb))
1492 		return 0;
1493 
1494 	lockres = &OCFS2_I(inode)->ip_rw_lockres;
1495 
1496 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1497 
1498 	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
1499 				    0);
1500 	if (status < 0)
1501 		mlog_errno(status);
1502 
1503 	mlog_exit(status);
1504 	return status;
1505 }
1506 
1507 void ocfs2_rw_unlock(struct inode *inode, int write)
1508 {
1509 	int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1510 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
1511 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1512 
1513 	mlog_entry_void();
1514 
1515 	mlog(0, "inode %llu drop %s RW lock\n",
1516 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1517 	     write ? "EXMODE" : "PRMODE");
1518 
1519 	if (!ocfs2_mount_local(osb))
1520 		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1521 
1522 	mlog_exit_void();
1523 }
1524 
1525 /*
1526  * ocfs2_open_lock always get PR mode lock.
1527  */
1528 int ocfs2_open_lock(struct inode *inode)
1529 {
1530 	int status = 0;
1531 	struct ocfs2_lock_res *lockres;
1532 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1533 
1534 	BUG_ON(!inode);
1535 
1536 	mlog_entry_void();
1537 
1538 	mlog(0, "inode %llu take PRMODE open lock\n",
1539 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1540 
1541 	if (ocfs2_mount_local(osb))
1542 		goto out;
1543 
1544 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1545 
1546 	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
1547 				    DLM_LOCK_PR, 0, 0);
1548 	if (status < 0)
1549 		mlog_errno(status);
1550 
1551 out:
1552 	mlog_exit(status);
1553 	return status;
1554 }
1555 
1556 int ocfs2_try_open_lock(struct inode *inode, int write)
1557 {
1558 	int status = 0, level;
1559 	struct ocfs2_lock_res *lockres;
1560 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1561 
1562 	BUG_ON(!inode);
1563 
1564 	mlog_entry_void();
1565 
1566 	mlog(0, "inode %llu try to take %s open lock\n",
1567 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1568 	     write ? "EXMODE" : "PRMODE");
1569 
1570 	if (ocfs2_mount_local(osb))
1571 		goto out;
1572 
1573 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1574 
1575 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1576 
1577 	/*
1578 	 * The file system may already holding a PRMODE/EXMODE open lock.
1579 	 * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
1580 	 * other nodes and the -EAGAIN will indicate to the caller that
1581 	 * this inode is still in use.
1582 	 */
1583 	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
1584 				    level, DLM_LKF_NOQUEUE, 0);
1585 
1586 out:
1587 	mlog_exit(status);
1588 	return status;
1589 }
1590 
1591 /*
1592  * ocfs2_open_unlock unlock PR and EX mode open locks.
1593  */
1594 void ocfs2_open_unlock(struct inode *inode)
1595 {
1596 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
1597 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1598 
1599 	mlog_entry_void();
1600 
1601 	mlog(0, "inode %llu drop open lock\n",
1602 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1603 
1604 	if (ocfs2_mount_local(osb))
1605 		goto out;
1606 
1607 	if(lockres->l_ro_holders)
1608 		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
1609 				     DLM_LOCK_PR);
1610 	if(lockres->l_ex_holders)
1611 		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
1612 				     DLM_LOCK_EX);
1613 
1614 out:
1615 	mlog_exit_void();
1616 }
1617 
1618 static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
1619 				     int level)
1620 {
1621 	int ret;
1622 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1623 	unsigned long flags;
1624 	struct ocfs2_mask_waiter mw;
1625 
1626 	ocfs2_init_mask_waiter(&mw);
1627 
1628 retry_cancel:
1629 	spin_lock_irqsave(&lockres->l_lock, flags);
1630 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
1631 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
1632 		if (ret) {
1633 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1634 			ret = ocfs2_cancel_convert(osb, lockres);
1635 			if (ret < 0) {
1636 				mlog_errno(ret);
1637 				goto out;
1638 			}
1639 			goto retry_cancel;
1640 		}
1641 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1642 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1643 
1644 		ocfs2_wait_for_mask(&mw);
1645 		goto retry_cancel;
1646 	}
1647 
1648 	ret = -ERESTARTSYS;
1649 	/*
1650 	 * We may still have gotten the lock, in which case there's no
1651 	 * point to restarting the syscall.
1652 	 */
1653 	if (lockres->l_level == level)
1654 		ret = 0;
1655 
1656 	mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
1657 	     lockres->l_flags, lockres->l_level, lockres->l_action);
1658 
1659 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1660 
1661 out:
1662 	return ret;
1663 }
1664 
1665 /*
1666  * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
1667  * flock() calls. The locking approach this requires is sufficiently
1668  * different from all other cluster lock types that we implement a
1669  * seperate path to the "low-level" dlm calls. In particular:
1670  *
1671  * - No optimization of lock levels is done - we take at exactly
1672  *   what's been requested.
1673  *
1674  * - No lock caching is employed. We immediately downconvert to
1675  *   no-lock at unlock time. This also means flock locks never go on
1676  *   the blocking list).
1677  *
1678  * - Since userspace can trivially deadlock itself with flock, we make
1679  *   sure to allow cancellation of a misbehaving applications flock()
1680  *   request.
1681  *
1682  * - Access to any flock lockres doesn't require concurrency, so we
1683  *   can simplify the code by requiring the caller to guarantee
1684  *   serialization of dlmglue flock calls.
1685  */
1686 int ocfs2_file_lock(struct file *file, int ex, int trylock)
1687 {
1688 	int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1689 	unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0;
1690 	unsigned long flags;
1691 	struct ocfs2_file_private *fp = file->private_data;
1692 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
1693 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1694 	struct ocfs2_mask_waiter mw;
1695 
1696 	ocfs2_init_mask_waiter(&mw);
1697 
1698 	if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
1699 	    (lockres->l_level > DLM_LOCK_NL)) {
1700 		mlog(ML_ERROR,
1701 		     "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
1702 		     "level: %u\n", lockres->l_name, lockres->l_flags,
1703 		     lockres->l_level);
1704 		return -EINVAL;
1705 	}
1706 
1707 	spin_lock_irqsave(&lockres->l_lock, flags);
1708 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1709 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1710 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1711 
1712 		/*
1713 		 * Get the lock at NLMODE to start - that way we
1714 		 * can cancel the upconvert request if need be.
1715 		 */
1716 		ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0);
1717 		if (ret < 0) {
1718 			mlog_errno(ret);
1719 			goto out;
1720 		}
1721 
1722 		ret = ocfs2_wait_for_mask(&mw);
1723 		if (ret) {
1724 			mlog_errno(ret);
1725 			goto out;
1726 		}
1727 		spin_lock_irqsave(&lockres->l_lock, flags);
1728 	}
1729 
1730 	lockres->l_action = OCFS2_AST_CONVERT;
1731 	lkm_flags |= DLM_LKF_CONVERT;
1732 	lockres->l_requested = level;
1733 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1734 
1735 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1736 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1737 
1738 	ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags,
1739 			     lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1,
1740 			     lockres);
1741 	if (ret) {
1742 		if (!trylock || (ret != -EAGAIN)) {
1743 			ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
1744 			ret = -EINVAL;
1745 		}
1746 
1747 		ocfs2_recover_from_dlm_error(lockres, 1);
1748 		lockres_remove_mask_waiter(lockres, &mw);
1749 		goto out;
1750 	}
1751 
1752 	ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
1753 	if (ret == -ERESTARTSYS) {
1754 		/*
1755 		 * Userspace can cause deadlock itself with
1756 		 * flock(). Current behavior locally is to allow the
1757 		 * deadlock, but abort the system call if a signal is
1758 		 * received. We follow this example, otherwise a
1759 		 * poorly written program could sit in kernel until
1760 		 * reboot.
1761 		 *
1762 		 * Handling this is a bit more complicated for Ocfs2
1763 		 * though. We can't exit this function with an
1764 		 * outstanding lock request, so a cancel convert is
1765 		 * required. We intentionally overwrite 'ret' - if the
1766 		 * cancel fails and the lock was granted, it's easier
1767 		 * to just bubble sucess back up to the user.
1768 		 */
1769 		ret = ocfs2_flock_handle_signal(lockres, level);
1770 	} else if (!ret && (level > lockres->l_level)) {
1771 		/* Trylock failed asynchronously */
1772 		BUG_ON(!trylock);
1773 		ret = -EAGAIN;
1774 	}
1775 
1776 out:
1777 
1778 	mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
1779 	     lockres->l_name, ex, trylock, ret);
1780 	return ret;
1781 }
1782 
1783 void ocfs2_file_unlock(struct file *file)
1784 {
1785 	int ret;
1786 	unsigned int gen;
1787 	unsigned long flags;
1788 	struct ocfs2_file_private *fp = file->private_data;
1789 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
1790 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1791 	struct ocfs2_mask_waiter mw;
1792 
1793 	ocfs2_init_mask_waiter(&mw);
1794 
1795 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
1796 		return;
1797 
1798 	if (lockres->l_level == DLM_LOCK_NL)
1799 		return;
1800 
1801 	mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
1802 	     lockres->l_name, lockres->l_flags, lockres->l_level,
1803 	     lockres->l_action);
1804 
1805 	spin_lock_irqsave(&lockres->l_lock, flags);
1806 	/*
1807 	 * Fake a blocking ast for the downconvert code.
1808 	 */
1809 	lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
1810 	lockres->l_blocking = DLM_LOCK_EX;
1811 
1812 	gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL);
1813 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1814 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1815 
1816 	ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen);
1817 	if (ret) {
1818 		mlog_errno(ret);
1819 		return;
1820 	}
1821 
1822 	ret = ocfs2_wait_for_mask(&mw);
1823 	if (ret)
1824 		mlog_errno(ret);
1825 }
1826 
1827 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
1828 					struct ocfs2_lock_res *lockres)
1829 {
1830 	int kick = 0;
1831 
1832 	mlog_entry_void();
1833 
1834 	/* If we know that another node is waiting on our lock, kick
1835 	 * the downconvert thread * pre-emptively when we reach a release
1836 	 * condition. */
1837 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
1838 		switch(lockres->l_blocking) {
1839 		case DLM_LOCK_EX:
1840 			if (!lockres->l_ex_holders && !lockres->l_ro_holders)
1841 				kick = 1;
1842 			break;
1843 		case DLM_LOCK_PR:
1844 			if (!lockres->l_ex_holders)
1845 				kick = 1;
1846 			break;
1847 		default:
1848 			BUG();
1849 		}
1850 	}
1851 
1852 	if (kick)
1853 		ocfs2_wake_downconvert_thread(osb);
1854 
1855 	mlog_exit_void();
1856 }
1857 
1858 #define OCFS2_SEC_BITS   34
1859 #define OCFS2_SEC_SHIFT  (64 - 34)
1860 #define OCFS2_NSEC_MASK  ((1ULL << OCFS2_SEC_SHIFT) - 1)
1861 
1862 /* LVB only has room for 64 bits of time here so we pack it for
1863  * now. */
1864 static u64 ocfs2_pack_timespec(struct timespec *spec)
1865 {
1866 	u64 res;
1867 	u64 sec = spec->tv_sec;
1868 	u32 nsec = spec->tv_nsec;
1869 
1870 	res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
1871 
1872 	return res;
1873 }
1874 
1875 /* Call this with the lockres locked. I am reasonably sure we don't
1876  * need ip_lock in this function as anyone who would be changing those
1877  * values is supposed to be blocked in ocfs2_inode_lock right now. */
1878 static void __ocfs2_stuff_meta_lvb(struct inode *inode)
1879 {
1880 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1881 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
1882 	struct ocfs2_meta_lvb *lvb;
1883 
1884 	mlog_entry_void();
1885 
1886 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
1887 
1888 	/*
1889 	 * Invalidate the LVB of a deleted inode - this way other
1890 	 * nodes are forced to go to disk and discover the new inode
1891 	 * status.
1892 	 */
1893 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
1894 		lvb->lvb_version = 0;
1895 		goto out;
1896 	}
1897 
1898 	lvb->lvb_version   = OCFS2_LVB_VERSION;
1899 	lvb->lvb_isize	   = cpu_to_be64(i_size_read(inode));
1900 	lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
1901 	lvb->lvb_iuid      = cpu_to_be32(inode->i_uid);
1902 	lvb->lvb_igid      = cpu_to_be32(inode->i_gid);
1903 	lvb->lvb_imode     = cpu_to_be16(inode->i_mode);
1904 	lvb->lvb_inlink    = cpu_to_be16(inode->i_nlink);
1905 	lvb->lvb_iatime_packed  =
1906 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
1907 	lvb->lvb_ictime_packed =
1908 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
1909 	lvb->lvb_imtime_packed =
1910 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
1911 	lvb->lvb_iattr    = cpu_to_be32(oi->ip_attr);
1912 	lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
1913 	lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
1914 
1915 out:
1916 	mlog_meta_lvb(0, lockres);
1917 
1918 	mlog_exit_void();
1919 }
1920 
1921 static void ocfs2_unpack_timespec(struct timespec *spec,
1922 				  u64 packed_time)
1923 {
1924 	spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
1925 	spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
1926 }
1927 
1928 static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
1929 {
1930 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1931 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
1932 	struct ocfs2_meta_lvb *lvb;
1933 
1934 	mlog_entry_void();
1935 
1936 	mlog_meta_lvb(0, lockres);
1937 
1938 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
1939 
1940 	/* We're safe here without the lockres lock... */
1941 	spin_lock(&oi->ip_lock);
1942 	oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
1943 	i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
1944 
1945 	oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
1946 	oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
1947 	ocfs2_set_inode_flags(inode);
1948 
1949 	/* fast-symlinks are a special case */
1950 	if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
1951 		inode->i_blocks = 0;
1952 	else
1953 		inode->i_blocks = ocfs2_inode_sector_count(inode);
1954 
1955 	inode->i_uid     = be32_to_cpu(lvb->lvb_iuid);
1956 	inode->i_gid     = be32_to_cpu(lvb->lvb_igid);
1957 	inode->i_mode    = be16_to_cpu(lvb->lvb_imode);
1958 	inode->i_nlink   = be16_to_cpu(lvb->lvb_inlink);
1959 	ocfs2_unpack_timespec(&inode->i_atime,
1960 			      be64_to_cpu(lvb->lvb_iatime_packed));
1961 	ocfs2_unpack_timespec(&inode->i_mtime,
1962 			      be64_to_cpu(lvb->lvb_imtime_packed));
1963 	ocfs2_unpack_timespec(&inode->i_ctime,
1964 			      be64_to_cpu(lvb->lvb_ictime_packed));
1965 	spin_unlock(&oi->ip_lock);
1966 
1967 	mlog_exit_void();
1968 }
1969 
1970 static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
1971 					      struct ocfs2_lock_res *lockres)
1972 {
1973 	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
1974 
1975 	if (lvb->lvb_version == OCFS2_LVB_VERSION
1976 	    && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
1977 		return 1;
1978 	return 0;
1979 }
1980 
1981 /* Determine whether a lock resource needs to be refreshed, and
1982  * arbitrate who gets to refresh it.
1983  *
1984  *   0 means no refresh needed.
1985  *
1986  *   > 0 means you need to refresh this and you MUST call
1987  *   ocfs2_complete_lock_res_refresh afterwards. */
1988 static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
1989 {
1990 	unsigned long flags;
1991 	int status = 0;
1992 
1993 	mlog_entry_void();
1994 
1995 refresh_check:
1996 	spin_lock_irqsave(&lockres->l_lock, flags);
1997 	if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
1998 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1999 		goto bail;
2000 	}
2001 
2002 	if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
2003 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2004 
2005 		ocfs2_wait_on_refreshing_lock(lockres);
2006 		goto refresh_check;
2007 	}
2008 
2009 	/* Ok, I'll be the one to refresh this lock. */
2010 	lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
2011 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2012 
2013 	status = 1;
2014 bail:
2015 	mlog_exit(status);
2016 	return status;
2017 }
2018 
2019 /* If status is non zero, I'll mark it as not being in refresh
2020  * anymroe, but i won't clear the needs refresh flag. */
2021 static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
2022 						   int status)
2023 {
2024 	unsigned long flags;
2025 	mlog_entry_void();
2026 
2027 	spin_lock_irqsave(&lockres->l_lock, flags);
2028 	lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
2029 	if (!status)
2030 		lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
2031 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2032 
2033 	wake_up(&lockres->l_event);
2034 
2035 	mlog_exit_void();
2036 }
2037 
2038 /* may or may not return a bh if it went to disk. */
2039 static int ocfs2_inode_lock_update(struct inode *inode,
2040 				  struct buffer_head **bh)
2041 {
2042 	int status = 0;
2043 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2044 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2045 	struct ocfs2_dinode *fe;
2046 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2047 
2048 	mlog_entry_void();
2049 
2050 	if (ocfs2_mount_local(osb))
2051 		goto bail;
2052 
2053 	spin_lock(&oi->ip_lock);
2054 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
2055 		mlog(0, "Orphaned inode %llu was deleted while we "
2056 		     "were waiting on a lock. ip_flags = 0x%x\n",
2057 		     (unsigned long long)oi->ip_blkno, oi->ip_flags);
2058 		spin_unlock(&oi->ip_lock);
2059 		status = -ENOENT;
2060 		goto bail;
2061 	}
2062 	spin_unlock(&oi->ip_lock);
2063 
2064 	if (!ocfs2_should_refresh_lock_res(lockres))
2065 		goto bail;
2066 
2067 	/* This will discard any caching information we might have had
2068 	 * for the inode metadata. */
2069 	ocfs2_metadata_cache_purge(inode);
2070 
2071 	ocfs2_extent_map_trunc(inode, 0);
2072 
2073 	if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
2074 		mlog(0, "Trusting LVB on inode %llu\n",
2075 		     (unsigned long long)oi->ip_blkno);
2076 		ocfs2_refresh_inode_from_lvb(inode);
2077 	} else {
2078 		/* Boo, we have to go to disk. */
2079 		/* read bh, cast, ocfs2_refresh_inode */
2080 		status = ocfs2_read_inode_block(inode, bh);
2081 		if (status < 0) {
2082 			mlog_errno(status);
2083 			goto bail_refresh;
2084 		}
2085 		fe = (struct ocfs2_dinode *) (*bh)->b_data;
2086 
2087 		/* This is a good chance to make sure we're not
2088 		 * locking an invalid object.  ocfs2_read_inode_block()
2089 		 * already checked that the inode block is sane.
2090 		 *
2091 		 * We bug on a stale inode here because we checked
2092 		 * above whether it was wiped from disk. The wiping
2093 		 * node provides a guarantee that we receive that
2094 		 * message and can mark the inode before dropping any
2095 		 * locks associated with it. */
2096 		mlog_bug_on_msg(inode->i_generation !=
2097 				le32_to_cpu(fe->i_generation),
2098 				"Invalid dinode %llu disk generation: %u "
2099 				"inode->i_generation: %u\n",
2100 				(unsigned long long)oi->ip_blkno,
2101 				le32_to_cpu(fe->i_generation),
2102 				inode->i_generation);
2103 		mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
2104 				!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
2105 				"Stale dinode %llu dtime: %llu flags: 0x%x\n",
2106 				(unsigned long long)oi->ip_blkno,
2107 				(unsigned long long)le64_to_cpu(fe->i_dtime),
2108 				le32_to_cpu(fe->i_flags));
2109 
2110 		ocfs2_refresh_inode(inode, fe);
2111 		ocfs2_track_lock_refresh(lockres);
2112 	}
2113 
2114 	status = 0;
2115 bail_refresh:
2116 	ocfs2_complete_lock_res_refresh(lockres, status);
2117 bail:
2118 	mlog_exit(status);
2119 	return status;
2120 }
2121 
2122 static int ocfs2_assign_bh(struct inode *inode,
2123 			   struct buffer_head **ret_bh,
2124 			   struct buffer_head *passed_bh)
2125 {
2126 	int status;
2127 
2128 	if (passed_bh) {
2129 		/* Ok, the update went to disk for us, use the
2130 		 * returned bh. */
2131 		*ret_bh = passed_bh;
2132 		get_bh(*ret_bh);
2133 
2134 		return 0;
2135 	}
2136 
2137 	status = ocfs2_read_inode_block(inode, ret_bh);
2138 	if (status < 0)
2139 		mlog_errno(status);
2140 
2141 	return status;
2142 }
2143 
2144 /*
2145  * returns < 0 error if the callback will never be called, otherwise
2146  * the result of the lock will be communicated via the callback.
2147  */
2148 int ocfs2_inode_lock_full(struct inode *inode,
2149 			 struct buffer_head **ret_bh,
2150 			 int ex,
2151 			 int arg_flags)
2152 {
2153 	int status, level, acquired;
2154 	u32 dlm_flags;
2155 	struct ocfs2_lock_res *lockres = NULL;
2156 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2157 	struct buffer_head *local_bh = NULL;
2158 
2159 	BUG_ON(!inode);
2160 
2161 	mlog_entry_void();
2162 
2163 	mlog(0, "inode %llu, take %s META lock\n",
2164 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2165 	     ex ? "EXMODE" : "PRMODE");
2166 
2167 	status = 0;
2168 	acquired = 0;
2169 	/* We'll allow faking a readonly metadata lock for
2170 	 * rodevices. */
2171 	if (ocfs2_is_hard_readonly(osb)) {
2172 		if (ex)
2173 			status = -EROFS;
2174 		goto bail;
2175 	}
2176 
2177 	if (ocfs2_mount_local(osb))
2178 		goto local;
2179 
2180 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2181 		ocfs2_wait_for_recovery(osb);
2182 
2183 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2184 	level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2185 	dlm_flags = 0;
2186 	if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
2187 		dlm_flags |= DLM_LKF_NOQUEUE;
2188 
2189 	status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags);
2190 	if (status < 0) {
2191 		if (status != -EAGAIN && status != -EIOCBRETRY)
2192 			mlog_errno(status);
2193 		goto bail;
2194 	}
2195 
2196 	/* Notify the error cleanup path to drop the cluster lock. */
2197 	acquired = 1;
2198 
2199 	/* We wait twice because a node may have died while we were in
2200 	 * the lower dlm layers. The second time though, we've
2201 	 * committed to owning this lock so we don't allow signals to
2202 	 * abort the operation. */
2203 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2204 		ocfs2_wait_for_recovery(osb);
2205 
2206 local:
2207 	/*
2208 	 * We only see this flag if we're being called from
2209 	 * ocfs2_read_locked_inode(). It means we're locking an inode
2210 	 * which hasn't been populated yet, so clear the refresh flag
2211 	 * and let the caller handle it.
2212 	 */
2213 	if (inode->i_state & I_NEW) {
2214 		status = 0;
2215 		if (lockres)
2216 			ocfs2_complete_lock_res_refresh(lockres, 0);
2217 		goto bail;
2218 	}
2219 
2220 	/* This is fun. The caller may want a bh back, or it may
2221 	 * not. ocfs2_inode_lock_update definitely wants one in, but
2222 	 * may or may not read one, depending on what's in the
2223 	 * LVB. The result of all of this is that we've *only* gone to
2224 	 * disk if we have to, so the complexity is worthwhile. */
2225 	status = ocfs2_inode_lock_update(inode, &local_bh);
2226 	if (status < 0) {
2227 		if (status != -ENOENT)
2228 			mlog_errno(status);
2229 		goto bail;
2230 	}
2231 
2232 	if (ret_bh) {
2233 		status = ocfs2_assign_bh(inode, ret_bh, local_bh);
2234 		if (status < 0) {
2235 			mlog_errno(status);
2236 			goto bail;
2237 		}
2238 	}
2239 
2240 bail:
2241 	if (status < 0) {
2242 		if (ret_bh && (*ret_bh)) {
2243 			brelse(*ret_bh);
2244 			*ret_bh = NULL;
2245 		}
2246 		if (acquired)
2247 			ocfs2_inode_unlock(inode, ex);
2248 	}
2249 
2250 	if (local_bh)
2251 		brelse(local_bh);
2252 
2253 	mlog_exit(status);
2254 	return status;
2255 }
2256 
2257 /*
2258  * This is working around a lock inversion between tasks acquiring DLM
2259  * locks while holding a page lock and the downconvert thread which
2260  * blocks dlm lock acquiry while acquiring page locks.
2261  *
2262  * ** These _with_page variantes are only intended to be called from aop
2263  * methods that hold page locks and return a very specific *positive* error
2264  * code that aop methods pass up to the VFS -- test for errors with != 0. **
2265  *
2266  * The DLM is called such that it returns -EAGAIN if it would have
2267  * blocked waiting for the downconvert thread.  In that case we unlock
2268  * our page so the downconvert thread can make progress.  Once we've
2269  * done this we have to return AOP_TRUNCATED_PAGE so the aop method
2270  * that called us can bubble that back up into the VFS who will then
2271  * immediately retry the aop call.
2272  *
2273  * We do a blocking lock and immediate unlock before returning, though, so that
2274  * the lock has a great chance of being cached on this node by the time the VFS
2275  * calls back to retry the aop.    This has a potential to livelock as nodes
2276  * ping locks back and forth, but that's a risk we're willing to take to avoid
2277  * the lock inversion simply.
2278  */
2279 int ocfs2_inode_lock_with_page(struct inode *inode,
2280 			      struct buffer_head **ret_bh,
2281 			      int ex,
2282 			      struct page *page)
2283 {
2284 	int ret;
2285 
2286 	ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
2287 	if (ret == -EAGAIN) {
2288 		unlock_page(page);
2289 		if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
2290 			ocfs2_inode_unlock(inode, ex);
2291 		ret = AOP_TRUNCATED_PAGE;
2292 	}
2293 
2294 	return ret;
2295 }
2296 
2297 int ocfs2_inode_lock_atime(struct inode *inode,
2298 			  struct vfsmount *vfsmnt,
2299 			  int *level)
2300 {
2301 	int ret;
2302 
2303 	mlog_entry_void();
2304 	ret = ocfs2_inode_lock(inode, NULL, 0);
2305 	if (ret < 0) {
2306 		mlog_errno(ret);
2307 		return ret;
2308 	}
2309 
2310 	/*
2311 	 * If we should update atime, we will get EX lock,
2312 	 * otherwise we just get PR lock.
2313 	 */
2314 	if (ocfs2_should_update_atime(inode, vfsmnt)) {
2315 		struct buffer_head *bh = NULL;
2316 
2317 		ocfs2_inode_unlock(inode, 0);
2318 		ret = ocfs2_inode_lock(inode, &bh, 1);
2319 		if (ret < 0) {
2320 			mlog_errno(ret);
2321 			return ret;
2322 		}
2323 		*level = 1;
2324 		if (ocfs2_should_update_atime(inode, vfsmnt))
2325 			ocfs2_update_inode_atime(inode, bh);
2326 		if (bh)
2327 			brelse(bh);
2328 	} else
2329 		*level = 0;
2330 
2331 	mlog_exit(ret);
2332 	return ret;
2333 }
2334 
2335 void ocfs2_inode_unlock(struct inode *inode,
2336 		       int ex)
2337 {
2338 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2339 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
2340 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2341 
2342 	mlog_entry_void();
2343 
2344 	mlog(0, "inode %llu drop %s META lock\n",
2345 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2346 	     ex ? "EXMODE" : "PRMODE");
2347 
2348 	if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) &&
2349 	    !ocfs2_mount_local(osb))
2350 		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
2351 
2352 	mlog_exit_void();
2353 }
2354 
2355 int ocfs2_super_lock(struct ocfs2_super *osb,
2356 		     int ex)
2357 {
2358 	int status = 0;
2359 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2360 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2361 
2362 	mlog_entry_void();
2363 
2364 	if (ocfs2_is_hard_readonly(osb))
2365 		return -EROFS;
2366 
2367 	if (ocfs2_mount_local(osb))
2368 		goto bail;
2369 
2370 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
2371 	if (status < 0) {
2372 		mlog_errno(status);
2373 		goto bail;
2374 	}
2375 
2376 	/* The super block lock path is really in the best position to
2377 	 * know when resources covered by the lock need to be
2378 	 * refreshed, so we do it here. Of course, making sense of
2379 	 * everything is up to the caller :) */
2380 	status = ocfs2_should_refresh_lock_res(lockres);
2381 	if (status < 0) {
2382 		mlog_errno(status);
2383 		goto bail;
2384 	}
2385 	if (status) {
2386 		status = ocfs2_refresh_slot_info(osb);
2387 
2388 		ocfs2_complete_lock_res_refresh(lockres, status);
2389 
2390 		if (status < 0)
2391 			mlog_errno(status);
2392 		ocfs2_track_lock_refresh(lockres);
2393 	}
2394 bail:
2395 	mlog_exit(status);
2396 	return status;
2397 }
2398 
2399 void ocfs2_super_unlock(struct ocfs2_super *osb,
2400 			int ex)
2401 {
2402 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2403 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2404 
2405 	if (!ocfs2_mount_local(osb))
2406 		ocfs2_cluster_unlock(osb, lockres, level);
2407 }
2408 
2409 int ocfs2_rename_lock(struct ocfs2_super *osb)
2410 {
2411 	int status;
2412 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2413 
2414 	if (ocfs2_is_hard_readonly(osb))
2415 		return -EROFS;
2416 
2417 	if (ocfs2_mount_local(osb))
2418 		return 0;
2419 
2420 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2421 	if (status < 0)
2422 		mlog_errno(status);
2423 
2424 	return status;
2425 }
2426 
2427 void ocfs2_rename_unlock(struct ocfs2_super *osb)
2428 {
2429 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2430 
2431 	if (!ocfs2_mount_local(osb))
2432 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2433 }
2434 
2435 int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex)
2436 {
2437 	int status;
2438 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2439 
2440 	if (ocfs2_is_hard_readonly(osb))
2441 		return -EROFS;
2442 
2443 	if (ocfs2_mount_local(osb))
2444 		return 0;
2445 
2446 	status = ocfs2_cluster_lock(osb, lockres, ex ? LKM_EXMODE : LKM_PRMODE,
2447 				    0, 0);
2448 	if (status < 0)
2449 		mlog(ML_ERROR, "lock on nfs sync lock failed %d\n", status);
2450 
2451 	return status;
2452 }
2453 
2454 void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex)
2455 {
2456 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2457 
2458 	if (!ocfs2_mount_local(osb))
2459 		ocfs2_cluster_unlock(osb, lockres,
2460 				     ex ? LKM_EXMODE : LKM_PRMODE);
2461 }
2462 
2463 int ocfs2_dentry_lock(struct dentry *dentry, int ex)
2464 {
2465 	int ret;
2466 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2467 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2468 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2469 
2470 	BUG_ON(!dl);
2471 
2472 	if (ocfs2_is_hard_readonly(osb))
2473 		return -EROFS;
2474 
2475 	if (ocfs2_mount_local(osb))
2476 		return 0;
2477 
2478 	ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2479 	if (ret < 0)
2480 		mlog_errno(ret);
2481 
2482 	return ret;
2483 }
2484 
2485 void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2486 {
2487 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2488 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2489 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2490 
2491 	if (!ocfs2_mount_local(osb))
2492 		ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
2493 }
2494 
2495 /* Reference counting of the dlm debug structure. We want this because
2496  * open references on the debug inodes can live on after a mount, so
2497  * we can't rely on the ocfs2_super to always exist. */
2498 static void ocfs2_dlm_debug_free(struct kref *kref)
2499 {
2500 	struct ocfs2_dlm_debug *dlm_debug;
2501 
2502 	dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
2503 
2504 	kfree(dlm_debug);
2505 }
2506 
2507 void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
2508 {
2509 	if (dlm_debug)
2510 		kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
2511 }
2512 
2513 static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
2514 {
2515 	kref_get(&debug->d_refcnt);
2516 }
2517 
2518 struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
2519 {
2520 	struct ocfs2_dlm_debug *dlm_debug;
2521 
2522 	dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
2523 	if (!dlm_debug) {
2524 		mlog_errno(-ENOMEM);
2525 		goto out;
2526 	}
2527 
2528 	kref_init(&dlm_debug->d_refcnt);
2529 	INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
2530 	dlm_debug->d_locking_state = NULL;
2531 out:
2532 	return dlm_debug;
2533 }
2534 
2535 /* Access to this is arbitrated for us via seq_file->sem. */
2536 struct ocfs2_dlm_seq_priv {
2537 	struct ocfs2_dlm_debug *p_dlm_debug;
2538 	struct ocfs2_lock_res p_iter_res;
2539 	struct ocfs2_lock_res p_tmp_res;
2540 };
2541 
2542 static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
2543 						 struct ocfs2_dlm_seq_priv *priv)
2544 {
2545 	struct ocfs2_lock_res *iter, *ret = NULL;
2546 	struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
2547 
2548 	assert_spin_locked(&ocfs2_dlm_tracking_lock);
2549 
2550 	list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
2551 		/* discover the head of the list */
2552 		if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
2553 			mlog(0, "End of list found, %p\n", ret);
2554 			break;
2555 		}
2556 
2557 		/* We track our "dummy" iteration lockres' by a NULL
2558 		 * l_ops field. */
2559 		if (iter->l_ops != NULL) {
2560 			ret = iter;
2561 			break;
2562 		}
2563 	}
2564 
2565 	return ret;
2566 }
2567 
2568 static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
2569 {
2570 	struct ocfs2_dlm_seq_priv *priv = m->private;
2571 	struct ocfs2_lock_res *iter;
2572 
2573 	spin_lock(&ocfs2_dlm_tracking_lock);
2574 	iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
2575 	if (iter) {
2576 		/* Since lockres' have the lifetime of their container
2577 		 * (which can be inodes, ocfs2_supers, etc) we want to
2578 		 * copy this out to a temporary lockres while still
2579 		 * under the spinlock. Obviously after this we can't
2580 		 * trust any pointers on the copy returned, but that's
2581 		 * ok as the information we want isn't typically held
2582 		 * in them. */
2583 		priv->p_tmp_res = *iter;
2584 		iter = &priv->p_tmp_res;
2585 	}
2586 	spin_unlock(&ocfs2_dlm_tracking_lock);
2587 
2588 	return iter;
2589 }
2590 
2591 static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
2592 {
2593 }
2594 
2595 static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
2596 {
2597 	struct ocfs2_dlm_seq_priv *priv = m->private;
2598 	struct ocfs2_lock_res *iter = v;
2599 	struct ocfs2_lock_res *dummy = &priv->p_iter_res;
2600 
2601 	spin_lock(&ocfs2_dlm_tracking_lock);
2602 	iter = ocfs2_dlm_next_res(iter, priv);
2603 	list_del_init(&dummy->l_debug_list);
2604 	if (iter) {
2605 		list_add(&dummy->l_debug_list, &iter->l_debug_list);
2606 		priv->p_tmp_res = *iter;
2607 		iter = &priv->p_tmp_res;
2608 	}
2609 	spin_unlock(&ocfs2_dlm_tracking_lock);
2610 
2611 	return iter;
2612 }
2613 
2614 /* So that debugfs.ocfs2 can determine which format is being used */
2615 #define OCFS2_DLM_DEBUG_STR_VERSION 2
2616 static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
2617 {
2618 	int i;
2619 	char *lvb;
2620 	struct ocfs2_lock_res *lockres = v;
2621 
2622 	if (!lockres)
2623 		return -EINVAL;
2624 
2625 	seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
2626 
2627 	if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
2628 		seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
2629 			   lockres->l_name,
2630 			   (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
2631 	else
2632 		seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
2633 
2634 	seq_printf(m, "%d\t"
2635 		   "0x%lx\t"
2636 		   "0x%x\t"
2637 		   "0x%x\t"
2638 		   "%u\t"
2639 		   "%u\t"
2640 		   "%d\t"
2641 		   "%d\t",
2642 		   lockres->l_level,
2643 		   lockres->l_flags,
2644 		   lockres->l_action,
2645 		   lockres->l_unlock_action,
2646 		   lockres->l_ro_holders,
2647 		   lockres->l_ex_holders,
2648 		   lockres->l_requested,
2649 		   lockres->l_blocking);
2650 
2651 	/* Dump the raw LVB */
2652 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2653 	for(i = 0; i < DLM_LVB_LEN; i++)
2654 		seq_printf(m, "0x%x\t", lvb[i]);
2655 
2656 #ifdef CONFIG_OCFS2_FS_STATS
2657 # define lock_num_prmode(_l)		(_l)->l_lock_num_prmode
2658 # define lock_num_exmode(_l)		(_l)->l_lock_num_exmode
2659 # define lock_num_prmode_failed(_l)	(_l)->l_lock_num_prmode_failed
2660 # define lock_num_exmode_failed(_l)	(_l)->l_lock_num_exmode_failed
2661 # define lock_total_prmode(_l)		(_l)->l_lock_total_prmode
2662 # define lock_total_exmode(_l)		(_l)->l_lock_total_exmode
2663 # define lock_max_prmode(_l)		(_l)->l_lock_max_prmode
2664 # define lock_max_exmode(_l)		(_l)->l_lock_max_exmode
2665 # define lock_refresh(_l)		(_l)->l_lock_refresh
2666 #else
2667 # define lock_num_prmode(_l)		(0ULL)
2668 # define lock_num_exmode(_l)		(0ULL)
2669 # define lock_num_prmode_failed(_l)	(0)
2670 # define lock_num_exmode_failed(_l)	(0)
2671 # define lock_total_prmode(_l)		(0ULL)
2672 # define lock_total_exmode(_l)		(0ULL)
2673 # define lock_max_prmode(_l)		(0)
2674 # define lock_max_exmode(_l)		(0)
2675 # define lock_refresh(_l)		(0)
2676 #endif
2677 	/* The following seq_print was added in version 2 of this output */
2678 	seq_printf(m, "%llu\t"
2679 		   "%llu\t"
2680 		   "%u\t"
2681 		   "%u\t"
2682 		   "%llu\t"
2683 		   "%llu\t"
2684 		   "%u\t"
2685 		   "%u\t"
2686 		   "%u\t",
2687 		   lock_num_prmode(lockres),
2688 		   lock_num_exmode(lockres),
2689 		   lock_num_prmode_failed(lockres),
2690 		   lock_num_exmode_failed(lockres),
2691 		   lock_total_prmode(lockres),
2692 		   lock_total_exmode(lockres),
2693 		   lock_max_prmode(lockres),
2694 		   lock_max_exmode(lockres),
2695 		   lock_refresh(lockres));
2696 
2697 	/* End the line */
2698 	seq_printf(m, "\n");
2699 	return 0;
2700 }
2701 
2702 static const struct seq_operations ocfs2_dlm_seq_ops = {
2703 	.start =	ocfs2_dlm_seq_start,
2704 	.stop =		ocfs2_dlm_seq_stop,
2705 	.next =		ocfs2_dlm_seq_next,
2706 	.show =		ocfs2_dlm_seq_show,
2707 };
2708 
2709 static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
2710 {
2711 	struct seq_file *seq = (struct seq_file *) file->private_data;
2712 	struct ocfs2_dlm_seq_priv *priv = seq->private;
2713 	struct ocfs2_lock_res *res = &priv->p_iter_res;
2714 
2715 	ocfs2_remove_lockres_tracking(res);
2716 	ocfs2_put_dlm_debug(priv->p_dlm_debug);
2717 	return seq_release_private(inode, file);
2718 }
2719 
2720 static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
2721 {
2722 	int ret;
2723 	struct ocfs2_dlm_seq_priv *priv;
2724 	struct seq_file *seq;
2725 	struct ocfs2_super *osb;
2726 
2727 	priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
2728 	if (!priv) {
2729 		ret = -ENOMEM;
2730 		mlog_errno(ret);
2731 		goto out;
2732 	}
2733 	osb = inode->i_private;
2734 	ocfs2_get_dlm_debug(osb->osb_dlm_debug);
2735 	priv->p_dlm_debug = osb->osb_dlm_debug;
2736 	INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
2737 
2738 	ret = seq_open(file, &ocfs2_dlm_seq_ops);
2739 	if (ret) {
2740 		kfree(priv);
2741 		mlog_errno(ret);
2742 		goto out;
2743 	}
2744 
2745 	seq = (struct seq_file *) file->private_data;
2746 	seq->private = priv;
2747 
2748 	ocfs2_add_lockres_tracking(&priv->p_iter_res,
2749 				   priv->p_dlm_debug);
2750 
2751 out:
2752 	return ret;
2753 }
2754 
2755 static const struct file_operations ocfs2_dlm_debug_fops = {
2756 	.open =		ocfs2_dlm_debug_open,
2757 	.release =	ocfs2_dlm_debug_release,
2758 	.read =		seq_read,
2759 	.llseek =	seq_lseek,
2760 };
2761 
2762 static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
2763 {
2764 	int ret = 0;
2765 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2766 
2767 	dlm_debug->d_locking_state = debugfs_create_file("locking_state",
2768 							 S_IFREG|S_IRUSR,
2769 							 osb->osb_debug_root,
2770 							 osb,
2771 							 &ocfs2_dlm_debug_fops);
2772 	if (!dlm_debug->d_locking_state) {
2773 		ret = -EINVAL;
2774 		mlog(ML_ERROR,
2775 		     "Unable to create locking state debugfs file.\n");
2776 		goto out;
2777 	}
2778 
2779 	ocfs2_get_dlm_debug(dlm_debug);
2780 out:
2781 	return ret;
2782 }
2783 
2784 static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
2785 {
2786 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2787 
2788 	if (dlm_debug) {
2789 		debugfs_remove(dlm_debug->d_locking_state);
2790 		ocfs2_put_dlm_debug(dlm_debug);
2791 	}
2792 }
2793 
2794 int ocfs2_dlm_init(struct ocfs2_super *osb)
2795 {
2796 	int status = 0;
2797 	struct ocfs2_cluster_connection *conn = NULL;
2798 
2799 	mlog_entry_void();
2800 
2801 	if (ocfs2_mount_local(osb)) {
2802 		osb->node_num = 0;
2803 		goto local;
2804 	}
2805 
2806 	status = ocfs2_dlm_init_debug(osb);
2807 	if (status < 0) {
2808 		mlog_errno(status);
2809 		goto bail;
2810 	}
2811 
2812 	/* launch downconvert thread */
2813 	osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc");
2814 	if (IS_ERR(osb->dc_task)) {
2815 		status = PTR_ERR(osb->dc_task);
2816 		osb->dc_task = NULL;
2817 		mlog_errno(status);
2818 		goto bail;
2819 	}
2820 
2821 	/* for now, uuid == domain */
2822 	status = ocfs2_cluster_connect(osb->osb_cluster_stack,
2823 				       osb->uuid_str,
2824 				       strlen(osb->uuid_str),
2825 				       ocfs2_do_node_down, osb,
2826 				       &conn);
2827 	if (status) {
2828 		mlog_errno(status);
2829 		goto bail;
2830 	}
2831 
2832 	status = ocfs2_cluster_this_node(&osb->node_num);
2833 	if (status < 0) {
2834 		mlog_errno(status);
2835 		mlog(ML_ERROR,
2836 		     "could not find this host's node number\n");
2837 		ocfs2_cluster_disconnect(conn, 0);
2838 		goto bail;
2839 	}
2840 
2841 local:
2842 	ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
2843 	ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
2844 	ocfs2_nfs_sync_lock_res_init(&osb->osb_nfs_sync_lockres, osb);
2845 
2846 	osb->cconn = conn;
2847 
2848 	status = 0;
2849 bail:
2850 	if (status < 0) {
2851 		ocfs2_dlm_shutdown_debug(osb);
2852 		if (osb->dc_task)
2853 			kthread_stop(osb->dc_task);
2854 	}
2855 
2856 	mlog_exit(status);
2857 	return status;
2858 }
2859 
2860 void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
2861 			int hangup_pending)
2862 {
2863 	mlog_entry_void();
2864 
2865 	ocfs2_drop_osb_locks(osb);
2866 
2867 	/*
2868 	 * Now that we have dropped all locks and ocfs2_dismount_volume()
2869 	 * has disabled recovery, the DLM won't be talking to us.  It's
2870 	 * safe to tear things down before disconnecting the cluster.
2871 	 */
2872 
2873 	if (osb->dc_task) {
2874 		kthread_stop(osb->dc_task);
2875 		osb->dc_task = NULL;
2876 	}
2877 
2878 	ocfs2_lock_res_free(&osb->osb_super_lockres);
2879 	ocfs2_lock_res_free(&osb->osb_rename_lockres);
2880 	ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres);
2881 
2882 	ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
2883 	osb->cconn = NULL;
2884 
2885 	ocfs2_dlm_shutdown_debug(osb);
2886 
2887 	mlog_exit_void();
2888 }
2889 
2890 static void ocfs2_unlock_ast(void *opaque, int error)
2891 {
2892 	struct ocfs2_lock_res *lockres = opaque;
2893 	unsigned long flags;
2894 
2895 	mlog_entry_void();
2896 
2897 	mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name,
2898 	     lockres->l_unlock_action);
2899 
2900 	spin_lock_irqsave(&lockres->l_lock, flags);
2901 	if (error) {
2902 		mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
2903 		     "unlock_action %d\n", error, lockres->l_name,
2904 		     lockres->l_unlock_action);
2905 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2906 		return;
2907 	}
2908 
2909 	switch(lockres->l_unlock_action) {
2910 	case OCFS2_UNLOCK_CANCEL_CONVERT:
2911 		mlog(0, "Cancel convert success for %s\n", lockres->l_name);
2912 		lockres->l_action = OCFS2_AST_INVALID;
2913 		/* Downconvert thread may have requeued this lock, we
2914 		 * need to wake it. */
2915 		if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
2916 			ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres));
2917 		break;
2918 	case OCFS2_UNLOCK_DROP_LOCK:
2919 		lockres->l_level = DLM_LOCK_IV;
2920 		break;
2921 	default:
2922 		BUG();
2923 	}
2924 
2925 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
2926 	lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
2927 	wake_up(&lockres->l_event);
2928 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2929 
2930 	mlog_exit_void();
2931 }
2932 
2933 static int ocfs2_drop_lock(struct ocfs2_super *osb,
2934 			   struct ocfs2_lock_res *lockres)
2935 {
2936 	int ret;
2937 	unsigned long flags;
2938 	u32 lkm_flags = 0;
2939 
2940 	/* We didn't get anywhere near actually using this lockres. */
2941 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
2942 		goto out;
2943 
2944 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
2945 		lkm_flags |= DLM_LKF_VALBLK;
2946 
2947 	spin_lock_irqsave(&lockres->l_lock, flags);
2948 
2949 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
2950 			"lockres %s, flags 0x%lx\n",
2951 			lockres->l_name, lockres->l_flags);
2952 
2953 	while (lockres->l_flags & OCFS2_LOCK_BUSY) {
2954 		mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
2955 		     "%u, unlock_action = %u\n",
2956 		     lockres->l_name, lockres->l_flags, lockres->l_action,
2957 		     lockres->l_unlock_action);
2958 
2959 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2960 
2961 		/* XXX: Today we just wait on any busy
2962 		 * locks... Perhaps we need to cancel converts in the
2963 		 * future? */
2964 		ocfs2_wait_on_busy_lock(lockres);
2965 
2966 		spin_lock_irqsave(&lockres->l_lock, flags);
2967 	}
2968 
2969 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
2970 		if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
2971 		    lockres->l_level == DLM_LOCK_EX &&
2972 		    !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
2973 			lockres->l_ops->set_lvb(lockres);
2974 	}
2975 
2976 	if (lockres->l_flags & OCFS2_LOCK_BUSY)
2977 		mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
2978 		     lockres->l_name);
2979 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
2980 		mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
2981 
2982 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
2983 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2984 		goto out;
2985 	}
2986 
2987 	lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
2988 
2989 	/* make sure we never get here while waiting for an ast to
2990 	 * fire. */
2991 	BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
2992 
2993 	/* is this necessary? */
2994 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2995 	lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
2996 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2997 
2998 	mlog(0, "lock %s\n", lockres->l_name);
2999 
3000 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags,
3001 			       lockres);
3002 	if (ret) {
3003 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3004 		mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
3005 		ocfs2_dlm_dump_lksb(&lockres->l_lksb);
3006 		BUG();
3007 	}
3008 	mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n",
3009 	     lockres->l_name);
3010 
3011 	ocfs2_wait_on_busy_lock(lockres);
3012 out:
3013 	mlog_exit(0);
3014 	return 0;
3015 }
3016 
3017 /* Mark the lockres as being dropped. It will no longer be
3018  * queued if blocking, but we still may have to wait on it
3019  * being dequeued from the downconvert thread before we can consider
3020  * it safe to drop.
3021  *
3022  * You can *not* attempt to call cluster_lock on this lockres anymore. */
3023 void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
3024 {
3025 	int status;
3026 	struct ocfs2_mask_waiter mw;
3027 	unsigned long flags;
3028 
3029 	ocfs2_init_mask_waiter(&mw);
3030 
3031 	spin_lock_irqsave(&lockres->l_lock, flags);
3032 	lockres->l_flags |= OCFS2_LOCK_FREEING;
3033 	while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
3034 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
3035 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3036 
3037 		mlog(0, "Waiting on lockres %s\n", lockres->l_name);
3038 
3039 		status = ocfs2_wait_for_mask(&mw);
3040 		if (status)
3041 			mlog_errno(status);
3042 
3043 		spin_lock_irqsave(&lockres->l_lock, flags);
3044 	}
3045 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3046 }
3047 
3048 void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
3049 			       struct ocfs2_lock_res *lockres)
3050 {
3051 	int ret;
3052 
3053 	ocfs2_mark_lockres_freeing(lockres);
3054 	ret = ocfs2_drop_lock(osb, lockres);
3055 	if (ret)
3056 		mlog_errno(ret);
3057 }
3058 
3059 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
3060 {
3061 	ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
3062 	ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
3063 	ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres);
3064 }
3065 
3066 int ocfs2_drop_inode_locks(struct inode *inode)
3067 {
3068 	int status, err;
3069 
3070 	mlog_entry_void();
3071 
3072 	/* No need to call ocfs2_mark_lockres_freeing here -
3073 	 * ocfs2_clear_inode has done it for us. */
3074 
3075 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3076 			      &OCFS2_I(inode)->ip_open_lockres);
3077 	if (err < 0)
3078 		mlog_errno(err);
3079 
3080 	status = err;
3081 
3082 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3083 			      &OCFS2_I(inode)->ip_inode_lockres);
3084 	if (err < 0)
3085 		mlog_errno(err);
3086 	if (err < 0 && !status)
3087 		status = err;
3088 
3089 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3090 			      &OCFS2_I(inode)->ip_rw_lockres);
3091 	if (err < 0)
3092 		mlog_errno(err);
3093 	if (err < 0 && !status)
3094 		status = err;
3095 
3096 	mlog_exit(status);
3097 	return status;
3098 }
3099 
3100 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
3101 					      int new_level)
3102 {
3103 	assert_spin_locked(&lockres->l_lock);
3104 
3105 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
3106 
3107 	if (lockres->l_level <= new_level) {
3108 		mlog(ML_ERROR, "lockres->l_level (%d) <= new_level (%d)\n",
3109 		     lockres->l_level, new_level);
3110 		BUG();
3111 	}
3112 
3113 	mlog(0, "lock %s, new_level = %d, l_blocking = %d\n",
3114 	     lockres->l_name, new_level, lockres->l_blocking);
3115 
3116 	lockres->l_action = OCFS2_AST_DOWNCONVERT;
3117 	lockres->l_requested = new_level;
3118 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3119 	return lockres_set_pending(lockres);
3120 }
3121 
3122 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
3123 				  struct ocfs2_lock_res *lockres,
3124 				  int new_level,
3125 				  int lvb,
3126 				  unsigned int generation)
3127 {
3128 	int ret;
3129 	u32 dlm_flags = DLM_LKF_CONVERT;
3130 
3131 	mlog_entry_void();
3132 
3133 	if (lvb)
3134 		dlm_flags |= DLM_LKF_VALBLK;
3135 
3136 	ret = ocfs2_dlm_lock(osb->cconn,
3137 			     new_level,
3138 			     &lockres->l_lksb,
3139 			     dlm_flags,
3140 			     lockres->l_name,
3141 			     OCFS2_LOCK_ID_MAX_LEN - 1,
3142 			     lockres);
3143 	lockres_clear_pending(lockres, generation, osb);
3144 	if (ret) {
3145 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
3146 		ocfs2_recover_from_dlm_error(lockres, 1);
3147 		goto bail;
3148 	}
3149 
3150 	ret = 0;
3151 bail:
3152 	mlog_exit(ret);
3153 	return ret;
3154 }
3155 
3156 /* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
3157 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
3158 				        struct ocfs2_lock_res *lockres)
3159 {
3160 	assert_spin_locked(&lockres->l_lock);
3161 
3162 	mlog_entry_void();
3163 	mlog(0, "lock %s\n", lockres->l_name);
3164 
3165 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
3166 		/* If we're already trying to cancel a lock conversion
3167 		 * then just drop the spinlock and allow the caller to
3168 		 * requeue this lock. */
3169 
3170 		mlog(0, "Lockres %s, skip convert\n", lockres->l_name);
3171 		return 0;
3172 	}
3173 
3174 	/* were we in a convert when we got the bast fire? */
3175 	BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
3176 	       lockres->l_action != OCFS2_AST_DOWNCONVERT);
3177 	/* set things up for the unlockast to know to just
3178 	 * clear out the ast_action and unset busy, etc. */
3179 	lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
3180 
3181 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
3182 			"lock %s, invalid flags: 0x%lx\n",
3183 			lockres->l_name, lockres->l_flags);
3184 
3185 	return 1;
3186 }
3187 
3188 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
3189 				struct ocfs2_lock_res *lockres)
3190 {
3191 	int ret;
3192 
3193 	mlog_entry_void();
3194 	mlog(0, "lock %s\n", lockres->l_name);
3195 
3196 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
3197 			       DLM_LKF_CANCEL, lockres);
3198 	if (ret) {
3199 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3200 		ocfs2_recover_from_dlm_error(lockres, 0);
3201 	}
3202 
3203 	mlog(0, "lock %s return from ocfs2_dlm_unlock\n", lockres->l_name);
3204 
3205 	mlog_exit(ret);
3206 	return ret;
3207 }
3208 
3209 static int ocfs2_unblock_lock(struct ocfs2_super *osb,
3210 			      struct ocfs2_lock_res *lockres,
3211 			      struct ocfs2_unblock_ctl *ctl)
3212 {
3213 	unsigned long flags;
3214 	int blocking;
3215 	int new_level;
3216 	int ret = 0;
3217 	int set_lvb = 0;
3218 	unsigned int gen;
3219 
3220 	mlog_entry_void();
3221 
3222 	spin_lock_irqsave(&lockres->l_lock, flags);
3223 
3224 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
3225 
3226 recheck:
3227 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
3228 		/* XXX
3229 		 * This is a *big* race.  The OCFS2_LOCK_PENDING flag
3230 		 * exists entirely for one reason - another thread has set
3231 		 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
3232 		 *
3233 		 * If we do ocfs2_cancel_convert() before the other thread
3234 		 * calls dlm_lock(), our cancel will do nothing.  We will
3235 		 * get no ast, and we will have no way of knowing the
3236 		 * cancel failed.  Meanwhile, the other thread will call
3237 		 * into dlm_lock() and wait...forever.
3238 		 *
3239 		 * Why forever?  Because another node has asked for the
3240 		 * lock first; that's why we're here in unblock_lock().
3241 		 *
3242 		 * The solution is OCFS2_LOCK_PENDING.  When PENDING is
3243 		 * set, we just requeue the unblock.  Only when the other
3244 		 * thread has called dlm_lock() and cleared PENDING will
3245 		 * we then cancel their request.
3246 		 *
3247 		 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
3248 		 * at the same time they set OCFS2_DLM_BUSY.  They must
3249 		 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
3250 		 */
3251 		if (lockres->l_flags & OCFS2_LOCK_PENDING)
3252 			goto leave_requeue;
3253 
3254 		ctl->requeue = 1;
3255 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
3256 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3257 		if (ret) {
3258 			ret = ocfs2_cancel_convert(osb, lockres);
3259 			if (ret < 0)
3260 				mlog_errno(ret);
3261 		}
3262 		goto leave;
3263 	}
3264 
3265 	/* if we're blocking an exclusive and we have *any* holders,
3266 	 * then requeue. */
3267 	if ((lockres->l_blocking == DLM_LOCK_EX)
3268 	    && (lockres->l_ex_holders || lockres->l_ro_holders))
3269 		goto leave_requeue;
3270 
3271 	/* If it's a PR we're blocking, then only
3272 	 * requeue if we've got any EX holders */
3273 	if (lockres->l_blocking == DLM_LOCK_PR &&
3274 	    lockres->l_ex_holders)
3275 		goto leave_requeue;
3276 
3277 	/*
3278 	 * Can we get a lock in this state if the holder counts are
3279 	 * zero? The meta data unblock code used to check this.
3280 	 */
3281 	if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
3282 	    && (lockres->l_flags & OCFS2_LOCK_REFRESHING))
3283 		goto leave_requeue;
3284 
3285 	new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
3286 
3287 	if (lockres->l_ops->check_downconvert
3288 	    && !lockres->l_ops->check_downconvert(lockres, new_level))
3289 		goto leave_requeue;
3290 
3291 	/* If we get here, then we know that there are no more
3292 	 * incompatible holders (and anyone asking for an incompatible
3293 	 * lock is blocked). We can now downconvert the lock */
3294 	if (!lockres->l_ops->downconvert_worker)
3295 		goto downconvert;
3296 
3297 	/* Some lockres types want to do a bit of work before
3298 	 * downconverting a lock. Allow that here. The worker function
3299 	 * may sleep, so we save off a copy of what we're blocking as
3300 	 * it may change while we're not holding the spin lock. */
3301 	blocking = lockres->l_blocking;
3302 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3303 
3304 	ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
3305 
3306 	if (ctl->unblock_action == UNBLOCK_STOP_POST)
3307 		goto leave;
3308 
3309 	spin_lock_irqsave(&lockres->l_lock, flags);
3310 	if (blocking != lockres->l_blocking) {
3311 		/* If this changed underneath us, then we can't drop
3312 		 * it just yet. */
3313 		goto recheck;
3314 	}
3315 
3316 downconvert:
3317 	ctl->requeue = 0;
3318 
3319 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3320 		if (lockres->l_level == DLM_LOCK_EX)
3321 			set_lvb = 1;
3322 
3323 		/*
3324 		 * We only set the lvb if the lock has been fully
3325 		 * refreshed - otherwise we risk setting stale
3326 		 * data. Otherwise, there's no need to actually clear
3327 		 * out the lvb here as it's value is still valid.
3328 		 */
3329 		if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3330 			lockres->l_ops->set_lvb(lockres);
3331 	}
3332 
3333 	gen = ocfs2_prepare_downconvert(lockres, new_level);
3334 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3335 	ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
3336 				     gen);
3337 
3338 leave:
3339 	mlog_exit(ret);
3340 	return ret;
3341 
3342 leave_requeue:
3343 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3344 	ctl->requeue = 1;
3345 
3346 	mlog_exit(0);
3347 	return 0;
3348 }
3349 
3350 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
3351 				     int blocking)
3352 {
3353 	struct inode *inode;
3354 	struct address_space *mapping;
3355 
3356        	inode = ocfs2_lock_res_inode(lockres);
3357 	mapping = inode->i_mapping;
3358 
3359 	if (!S_ISREG(inode->i_mode))
3360 		goto out;
3361 
3362 	/*
3363 	 * We need this before the filemap_fdatawrite() so that it can
3364 	 * transfer the dirty bit from the PTE to the
3365 	 * page. Unfortunately this means that even for EX->PR
3366 	 * downconverts, we'll lose our mappings and have to build
3367 	 * them up again.
3368 	 */
3369 	unmap_mapping_range(mapping, 0, 0, 0);
3370 
3371 	if (filemap_fdatawrite(mapping)) {
3372 		mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
3373 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
3374 	}
3375 	sync_mapping_buffers(mapping);
3376 	if (blocking == DLM_LOCK_EX) {
3377 		truncate_inode_pages(mapping, 0);
3378 	} else {
3379 		/* We only need to wait on the I/O if we're not also
3380 		 * truncating pages because truncate_inode_pages waits
3381 		 * for us above. We don't truncate pages if we're
3382 		 * blocking anything < EXMODE because we want to keep
3383 		 * them around in that case. */
3384 		filemap_fdatawait(mapping);
3385 	}
3386 
3387 out:
3388 	return UNBLOCK_CONTINUE;
3389 }
3390 
3391 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
3392 					int new_level)
3393 {
3394 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3395 	int checkpointed = ocfs2_inode_fully_checkpointed(inode);
3396 
3397 	BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
3398 	BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
3399 
3400 	if (checkpointed)
3401 		return 1;
3402 
3403 	ocfs2_start_checkpoint(OCFS2_SB(inode->i_sb));
3404 	return 0;
3405 }
3406 
3407 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
3408 {
3409 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3410 
3411 	__ocfs2_stuff_meta_lvb(inode);
3412 }
3413 
3414 /*
3415  * Does the final reference drop on our dentry lock. Right now this
3416  * happens in the downconvert thread, but we could choose to simplify the
3417  * dlmglue API and push these off to the ocfs2_wq in the future.
3418  */
3419 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
3420 				     struct ocfs2_lock_res *lockres)
3421 {
3422 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3423 	ocfs2_dentry_lock_put(osb, dl);
3424 }
3425 
3426 /*
3427  * d_delete() matching dentries before the lock downconvert.
3428  *
3429  * At this point, any process waiting to destroy the
3430  * dentry_lock due to last ref count is stopped by the
3431  * OCFS2_LOCK_QUEUED flag.
3432  *
3433  * We have two potential problems
3434  *
3435  * 1) If we do the last reference drop on our dentry_lock (via dput)
3436  *    we'll wind up in ocfs2_release_dentry_lock(), waiting on
3437  *    the downconvert to finish. Instead we take an elevated
3438  *    reference and push the drop until after we've completed our
3439  *    unblock processing.
3440  *
3441  * 2) There might be another process with a final reference,
3442  *    waiting on us to finish processing. If this is the case, we
3443  *    detect it and exit out - there's no more dentries anyway.
3444  */
3445 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
3446 				       int blocking)
3447 {
3448 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3449 	struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
3450 	struct dentry *dentry;
3451 	unsigned long flags;
3452 	int extra_ref = 0;
3453 
3454 	/*
3455 	 * This node is blocking another node from getting a read
3456 	 * lock. This happens when we've renamed within a
3457 	 * directory. We've forced the other nodes to d_delete(), but
3458 	 * we never actually dropped our lock because it's still
3459 	 * valid. The downconvert code will retain a PR for this node,
3460 	 * so there's no further work to do.
3461 	 */
3462 	if (blocking == DLM_LOCK_PR)
3463 		return UNBLOCK_CONTINUE;
3464 
3465 	/*
3466 	 * Mark this inode as potentially orphaned. The code in
3467 	 * ocfs2_delete_inode() will figure out whether it actually
3468 	 * needs to be freed or not.
3469 	 */
3470 	spin_lock(&oi->ip_lock);
3471 	oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
3472 	spin_unlock(&oi->ip_lock);
3473 
3474 	/*
3475 	 * Yuck. We need to make sure however that the check of
3476 	 * OCFS2_LOCK_FREEING and the extra reference are atomic with
3477 	 * respect to a reference decrement or the setting of that
3478 	 * flag.
3479 	 */
3480 	spin_lock_irqsave(&lockres->l_lock, flags);
3481 	spin_lock(&dentry_attach_lock);
3482 	if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
3483 	    && dl->dl_count) {
3484 		dl->dl_count++;
3485 		extra_ref = 1;
3486 	}
3487 	spin_unlock(&dentry_attach_lock);
3488 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3489 
3490 	mlog(0, "extra_ref = %d\n", extra_ref);
3491 
3492 	/*
3493 	 * We have a process waiting on us in ocfs2_dentry_iput(),
3494 	 * which means we can't have any more outstanding
3495 	 * aliases. There's no need to do any more work.
3496 	 */
3497 	if (!extra_ref)
3498 		return UNBLOCK_CONTINUE;
3499 
3500 	spin_lock(&dentry_attach_lock);
3501 	while (1) {
3502 		dentry = ocfs2_find_local_alias(dl->dl_inode,
3503 						dl->dl_parent_blkno, 1);
3504 		if (!dentry)
3505 			break;
3506 		spin_unlock(&dentry_attach_lock);
3507 
3508 		mlog(0, "d_delete(%.*s);\n", dentry->d_name.len,
3509 		     dentry->d_name.name);
3510 
3511 		/*
3512 		 * The following dcache calls may do an
3513 		 * iput(). Normally we don't want that from the
3514 		 * downconverting thread, but in this case it's ok
3515 		 * because the requesting node already has an
3516 		 * exclusive lock on the inode, so it can't be queued
3517 		 * for a downconvert.
3518 		 */
3519 		d_delete(dentry);
3520 		dput(dentry);
3521 
3522 		spin_lock(&dentry_attach_lock);
3523 	}
3524 	spin_unlock(&dentry_attach_lock);
3525 
3526 	/*
3527 	 * If we are the last holder of this dentry lock, there is no
3528 	 * reason to downconvert so skip straight to the unlock.
3529 	 */
3530 	if (dl->dl_count == 1)
3531 		return UNBLOCK_STOP_POST;
3532 
3533 	return UNBLOCK_CONTINUE_POST;
3534 }
3535 
3536 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
3537 {
3538 	struct ocfs2_qinfo_lvb *lvb;
3539 	struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres);
3540 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
3541 					    oinfo->dqi_gi.dqi_type);
3542 
3543 	mlog_entry_void();
3544 
3545 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
3546 	lvb->lvb_version = OCFS2_QINFO_LVB_VERSION;
3547 	lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace);
3548 	lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace);
3549 	lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms);
3550 	lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks);
3551 	lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk);
3552 	lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry);
3553 
3554 	mlog_exit_void();
3555 }
3556 
3557 void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex)
3558 {
3559 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3560 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
3561 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3562 
3563 	mlog_entry_void();
3564 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
3565 		ocfs2_cluster_unlock(osb, lockres, level);
3566 	mlog_exit_void();
3567 }
3568 
3569 static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo)
3570 {
3571 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
3572 					    oinfo->dqi_gi.dqi_type);
3573 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3574 	struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
3575 	struct buffer_head *bh = NULL;
3576 	struct ocfs2_global_disk_dqinfo *gdinfo;
3577 	int status = 0;
3578 
3579 	if (lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) {
3580 		info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace);
3581 		info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace);
3582 		oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms);
3583 		oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks);
3584 		oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk);
3585 		oinfo->dqi_gi.dqi_free_entry =
3586 					be32_to_cpu(lvb->lvb_free_entry);
3587 	} else {
3588 		status = ocfs2_read_quota_block(oinfo->dqi_gqinode, 0, &bh);
3589 		if (status) {
3590 			mlog_errno(status);
3591 			goto bail;
3592 		}
3593 		gdinfo = (struct ocfs2_global_disk_dqinfo *)
3594 					(bh->b_data + OCFS2_GLOBAL_INFO_OFF);
3595 		info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace);
3596 		info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace);
3597 		oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms);
3598 		oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks);
3599 		oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk);
3600 		oinfo->dqi_gi.dqi_free_entry =
3601 					le32_to_cpu(gdinfo->dqi_free_entry);
3602 		brelse(bh);
3603 		ocfs2_track_lock_refresh(lockres);
3604 	}
3605 
3606 bail:
3607 	return status;
3608 }
3609 
3610 /* Lock quota info, this function expects at least shared lock on the quota file
3611  * so that we can safely refresh quota info from disk. */
3612 int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex)
3613 {
3614 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3615 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
3616 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3617 	int status = 0;
3618 
3619 	mlog_entry_void();
3620 
3621 	/* On RO devices, locking really isn't needed... */
3622 	if (ocfs2_is_hard_readonly(osb)) {
3623 		if (ex)
3624 			status = -EROFS;
3625 		goto bail;
3626 	}
3627 	if (ocfs2_mount_local(osb))
3628 		goto bail;
3629 
3630 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
3631 	if (status < 0) {
3632 		mlog_errno(status);
3633 		goto bail;
3634 	}
3635 	if (!ocfs2_should_refresh_lock_res(lockres))
3636 		goto bail;
3637 	/* OK, we have the lock but we need to refresh the quota info */
3638 	status = ocfs2_refresh_qinfo(oinfo);
3639 	if (status)
3640 		ocfs2_qinfo_unlock(oinfo, ex);
3641 	ocfs2_complete_lock_res_refresh(lockres, status);
3642 bail:
3643 	mlog_exit(status);
3644 	return status;
3645 }
3646 
3647 /*
3648  * This is the filesystem locking protocol.  It provides the lock handling
3649  * hooks for the underlying DLM.  It has a maximum version number.
3650  * The version number allows interoperability with systems running at
3651  * the same major number and an equal or smaller minor number.
3652  *
3653  * Whenever the filesystem does new things with locks (adds or removes a
3654  * lock, orders them differently, does different things underneath a lock),
3655  * the version must be changed.  The protocol is negotiated when joining
3656  * the dlm domain.  A node may join the domain if its major version is
3657  * identical to all other nodes and its minor version is greater than
3658  * or equal to all other nodes.  When its minor version is greater than
3659  * the other nodes, it will run at the minor version specified by the
3660  * other nodes.
3661  *
3662  * If a locking change is made that will not be compatible with older
3663  * versions, the major number must be increased and the minor version set
3664  * to zero.  If a change merely adds a behavior that can be disabled when
3665  * speaking to older versions, the minor version must be increased.  If a
3666  * change adds a fully backwards compatible change (eg, LVB changes that
3667  * are just ignored by older versions), the version does not need to be
3668  * updated.
3669  */
3670 static struct ocfs2_locking_protocol lproto = {
3671 	.lp_max_version = {
3672 		.pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
3673 		.pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
3674 	},
3675 	.lp_lock_ast		= ocfs2_locking_ast,
3676 	.lp_blocking_ast	= ocfs2_blocking_ast,
3677 	.lp_unlock_ast		= ocfs2_unlock_ast,
3678 };
3679 
3680 void ocfs2_set_locking_protocol(void)
3681 {
3682 	ocfs2_stack_glue_set_locking_protocol(&lproto);
3683 }
3684 
3685 
3686 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3687 				       struct ocfs2_lock_res *lockres)
3688 {
3689 	int status;
3690 	struct ocfs2_unblock_ctl ctl = {0, 0,};
3691 	unsigned long flags;
3692 
3693 	/* Our reference to the lockres in this function can be
3694 	 * considered valid until we remove the OCFS2_LOCK_QUEUED
3695 	 * flag. */
3696 
3697 	mlog_entry_void();
3698 
3699 	BUG_ON(!lockres);
3700 	BUG_ON(!lockres->l_ops);
3701 
3702 	mlog(0, "lockres %s blocked.\n", lockres->l_name);
3703 
3704 	/* Detect whether a lock has been marked as going away while
3705 	 * the downconvert thread was processing other things. A lock can
3706 	 * still be marked with OCFS2_LOCK_FREEING after this check,
3707 	 * but short circuiting here will still save us some
3708 	 * performance. */
3709 	spin_lock_irqsave(&lockres->l_lock, flags);
3710 	if (lockres->l_flags & OCFS2_LOCK_FREEING)
3711 		goto unqueue;
3712 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3713 
3714 	status = ocfs2_unblock_lock(osb, lockres, &ctl);
3715 	if (status < 0)
3716 		mlog_errno(status);
3717 
3718 	spin_lock_irqsave(&lockres->l_lock, flags);
3719 unqueue:
3720 	if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
3721 		lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
3722 	} else
3723 		ocfs2_schedule_blocked_lock(osb, lockres);
3724 
3725 	mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name,
3726 	     ctl.requeue ? "yes" : "no");
3727 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3728 
3729 	if (ctl.unblock_action != UNBLOCK_CONTINUE
3730 	    && lockres->l_ops->post_unlock)
3731 		lockres->l_ops->post_unlock(osb, lockres);
3732 
3733 	mlog_exit_void();
3734 }
3735 
3736 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
3737 					struct ocfs2_lock_res *lockres)
3738 {
3739 	mlog_entry_void();
3740 
3741 	assert_spin_locked(&lockres->l_lock);
3742 
3743 	if (lockres->l_flags & OCFS2_LOCK_FREEING) {
3744 		/* Do not schedule a lock for downconvert when it's on
3745 		 * the way to destruction - any nodes wanting access
3746 		 * to the resource will get it soon. */
3747 		mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n",
3748 		     lockres->l_name, lockres->l_flags);
3749 		return;
3750 	}
3751 
3752 	lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
3753 
3754 	spin_lock(&osb->dc_task_lock);
3755 	if (list_empty(&lockres->l_blocked_list)) {
3756 		list_add_tail(&lockres->l_blocked_list,
3757 			      &osb->blocked_lock_list);
3758 		osb->blocked_lock_count++;
3759 	}
3760 	spin_unlock(&osb->dc_task_lock);
3761 
3762 	mlog_exit_void();
3763 }
3764 
3765 static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
3766 {
3767 	unsigned long processed;
3768 	struct ocfs2_lock_res *lockres;
3769 
3770 	mlog_entry_void();
3771 
3772 	spin_lock(&osb->dc_task_lock);
3773 	/* grab this early so we know to try again if a state change and
3774 	 * wake happens part-way through our work  */
3775 	osb->dc_work_sequence = osb->dc_wake_sequence;
3776 
3777 	processed = osb->blocked_lock_count;
3778 	while (processed) {
3779 		BUG_ON(list_empty(&osb->blocked_lock_list));
3780 
3781 		lockres = list_entry(osb->blocked_lock_list.next,
3782 				     struct ocfs2_lock_res, l_blocked_list);
3783 		list_del_init(&lockres->l_blocked_list);
3784 		osb->blocked_lock_count--;
3785 		spin_unlock(&osb->dc_task_lock);
3786 
3787 		BUG_ON(!processed);
3788 		processed--;
3789 
3790 		ocfs2_process_blocked_lock(osb, lockres);
3791 
3792 		spin_lock(&osb->dc_task_lock);
3793 	}
3794 	spin_unlock(&osb->dc_task_lock);
3795 
3796 	mlog_exit_void();
3797 }
3798 
3799 static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
3800 {
3801 	int empty = 0;
3802 
3803 	spin_lock(&osb->dc_task_lock);
3804 	if (list_empty(&osb->blocked_lock_list))
3805 		empty = 1;
3806 
3807 	spin_unlock(&osb->dc_task_lock);
3808 	return empty;
3809 }
3810 
3811 static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
3812 {
3813 	int should_wake = 0;
3814 
3815 	spin_lock(&osb->dc_task_lock);
3816 	if (osb->dc_work_sequence != osb->dc_wake_sequence)
3817 		should_wake = 1;
3818 	spin_unlock(&osb->dc_task_lock);
3819 
3820 	return should_wake;
3821 }
3822 
3823 static int ocfs2_downconvert_thread(void *arg)
3824 {
3825 	int status = 0;
3826 	struct ocfs2_super *osb = arg;
3827 
3828 	/* only quit once we've been asked to stop and there is no more
3829 	 * work available */
3830 	while (!(kthread_should_stop() &&
3831 		ocfs2_downconvert_thread_lists_empty(osb))) {
3832 
3833 		wait_event_interruptible(osb->dc_event,
3834 					 ocfs2_downconvert_thread_should_wake(osb) ||
3835 					 kthread_should_stop());
3836 
3837 		mlog(0, "downconvert_thread: awoken\n");
3838 
3839 		ocfs2_downconvert_thread_do_work(osb);
3840 	}
3841 
3842 	osb->dc_task = NULL;
3843 	return status;
3844 }
3845 
3846 void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
3847 {
3848 	spin_lock(&osb->dc_task_lock);
3849 	/* make sure the voting thread gets a swipe at whatever changes
3850 	 * the caller may have made to the voting state */
3851 	osb->dc_wake_sequence++;
3852 	spin_unlock(&osb->dc_task_lock);
3853 	wake_up(&osb->dc_event);
3854 }
3855