xref: /openbmc/linux/fs/ocfs2/dlmglue.c (revision f220d3eb)
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 #include <linux/sched/signal.h>
37 
38 #define MLOG_MASK_PREFIX ML_DLM_GLUE
39 #include <cluster/masklog.h>
40 
41 #include "ocfs2.h"
42 #include "ocfs2_lockingver.h"
43 
44 #include "alloc.h"
45 #include "dcache.h"
46 #include "dlmglue.h"
47 #include "extent_map.h"
48 #include "file.h"
49 #include "heartbeat.h"
50 #include "inode.h"
51 #include "journal.h"
52 #include "stackglue.h"
53 #include "slot_map.h"
54 #include "super.h"
55 #include "uptodate.h"
56 #include "quota.h"
57 #include "refcounttree.h"
58 #include "acl.h"
59 
60 #include "buffer_head_io.h"
61 
62 struct ocfs2_mask_waiter {
63 	struct list_head	mw_item;
64 	int			mw_status;
65 	struct completion	mw_complete;
66 	unsigned long		mw_mask;
67 	unsigned long		mw_goal;
68 #ifdef CONFIG_OCFS2_FS_STATS
69 	ktime_t			mw_lock_start;
70 #endif
71 };
72 
73 static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres);
74 static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres);
75 static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres);
76 static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres);
77 
78 /*
79  * Return value from ->downconvert_worker functions.
80  *
81  * These control the precise actions of ocfs2_unblock_lock()
82  * and ocfs2_process_blocked_lock()
83  *
84  */
85 enum ocfs2_unblock_action {
86 	UNBLOCK_CONTINUE	= 0, /* Continue downconvert */
87 	UNBLOCK_CONTINUE_POST	= 1, /* Continue downconvert, fire
88 				      * ->post_unlock callback */
89 	UNBLOCK_STOP_POST	= 2, /* Do not downconvert, fire
90 				      * ->post_unlock() callback. */
91 };
92 
93 struct ocfs2_unblock_ctl {
94 	int requeue;
95 	enum ocfs2_unblock_action unblock_action;
96 };
97 
98 /* Lockdep class keys */
99 static struct lock_class_key lockdep_keys[OCFS2_NUM_LOCK_TYPES];
100 
101 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
102 					int new_level);
103 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);
104 
105 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
106 				     int blocking);
107 
108 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
109 				       int blocking);
110 
111 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
112 				     struct ocfs2_lock_res *lockres);
113 
114 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres);
115 
116 static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
117 					    int new_level);
118 static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
119 					 int blocking);
120 
121 #define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
122 
123 /* This aids in debugging situations where a bad LVB might be involved. */
124 static void ocfs2_dump_meta_lvb_info(u64 level,
125 				     const char *function,
126 				     unsigned int line,
127 				     struct ocfs2_lock_res *lockres)
128 {
129 	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
130 
131 	mlog(level, "LVB information for %s (called from %s:%u):\n",
132 	     lockres->l_name, function, line);
133 	mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
134 	     lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
135 	     be32_to_cpu(lvb->lvb_igeneration));
136 	mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
137 	     (unsigned long long)be64_to_cpu(lvb->lvb_isize),
138 	     be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
139 	     be16_to_cpu(lvb->lvb_imode));
140 	mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
141 	     "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
142 	     (long long)be64_to_cpu(lvb->lvb_iatime_packed),
143 	     (long long)be64_to_cpu(lvb->lvb_ictime_packed),
144 	     (long long)be64_to_cpu(lvb->lvb_imtime_packed),
145 	     be32_to_cpu(lvb->lvb_iattr));
146 }
147 
148 
149 /*
150  * OCFS2 Lock Resource Operations
151  *
152  * These fine tune the behavior of the generic dlmglue locking infrastructure.
153  *
154  * The most basic of lock types can point ->l_priv to their respective
155  * struct ocfs2_super and allow the default actions to manage things.
156  *
157  * Right now, each lock type also needs to implement an init function,
158  * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
159  * should be called when the lock is no longer needed (i.e., object
160  * destruction time).
161  */
162 struct ocfs2_lock_res_ops {
163 	/*
164 	 * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
165 	 * this callback if ->l_priv is not an ocfs2_super pointer
166 	 */
167 	struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
168 
169 	/*
170 	 * Optionally called in the downconvert thread after a
171 	 * successful downconvert. The lockres will not be referenced
172 	 * after this callback is called, so it is safe to free
173 	 * memory, etc.
174 	 *
175 	 * The exact semantics of when this is called are controlled
176 	 * by ->downconvert_worker()
177 	 */
178 	void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
179 
180 	/*
181 	 * Allow a lock type to add checks to determine whether it is
182 	 * safe to downconvert a lock. Return 0 to re-queue the
183 	 * downconvert at a later time, nonzero to continue.
184 	 *
185 	 * For most locks, the default checks that there are no
186 	 * incompatible holders are sufficient.
187 	 *
188 	 * Called with the lockres spinlock held.
189 	 */
190 	int (*check_downconvert)(struct ocfs2_lock_res *, int);
191 
192 	/*
193 	 * Allows a lock type to populate the lock value block. This
194 	 * is called on downconvert, and when we drop a lock.
195 	 *
196 	 * Locks that want to use this should set LOCK_TYPE_USES_LVB
197 	 * in the flags field.
198 	 *
199 	 * Called with the lockres spinlock held.
200 	 */
201 	void (*set_lvb)(struct ocfs2_lock_res *);
202 
203 	/*
204 	 * Called from the downconvert thread when it is determined
205 	 * that a lock will be downconverted. This is called without
206 	 * any locks held so the function can do work that might
207 	 * schedule (syncing out data, etc).
208 	 *
209 	 * This should return any one of the ocfs2_unblock_action
210 	 * values, depending on what it wants the thread to do.
211 	 */
212 	int (*downconvert_worker)(struct ocfs2_lock_res *, int);
213 
214 	/*
215 	 * LOCK_TYPE_* flags which describe the specific requirements
216 	 * of a lock type. Descriptions of each individual flag follow.
217 	 */
218 	int flags;
219 };
220 
221 /*
222  * Some locks want to "refresh" potentially stale data when a
223  * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
224  * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
225  * individual lockres l_flags member from the ast function. It is
226  * expected that the locking wrapper will clear the
227  * OCFS2_LOCK_NEEDS_REFRESH flag when done.
228  */
229 #define LOCK_TYPE_REQUIRES_REFRESH 0x1
230 
231 /*
232  * Indicate that a lock type makes use of the lock value block. The
233  * ->set_lvb lock type callback must be defined.
234  */
235 #define LOCK_TYPE_USES_LVB		0x2
236 
237 static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
238 	.get_osb	= ocfs2_get_inode_osb,
239 	.flags		= 0,
240 };
241 
242 static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = {
243 	.get_osb	= ocfs2_get_inode_osb,
244 	.check_downconvert = ocfs2_check_meta_downconvert,
245 	.set_lvb	= ocfs2_set_meta_lvb,
246 	.downconvert_worker = ocfs2_data_convert_worker,
247 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
248 };
249 
250 static struct ocfs2_lock_res_ops ocfs2_super_lops = {
251 	.flags		= LOCK_TYPE_REQUIRES_REFRESH,
252 };
253 
254 static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
255 	.flags		= 0,
256 };
257 
258 static struct ocfs2_lock_res_ops ocfs2_nfs_sync_lops = {
259 	.flags		= 0,
260 };
261 
262 static struct ocfs2_lock_res_ops ocfs2_trim_fs_lops = {
263 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
264 };
265 
266 static struct ocfs2_lock_res_ops ocfs2_orphan_scan_lops = {
267 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
268 };
269 
270 static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
271 	.get_osb	= ocfs2_get_dentry_osb,
272 	.post_unlock	= ocfs2_dentry_post_unlock,
273 	.downconvert_worker = ocfs2_dentry_convert_worker,
274 	.flags		= 0,
275 };
276 
277 static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
278 	.get_osb	= ocfs2_get_inode_osb,
279 	.flags		= 0,
280 };
281 
282 static struct ocfs2_lock_res_ops ocfs2_flock_lops = {
283 	.get_osb	= ocfs2_get_file_osb,
284 	.flags		= 0,
285 };
286 
287 static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = {
288 	.set_lvb	= ocfs2_set_qinfo_lvb,
289 	.get_osb	= ocfs2_get_qinfo_osb,
290 	.flags		= LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB,
291 };
292 
293 static struct ocfs2_lock_res_ops ocfs2_refcount_block_lops = {
294 	.check_downconvert = ocfs2_check_refcount_downconvert,
295 	.downconvert_worker = ocfs2_refcount_convert_worker,
296 	.flags		= 0,
297 };
298 
299 static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
300 {
301 	return lockres->l_type == OCFS2_LOCK_TYPE_META ||
302 		lockres->l_type == OCFS2_LOCK_TYPE_RW ||
303 		lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
304 }
305 
306 static inline struct ocfs2_lock_res *ocfs2_lksb_to_lock_res(struct ocfs2_dlm_lksb *lksb)
307 {
308 	return container_of(lksb, struct ocfs2_lock_res, l_lksb);
309 }
310 
311 static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
312 {
313 	BUG_ON(!ocfs2_is_inode_lock(lockres));
314 
315 	return (struct inode *) lockres->l_priv;
316 }
317 
318 static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
319 {
320 	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
321 
322 	return (struct ocfs2_dentry_lock *)lockres->l_priv;
323 }
324 
325 static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres)
326 {
327 	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO);
328 
329 	return (struct ocfs2_mem_dqinfo *)lockres->l_priv;
330 }
331 
332 static inline struct ocfs2_refcount_tree *
333 ocfs2_lock_res_refcount_tree(struct ocfs2_lock_res *res)
334 {
335 	return container_of(res, struct ocfs2_refcount_tree, rf_lockres);
336 }
337 
338 static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
339 {
340 	if (lockres->l_ops->get_osb)
341 		return lockres->l_ops->get_osb(lockres);
342 
343 	return (struct ocfs2_super *)lockres->l_priv;
344 }
345 
346 static int ocfs2_lock_create(struct ocfs2_super *osb,
347 			     struct ocfs2_lock_res *lockres,
348 			     int level,
349 			     u32 dlm_flags);
350 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
351 						     int wanted);
352 static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
353 				   struct ocfs2_lock_res *lockres,
354 				   int level, unsigned long caller_ip);
355 static inline void ocfs2_cluster_unlock(struct ocfs2_super *osb,
356 					struct ocfs2_lock_res *lockres,
357 					int level)
358 {
359 	__ocfs2_cluster_unlock(osb, lockres, level, _RET_IP_);
360 }
361 
362 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
363 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
364 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
365 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
366 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
367 					struct ocfs2_lock_res *lockres);
368 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
369 						int convert);
370 #define ocfs2_log_dlm_error(_func, _err, _lockres) do {					\
371 	if ((_lockres)->l_type != OCFS2_LOCK_TYPE_DENTRY)				\
372 		mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n",	\
373 		     _err, _func, _lockres->l_name);					\
374 	else										\
375 		mlog(ML_ERROR, "DLM error %d while calling %s on resource %.*s%08x\n",	\
376 		     _err, _func, OCFS2_DENTRY_LOCK_INO_START - 1, (_lockres)->l_name,	\
377 		     (unsigned int)ocfs2_get_dentry_lock_ino(_lockres));		\
378 } while (0)
379 static int ocfs2_downconvert_thread(void *arg);
380 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
381 					struct ocfs2_lock_res *lockres);
382 static int ocfs2_inode_lock_update(struct inode *inode,
383 				  struct buffer_head **bh);
384 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
385 static inline int ocfs2_highest_compat_lock_level(int level);
386 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
387 					      int new_level);
388 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
389 				  struct ocfs2_lock_res *lockres,
390 				  int new_level,
391 				  int lvb,
392 				  unsigned int generation);
393 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
394 				        struct ocfs2_lock_res *lockres);
395 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
396 				struct ocfs2_lock_res *lockres);
397 
398 
399 static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
400 				  u64 blkno,
401 				  u32 generation,
402 				  char *name)
403 {
404 	int len;
405 
406 	BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
407 
408 	len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
409 		       ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
410 		       (long long)blkno, generation);
411 
412 	BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
413 
414 	mlog(0, "built lock resource with name: %s\n", name);
415 }
416 
417 static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
418 
419 static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
420 				       struct ocfs2_dlm_debug *dlm_debug)
421 {
422 	mlog(0, "Add tracking for lockres %s\n", res->l_name);
423 
424 	spin_lock(&ocfs2_dlm_tracking_lock);
425 	list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
426 	spin_unlock(&ocfs2_dlm_tracking_lock);
427 }
428 
429 static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
430 {
431 	spin_lock(&ocfs2_dlm_tracking_lock);
432 	if (!list_empty(&res->l_debug_list))
433 		list_del_init(&res->l_debug_list);
434 	spin_unlock(&ocfs2_dlm_tracking_lock);
435 }
436 
437 #ifdef CONFIG_OCFS2_FS_STATS
438 static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
439 {
440 	res->l_lock_refresh = 0;
441 	memset(&res->l_lock_prmode, 0, sizeof(struct ocfs2_lock_stats));
442 	memset(&res->l_lock_exmode, 0, sizeof(struct ocfs2_lock_stats));
443 }
444 
445 static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level,
446 				    struct ocfs2_mask_waiter *mw, int ret)
447 {
448 	u32 usec;
449 	ktime_t kt;
450 	struct ocfs2_lock_stats *stats;
451 
452 	if (level == LKM_PRMODE)
453 		stats = &res->l_lock_prmode;
454 	else if (level == LKM_EXMODE)
455 		stats = &res->l_lock_exmode;
456 	else
457 		return;
458 
459 	kt = ktime_sub(ktime_get(), mw->mw_lock_start);
460 	usec = ktime_to_us(kt);
461 
462 	stats->ls_gets++;
463 	stats->ls_total += ktime_to_ns(kt);
464 	/* overflow */
465 	if (unlikely(stats->ls_gets == 0)) {
466 		stats->ls_gets++;
467 		stats->ls_total = ktime_to_ns(kt);
468 	}
469 
470 	if (stats->ls_max < usec)
471 		stats->ls_max = usec;
472 
473 	if (ret)
474 		stats->ls_fail++;
475 }
476 
477 static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
478 {
479 	lockres->l_lock_refresh++;
480 }
481 
482 static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
483 {
484 	mw->mw_lock_start = ktime_get();
485 }
486 #else
487 static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
488 {
489 }
490 static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res,
491 			   int level, struct ocfs2_mask_waiter *mw, int ret)
492 {
493 }
494 static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
495 {
496 }
497 static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
498 {
499 }
500 #endif
501 
502 static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
503 				       struct ocfs2_lock_res *res,
504 				       enum ocfs2_lock_type type,
505 				       struct ocfs2_lock_res_ops *ops,
506 				       void *priv)
507 {
508 	res->l_type          = type;
509 	res->l_ops           = ops;
510 	res->l_priv          = priv;
511 
512 	res->l_level         = DLM_LOCK_IV;
513 	res->l_requested     = DLM_LOCK_IV;
514 	res->l_blocking      = DLM_LOCK_IV;
515 	res->l_action        = OCFS2_AST_INVALID;
516 	res->l_unlock_action = OCFS2_UNLOCK_INVALID;
517 
518 	res->l_flags         = OCFS2_LOCK_INITIALIZED;
519 
520 	ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
521 
522 	ocfs2_init_lock_stats(res);
523 #ifdef CONFIG_DEBUG_LOCK_ALLOC
524 	if (type != OCFS2_LOCK_TYPE_OPEN)
525 		lockdep_init_map(&res->l_lockdep_map, ocfs2_lock_type_strings[type],
526 				 &lockdep_keys[type], 0);
527 	else
528 		res->l_lockdep_map.key = NULL;
529 #endif
530 }
531 
532 void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
533 {
534 	/* This also clears out the lock status block */
535 	memset(res, 0, sizeof(struct ocfs2_lock_res));
536 	spin_lock_init(&res->l_lock);
537 	init_waitqueue_head(&res->l_event);
538 	INIT_LIST_HEAD(&res->l_blocked_list);
539 	INIT_LIST_HEAD(&res->l_mask_waiters);
540 	INIT_LIST_HEAD(&res->l_holders);
541 }
542 
543 void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
544 			       enum ocfs2_lock_type type,
545 			       unsigned int generation,
546 			       struct inode *inode)
547 {
548 	struct ocfs2_lock_res_ops *ops;
549 
550 	switch(type) {
551 		case OCFS2_LOCK_TYPE_RW:
552 			ops = &ocfs2_inode_rw_lops;
553 			break;
554 		case OCFS2_LOCK_TYPE_META:
555 			ops = &ocfs2_inode_inode_lops;
556 			break;
557 		case OCFS2_LOCK_TYPE_OPEN:
558 			ops = &ocfs2_inode_open_lops;
559 			break;
560 		default:
561 			mlog_bug_on_msg(1, "type: %d\n", type);
562 			ops = NULL; /* thanks, gcc */
563 			break;
564 	};
565 
566 	ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
567 			      generation, res->l_name);
568 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
569 }
570 
571 static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
572 {
573 	struct inode *inode = ocfs2_lock_res_inode(lockres);
574 
575 	return OCFS2_SB(inode->i_sb);
576 }
577 
578 static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres)
579 {
580 	struct ocfs2_mem_dqinfo *info = lockres->l_priv;
581 
582 	return OCFS2_SB(info->dqi_gi.dqi_sb);
583 }
584 
585 static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres)
586 {
587 	struct ocfs2_file_private *fp = lockres->l_priv;
588 
589 	return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb);
590 }
591 
592 static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
593 {
594 	__be64 inode_blkno_be;
595 
596 	memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
597 	       sizeof(__be64));
598 
599 	return be64_to_cpu(inode_blkno_be);
600 }
601 
602 static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
603 {
604 	struct ocfs2_dentry_lock *dl = lockres->l_priv;
605 
606 	return OCFS2_SB(dl->dl_inode->i_sb);
607 }
608 
609 void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
610 				u64 parent, struct inode *inode)
611 {
612 	int len;
613 	u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
614 	__be64 inode_blkno_be = cpu_to_be64(inode_blkno);
615 	struct ocfs2_lock_res *lockres = &dl->dl_lockres;
616 
617 	ocfs2_lock_res_init_once(lockres);
618 
619 	/*
620 	 * Unfortunately, the standard lock naming scheme won't work
621 	 * here because we have two 16 byte values to use. Instead,
622 	 * we'll stuff the inode number as a binary value. We still
623 	 * want error prints to show something without garbling the
624 	 * display, so drop a null byte in there before the inode
625 	 * number. A future version of OCFS2 will likely use all
626 	 * binary lock names. The stringified names have been a
627 	 * tremendous aid in debugging, but now that the debugfs
628 	 * interface exists, we can mangle things there if need be.
629 	 *
630 	 * NOTE: We also drop the standard "pad" value (the total lock
631 	 * name size stays the same though - the last part is all
632 	 * zeros due to the memset in ocfs2_lock_res_init_once()
633 	 */
634 	len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
635 		       "%c%016llx",
636 		       ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
637 		       (long long)parent);
638 
639 	BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
640 
641 	memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
642 	       sizeof(__be64));
643 
644 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
645 				   OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
646 				   dl);
647 }
648 
649 static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
650 				      struct ocfs2_super *osb)
651 {
652 	/* Superblock lockres doesn't come from a slab so we call init
653 	 * once on it manually.  */
654 	ocfs2_lock_res_init_once(res);
655 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
656 			      0, res->l_name);
657 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
658 				   &ocfs2_super_lops, osb);
659 }
660 
661 static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
662 				       struct ocfs2_super *osb)
663 {
664 	/* Rename lockres doesn't come from a slab so we call init
665 	 * once on it manually.  */
666 	ocfs2_lock_res_init_once(res);
667 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
668 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
669 				   &ocfs2_rename_lops, osb);
670 }
671 
672 static void ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res *res,
673 					 struct ocfs2_super *osb)
674 {
675 	/* nfs_sync lockres doesn't come from a slab so we call init
676 	 * once on it manually.  */
677 	ocfs2_lock_res_init_once(res);
678 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_NFS_SYNC, 0, 0, res->l_name);
679 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_NFS_SYNC,
680 				   &ocfs2_nfs_sync_lops, osb);
681 }
682 
683 void ocfs2_trim_fs_lock_res_init(struct ocfs2_super *osb)
684 {
685 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
686 
687 	ocfs2_lock_res_init_once(lockres);
688 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_TRIM_FS, 0, 0, lockres->l_name);
689 	ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_TRIM_FS,
690 				   &ocfs2_trim_fs_lops, osb);
691 }
692 
693 void ocfs2_trim_fs_lock_res_uninit(struct ocfs2_super *osb)
694 {
695 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
696 
697 	ocfs2_simple_drop_lockres(osb, lockres);
698 	ocfs2_lock_res_free(lockres);
699 }
700 
701 static void ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res *res,
702 					    struct ocfs2_super *osb)
703 {
704 	ocfs2_lock_res_init_once(res);
705 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_ORPHAN_SCAN, 0, 0, res->l_name);
706 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_ORPHAN_SCAN,
707 				   &ocfs2_orphan_scan_lops, osb);
708 }
709 
710 void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
711 			      struct ocfs2_file_private *fp)
712 {
713 	struct inode *inode = fp->fp_file->f_mapping->host;
714 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
715 
716 	ocfs2_lock_res_init_once(lockres);
717 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
718 			      inode->i_generation, lockres->l_name);
719 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
720 				   OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
721 				   fp);
722 	lockres->l_flags |= OCFS2_LOCK_NOCACHE;
723 }
724 
725 void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
726 			       struct ocfs2_mem_dqinfo *info)
727 {
728 	ocfs2_lock_res_init_once(lockres);
729 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type,
730 			      0, lockres->l_name);
731 	ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres,
732 				   OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops,
733 				   info);
734 }
735 
736 void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres,
737 				  struct ocfs2_super *osb, u64 ref_blkno,
738 				  unsigned int generation)
739 {
740 	ocfs2_lock_res_init_once(lockres);
741 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_REFCOUNT, ref_blkno,
742 			      generation, lockres->l_name);
743 	ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_REFCOUNT,
744 				   &ocfs2_refcount_block_lops, osb);
745 }
746 
747 void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
748 {
749 	if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
750 		return;
751 
752 	ocfs2_remove_lockres_tracking(res);
753 
754 	mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
755 			"Lockres %s is on the blocked list\n",
756 			res->l_name);
757 	mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
758 			"Lockres %s has mask waiters pending\n",
759 			res->l_name);
760 	mlog_bug_on_msg(spin_is_locked(&res->l_lock),
761 			"Lockres %s is locked\n",
762 			res->l_name);
763 	mlog_bug_on_msg(res->l_ro_holders,
764 			"Lockres %s has %u ro holders\n",
765 			res->l_name, res->l_ro_holders);
766 	mlog_bug_on_msg(res->l_ex_holders,
767 			"Lockres %s has %u ex holders\n",
768 			res->l_name, res->l_ex_holders);
769 
770 	/* Need to clear out the lock status block for the dlm */
771 	memset(&res->l_lksb, 0, sizeof(res->l_lksb));
772 
773 	res->l_flags = 0UL;
774 }
775 
776 /*
777  * Keep a list of processes who have interest in a lockres.
778  * Note: this is now only uesed for check recursive cluster locking.
779  */
780 static inline void ocfs2_add_holder(struct ocfs2_lock_res *lockres,
781 				   struct ocfs2_lock_holder *oh)
782 {
783 	INIT_LIST_HEAD(&oh->oh_list);
784 	oh->oh_owner_pid = get_pid(task_pid(current));
785 
786 	spin_lock(&lockres->l_lock);
787 	list_add_tail(&oh->oh_list, &lockres->l_holders);
788 	spin_unlock(&lockres->l_lock);
789 }
790 
791 static struct ocfs2_lock_holder *
792 ocfs2_pid_holder(struct ocfs2_lock_res *lockres,
793 		struct pid *pid)
794 {
795 	struct ocfs2_lock_holder *oh;
796 
797 	spin_lock(&lockres->l_lock);
798 	list_for_each_entry(oh, &lockres->l_holders, oh_list) {
799 		if (oh->oh_owner_pid == pid) {
800 			spin_unlock(&lockres->l_lock);
801 			return oh;
802 		}
803 	}
804 	spin_unlock(&lockres->l_lock);
805 	return NULL;
806 }
807 
808 static inline void ocfs2_remove_holder(struct ocfs2_lock_res *lockres,
809 				       struct ocfs2_lock_holder *oh)
810 {
811 	spin_lock(&lockres->l_lock);
812 	list_del(&oh->oh_list);
813 	spin_unlock(&lockres->l_lock);
814 
815 	put_pid(oh->oh_owner_pid);
816 }
817 
818 
819 static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
820 				     int level)
821 {
822 	BUG_ON(!lockres);
823 
824 	switch(level) {
825 	case DLM_LOCK_EX:
826 		lockres->l_ex_holders++;
827 		break;
828 	case DLM_LOCK_PR:
829 		lockres->l_ro_holders++;
830 		break;
831 	default:
832 		BUG();
833 	}
834 }
835 
836 static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
837 				     int level)
838 {
839 	BUG_ON(!lockres);
840 
841 	switch(level) {
842 	case DLM_LOCK_EX:
843 		BUG_ON(!lockres->l_ex_holders);
844 		lockres->l_ex_holders--;
845 		break;
846 	case DLM_LOCK_PR:
847 		BUG_ON(!lockres->l_ro_holders);
848 		lockres->l_ro_holders--;
849 		break;
850 	default:
851 		BUG();
852 	}
853 }
854 
855 /* WARNING: This function lives in a world where the only three lock
856  * levels are EX, PR, and NL. It *will* have to be adjusted when more
857  * lock types are added. */
858 static inline int ocfs2_highest_compat_lock_level(int level)
859 {
860 	int new_level = DLM_LOCK_EX;
861 
862 	if (level == DLM_LOCK_EX)
863 		new_level = DLM_LOCK_NL;
864 	else if (level == DLM_LOCK_PR)
865 		new_level = DLM_LOCK_PR;
866 	return new_level;
867 }
868 
869 static void lockres_set_flags(struct ocfs2_lock_res *lockres,
870 			      unsigned long newflags)
871 {
872 	struct ocfs2_mask_waiter *mw, *tmp;
873 
874  	assert_spin_locked(&lockres->l_lock);
875 
876 	lockres->l_flags = newflags;
877 
878 	list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
879 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
880 			continue;
881 
882 		list_del_init(&mw->mw_item);
883 		mw->mw_status = 0;
884 		complete(&mw->mw_complete);
885 	}
886 }
887 static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
888 {
889 	lockres_set_flags(lockres, lockres->l_flags | or);
890 }
891 static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
892 				unsigned long clear)
893 {
894 	lockres_set_flags(lockres, lockres->l_flags & ~clear);
895 }
896 
897 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
898 {
899 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
900 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
901 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
902 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
903 
904 	lockres->l_level = lockres->l_requested;
905 	if (lockres->l_level <=
906 	    ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
907 		lockres->l_blocking = DLM_LOCK_NL;
908 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
909 	}
910 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
911 }
912 
913 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
914 {
915 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
916 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
917 
918 	/* Convert from RO to EX doesn't really need anything as our
919 	 * information is already up to data. Convert from NL to
920 	 * *anything* however should mark ourselves as needing an
921 	 * update */
922 	if (lockres->l_level == DLM_LOCK_NL &&
923 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
924 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
925 
926 	lockres->l_level = lockres->l_requested;
927 
928 	/*
929 	 * We set the OCFS2_LOCK_UPCONVERT_FINISHING flag before clearing
930 	 * the OCFS2_LOCK_BUSY flag to prevent the dc thread from
931 	 * downconverting the lock before the upconvert has fully completed.
932 	 * Do not prevent the dc thread from downconverting if NONBLOCK lock
933 	 * had already returned.
934 	 */
935 	if (!(lockres->l_flags & OCFS2_LOCK_NONBLOCK_FINISHED))
936 		lockres_or_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
937 	else
938 		lockres_clear_flags(lockres, OCFS2_LOCK_NONBLOCK_FINISHED);
939 
940 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
941 }
942 
943 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
944 {
945 	BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
946 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
947 
948 	if (lockres->l_requested > DLM_LOCK_NL &&
949 	    !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
950 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
951 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
952 
953 	lockres->l_level = lockres->l_requested;
954 	lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
955 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
956 }
957 
958 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
959 				     int level)
960 {
961 	int needs_downconvert = 0;
962 
963 	assert_spin_locked(&lockres->l_lock);
964 
965 	if (level > lockres->l_blocking) {
966 		/* only schedule a downconvert if we haven't already scheduled
967 		 * one that goes low enough to satisfy the level we're
968 		 * blocking.  this also catches the case where we get
969 		 * duplicate BASTs */
970 		if (ocfs2_highest_compat_lock_level(level) <
971 		    ocfs2_highest_compat_lock_level(lockres->l_blocking))
972 			needs_downconvert = 1;
973 
974 		lockres->l_blocking = level;
975 	}
976 
977 	mlog(ML_BASTS, "lockres %s, block %d, level %d, l_block %d, dwn %d\n",
978 	     lockres->l_name, level, lockres->l_level, lockres->l_blocking,
979 	     needs_downconvert);
980 
981 	if (needs_downconvert)
982 		lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
983 	mlog(0, "needs_downconvert = %d\n", needs_downconvert);
984 	return needs_downconvert;
985 }
986 
987 /*
988  * OCFS2_LOCK_PENDING and l_pending_gen.
989  *
990  * Why does OCFS2_LOCK_PENDING exist?  To close a race between setting
991  * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock().  See ocfs2_unblock_lock()
992  * for more details on the race.
993  *
994  * OCFS2_LOCK_PENDING closes the race quite nicely.  However, it introduces
995  * a race on itself.  In o2dlm, we can get the ast before ocfs2_dlm_lock()
996  * returns.  The ast clears OCFS2_LOCK_BUSY, and must therefore clear
997  * OCFS2_LOCK_PENDING at the same time.  When ocfs2_dlm_lock() returns,
998  * the caller is going to try to clear PENDING again.  If nothing else is
999  * happening, __lockres_clear_pending() sees PENDING is unset and does
1000  * nothing.
1001  *
1002  * But what if another path (eg downconvert thread) has just started a
1003  * new locking action?  The other path has re-set PENDING.  Our path
1004  * cannot clear PENDING, because that will re-open the original race
1005  * window.
1006  *
1007  * [Example]
1008  *
1009  * ocfs2_meta_lock()
1010  *  ocfs2_cluster_lock()
1011  *   set BUSY
1012  *   set PENDING
1013  *   drop l_lock
1014  *   ocfs2_dlm_lock()
1015  *    ocfs2_locking_ast()		ocfs2_downconvert_thread()
1016  *     clear PENDING			 ocfs2_unblock_lock()
1017  *					  take_l_lock
1018  *					  !BUSY
1019  *					  ocfs2_prepare_downconvert()
1020  *					   set BUSY
1021  *					   set PENDING
1022  *					  drop l_lock
1023  *   take l_lock
1024  *   clear PENDING
1025  *   drop l_lock
1026  *			<window>
1027  *					  ocfs2_dlm_lock()
1028  *
1029  * So as you can see, we now have a window where l_lock is not held,
1030  * PENDING is not set, and ocfs2_dlm_lock() has not been called.
1031  *
1032  * The core problem is that ocfs2_cluster_lock() has cleared the PENDING
1033  * set by ocfs2_prepare_downconvert().  That wasn't nice.
1034  *
1035  * To solve this we introduce l_pending_gen.  A call to
1036  * lockres_clear_pending() will only do so when it is passed a generation
1037  * number that matches the lockres.  lockres_set_pending() will return the
1038  * current generation number.  When ocfs2_cluster_lock() goes to clear
1039  * PENDING, it passes the generation it got from set_pending().  In our
1040  * example above, the generation numbers will *not* match.  Thus,
1041  * ocfs2_cluster_lock() will not clear the PENDING set by
1042  * ocfs2_prepare_downconvert().
1043  */
1044 
1045 /* Unlocked version for ocfs2_locking_ast() */
1046 static void __lockres_clear_pending(struct ocfs2_lock_res *lockres,
1047 				    unsigned int generation,
1048 				    struct ocfs2_super *osb)
1049 {
1050 	assert_spin_locked(&lockres->l_lock);
1051 
1052 	/*
1053 	 * The ast and locking functions can race us here.  The winner
1054 	 * will clear pending, the loser will not.
1055 	 */
1056 	if (!(lockres->l_flags & OCFS2_LOCK_PENDING) ||
1057 	    (lockres->l_pending_gen != generation))
1058 		return;
1059 
1060 	lockres_clear_flags(lockres, OCFS2_LOCK_PENDING);
1061 	lockres->l_pending_gen++;
1062 
1063 	/*
1064 	 * The downconvert thread may have skipped us because we
1065 	 * were PENDING.  Wake it up.
1066 	 */
1067 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1068 		ocfs2_wake_downconvert_thread(osb);
1069 }
1070 
1071 /* Locked version for callers of ocfs2_dlm_lock() */
1072 static void lockres_clear_pending(struct ocfs2_lock_res *lockres,
1073 				  unsigned int generation,
1074 				  struct ocfs2_super *osb)
1075 {
1076 	unsigned long flags;
1077 
1078 	spin_lock_irqsave(&lockres->l_lock, flags);
1079 	__lockres_clear_pending(lockres, generation, osb);
1080 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1081 }
1082 
1083 static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres)
1084 {
1085 	assert_spin_locked(&lockres->l_lock);
1086 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
1087 
1088 	lockres_or_flags(lockres, OCFS2_LOCK_PENDING);
1089 
1090 	return lockres->l_pending_gen;
1091 }
1092 
1093 static void ocfs2_blocking_ast(struct ocfs2_dlm_lksb *lksb, int level)
1094 {
1095 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1096 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1097 	int needs_downconvert;
1098 	unsigned long flags;
1099 
1100 	BUG_ON(level <= DLM_LOCK_NL);
1101 
1102 	mlog(ML_BASTS, "BAST fired for lockres %s, blocking %d, level %d, "
1103 	     "type %s\n", lockres->l_name, level, lockres->l_level,
1104 	     ocfs2_lock_type_string(lockres->l_type));
1105 
1106 	/*
1107 	 * We can skip the bast for locks which don't enable caching -
1108 	 * they'll be dropped at the earliest possible time anyway.
1109 	 */
1110 	if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
1111 		return;
1112 
1113 	spin_lock_irqsave(&lockres->l_lock, flags);
1114 	needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
1115 	if (needs_downconvert)
1116 		ocfs2_schedule_blocked_lock(osb, lockres);
1117 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1118 
1119 	wake_up(&lockres->l_event);
1120 
1121 	ocfs2_wake_downconvert_thread(osb);
1122 }
1123 
1124 static void ocfs2_locking_ast(struct ocfs2_dlm_lksb *lksb)
1125 {
1126 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1127 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1128 	unsigned long flags;
1129 	int status;
1130 
1131 	spin_lock_irqsave(&lockres->l_lock, flags);
1132 
1133 	status = ocfs2_dlm_lock_status(&lockres->l_lksb);
1134 
1135 	if (status == -EAGAIN) {
1136 		lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1137 		goto out;
1138 	}
1139 
1140 	if (status) {
1141 		mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n",
1142 		     lockres->l_name, status);
1143 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1144 		return;
1145 	}
1146 
1147 	mlog(ML_BASTS, "AST fired for lockres %s, action %d, unlock %d, "
1148 	     "level %d => %d\n", lockres->l_name, lockres->l_action,
1149 	     lockres->l_unlock_action, lockres->l_level, lockres->l_requested);
1150 
1151 	switch(lockres->l_action) {
1152 	case OCFS2_AST_ATTACH:
1153 		ocfs2_generic_handle_attach_action(lockres);
1154 		lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
1155 		break;
1156 	case OCFS2_AST_CONVERT:
1157 		ocfs2_generic_handle_convert_action(lockres);
1158 		break;
1159 	case OCFS2_AST_DOWNCONVERT:
1160 		ocfs2_generic_handle_downconvert_action(lockres);
1161 		break;
1162 	default:
1163 		mlog(ML_ERROR, "lockres %s: AST fired with invalid action: %u, "
1164 		     "flags 0x%lx, unlock: %u\n",
1165 		     lockres->l_name, lockres->l_action, lockres->l_flags,
1166 		     lockres->l_unlock_action);
1167 		BUG();
1168 	}
1169 out:
1170 	/* set it to something invalid so if we get called again we
1171 	 * can catch it. */
1172 	lockres->l_action = OCFS2_AST_INVALID;
1173 
1174 	/* Did we try to cancel this lock?  Clear that state */
1175 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT)
1176 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1177 
1178 	/*
1179 	 * We may have beaten the locking functions here.  We certainly
1180 	 * know that dlm_lock() has been called :-)
1181 	 * Because we can't have two lock calls in flight at once, we
1182 	 * can use lockres->l_pending_gen.
1183 	 */
1184 	__lockres_clear_pending(lockres, lockres->l_pending_gen,  osb);
1185 
1186 	wake_up(&lockres->l_event);
1187 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1188 }
1189 
1190 static void ocfs2_unlock_ast(struct ocfs2_dlm_lksb *lksb, int error)
1191 {
1192 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1193 	unsigned long flags;
1194 
1195 	mlog(ML_BASTS, "UNLOCK AST fired for lockres %s, action = %d\n",
1196 	     lockres->l_name, lockres->l_unlock_action);
1197 
1198 	spin_lock_irqsave(&lockres->l_lock, flags);
1199 	if (error) {
1200 		mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
1201 		     "unlock_action %d\n", error, lockres->l_name,
1202 		     lockres->l_unlock_action);
1203 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1204 		return;
1205 	}
1206 
1207 	switch(lockres->l_unlock_action) {
1208 	case OCFS2_UNLOCK_CANCEL_CONVERT:
1209 		mlog(0, "Cancel convert success for %s\n", lockres->l_name);
1210 		lockres->l_action = OCFS2_AST_INVALID;
1211 		/* Downconvert thread may have requeued this lock, we
1212 		 * need to wake it. */
1213 		if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1214 			ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres));
1215 		break;
1216 	case OCFS2_UNLOCK_DROP_LOCK:
1217 		lockres->l_level = DLM_LOCK_IV;
1218 		break;
1219 	default:
1220 		BUG();
1221 	}
1222 
1223 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1224 	lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1225 	wake_up(&lockres->l_event);
1226 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1227 }
1228 
1229 /*
1230  * This is the filesystem locking protocol.  It provides the lock handling
1231  * hooks for the underlying DLM.  It has a maximum version number.
1232  * The version number allows interoperability with systems running at
1233  * the same major number and an equal or smaller minor number.
1234  *
1235  * Whenever the filesystem does new things with locks (adds or removes a
1236  * lock, orders them differently, does different things underneath a lock),
1237  * the version must be changed.  The protocol is negotiated when joining
1238  * the dlm domain.  A node may join the domain if its major version is
1239  * identical to all other nodes and its minor version is greater than
1240  * or equal to all other nodes.  When its minor version is greater than
1241  * the other nodes, it will run at the minor version specified by the
1242  * other nodes.
1243  *
1244  * If a locking change is made that will not be compatible with older
1245  * versions, the major number must be increased and the minor version set
1246  * to zero.  If a change merely adds a behavior that can be disabled when
1247  * speaking to older versions, the minor version must be increased.  If a
1248  * change adds a fully backwards compatible change (eg, LVB changes that
1249  * are just ignored by older versions), the version does not need to be
1250  * updated.
1251  */
1252 static struct ocfs2_locking_protocol lproto = {
1253 	.lp_max_version = {
1254 		.pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
1255 		.pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
1256 	},
1257 	.lp_lock_ast		= ocfs2_locking_ast,
1258 	.lp_blocking_ast	= ocfs2_blocking_ast,
1259 	.lp_unlock_ast		= ocfs2_unlock_ast,
1260 };
1261 
1262 void ocfs2_set_locking_protocol(void)
1263 {
1264 	ocfs2_stack_glue_set_max_proto_version(&lproto.lp_max_version);
1265 }
1266 
1267 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
1268 						int convert)
1269 {
1270 	unsigned long flags;
1271 
1272 	spin_lock_irqsave(&lockres->l_lock, flags);
1273 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1274 	lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1275 	if (convert)
1276 		lockres->l_action = OCFS2_AST_INVALID;
1277 	else
1278 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1279 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1280 
1281 	wake_up(&lockres->l_event);
1282 }
1283 
1284 /* Note: If we detect another process working on the lock (i.e.,
1285  * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
1286  * to do the right thing in that case.
1287  */
1288 static int ocfs2_lock_create(struct ocfs2_super *osb,
1289 			     struct ocfs2_lock_res *lockres,
1290 			     int level,
1291 			     u32 dlm_flags)
1292 {
1293 	int ret = 0;
1294 	unsigned long flags;
1295 	unsigned int gen;
1296 
1297 	mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
1298 	     dlm_flags);
1299 
1300 	spin_lock_irqsave(&lockres->l_lock, flags);
1301 	if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
1302 	    (lockres->l_flags & OCFS2_LOCK_BUSY)) {
1303 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1304 		goto bail;
1305 	}
1306 
1307 	lockres->l_action = OCFS2_AST_ATTACH;
1308 	lockres->l_requested = level;
1309 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1310 	gen = lockres_set_pending(lockres);
1311 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1312 
1313 	ret = ocfs2_dlm_lock(osb->cconn,
1314 			     level,
1315 			     &lockres->l_lksb,
1316 			     dlm_flags,
1317 			     lockres->l_name,
1318 			     OCFS2_LOCK_ID_MAX_LEN - 1);
1319 	lockres_clear_pending(lockres, gen, osb);
1320 	if (ret) {
1321 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
1322 		ocfs2_recover_from_dlm_error(lockres, 1);
1323 	}
1324 
1325 	mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
1326 
1327 bail:
1328 	return ret;
1329 }
1330 
1331 static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
1332 					int flag)
1333 {
1334 	unsigned long flags;
1335 	int ret;
1336 
1337 	spin_lock_irqsave(&lockres->l_lock, flags);
1338 	ret = lockres->l_flags & flag;
1339 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1340 
1341 	return ret;
1342 }
1343 
1344 static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
1345 
1346 {
1347 	wait_event(lockres->l_event,
1348 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
1349 }
1350 
1351 static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
1352 
1353 {
1354 	wait_event(lockres->l_event,
1355 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
1356 }
1357 
1358 /* predict what lock level we'll be dropping down to on behalf
1359  * of another node, and return true if the currently wanted
1360  * level will be compatible with it. */
1361 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
1362 						     int wanted)
1363 {
1364 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
1365 
1366 	return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
1367 }
1368 
1369 static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
1370 {
1371 	INIT_LIST_HEAD(&mw->mw_item);
1372 	init_completion(&mw->mw_complete);
1373 	ocfs2_init_start_time(mw);
1374 }
1375 
1376 static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
1377 {
1378 	wait_for_completion(&mw->mw_complete);
1379 	/* Re-arm the completion in case we want to wait on it again */
1380 	reinit_completion(&mw->mw_complete);
1381 	return mw->mw_status;
1382 }
1383 
1384 static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
1385 				    struct ocfs2_mask_waiter *mw,
1386 				    unsigned long mask,
1387 				    unsigned long goal)
1388 {
1389 	BUG_ON(!list_empty(&mw->mw_item));
1390 
1391 	assert_spin_locked(&lockres->l_lock);
1392 
1393 	list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
1394 	mw->mw_mask = mask;
1395 	mw->mw_goal = goal;
1396 }
1397 
1398 /* returns 0 if the mw that was removed was already satisfied, -EBUSY
1399  * if the mask still hadn't reached its goal */
1400 static int __lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1401 				      struct ocfs2_mask_waiter *mw)
1402 {
1403 	int ret = 0;
1404 
1405 	assert_spin_locked(&lockres->l_lock);
1406 	if (!list_empty(&mw->mw_item)) {
1407 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
1408 			ret = -EBUSY;
1409 
1410 		list_del_init(&mw->mw_item);
1411 		init_completion(&mw->mw_complete);
1412 	}
1413 
1414 	return ret;
1415 }
1416 
1417 static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1418 				      struct ocfs2_mask_waiter *mw)
1419 {
1420 	unsigned long flags;
1421 	int ret = 0;
1422 
1423 	spin_lock_irqsave(&lockres->l_lock, flags);
1424 	ret = __lockres_remove_mask_waiter(lockres, mw);
1425 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1426 
1427 	return ret;
1428 
1429 }
1430 
1431 static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
1432 					     struct ocfs2_lock_res *lockres)
1433 {
1434 	int ret;
1435 
1436 	ret = wait_for_completion_interruptible(&mw->mw_complete);
1437 	if (ret)
1438 		lockres_remove_mask_waiter(lockres, mw);
1439 	else
1440 		ret = mw->mw_status;
1441 	/* Re-arm the completion in case we want to wait on it again */
1442 	reinit_completion(&mw->mw_complete);
1443 	return ret;
1444 }
1445 
1446 static int __ocfs2_cluster_lock(struct ocfs2_super *osb,
1447 				struct ocfs2_lock_res *lockres,
1448 				int level,
1449 				u32 lkm_flags,
1450 				int arg_flags,
1451 				int l_subclass,
1452 				unsigned long caller_ip)
1453 {
1454 	struct ocfs2_mask_waiter mw;
1455 	int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
1456 	int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1457 	unsigned long flags;
1458 	unsigned int gen;
1459 	int noqueue_attempted = 0;
1460 	int dlm_locked = 0;
1461 	int kick_dc = 0;
1462 
1463 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED)) {
1464 		mlog_errno(-EINVAL);
1465 		return -EINVAL;
1466 	}
1467 
1468 	ocfs2_init_mask_waiter(&mw);
1469 
1470 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
1471 		lkm_flags |= DLM_LKF_VALBLK;
1472 
1473 again:
1474 	wait = 0;
1475 
1476 	spin_lock_irqsave(&lockres->l_lock, flags);
1477 
1478 	if (catch_signals && signal_pending(current)) {
1479 		ret = -ERESTARTSYS;
1480 		goto unlock;
1481 	}
1482 
1483 	mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
1484 			"Cluster lock called on freeing lockres %s! flags "
1485 			"0x%lx\n", lockres->l_name, lockres->l_flags);
1486 
1487 	/* We only compare against the currently granted level
1488 	 * here. If the lock is blocked waiting on a downconvert,
1489 	 * we'll get caught below. */
1490 	if (lockres->l_flags & OCFS2_LOCK_BUSY &&
1491 	    level > lockres->l_level) {
1492 		/* is someone sitting in dlm_lock? If so, wait on
1493 		 * them. */
1494 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1495 		wait = 1;
1496 		goto unlock;
1497 	}
1498 
1499 	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING) {
1500 		/*
1501 		 * We've upconverted. If the lock now has a level we can
1502 		 * work with, we take it. If, however, the lock is not at the
1503 		 * required level, we go thru the full cycle. One way this could
1504 		 * happen is if a process requesting an upconvert to PR is
1505 		 * closely followed by another requesting upconvert to an EX.
1506 		 * If the process requesting EX lands here, we want it to
1507 		 * continue attempting to upconvert and let the process
1508 		 * requesting PR take the lock.
1509 		 * If multiple processes request upconvert to PR, the first one
1510 		 * here will take the lock. The others will have to go thru the
1511 		 * OCFS2_LOCK_BLOCKED check to ensure that there is no pending
1512 		 * downconvert request.
1513 		 */
1514 		if (level <= lockres->l_level)
1515 			goto update_holders;
1516 	}
1517 
1518 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
1519 	    !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
1520 		/* is the lock is currently blocked on behalf of
1521 		 * another node */
1522 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
1523 		wait = 1;
1524 		goto unlock;
1525 	}
1526 
1527 	if (level > lockres->l_level) {
1528 		if (noqueue_attempted > 0) {
1529 			ret = -EAGAIN;
1530 			goto unlock;
1531 		}
1532 		if (lkm_flags & DLM_LKF_NOQUEUE)
1533 			noqueue_attempted = 1;
1534 
1535 		if (lockres->l_action != OCFS2_AST_INVALID)
1536 			mlog(ML_ERROR, "lockres %s has action %u pending\n",
1537 			     lockres->l_name, lockres->l_action);
1538 
1539 		if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1540 			lockres->l_action = OCFS2_AST_ATTACH;
1541 			lkm_flags &= ~DLM_LKF_CONVERT;
1542 		} else {
1543 			lockres->l_action = OCFS2_AST_CONVERT;
1544 			lkm_flags |= DLM_LKF_CONVERT;
1545 		}
1546 
1547 		lockres->l_requested = level;
1548 		lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1549 		gen = lockres_set_pending(lockres);
1550 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1551 
1552 		BUG_ON(level == DLM_LOCK_IV);
1553 		BUG_ON(level == DLM_LOCK_NL);
1554 
1555 		mlog(ML_BASTS, "lockres %s, convert from %d to %d\n",
1556 		     lockres->l_name, lockres->l_level, level);
1557 
1558 		/* call dlm_lock to upgrade lock now */
1559 		ret = ocfs2_dlm_lock(osb->cconn,
1560 				     level,
1561 				     &lockres->l_lksb,
1562 				     lkm_flags,
1563 				     lockres->l_name,
1564 				     OCFS2_LOCK_ID_MAX_LEN - 1);
1565 		lockres_clear_pending(lockres, gen, osb);
1566 		if (ret) {
1567 			if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
1568 			    (ret != -EAGAIN)) {
1569 				ocfs2_log_dlm_error("ocfs2_dlm_lock",
1570 						    ret, lockres);
1571 			}
1572 			ocfs2_recover_from_dlm_error(lockres, 1);
1573 			goto out;
1574 		}
1575 		dlm_locked = 1;
1576 
1577 		mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n",
1578 		     lockres->l_name);
1579 
1580 		/* At this point we've gone inside the dlm and need to
1581 		 * complete our work regardless. */
1582 		catch_signals = 0;
1583 
1584 		/* wait for busy to clear and carry on */
1585 		goto again;
1586 	}
1587 
1588 update_holders:
1589 	/* Ok, if we get here then we're good to go. */
1590 	ocfs2_inc_holders(lockres, level);
1591 
1592 	ret = 0;
1593 unlock:
1594 	lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1595 
1596 	/* ocfs2_unblock_lock reques on seeing OCFS2_LOCK_UPCONVERT_FINISHING */
1597 	kick_dc = (lockres->l_flags & OCFS2_LOCK_BLOCKED);
1598 
1599 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1600 	if (kick_dc)
1601 		ocfs2_wake_downconvert_thread(osb);
1602 out:
1603 	/*
1604 	 * This is helping work around a lock inversion between the page lock
1605 	 * and dlm locks.  One path holds the page lock while calling aops
1606 	 * which block acquiring dlm locks.  The voting thread holds dlm
1607 	 * locks while acquiring page locks while down converting data locks.
1608 	 * This block is helping an aop path notice the inversion and back
1609 	 * off to unlock its page lock before trying the dlm lock again.
1610 	 */
1611 	if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1612 	    mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1613 		wait = 0;
1614 		spin_lock_irqsave(&lockres->l_lock, flags);
1615 		if (__lockres_remove_mask_waiter(lockres, &mw)) {
1616 			if (dlm_locked)
1617 				lockres_or_flags(lockres,
1618 					OCFS2_LOCK_NONBLOCK_FINISHED);
1619 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1620 			ret = -EAGAIN;
1621 		} else {
1622 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1623 			goto again;
1624 		}
1625 	}
1626 	if (wait) {
1627 		ret = ocfs2_wait_for_mask(&mw);
1628 		if (ret == 0)
1629 			goto again;
1630 		mlog_errno(ret);
1631 	}
1632 	ocfs2_update_lock_stats(lockres, level, &mw, ret);
1633 
1634 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1635 	if (!ret && lockres->l_lockdep_map.key != NULL) {
1636 		if (level == DLM_LOCK_PR)
1637 			rwsem_acquire_read(&lockres->l_lockdep_map, l_subclass,
1638 				!!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1639 				caller_ip);
1640 		else
1641 			rwsem_acquire(&lockres->l_lockdep_map, l_subclass,
1642 				!!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1643 				caller_ip);
1644 	}
1645 #endif
1646 	return ret;
1647 }
1648 
1649 static inline int ocfs2_cluster_lock(struct ocfs2_super *osb,
1650 				     struct ocfs2_lock_res *lockres,
1651 				     int level,
1652 				     u32 lkm_flags,
1653 				     int arg_flags)
1654 {
1655 	return __ocfs2_cluster_lock(osb, lockres, level, lkm_flags, arg_flags,
1656 				    0, _RET_IP_);
1657 }
1658 
1659 
1660 static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
1661 				   struct ocfs2_lock_res *lockres,
1662 				   int level,
1663 				   unsigned long caller_ip)
1664 {
1665 	unsigned long flags;
1666 
1667 	spin_lock_irqsave(&lockres->l_lock, flags);
1668 	ocfs2_dec_holders(lockres, level);
1669 	ocfs2_downconvert_on_unlock(osb, lockres);
1670 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1671 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1672 	if (lockres->l_lockdep_map.key != NULL)
1673 		rwsem_release(&lockres->l_lockdep_map, 1, caller_ip);
1674 #endif
1675 }
1676 
1677 static int ocfs2_create_new_lock(struct ocfs2_super *osb,
1678 				 struct ocfs2_lock_res *lockres,
1679 				 int ex,
1680 				 int local)
1681 {
1682 	int level =  ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1683 	unsigned long flags;
1684 	u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
1685 
1686 	spin_lock_irqsave(&lockres->l_lock, flags);
1687 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1688 	lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1689 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1690 
1691 	return ocfs2_lock_create(osb, lockres, level, lkm_flags);
1692 }
1693 
1694 /* Grants us an EX lock on the data and metadata resources, skipping
1695  * the normal cluster directory lookup. Use this ONLY on newly created
1696  * inodes which other nodes can't possibly see, and which haven't been
1697  * hashed in the inode hash yet. This can give us a good performance
1698  * increase as it'll skip the network broadcast normally associated
1699  * with creating a new lock resource. */
1700 int ocfs2_create_new_inode_locks(struct inode *inode)
1701 {
1702 	int ret;
1703 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1704 
1705 	BUG_ON(!ocfs2_inode_is_new(inode));
1706 
1707 	mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
1708 
1709 	/* NOTE: That we don't increment any of the holder counts, nor
1710 	 * do we add anything to a journal handle. Since this is
1711 	 * supposed to be a new inode which the cluster doesn't know
1712 	 * about yet, there is no need to.  As far as the LVB handling
1713 	 * is concerned, this is basically like acquiring an EX lock
1714 	 * on a resource which has an invalid one -- we'll set it
1715 	 * valid when we release the EX. */
1716 
1717 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
1718 	if (ret) {
1719 		mlog_errno(ret);
1720 		goto bail;
1721 	}
1722 
1723 	/*
1724 	 * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
1725 	 * don't use a generation in their lock names.
1726 	 */
1727 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
1728 	if (ret) {
1729 		mlog_errno(ret);
1730 		goto bail;
1731 	}
1732 
1733 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
1734 	if (ret)
1735 		mlog_errno(ret);
1736 
1737 bail:
1738 	return ret;
1739 }
1740 
1741 int ocfs2_rw_lock(struct inode *inode, int write)
1742 {
1743 	int status, level;
1744 	struct ocfs2_lock_res *lockres;
1745 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1746 
1747 	mlog(0, "inode %llu take %s RW lock\n",
1748 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1749 	     write ? "EXMODE" : "PRMODE");
1750 
1751 	if (ocfs2_mount_local(osb))
1752 		return 0;
1753 
1754 	lockres = &OCFS2_I(inode)->ip_rw_lockres;
1755 
1756 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1757 
1758 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
1759 	if (status < 0)
1760 		mlog_errno(status);
1761 
1762 	return status;
1763 }
1764 
1765 int ocfs2_try_rw_lock(struct inode *inode, int write)
1766 {
1767 	int status, level;
1768 	struct ocfs2_lock_res *lockres;
1769 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1770 
1771 	mlog(0, "inode %llu try to take %s RW lock\n",
1772 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1773 	     write ? "EXMODE" : "PRMODE");
1774 
1775 	if (ocfs2_mount_local(osb))
1776 		return 0;
1777 
1778 	lockres = &OCFS2_I(inode)->ip_rw_lockres;
1779 
1780 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1781 
1782 	status = ocfs2_cluster_lock(osb, lockres, level, DLM_LKF_NOQUEUE, 0);
1783 	return status;
1784 }
1785 
1786 void ocfs2_rw_unlock(struct inode *inode, int write)
1787 {
1788 	int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1789 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
1790 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1791 
1792 	mlog(0, "inode %llu drop %s RW lock\n",
1793 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1794 	     write ? "EXMODE" : "PRMODE");
1795 
1796 	if (!ocfs2_mount_local(osb))
1797 		ocfs2_cluster_unlock(osb, lockres, level);
1798 }
1799 
1800 /*
1801  * ocfs2_open_lock always get PR mode lock.
1802  */
1803 int ocfs2_open_lock(struct inode *inode)
1804 {
1805 	int status = 0;
1806 	struct ocfs2_lock_res *lockres;
1807 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1808 
1809 	mlog(0, "inode %llu take PRMODE open lock\n",
1810 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1811 
1812 	if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1813 		goto out;
1814 
1815 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1816 
1817 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_PR, 0, 0);
1818 	if (status < 0)
1819 		mlog_errno(status);
1820 
1821 out:
1822 	return status;
1823 }
1824 
1825 int ocfs2_try_open_lock(struct inode *inode, int write)
1826 {
1827 	int status = 0, level;
1828 	struct ocfs2_lock_res *lockres;
1829 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1830 
1831 	mlog(0, "inode %llu try to take %s open lock\n",
1832 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1833 	     write ? "EXMODE" : "PRMODE");
1834 
1835 	if (ocfs2_is_hard_readonly(osb)) {
1836 		if (write)
1837 			status = -EROFS;
1838 		goto out;
1839 	}
1840 
1841 	if (ocfs2_mount_local(osb))
1842 		goto out;
1843 
1844 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1845 
1846 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1847 
1848 	/*
1849 	 * The file system may already holding a PRMODE/EXMODE open lock.
1850 	 * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
1851 	 * other nodes and the -EAGAIN will indicate to the caller that
1852 	 * this inode is still in use.
1853 	 */
1854 	status = ocfs2_cluster_lock(osb, lockres, level, DLM_LKF_NOQUEUE, 0);
1855 
1856 out:
1857 	return status;
1858 }
1859 
1860 /*
1861  * ocfs2_open_unlock unlock PR and EX mode open locks.
1862  */
1863 void ocfs2_open_unlock(struct inode *inode)
1864 {
1865 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
1866 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1867 
1868 	mlog(0, "inode %llu drop open lock\n",
1869 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1870 
1871 	if (ocfs2_mount_local(osb))
1872 		goto out;
1873 
1874 	if(lockres->l_ro_holders)
1875 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_PR);
1876 	if(lockres->l_ex_holders)
1877 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
1878 
1879 out:
1880 	return;
1881 }
1882 
1883 static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
1884 				     int level)
1885 {
1886 	int ret;
1887 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1888 	unsigned long flags;
1889 	struct ocfs2_mask_waiter mw;
1890 
1891 	ocfs2_init_mask_waiter(&mw);
1892 
1893 retry_cancel:
1894 	spin_lock_irqsave(&lockres->l_lock, flags);
1895 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
1896 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
1897 		if (ret) {
1898 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1899 			ret = ocfs2_cancel_convert(osb, lockres);
1900 			if (ret < 0) {
1901 				mlog_errno(ret);
1902 				goto out;
1903 			}
1904 			goto retry_cancel;
1905 		}
1906 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1907 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1908 
1909 		ocfs2_wait_for_mask(&mw);
1910 		goto retry_cancel;
1911 	}
1912 
1913 	ret = -ERESTARTSYS;
1914 	/*
1915 	 * We may still have gotten the lock, in which case there's no
1916 	 * point to restarting the syscall.
1917 	 */
1918 	if (lockres->l_level == level)
1919 		ret = 0;
1920 
1921 	mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
1922 	     lockres->l_flags, lockres->l_level, lockres->l_action);
1923 
1924 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1925 
1926 out:
1927 	return ret;
1928 }
1929 
1930 /*
1931  * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
1932  * flock() calls. The locking approach this requires is sufficiently
1933  * different from all other cluster lock types that we implement a
1934  * separate path to the "low-level" dlm calls. In particular:
1935  *
1936  * - No optimization of lock levels is done - we take at exactly
1937  *   what's been requested.
1938  *
1939  * - No lock caching is employed. We immediately downconvert to
1940  *   no-lock at unlock time. This also means flock locks never go on
1941  *   the blocking list).
1942  *
1943  * - Since userspace can trivially deadlock itself with flock, we make
1944  *   sure to allow cancellation of a misbehaving applications flock()
1945  *   request.
1946  *
1947  * - Access to any flock lockres doesn't require concurrency, so we
1948  *   can simplify the code by requiring the caller to guarantee
1949  *   serialization of dlmglue flock calls.
1950  */
1951 int ocfs2_file_lock(struct file *file, int ex, int trylock)
1952 {
1953 	int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1954 	unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0;
1955 	unsigned long flags;
1956 	struct ocfs2_file_private *fp = file->private_data;
1957 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
1958 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1959 	struct ocfs2_mask_waiter mw;
1960 
1961 	ocfs2_init_mask_waiter(&mw);
1962 
1963 	if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
1964 	    (lockres->l_level > DLM_LOCK_NL)) {
1965 		mlog(ML_ERROR,
1966 		     "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
1967 		     "level: %u\n", lockres->l_name, lockres->l_flags,
1968 		     lockres->l_level);
1969 		return -EINVAL;
1970 	}
1971 
1972 	spin_lock_irqsave(&lockres->l_lock, flags);
1973 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1974 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1975 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1976 
1977 		/*
1978 		 * Get the lock at NLMODE to start - that way we
1979 		 * can cancel the upconvert request if need be.
1980 		 */
1981 		ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0);
1982 		if (ret < 0) {
1983 			mlog_errno(ret);
1984 			goto out;
1985 		}
1986 
1987 		ret = ocfs2_wait_for_mask(&mw);
1988 		if (ret) {
1989 			mlog_errno(ret);
1990 			goto out;
1991 		}
1992 		spin_lock_irqsave(&lockres->l_lock, flags);
1993 	}
1994 
1995 	lockres->l_action = OCFS2_AST_CONVERT;
1996 	lkm_flags |= DLM_LKF_CONVERT;
1997 	lockres->l_requested = level;
1998 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1999 
2000 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
2001 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2002 
2003 	ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags,
2004 			     lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1);
2005 	if (ret) {
2006 		if (!trylock || (ret != -EAGAIN)) {
2007 			ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
2008 			ret = -EINVAL;
2009 		}
2010 
2011 		ocfs2_recover_from_dlm_error(lockres, 1);
2012 		lockres_remove_mask_waiter(lockres, &mw);
2013 		goto out;
2014 	}
2015 
2016 	ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
2017 	if (ret == -ERESTARTSYS) {
2018 		/*
2019 		 * Userspace can cause deadlock itself with
2020 		 * flock(). Current behavior locally is to allow the
2021 		 * deadlock, but abort the system call if a signal is
2022 		 * received. We follow this example, otherwise a
2023 		 * poorly written program could sit in kernel until
2024 		 * reboot.
2025 		 *
2026 		 * Handling this is a bit more complicated for Ocfs2
2027 		 * though. We can't exit this function with an
2028 		 * outstanding lock request, so a cancel convert is
2029 		 * required. We intentionally overwrite 'ret' - if the
2030 		 * cancel fails and the lock was granted, it's easier
2031 		 * to just bubble success back up to the user.
2032 		 */
2033 		ret = ocfs2_flock_handle_signal(lockres, level);
2034 	} else if (!ret && (level > lockres->l_level)) {
2035 		/* Trylock failed asynchronously */
2036 		BUG_ON(!trylock);
2037 		ret = -EAGAIN;
2038 	}
2039 
2040 out:
2041 
2042 	mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
2043 	     lockres->l_name, ex, trylock, ret);
2044 	return ret;
2045 }
2046 
2047 void ocfs2_file_unlock(struct file *file)
2048 {
2049 	int ret;
2050 	unsigned int gen;
2051 	unsigned long flags;
2052 	struct ocfs2_file_private *fp = file->private_data;
2053 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
2054 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
2055 	struct ocfs2_mask_waiter mw;
2056 
2057 	ocfs2_init_mask_waiter(&mw);
2058 
2059 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
2060 		return;
2061 
2062 	if (lockres->l_level == DLM_LOCK_NL)
2063 		return;
2064 
2065 	mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
2066 	     lockres->l_name, lockres->l_flags, lockres->l_level,
2067 	     lockres->l_action);
2068 
2069 	spin_lock_irqsave(&lockres->l_lock, flags);
2070 	/*
2071 	 * Fake a blocking ast for the downconvert code.
2072 	 */
2073 	lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
2074 	lockres->l_blocking = DLM_LOCK_EX;
2075 
2076 	gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL);
2077 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
2078 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2079 
2080 	ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen);
2081 	if (ret) {
2082 		mlog_errno(ret);
2083 		return;
2084 	}
2085 
2086 	ret = ocfs2_wait_for_mask(&mw);
2087 	if (ret)
2088 		mlog_errno(ret);
2089 }
2090 
2091 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
2092 					struct ocfs2_lock_res *lockres)
2093 {
2094 	int kick = 0;
2095 
2096 	/* If we know that another node is waiting on our lock, kick
2097 	 * the downconvert thread * pre-emptively when we reach a release
2098 	 * condition. */
2099 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
2100 		switch(lockres->l_blocking) {
2101 		case DLM_LOCK_EX:
2102 			if (!lockres->l_ex_holders && !lockres->l_ro_holders)
2103 				kick = 1;
2104 			break;
2105 		case DLM_LOCK_PR:
2106 			if (!lockres->l_ex_holders)
2107 				kick = 1;
2108 			break;
2109 		default:
2110 			BUG();
2111 		}
2112 	}
2113 
2114 	if (kick)
2115 		ocfs2_wake_downconvert_thread(osb);
2116 }
2117 
2118 #define OCFS2_SEC_BITS   34
2119 #define OCFS2_SEC_SHIFT  (64 - 34)
2120 #define OCFS2_NSEC_MASK  ((1ULL << OCFS2_SEC_SHIFT) - 1)
2121 
2122 /* LVB only has room for 64 bits of time here so we pack it for
2123  * now. */
2124 static u64 ocfs2_pack_timespec(struct timespec *spec)
2125 {
2126 	u64 res;
2127 	u64 sec = spec->tv_sec;
2128 	u32 nsec = spec->tv_nsec;
2129 
2130 	res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
2131 
2132 	return res;
2133 }
2134 
2135 /* Call this with the lockres locked. I am reasonably sure we don't
2136  * need ip_lock in this function as anyone who would be changing those
2137  * values is supposed to be blocked in ocfs2_inode_lock right now. */
2138 static void __ocfs2_stuff_meta_lvb(struct inode *inode)
2139 {
2140 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2141 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2142 	struct ocfs2_meta_lvb *lvb;
2143 	struct timespec ts;
2144 
2145 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2146 
2147 	/*
2148 	 * Invalidate the LVB of a deleted inode - this way other
2149 	 * nodes are forced to go to disk and discover the new inode
2150 	 * status.
2151 	 */
2152 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
2153 		lvb->lvb_version = 0;
2154 		goto out;
2155 	}
2156 
2157 	lvb->lvb_version   = OCFS2_LVB_VERSION;
2158 	lvb->lvb_isize	   = cpu_to_be64(i_size_read(inode));
2159 	lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
2160 	lvb->lvb_iuid      = cpu_to_be32(i_uid_read(inode));
2161 	lvb->lvb_igid      = cpu_to_be32(i_gid_read(inode));
2162 	lvb->lvb_imode     = cpu_to_be16(inode->i_mode);
2163 	lvb->lvb_inlink    = cpu_to_be16(inode->i_nlink);
2164 	ts = timespec64_to_timespec(inode->i_atime);
2165 	lvb->lvb_iatime_packed  =
2166 		cpu_to_be64(ocfs2_pack_timespec(&ts));
2167 	ts = timespec64_to_timespec(inode->i_ctime);
2168 	lvb->lvb_ictime_packed =
2169 		cpu_to_be64(ocfs2_pack_timespec(&ts));
2170 	ts = timespec64_to_timespec(inode->i_mtime);
2171 	lvb->lvb_imtime_packed =
2172 		cpu_to_be64(ocfs2_pack_timespec(&ts));
2173 	lvb->lvb_iattr    = cpu_to_be32(oi->ip_attr);
2174 	lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
2175 	lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
2176 
2177 out:
2178 	mlog_meta_lvb(0, lockres);
2179 }
2180 
2181 static void ocfs2_unpack_timespec(struct timespec *spec,
2182 				  u64 packed_time)
2183 {
2184 	spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
2185 	spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
2186 }
2187 
2188 static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
2189 {
2190 	struct timespec ts;
2191 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2192 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2193 	struct ocfs2_meta_lvb *lvb;
2194 
2195 	mlog_meta_lvb(0, lockres);
2196 
2197 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2198 
2199 	/* We're safe here without the lockres lock... */
2200 	spin_lock(&oi->ip_lock);
2201 	oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
2202 	i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
2203 
2204 	oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
2205 	oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
2206 	ocfs2_set_inode_flags(inode);
2207 
2208 	/* fast-symlinks are a special case */
2209 	if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
2210 		inode->i_blocks = 0;
2211 	else
2212 		inode->i_blocks = ocfs2_inode_sector_count(inode);
2213 
2214 	i_uid_write(inode, be32_to_cpu(lvb->lvb_iuid));
2215 	i_gid_write(inode, be32_to_cpu(lvb->lvb_igid));
2216 	inode->i_mode    = be16_to_cpu(lvb->lvb_imode);
2217 	set_nlink(inode, be16_to_cpu(lvb->lvb_inlink));
2218 	ocfs2_unpack_timespec(&ts,
2219 			      be64_to_cpu(lvb->lvb_iatime_packed));
2220 	inode->i_atime = timespec_to_timespec64(ts);
2221 	ocfs2_unpack_timespec(&ts,
2222 			      be64_to_cpu(lvb->lvb_imtime_packed));
2223 	inode->i_mtime = timespec_to_timespec64(ts);
2224 	ocfs2_unpack_timespec(&ts,
2225 			      be64_to_cpu(lvb->lvb_ictime_packed));
2226 	inode->i_ctime = timespec_to_timespec64(ts);
2227 	spin_unlock(&oi->ip_lock);
2228 }
2229 
2230 static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
2231 					      struct ocfs2_lock_res *lockres)
2232 {
2233 	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2234 
2235 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)
2236 	    && lvb->lvb_version == OCFS2_LVB_VERSION
2237 	    && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
2238 		return 1;
2239 	return 0;
2240 }
2241 
2242 /* Determine whether a lock resource needs to be refreshed, and
2243  * arbitrate who gets to refresh it.
2244  *
2245  *   0 means no refresh needed.
2246  *
2247  *   > 0 means you need to refresh this and you MUST call
2248  *   ocfs2_complete_lock_res_refresh afterwards. */
2249 static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
2250 {
2251 	unsigned long flags;
2252 	int status = 0;
2253 
2254 refresh_check:
2255 	spin_lock_irqsave(&lockres->l_lock, flags);
2256 	if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
2257 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2258 		goto bail;
2259 	}
2260 
2261 	if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
2262 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2263 
2264 		ocfs2_wait_on_refreshing_lock(lockres);
2265 		goto refresh_check;
2266 	}
2267 
2268 	/* Ok, I'll be the one to refresh this lock. */
2269 	lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
2270 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2271 
2272 	status = 1;
2273 bail:
2274 	mlog(0, "status %d\n", status);
2275 	return status;
2276 }
2277 
2278 /* If status is non zero, I'll mark it as not being in refresh
2279  * anymroe, but i won't clear the needs refresh flag. */
2280 static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
2281 						   int status)
2282 {
2283 	unsigned long flags;
2284 
2285 	spin_lock_irqsave(&lockres->l_lock, flags);
2286 	lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
2287 	if (!status)
2288 		lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
2289 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2290 
2291 	wake_up(&lockres->l_event);
2292 }
2293 
2294 /* may or may not return a bh if it went to disk. */
2295 static int ocfs2_inode_lock_update(struct inode *inode,
2296 				  struct buffer_head **bh)
2297 {
2298 	int status = 0;
2299 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2300 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2301 	struct ocfs2_dinode *fe;
2302 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2303 
2304 	if (ocfs2_mount_local(osb))
2305 		goto bail;
2306 
2307 	spin_lock(&oi->ip_lock);
2308 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
2309 		mlog(0, "Orphaned inode %llu was deleted while we "
2310 		     "were waiting on a lock. ip_flags = 0x%x\n",
2311 		     (unsigned long long)oi->ip_blkno, oi->ip_flags);
2312 		spin_unlock(&oi->ip_lock);
2313 		status = -ENOENT;
2314 		goto bail;
2315 	}
2316 	spin_unlock(&oi->ip_lock);
2317 
2318 	if (!ocfs2_should_refresh_lock_res(lockres))
2319 		goto bail;
2320 
2321 	/* This will discard any caching information we might have had
2322 	 * for the inode metadata. */
2323 	ocfs2_metadata_cache_purge(INODE_CACHE(inode));
2324 
2325 	ocfs2_extent_map_trunc(inode, 0);
2326 
2327 	if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
2328 		mlog(0, "Trusting LVB on inode %llu\n",
2329 		     (unsigned long long)oi->ip_blkno);
2330 		ocfs2_refresh_inode_from_lvb(inode);
2331 	} else {
2332 		/* Boo, we have to go to disk. */
2333 		/* read bh, cast, ocfs2_refresh_inode */
2334 		status = ocfs2_read_inode_block(inode, bh);
2335 		if (status < 0) {
2336 			mlog_errno(status);
2337 			goto bail_refresh;
2338 		}
2339 		fe = (struct ocfs2_dinode *) (*bh)->b_data;
2340 
2341 		/* This is a good chance to make sure we're not
2342 		 * locking an invalid object.  ocfs2_read_inode_block()
2343 		 * already checked that the inode block is sane.
2344 		 *
2345 		 * We bug on a stale inode here because we checked
2346 		 * above whether it was wiped from disk. The wiping
2347 		 * node provides a guarantee that we receive that
2348 		 * message and can mark the inode before dropping any
2349 		 * locks associated with it. */
2350 		mlog_bug_on_msg(inode->i_generation !=
2351 				le32_to_cpu(fe->i_generation),
2352 				"Invalid dinode %llu disk generation: %u "
2353 				"inode->i_generation: %u\n",
2354 				(unsigned long long)oi->ip_blkno,
2355 				le32_to_cpu(fe->i_generation),
2356 				inode->i_generation);
2357 		mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
2358 				!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
2359 				"Stale dinode %llu dtime: %llu flags: 0x%x\n",
2360 				(unsigned long long)oi->ip_blkno,
2361 				(unsigned long long)le64_to_cpu(fe->i_dtime),
2362 				le32_to_cpu(fe->i_flags));
2363 
2364 		ocfs2_refresh_inode(inode, fe);
2365 		ocfs2_track_lock_refresh(lockres);
2366 	}
2367 
2368 	status = 0;
2369 bail_refresh:
2370 	ocfs2_complete_lock_res_refresh(lockres, status);
2371 bail:
2372 	return status;
2373 }
2374 
2375 static int ocfs2_assign_bh(struct inode *inode,
2376 			   struct buffer_head **ret_bh,
2377 			   struct buffer_head *passed_bh)
2378 {
2379 	int status;
2380 
2381 	if (passed_bh) {
2382 		/* Ok, the update went to disk for us, use the
2383 		 * returned bh. */
2384 		*ret_bh = passed_bh;
2385 		get_bh(*ret_bh);
2386 
2387 		return 0;
2388 	}
2389 
2390 	status = ocfs2_read_inode_block(inode, ret_bh);
2391 	if (status < 0)
2392 		mlog_errno(status);
2393 
2394 	return status;
2395 }
2396 
2397 /*
2398  * returns < 0 error if the callback will never be called, otherwise
2399  * the result of the lock will be communicated via the callback.
2400  */
2401 int ocfs2_inode_lock_full_nested(struct inode *inode,
2402 				 struct buffer_head **ret_bh,
2403 				 int ex,
2404 				 int arg_flags,
2405 				 int subclass)
2406 {
2407 	int status, level, acquired;
2408 	u32 dlm_flags;
2409 	struct ocfs2_lock_res *lockres = NULL;
2410 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2411 	struct buffer_head *local_bh = NULL;
2412 
2413 	mlog(0, "inode %llu, take %s META lock\n",
2414 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2415 	     ex ? "EXMODE" : "PRMODE");
2416 
2417 	status = 0;
2418 	acquired = 0;
2419 	/* We'll allow faking a readonly metadata lock for
2420 	 * rodevices. */
2421 	if (ocfs2_is_hard_readonly(osb)) {
2422 		if (ex)
2423 			status = -EROFS;
2424 		goto getbh;
2425 	}
2426 
2427 	if ((arg_flags & OCFS2_META_LOCK_GETBH) ||
2428 	    ocfs2_mount_local(osb))
2429 		goto update;
2430 
2431 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2432 		ocfs2_wait_for_recovery(osb);
2433 
2434 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2435 	level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2436 	dlm_flags = 0;
2437 	if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
2438 		dlm_flags |= DLM_LKF_NOQUEUE;
2439 
2440 	status = __ocfs2_cluster_lock(osb, lockres, level, dlm_flags,
2441 				      arg_flags, subclass, _RET_IP_);
2442 	if (status < 0) {
2443 		if (status != -EAGAIN)
2444 			mlog_errno(status);
2445 		goto bail;
2446 	}
2447 
2448 	/* Notify the error cleanup path to drop the cluster lock. */
2449 	acquired = 1;
2450 
2451 	/* We wait twice because a node may have died while we were in
2452 	 * the lower dlm layers. The second time though, we've
2453 	 * committed to owning this lock so we don't allow signals to
2454 	 * abort the operation. */
2455 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2456 		ocfs2_wait_for_recovery(osb);
2457 
2458 update:
2459 	/*
2460 	 * We only see this flag if we're being called from
2461 	 * ocfs2_read_locked_inode(). It means we're locking an inode
2462 	 * which hasn't been populated yet, so clear the refresh flag
2463 	 * and let the caller handle it.
2464 	 */
2465 	if (inode->i_state & I_NEW) {
2466 		status = 0;
2467 		if (lockres)
2468 			ocfs2_complete_lock_res_refresh(lockres, 0);
2469 		goto bail;
2470 	}
2471 
2472 	/* This is fun. The caller may want a bh back, or it may
2473 	 * not. ocfs2_inode_lock_update definitely wants one in, but
2474 	 * may or may not read one, depending on what's in the
2475 	 * LVB. The result of all of this is that we've *only* gone to
2476 	 * disk if we have to, so the complexity is worthwhile. */
2477 	status = ocfs2_inode_lock_update(inode, &local_bh);
2478 	if (status < 0) {
2479 		if (status != -ENOENT)
2480 			mlog_errno(status);
2481 		goto bail;
2482 	}
2483 getbh:
2484 	if (ret_bh) {
2485 		status = ocfs2_assign_bh(inode, ret_bh, local_bh);
2486 		if (status < 0) {
2487 			mlog_errno(status);
2488 			goto bail;
2489 		}
2490 	}
2491 
2492 bail:
2493 	if (status < 0) {
2494 		if (ret_bh && (*ret_bh)) {
2495 			brelse(*ret_bh);
2496 			*ret_bh = NULL;
2497 		}
2498 		if (acquired)
2499 			ocfs2_inode_unlock(inode, ex);
2500 	}
2501 
2502 	if (local_bh)
2503 		brelse(local_bh);
2504 
2505 	return status;
2506 }
2507 
2508 /*
2509  * This is working around a lock inversion between tasks acquiring DLM
2510  * locks while holding a page lock and the downconvert thread which
2511  * blocks dlm lock acquiry while acquiring page locks.
2512  *
2513  * ** These _with_page variantes are only intended to be called from aop
2514  * methods that hold page locks and return a very specific *positive* error
2515  * code that aop methods pass up to the VFS -- test for errors with != 0. **
2516  *
2517  * The DLM is called such that it returns -EAGAIN if it would have
2518  * blocked waiting for the downconvert thread.  In that case we unlock
2519  * our page so the downconvert thread can make progress.  Once we've
2520  * done this we have to return AOP_TRUNCATED_PAGE so the aop method
2521  * that called us can bubble that back up into the VFS who will then
2522  * immediately retry the aop call.
2523  */
2524 int ocfs2_inode_lock_with_page(struct inode *inode,
2525 			      struct buffer_head **ret_bh,
2526 			      int ex,
2527 			      struct page *page)
2528 {
2529 	int ret;
2530 
2531 	ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
2532 	if (ret == -EAGAIN) {
2533 		unlock_page(page);
2534 		/*
2535 		 * If we can't get inode lock immediately, we should not return
2536 		 * directly here, since this will lead to a softlockup problem.
2537 		 * The method is to get a blocking lock and immediately unlock
2538 		 * before returning, this can avoid CPU resource waste due to
2539 		 * lots of retries, and benefits fairness in getting lock.
2540 		 */
2541 		if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
2542 			ocfs2_inode_unlock(inode, ex);
2543 		ret = AOP_TRUNCATED_PAGE;
2544 	}
2545 
2546 	return ret;
2547 }
2548 
2549 int ocfs2_inode_lock_atime(struct inode *inode,
2550 			  struct vfsmount *vfsmnt,
2551 			  int *level, int wait)
2552 {
2553 	int ret;
2554 
2555 	if (wait)
2556 		ret = ocfs2_inode_lock(inode, NULL, 0);
2557 	else
2558 		ret = ocfs2_try_inode_lock(inode, NULL, 0);
2559 
2560 	if (ret < 0) {
2561 		if (ret != -EAGAIN)
2562 			mlog_errno(ret);
2563 		return ret;
2564 	}
2565 
2566 	/*
2567 	 * If we should update atime, we will get EX lock,
2568 	 * otherwise we just get PR lock.
2569 	 */
2570 	if (ocfs2_should_update_atime(inode, vfsmnt)) {
2571 		struct buffer_head *bh = NULL;
2572 
2573 		ocfs2_inode_unlock(inode, 0);
2574 		if (wait)
2575 			ret = ocfs2_inode_lock(inode, &bh, 1);
2576 		else
2577 			ret = ocfs2_try_inode_lock(inode, &bh, 1);
2578 
2579 		if (ret < 0) {
2580 			if (ret != -EAGAIN)
2581 				mlog_errno(ret);
2582 			return ret;
2583 		}
2584 		*level = 1;
2585 		if (ocfs2_should_update_atime(inode, vfsmnt))
2586 			ocfs2_update_inode_atime(inode, bh);
2587 		if (bh)
2588 			brelse(bh);
2589 	} else
2590 		*level = 0;
2591 
2592 	return ret;
2593 }
2594 
2595 void ocfs2_inode_unlock(struct inode *inode,
2596 		       int ex)
2597 {
2598 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2599 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
2600 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2601 
2602 	mlog(0, "inode %llu drop %s META lock\n",
2603 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2604 	     ex ? "EXMODE" : "PRMODE");
2605 
2606 	if (!ocfs2_is_hard_readonly(osb) &&
2607 	    !ocfs2_mount_local(osb))
2608 		ocfs2_cluster_unlock(osb, lockres, level);
2609 }
2610 
2611 /*
2612  * This _tracker variantes are introduced to deal with the recursive cluster
2613  * locking issue. The idea is to keep track of a lock holder on the stack of
2614  * the current process. If there's a lock holder on the stack, we know the
2615  * task context is already protected by cluster locking. Currently, they're
2616  * used in some VFS entry routines.
2617  *
2618  * return < 0 on error, return == 0 if there's no lock holder on the stack
2619  * before this call, return == 1 if this call would be a recursive locking.
2620  * return == -1 if this lock attempt will cause an upgrade which is forbidden.
2621  *
2622  * When taking lock levels into account,we face some different situations.
2623  *
2624  * 1. no lock is held
2625  *    In this case, just lock the inode as requested and return 0
2626  *
2627  * 2. We are holding a lock
2628  *    For this situation, things diverges into several cases
2629  *
2630  *    wanted     holding	     what to do
2631  *    ex		ex	    see 2.1 below
2632  *    ex		pr	    see 2.2 below
2633  *    pr		ex	    see 2.1 below
2634  *    pr		pr	    see 2.1 below
2635  *
2636  *    2.1 lock level that is been held is compatible
2637  *    with the wanted level, so no lock action will be tacken.
2638  *
2639  *    2.2 Otherwise, an upgrade is needed, but it is forbidden.
2640  *
2641  * Reason why upgrade within a process is forbidden is that
2642  * lock upgrade may cause dead lock. The following illustrates
2643  * how it happens.
2644  *
2645  *         thread on node1                             thread on node2
2646  * ocfs2_inode_lock_tracker(ex=0)
2647  *
2648  *                                <======   ocfs2_inode_lock_tracker(ex=1)
2649  *
2650  * ocfs2_inode_lock_tracker(ex=1)
2651  */
2652 int ocfs2_inode_lock_tracker(struct inode *inode,
2653 			     struct buffer_head **ret_bh,
2654 			     int ex,
2655 			     struct ocfs2_lock_holder *oh)
2656 {
2657 	int status = 0;
2658 	struct ocfs2_lock_res *lockres;
2659 	struct ocfs2_lock_holder *tmp_oh;
2660 	struct pid *pid = task_pid(current);
2661 
2662 
2663 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2664 	tmp_oh = ocfs2_pid_holder(lockres, pid);
2665 
2666 	if (!tmp_oh) {
2667 		/*
2668 		 * This corresponds to the case 1.
2669 		 * We haven't got any lock before.
2670 		 */
2671 		status = ocfs2_inode_lock_full(inode, ret_bh, ex, 0);
2672 		if (status < 0) {
2673 			if (status != -ENOENT)
2674 				mlog_errno(status);
2675 			return status;
2676 		}
2677 
2678 		oh->oh_ex = ex;
2679 		ocfs2_add_holder(lockres, oh);
2680 		return 0;
2681 	}
2682 
2683 	if (unlikely(ex && !tmp_oh->oh_ex)) {
2684 		/*
2685 		 * case 2.2 upgrade may cause dead lock, forbid it.
2686 		 */
2687 		mlog(ML_ERROR, "Recursive locking is not permitted to "
2688 		     "upgrade to EX level from PR level.\n");
2689 		dump_stack();
2690 		return -EINVAL;
2691 	}
2692 
2693 	/*
2694 	 *  case 2.1 OCFS2_META_LOCK_GETBH flag make ocfs2_inode_lock_full.
2695 	 *  ignore the lock level and just update it.
2696 	 */
2697 	if (ret_bh) {
2698 		status = ocfs2_inode_lock_full(inode, ret_bh, ex,
2699 					       OCFS2_META_LOCK_GETBH);
2700 		if (status < 0) {
2701 			if (status != -ENOENT)
2702 				mlog_errno(status);
2703 			return status;
2704 		}
2705 	}
2706 	return tmp_oh ? 1 : 0;
2707 }
2708 
2709 void ocfs2_inode_unlock_tracker(struct inode *inode,
2710 				int ex,
2711 				struct ocfs2_lock_holder *oh,
2712 				int had_lock)
2713 {
2714 	struct ocfs2_lock_res *lockres;
2715 
2716 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2717 	/* had_lock means that the currect process already takes the cluster
2718 	 * lock previously.
2719 	 * If had_lock is 1, we have nothing to do here.
2720 	 * If had_lock is 0, we will release the lock.
2721 	 */
2722 	if (!had_lock) {
2723 		ocfs2_inode_unlock(inode, oh->oh_ex);
2724 		ocfs2_remove_holder(lockres, oh);
2725 	}
2726 }
2727 
2728 int ocfs2_orphan_scan_lock(struct ocfs2_super *osb, u32 *seqno)
2729 {
2730 	struct ocfs2_lock_res *lockres;
2731 	struct ocfs2_orphan_scan_lvb *lvb;
2732 	int status = 0;
2733 
2734 	if (ocfs2_is_hard_readonly(osb))
2735 		return -EROFS;
2736 
2737 	if (ocfs2_mount_local(osb))
2738 		return 0;
2739 
2740 	lockres = &osb->osb_orphan_scan.os_lockres;
2741 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2742 	if (status < 0)
2743 		return status;
2744 
2745 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2746 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2747 	    lvb->lvb_version == OCFS2_ORPHAN_LVB_VERSION)
2748 		*seqno = be32_to_cpu(lvb->lvb_os_seqno);
2749 	else
2750 		*seqno = osb->osb_orphan_scan.os_seqno + 1;
2751 
2752 	return status;
2753 }
2754 
2755 void ocfs2_orphan_scan_unlock(struct ocfs2_super *osb, u32 seqno)
2756 {
2757 	struct ocfs2_lock_res *lockres;
2758 	struct ocfs2_orphan_scan_lvb *lvb;
2759 
2760 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) {
2761 		lockres = &osb->osb_orphan_scan.os_lockres;
2762 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2763 		lvb->lvb_version = OCFS2_ORPHAN_LVB_VERSION;
2764 		lvb->lvb_os_seqno = cpu_to_be32(seqno);
2765 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2766 	}
2767 }
2768 
2769 int ocfs2_super_lock(struct ocfs2_super *osb,
2770 		     int ex)
2771 {
2772 	int status = 0;
2773 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2774 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2775 
2776 	if (ocfs2_is_hard_readonly(osb))
2777 		return -EROFS;
2778 
2779 	if (ocfs2_mount_local(osb))
2780 		goto bail;
2781 
2782 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
2783 	if (status < 0) {
2784 		mlog_errno(status);
2785 		goto bail;
2786 	}
2787 
2788 	/* The super block lock path is really in the best position to
2789 	 * know when resources covered by the lock need to be
2790 	 * refreshed, so we do it here. Of course, making sense of
2791 	 * everything is up to the caller :) */
2792 	status = ocfs2_should_refresh_lock_res(lockres);
2793 	if (status) {
2794 		status = ocfs2_refresh_slot_info(osb);
2795 
2796 		ocfs2_complete_lock_res_refresh(lockres, status);
2797 
2798 		if (status < 0) {
2799 			ocfs2_cluster_unlock(osb, lockres, level);
2800 			mlog_errno(status);
2801 		}
2802 		ocfs2_track_lock_refresh(lockres);
2803 	}
2804 bail:
2805 	return status;
2806 }
2807 
2808 void ocfs2_super_unlock(struct ocfs2_super *osb,
2809 			int ex)
2810 {
2811 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2812 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2813 
2814 	if (!ocfs2_mount_local(osb))
2815 		ocfs2_cluster_unlock(osb, lockres, level);
2816 }
2817 
2818 int ocfs2_rename_lock(struct ocfs2_super *osb)
2819 {
2820 	int status;
2821 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2822 
2823 	if (ocfs2_is_hard_readonly(osb))
2824 		return -EROFS;
2825 
2826 	if (ocfs2_mount_local(osb))
2827 		return 0;
2828 
2829 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2830 	if (status < 0)
2831 		mlog_errno(status);
2832 
2833 	return status;
2834 }
2835 
2836 void ocfs2_rename_unlock(struct ocfs2_super *osb)
2837 {
2838 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2839 
2840 	if (!ocfs2_mount_local(osb))
2841 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2842 }
2843 
2844 int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex)
2845 {
2846 	int status;
2847 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2848 
2849 	if (ocfs2_is_hard_readonly(osb))
2850 		return -EROFS;
2851 
2852 	if (ocfs2_mount_local(osb))
2853 		return 0;
2854 
2855 	status = ocfs2_cluster_lock(osb, lockres, ex ? LKM_EXMODE : LKM_PRMODE,
2856 				    0, 0);
2857 	if (status < 0)
2858 		mlog(ML_ERROR, "lock on nfs sync lock failed %d\n", status);
2859 
2860 	return status;
2861 }
2862 
2863 void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex)
2864 {
2865 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2866 
2867 	if (!ocfs2_mount_local(osb))
2868 		ocfs2_cluster_unlock(osb, lockres,
2869 				     ex ? LKM_EXMODE : LKM_PRMODE);
2870 }
2871 
2872 int ocfs2_trim_fs_lock(struct ocfs2_super *osb,
2873 		       struct ocfs2_trim_fs_info *info, int trylock)
2874 {
2875 	int status;
2876 	struct ocfs2_trim_fs_lvb *lvb;
2877 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
2878 
2879 	if (info)
2880 		info->tf_valid = 0;
2881 
2882 	if (ocfs2_is_hard_readonly(osb))
2883 		return -EROFS;
2884 
2885 	if (ocfs2_mount_local(osb))
2886 		return 0;
2887 
2888 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX,
2889 				    trylock ? DLM_LKF_NOQUEUE : 0, 0);
2890 	if (status < 0) {
2891 		if (status != -EAGAIN)
2892 			mlog_errno(status);
2893 		return status;
2894 	}
2895 
2896 	if (info) {
2897 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2898 		if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2899 		    lvb->lvb_version == OCFS2_TRIMFS_LVB_VERSION) {
2900 			info->tf_valid = 1;
2901 			info->tf_success = lvb->lvb_success;
2902 			info->tf_nodenum = be32_to_cpu(lvb->lvb_nodenum);
2903 			info->tf_start = be64_to_cpu(lvb->lvb_start);
2904 			info->tf_len = be64_to_cpu(lvb->lvb_len);
2905 			info->tf_minlen = be64_to_cpu(lvb->lvb_minlen);
2906 			info->tf_trimlen = be64_to_cpu(lvb->lvb_trimlen);
2907 		}
2908 	}
2909 
2910 	return status;
2911 }
2912 
2913 void ocfs2_trim_fs_unlock(struct ocfs2_super *osb,
2914 			  struct ocfs2_trim_fs_info *info)
2915 {
2916 	struct ocfs2_trim_fs_lvb *lvb;
2917 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
2918 
2919 	if (ocfs2_mount_local(osb))
2920 		return;
2921 
2922 	if (info) {
2923 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2924 		lvb->lvb_version = OCFS2_TRIMFS_LVB_VERSION;
2925 		lvb->lvb_success = info->tf_success;
2926 		lvb->lvb_nodenum = cpu_to_be32(info->tf_nodenum);
2927 		lvb->lvb_start = cpu_to_be64(info->tf_start);
2928 		lvb->lvb_len = cpu_to_be64(info->tf_len);
2929 		lvb->lvb_minlen = cpu_to_be64(info->tf_minlen);
2930 		lvb->lvb_trimlen = cpu_to_be64(info->tf_trimlen);
2931 	}
2932 
2933 	ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2934 }
2935 
2936 int ocfs2_dentry_lock(struct dentry *dentry, int ex)
2937 {
2938 	int ret;
2939 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2940 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2941 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2942 
2943 	BUG_ON(!dl);
2944 
2945 	if (ocfs2_is_hard_readonly(osb)) {
2946 		if (ex)
2947 			return -EROFS;
2948 		return 0;
2949 	}
2950 
2951 	if (ocfs2_mount_local(osb))
2952 		return 0;
2953 
2954 	ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2955 	if (ret < 0)
2956 		mlog_errno(ret);
2957 
2958 	return ret;
2959 }
2960 
2961 void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2962 {
2963 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2964 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2965 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2966 
2967 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
2968 		ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
2969 }
2970 
2971 /* Reference counting of the dlm debug structure. We want this because
2972  * open references on the debug inodes can live on after a mount, so
2973  * we can't rely on the ocfs2_super to always exist. */
2974 static void ocfs2_dlm_debug_free(struct kref *kref)
2975 {
2976 	struct ocfs2_dlm_debug *dlm_debug;
2977 
2978 	dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
2979 
2980 	kfree(dlm_debug);
2981 }
2982 
2983 void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
2984 {
2985 	if (dlm_debug)
2986 		kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
2987 }
2988 
2989 static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
2990 {
2991 	kref_get(&debug->d_refcnt);
2992 }
2993 
2994 struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
2995 {
2996 	struct ocfs2_dlm_debug *dlm_debug;
2997 
2998 	dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
2999 	if (!dlm_debug) {
3000 		mlog_errno(-ENOMEM);
3001 		goto out;
3002 	}
3003 
3004 	kref_init(&dlm_debug->d_refcnt);
3005 	INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
3006 	dlm_debug->d_locking_state = NULL;
3007 out:
3008 	return dlm_debug;
3009 }
3010 
3011 /* Access to this is arbitrated for us via seq_file->sem. */
3012 struct ocfs2_dlm_seq_priv {
3013 	struct ocfs2_dlm_debug *p_dlm_debug;
3014 	struct ocfs2_lock_res p_iter_res;
3015 	struct ocfs2_lock_res p_tmp_res;
3016 };
3017 
3018 static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
3019 						 struct ocfs2_dlm_seq_priv *priv)
3020 {
3021 	struct ocfs2_lock_res *iter, *ret = NULL;
3022 	struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
3023 
3024 	assert_spin_locked(&ocfs2_dlm_tracking_lock);
3025 
3026 	list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
3027 		/* discover the head of the list */
3028 		if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
3029 			mlog(0, "End of list found, %p\n", ret);
3030 			break;
3031 		}
3032 
3033 		/* We track our "dummy" iteration lockres' by a NULL
3034 		 * l_ops field. */
3035 		if (iter->l_ops != NULL) {
3036 			ret = iter;
3037 			break;
3038 		}
3039 	}
3040 
3041 	return ret;
3042 }
3043 
3044 static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
3045 {
3046 	struct ocfs2_dlm_seq_priv *priv = m->private;
3047 	struct ocfs2_lock_res *iter;
3048 
3049 	spin_lock(&ocfs2_dlm_tracking_lock);
3050 	iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
3051 	if (iter) {
3052 		/* Since lockres' have the lifetime of their container
3053 		 * (which can be inodes, ocfs2_supers, etc) we want to
3054 		 * copy this out to a temporary lockres while still
3055 		 * under the spinlock. Obviously after this we can't
3056 		 * trust any pointers on the copy returned, but that's
3057 		 * ok as the information we want isn't typically held
3058 		 * in them. */
3059 		priv->p_tmp_res = *iter;
3060 		iter = &priv->p_tmp_res;
3061 	}
3062 	spin_unlock(&ocfs2_dlm_tracking_lock);
3063 
3064 	return iter;
3065 }
3066 
3067 static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
3068 {
3069 }
3070 
3071 static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
3072 {
3073 	struct ocfs2_dlm_seq_priv *priv = m->private;
3074 	struct ocfs2_lock_res *iter = v;
3075 	struct ocfs2_lock_res *dummy = &priv->p_iter_res;
3076 
3077 	spin_lock(&ocfs2_dlm_tracking_lock);
3078 	iter = ocfs2_dlm_next_res(iter, priv);
3079 	list_del_init(&dummy->l_debug_list);
3080 	if (iter) {
3081 		list_add(&dummy->l_debug_list, &iter->l_debug_list);
3082 		priv->p_tmp_res = *iter;
3083 		iter = &priv->p_tmp_res;
3084 	}
3085 	spin_unlock(&ocfs2_dlm_tracking_lock);
3086 
3087 	return iter;
3088 }
3089 
3090 /*
3091  * Version is used by debugfs.ocfs2 to determine the format being used
3092  *
3093  * New in version 2
3094  *	- Lock stats printed
3095  * New in version 3
3096  *	- Max time in lock stats is in usecs (instead of nsecs)
3097  */
3098 #define OCFS2_DLM_DEBUG_STR_VERSION 3
3099 static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
3100 {
3101 	int i;
3102 	char *lvb;
3103 	struct ocfs2_lock_res *lockres = v;
3104 
3105 	if (!lockres)
3106 		return -EINVAL;
3107 
3108 	seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
3109 
3110 	if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
3111 		seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
3112 			   lockres->l_name,
3113 			   (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
3114 	else
3115 		seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
3116 
3117 	seq_printf(m, "%d\t"
3118 		   "0x%lx\t"
3119 		   "0x%x\t"
3120 		   "0x%x\t"
3121 		   "%u\t"
3122 		   "%u\t"
3123 		   "%d\t"
3124 		   "%d\t",
3125 		   lockres->l_level,
3126 		   lockres->l_flags,
3127 		   lockres->l_action,
3128 		   lockres->l_unlock_action,
3129 		   lockres->l_ro_holders,
3130 		   lockres->l_ex_holders,
3131 		   lockres->l_requested,
3132 		   lockres->l_blocking);
3133 
3134 	/* Dump the raw LVB */
3135 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
3136 	for(i = 0; i < DLM_LVB_LEN; i++)
3137 		seq_printf(m, "0x%x\t", lvb[i]);
3138 
3139 #ifdef CONFIG_OCFS2_FS_STATS
3140 # define lock_num_prmode(_l)		((_l)->l_lock_prmode.ls_gets)
3141 # define lock_num_exmode(_l)		((_l)->l_lock_exmode.ls_gets)
3142 # define lock_num_prmode_failed(_l)	((_l)->l_lock_prmode.ls_fail)
3143 # define lock_num_exmode_failed(_l)	((_l)->l_lock_exmode.ls_fail)
3144 # define lock_total_prmode(_l)		((_l)->l_lock_prmode.ls_total)
3145 # define lock_total_exmode(_l)		((_l)->l_lock_exmode.ls_total)
3146 # define lock_max_prmode(_l)		((_l)->l_lock_prmode.ls_max)
3147 # define lock_max_exmode(_l)		((_l)->l_lock_exmode.ls_max)
3148 # define lock_refresh(_l)		((_l)->l_lock_refresh)
3149 #else
3150 # define lock_num_prmode(_l)		(0)
3151 # define lock_num_exmode(_l)		(0)
3152 # define lock_num_prmode_failed(_l)	(0)
3153 # define lock_num_exmode_failed(_l)	(0)
3154 # define lock_total_prmode(_l)		(0ULL)
3155 # define lock_total_exmode(_l)		(0ULL)
3156 # define lock_max_prmode(_l)		(0)
3157 # define lock_max_exmode(_l)		(0)
3158 # define lock_refresh(_l)		(0)
3159 #endif
3160 	/* The following seq_print was added in version 2 of this output */
3161 	seq_printf(m, "%u\t"
3162 		   "%u\t"
3163 		   "%u\t"
3164 		   "%u\t"
3165 		   "%llu\t"
3166 		   "%llu\t"
3167 		   "%u\t"
3168 		   "%u\t"
3169 		   "%u\t",
3170 		   lock_num_prmode(lockres),
3171 		   lock_num_exmode(lockres),
3172 		   lock_num_prmode_failed(lockres),
3173 		   lock_num_exmode_failed(lockres),
3174 		   lock_total_prmode(lockres),
3175 		   lock_total_exmode(lockres),
3176 		   lock_max_prmode(lockres),
3177 		   lock_max_exmode(lockres),
3178 		   lock_refresh(lockres));
3179 
3180 	/* End the line */
3181 	seq_printf(m, "\n");
3182 	return 0;
3183 }
3184 
3185 static const struct seq_operations ocfs2_dlm_seq_ops = {
3186 	.start =	ocfs2_dlm_seq_start,
3187 	.stop =		ocfs2_dlm_seq_stop,
3188 	.next =		ocfs2_dlm_seq_next,
3189 	.show =		ocfs2_dlm_seq_show,
3190 };
3191 
3192 static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
3193 {
3194 	struct seq_file *seq = file->private_data;
3195 	struct ocfs2_dlm_seq_priv *priv = seq->private;
3196 	struct ocfs2_lock_res *res = &priv->p_iter_res;
3197 
3198 	ocfs2_remove_lockres_tracking(res);
3199 	ocfs2_put_dlm_debug(priv->p_dlm_debug);
3200 	return seq_release_private(inode, file);
3201 }
3202 
3203 static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
3204 {
3205 	struct ocfs2_dlm_seq_priv *priv;
3206 	struct ocfs2_super *osb;
3207 
3208 	priv = __seq_open_private(file, &ocfs2_dlm_seq_ops, sizeof(*priv));
3209 	if (!priv) {
3210 		mlog_errno(-ENOMEM);
3211 		return -ENOMEM;
3212 	}
3213 
3214 	osb = inode->i_private;
3215 	ocfs2_get_dlm_debug(osb->osb_dlm_debug);
3216 	priv->p_dlm_debug = osb->osb_dlm_debug;
3217 	INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
3218 
3219 	ocfs2_add_lockres_tracking(&priv->p_iter_res,
3220 				   priv->p_dlm_debug);
3221 
3222 	return 0;
3223 }
3224 
3225 static const struct file_operations ocfs2_dlm_debug_fops = {
3226 	.open =		ocfs2_dlm_debug_open,
3227 	.release =	ocfs2_dlm_debug_release,
3228 	.read =		seq_read,
3229 	.llseek =	seq_lseek,
3230 };
3231 
3232 static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
3233 {
3234 	int ret = 0;
3235 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
3236 
3237 	dlm_debug->d_locking_state = debugfs_create_file("locking_state",
3238 							 S_IFREG|S_IRUSR,
3239 							 osb->osb_debug_root,
3240 							 osb,
3241 							 &ocfs2_dlm_debug_fops);
3242 	if (!dlm_debug->d_locking_state) {
3243 		ret = -EINVAL;
3244 		mlog(ML_ERROR,
3245 		     "Unable to create locking state debugfs file.\n");
3246 		goto out;
3247 	}
3248 
3249 	ocfs2_get_dlm_debug(dlm_debug);
3250 out:
3251 	return ret;
3252 }
3253 
3254 static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
3255 {
3256 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
3257 
3258 	if (dlm_debug) {
3259 		debugfs_remove(dlm_debug->d_locking_state);
3260 		ocfs2_put_dlm_debug(dlm_debug);
3261 	}
3262 }
3263 
3264 int ocfs2_dlm_init(struct ocfs2_super *osb)
3265 {
3266 	int status = 0;
3267 	struct ocfs2_cluster_connection *conn = NULL;
3268 
3269 	if (ocfs2_mount_local(osb)) {
3270 		osb->node_num = 0;
3271 		goto local;
3272 	}
3273 
3274 	status = ocfs2_dlm_init_debug(osb);
3275 	if (status < 0) {
3276 		mlog_errno(status);
3277 		goto bail;
3278 	}
3279 
3280 	/* launch downconvert thread */
3281 	osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc-%s",
3282 			osb->uuid_str);
3283 	if (IS_ERR(osb->dc_task)) {
3284 		status = PTR_ERR(osb->dc_task);
3285 		osb->dc_task = NULL;
3286 		mlog_errno(status);
3287 		goto bail;
3288 	}
3289 
3290 	/* for now, uuid == domain */
3291 	status = ocfs2_cluster_connect(osb->osb_cluster_stack,
3292 				       osb->osb_cluster_name,
3293 				       strlen(osb->osb_cluster_name),
3294 				       osb->uuid_str,
3295 				       strlen(osb->uuid_str),
3296 				       &lproto, ocfs2_do_node_down, osb,
3297 				       &conn);
3298 	if (status) {
3299 		mlog_errno(status);
3300 		goto bail;
3301 	}
3302 
3303 	status = ocfs2_cluster_this_node(conn, &osb->node_num);
3304 	if (status < 0) {
3305 		mlog_errno(status);
3306 		mlog(ML_ERROR,
3307 		     "could not find this host's node number\n");
3308 		ocfs2_cluster_disconnect(conn, 0);
3309 		goto bail;
3310 	}
3311 
3312 local:
3313 	ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
3314 	ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
3315 	ocfs2_nfs_sync_lock_res_init(&osb->osb_nfs_sync_lockres, osb);
3316 	ocfs2_orphan_scan_lock_res_init(&osb->osb_orphan_scan.os_lockres, osb);
3317 
3318 	osb->cconn = conn;
3319 bail:
3320 	if (status < 0) {
3321 		ocfs2_dlm_shutdown_debug(osb);
3322 		if (osb->dc_task)
3323 			kthread_stop(osb->dc_task);
3324 	}
3325 
3326 	return status;
3327 }
3328 
3329 void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
3330 			int hangup_pending)
3331 {
3332 	ocfs2_drop_osb_locks(osb);
3333 
3334 	/*
3335 	 * Now that we have dropped all locks and ocfs2_dismount_volume()
3336 	 * has disabled recovery, the DLM won't be talking to us.  It's
3337 	 * safe to tear things down before disconnecting the cluster.
3338 	 */
3339 
3340 	if (osb->dc_task) {
3341 		kthread_stop(osb->dc_task);
3342 		osb->dc_task = NULL;
3343 	}
3344 
3345 	ocfs2_lock_res_free(&osb->osb_super_lockres);
3346 	ocfs2_lock_res_free(&osb->osb_rename_lockres);
3347 	ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres);
3348 	ocfs2_lock_res_free(&osb->osb_orphan_scan.os_lockres);
3349 
3350 	ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
3351 	osb->cconn = NULL;
3352 
3353 	ocfs2_dlm_shutdown_debug(osb);
3354 }
3355 
3356 static int ocfs2_drop_lock(struct ocfs2_super *osb,
3357 			   struct ocfs2_lock_res *lockres)
3358 {
3359 	int ret;
3360 	unsigned long flags;
3361 	u32 lkm_flags = 0;
3362 
3363 	/* We didn't get anywhere near actually using this lockres. */
3364 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
3365 		goto out;
3366 
3367 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
3368 		lkm_flags |= DLM_LKF_VALBLK;
3369 
3370 	spin_lock_irqsave(&lockres->l_lock, flags);
3371 
3372 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
3373 			"lockres %s, flags 0x%lx\n",
3374 			lockres->l_name, lockres->l_flags);
3375 
3376 	while (lockres->l_flags & OCFS2_LOCK_BUSY) {
3377 		mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
3378 		     "%u, unlock_action = %u\n",
3379 		     lockres->l_name, lockres->l_flags, lockres->l_action,
3380 		     lockres->l_unlock_action);
3381 
3382 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3383 
3384 		/* XXX: Today we just wait on any busy
3385 		 * locks... Perhaps we need to cancel converts in the
3386 		 * future? */
3387 		ocfs2_wait_on_busy_lock(lockres);
3388 
3389 		spin_lock_irqsave(&lockres->l_lock, flags);
3390 	}
3391 
3392 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3393 		if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
3394 		    lockres->l_level == DLM_LOCK_EX &&
3395 		    !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3396 			lockres->l_ops->set_lvb(lockres);
3397 	}
3398 
3399 	if (lockres->l_flags & OCFS2_LOCK_BUSY)
3400 		mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
3401 		     lockres->l_name);
3402 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
3403 		mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
3404 
3405 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
3406 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3407 		goto out;
3408 	}
3409 
3410 	lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
3411 
3412 	/* make sure we never get here while waiting for an ast to
3413 	 * fire. */
3414 	BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
3415 
3416 	/* is this necessary? */
3417 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3418 	lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
3419 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3420 
3421 	mlog(0, "lock %s\n", lockres->l_name);
3422 
3423 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags);
3424 	if (ret) {
3425 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3426 		mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
3427 		ocfs2_dlm_dump_lksb(&lockres->l_lksb);
3428 		BUG();
3429 	}
3430 	mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n",
3431 	     lockres->l_name);
3432 
3433 	ocfs2_wait_on_busy_lock(lockres);
3434 out:
3435 	return 0;
3436 }
3437 
3438 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3439 				       struct ocfs2_lock_res *lockres);
3440 
3441 /* Mark the lockres as being dropped. It will no longer be
3442  * queued if blocking, but we still may have to wait on it
3443  * being dequeued from the downconvert thread before we can consider
3444  * it safe to drop.
3445  *
3446  * You can *not* attempt to call cluster_lock on this lockres anymore. */
3447 void ocfs2_mark_lockres_freeing(struct ocfs2_super *osb,
3448 				struct ocfs2_lock_res *lockres)
3449 {
3450 	int status;
3451 	struct ocfs2_mask_waiter mw;
3452 	unsigned long flags, flags2;
3453 
3454 	ocfs2_init_mask_waiter(&mw);
3455 
3456 	spin_lock_irqsave(&lockres->l_lock, flags);
3457 	lockres->l_flags |= OCFS2_LOCK_FREEING;
3458 	if (lockres->l_flags & OCFS2_LOCK_QUEUED && current == osb->dc_task) {
3459 		/*
3460 		 * We know the downconvert is queued but not in progress
3461 		 * because we are the downconvert thread and processing
3462 		 * different lock. So we can just remove the lock from the
3463 		 * queue. This is not only an optimization but also a way
3464 		 * to avoid the following deadlock:
3465 		 *   ocfs2_dentry_post_unlock()
3466 		 *     ocfs2_dentry_lock_put()
3467 		 *       ocfs2_drop_dentry_lock()
3468 		 *         iput()
3469 		 *           ocfs2_evict_inode()
3470 		 *             ocfs2_clear_inode()
3471 		 *               ocfs2_mark_lockres_freeing()
3472 		 *                 ... blocks waiting for OCFS2_LOCK_QUEUED
3473 		 *                 since we are the downconvert thread which
3474 		 *                 should clear the flag.
3475 		 */
3476 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3477 		spin_lock_irqsave(&osb->dc_task_lock, flags2);
3478 		list_del_init(&lockres->l_blocked_list);
3479 		osb->blocked_lock_count--;
3480 		spin_unlock_irqrestore(&osb->dc_task_lock, flags2);
3481 		/*
3482 		 * Warn if we recurse into another post_unlock call.  Strictly
3483 		 * speaking it isn't a problem but we need to be careful if
3484 		 * that happens (stack overflow, deadlocks, ...) so warn if
3485 		 * ocfs2 grows a path for which this can happen.
3486 		 */
3487 		WARN_ON_ONCE(lockres->l_ops->post_unlock);
3488 		/* Since the lock is freeing we don't do much in the fn below */
3489 		ocfs2_process_blocked_lock(osb, lockres);
3490 		return;
3491 	}
3492 	while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
3493 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
3494 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3495 
3496 		mlog(0, "Waiting on lockres %s\n", lockres->l_name);
3497 
3498 		status = ocfs2_wait_for_mask(&mw);
3499 		if (status)
3500 			mlog_errno(status);
3501 
3502 		spin_lock_irqsave(&lockres->l_lock, flags);
3503 	}
3504 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3505 }
3506 
3507 void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
3508 			       struct ocfs2_lock_res *lockres)
3509 {
3510 	int ret;
3511 
3512 	ocfs2_mark_lockres_freeing(osb, lockres);
3513 	ret = ocfs2_drop_lock(osb, lockres);
3514 	if (ret)
3515 		mlog_errno(ret);
3516 }
3517 
3518 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
3519 {
3520 	ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
3521 	ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
3522 	ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres);
3523 	ocfs2_simple_drop_lockres(osb, &osb->osb_orphan_scan.os_lockres);
3524 }
3525 
3526 int ocfs2_drop_inode_locks(struct inode *inode)
3527 {
3528 	int status, err;
3529 
3530 	/* No need to call ocfs2_mark_lockres_freeing here -
3531 	 * ocfs2_clear_inode has done it for us. */
3532 
3533 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3534 			      &OCFS2_I(inode)->ip_open_lockres);
3535 	if (err < 0)
3536 		mlog_errno(err);
3537 
3538 	status = err;
3539 
3540 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3541 			      &OCFS2_I(inode)->ip_inode_lockres);
3542 	if (err < 0)
3543 		mlog_errno(err);
3544 	if (err < 0 && !status)
3545 		status = err;
3546 
3547 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3548 			      &OCFS2_I(inode)->ip_rw_lockres);
3549 	if (err < 0)
3550 		mlog_errno(err);
3551 	if (err < 0 && !status)
3552 		status = err;
3553 
3554 	return status;
3555 }
3556 
3557 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
3558 					      int new_level)
3559 {
3560 	assert_spin_locked(&lockres->l_lock);
3561 
3562 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
3563 
3564 	if (lockres->l_level <= new_level) {
3565 		mlog(ML_ERROR, "lockres %s, lvl %d <= %d, blcklst %d, mask %d, "
3566 		     "type %d, flags 0x%lx, hold %d %d, act %d %d, req %d, "
3567 		     "block %d, pgen %d\n", lockres->l_name, lockres->l_level,
3568 		     new_level, list_empty(&lockres->l_blocked_list),
3569 		     list_empty(&lockres->l_mask_waiters), lockres->l_type,
3570 		     lockres->l_flags, lockres->l_ro_holders,
3571 		     lockres->l_ex_holders, lockres->l_action,
3572 		     lockres->l_unlock_action, lockres->l_requested,
3573 		     lockres->l_blocking, lockres->l_pending_gen);
3574 		BUG();
3575 	}
3576 
3577 	mlog(ML_BASTS, "lockres %s, level %d => %d, blocking %d\n",
3578 	     lockres->l_name, lockres->l_level, new_level, lockres->l_blocking);
3579 
3580 	lockres->l_action = OCFS2_AST_DOWNCONVERT;
3581 	lockres->l_requested = new_level;
3582 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3583 	return lockres_set_pending(lockres);
3584 }
3585 
3586 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
3587 				  struct ocfs2_lock_res *lockres,
3588 				  int new_level,
3589 				  int lvb,
3590 				  unsigned int generation)
3591 {
3592 	int ret;
3593 	u32 dlm_flags = DLM_LKF_CONVERT;
3594 
3595 	mlog(ML_BASTS, "lockres %s, level %d => %d\n", lockres->l_name,
3596 	     lockres->l_level, new_level);
3597 
3598 	/*
3599 	 * On DLM_LKF_VALBLK, fsdlm behaves differently with o2cb. It always
3600 	 * expects DLM_LKF_VALBLK being set if the LKB has LVB, so that
3601 	 * we can recover correctly from node failure. Otherwise, we may get
3602 	 * invalid LVB in LKB, but without DLM_SBF_VALNOTVALID being set.
3603 	 */
3604 	if (!ocfs2_is_o2cb_active() &&
3605 	    lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
3606 		lvb = 1;
3607 
3608 	if (lvb)
3609 		dlm_flags |= DLM_LKF_VALBLK;
3610 
3611 	ret = ocfs2_dlm_lock(osb->cconn,
3612 			     new_level,
3613 			     &lockres->l_lksb,
3614 			     dlm_flags,
3615 			     lockres->l_name,
3616 			     OCFS2_LOCK_ID_MAX_LEN - 1);
3617 	lockres_clear_pending(lockres, generation, osb);
3618 	if (ret) {
3619 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
3620 		ocfs2_recover_from_dlm_error(lockres, 1);
3621 		goto bail;
3622 	}
3623 
3624 	ret = 0;
3625 bail:
3626 	return ret;
3627 }
3628 
3629 /* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
3630 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
3631 				        struct ocfs2_lock_res *lockres)
3632 {
3633 	assert_spin_locked(&lockres->l_lock);
3634 
3635 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
3636 		/* If we're already trying to cancel a lock conversion
3637 		 * then just drop the spinlock and allow the caller to
3638 		 * requeue this lock. */
3639 		mlog(ML_BASTS, "lockres %s, skip convert\n", lockres->l_name);
3640 		return 0;
3641 	}
3642 
3643 	/* were we in a convert when we got the bast fire? */
3644 	BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
3645 	       lockres->l_action != OCFS2_AST_DOWNCONVERT);
3646 	/* set things up for the unlockast to know to just
3647 	 * clear out the ast_action and unset busy, etc. */
3648 	lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
3649 
3650 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
3651 			"lock %s, invalid flags: 0x%lx\n",
3652 			lockres->l_name, lockres->l_flags);
3653 
3654 	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3655 
3656 	return 1;
3657 }
3658 
3659 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
3660 				struct ocfs2_lock_res *lockres)
3661 {
3662 	int ret;
3663 
3664 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
3665 			       DLM_LKF_CANCEL);
3666 	if (ret) {
3667 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3668 		ocfs2_recover_from_dlm_error(lockres, 0);
3669 	}
3670 
3671 	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3672 
3673 	return ret;
3674 }
3675 
3676 static int ocfs2_unblock_lock(struct ocfs2_super *osb,
3677 			      struct ocfs2_lock_res *lockres,
3678 			      struct ocfs2_unblock_ctl *ctl)
3679 {
3680 	unsigned long flags;
3681 	int blocking;
3682 	int new_level;
3683 	int level;
3684 	int ret = 0;
3685 	int set_lvb = 0;
3686 	unsigned int gen;
3687 
3688 	spin_lock_irqsave(&lockres->l_lock, flags);
3689 
3690 recheck:
3691 	/*
3692 	 * Is it still blocking? If not, we have no more work to do.
3693 	 */
3694 	if (!(lockres->l_flags & OCFS2_LOCK_BLOCKED)) {
3695 		BUG_ON(lockres->l_blocking != DLM_LOCK_NL);
3696 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3697 		ret = 0;
3698 		goto leave;
3699 	}
3700 
3701 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
3702 		/* XXX
3703 		 * This is a *big* race.  The OCFS2_LOCK_PENDING flag
3704 		 * exists entirely for one reason - another thread has set
3705 		 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
3706 		 *
3707 		 * If we do ocfs2_cancel_convert() before the other thread
3708 		 * calls dlm_lock(), our cancel will do nothing.  We will
3709 		 * get no ast, and we will have no way of knowing the
3710 		 * cancel failed.  Meanwhile, the other thread will call
3711 		 * into dlm_lock() and wait...forever.
3712 		 *
3713 		 * Why forever?  Because another node has asked for the
3714 		 * lock first; that's why we're here in unblock_lock().
3715 		 *
3716 		 * The solution is OCFS2_LOCK_PENDING.  When PENDING is
3717 		 * set, we just requeue the unblock.  Only when the other
3718 		 * thread has called dlm_lock() and cleared PENDING will
3719 		 * we then cancel their request.
3720 		 *
3721 		 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
3722 		 * at the same time they set OCFS2_DLM_BUSY.  They must
3723 		 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
3724 		 */
3725 		if (lockres->l_flags & OCFS2_LOCK_PENDING) {
3726 			mlog(ML_BASTS, "lockres %s, ReQ: Pending\n",
3727 			     lockres->l_name);
3728 			goto leave_requeue;
3729 		}
3730 
3731 		ctl->requeue = 1;
3732 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
3733 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3734 		if (ret) {
3735 			ret = ocfs2_cancel_convert(osb, lockres);
3736 			if (ret < 0)
3737 				mlog_errno(ret);
3738 		}
3739 		goto leave;
3740 	}
3741 
3742 	/*
3743 	 * This prevents livelocks. OCFS2_LOCK_UPCONVERT_FINISHING flag is
3744 	 * set when the ast is received for an upconvert just before the
3745 	 * OCFS2_LOCK_BUSY flag is cleared. Now if the fs received a bast
3746 	 * on the heels of the ast, we want to delay the downconvert just
3747 	 * enough to allow the up requestor to do its task. Because this
3748 	 * lock is in the blocked queue, the lock will be downconverted
3749 	 * as soon as the requestor is done with the lock.
3750 	 */
3751 	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING)
3752 		goto leave_requeue;
3753 
3754 	/*
3755 	 * How can we block and yet be at NL?  We were trying to upconvert
3756 	 * from NL and got canceled.  The code comes back here, and now
3757 	 * we notice and clear BLOCKING.
3758 	 */
3759 	if (lockres->l_level == DLM_LOCK_NL) {
3760 		BUG_ON(lockres->l_ex_holders || lockres->l_ro_holders);
3761 		mlog(ML_BASTS, "lockres %s, Aborting dc\n", lockres->l_name);
3762 		lockres->l_blocking = DLM_LOCK_NL;
3763 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
3764 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3765 		goto leave;
3766 	}
3767 
3768 	/* if we're blocking an exclusive and we have *any* holders,
3769 	 * then requeue. */
3770 	if ((lockres->l_blocking == DLM_LOCK_EX)
3771 	    && (lockres->l_ex_holders || lockres->l_ro_holders)) {
3772 		mlog(ML_BASTS, "lockres %s, ReQ: EX/PR Holders %u,%u\n",
3773 		     lockres->l_name, lockres->l_ex_holders,
3774 		     lockres->l_ro_holders);
3775 		goto leave_requeue;
3776 	}
3777 
3778 	/* If it's a PR we're blocking, then only
3779 	 * requeue if we've got any EX holders */
3780 	if (lockres->l_blocking == DLM_LOCK_PR &&
3781 	    lockres->l_ex_holders) {
3782 		mlog(ML_BASTS, "lockres %s, ReQ: EX Holders %u\n",
3783 		     lockres->l_name, lockres->l_ex_holders);
3784 		goto leave_requeue;
3785 	}
3786 
3787 	/*
3788 	 * Can we get a lock in this state if the holder counts are
3789 	 * zero? The meta data unblock code used to check this.
3790 	 */
3791 	if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
3792 	    && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) {
3793 		mlog(ML_BASTS, "lockres %s, ReQ: Lock Refreshing\n",
3794 		     lockres->l_name);
3795 		goto leave_requeue;
3796 	}
3797 
3798 	new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
3799 
3800 	if (lockres->l_ops->check_downconvert
3801 	    && !lockres->l_ops->check_downconvert(lockres, new_level)) {
3802 		mlog(ML_BASTS, "lockres %s, ReQ: Checkpointing\n",
3803 		     lockres->l_name);
3804 		goto leave_requeue;
3805 	}
3806 
3807 	/* If we get here, then we know that there are no more
3808 	 * incompatible holders (and anyone asking for an incompatible
3809 	 * lock is blocked). We can now downconvert the lock */
3810 	if (!lockres->l_ops->downconvert_worker)
3811 		goto downconvert;
3812 
3813 	/* Some lockres types want to do a bit of work before
3814 	 * downconverting a lock. Allow that here. The worker function
3815 	 * may sleep, so we save off a copy of what we're blocking as
3816 	 * it may change while we're not holding the spin lock. */
3817 	blocking = lockres->l_blocking;
3818 	level = lockres->l_level;
3819 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3820 
3821 	ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
3822 
3823 	if (ctl->unblock_action == UNBLOCK_STOP_POST) {
3824 		mlog(ML_BASTS, "lockres %s, UNBLOCK_STOP_POST\n",
3825 		     lockres->l_name);
3826 		goto leave;
3827 	}
3828 
3829 	spin_lock_irqsave(&lockres->l_lock, flags);
3830 	if ((blocking != lockres->l_blocking) || (level != lockres->l_level)) {
3831 		/* If this changed underneath us, then we can't drop
3832 		 * it just yet. */
3833 		mlog(ML_BASTS, "lockres %s, block=%d:%d, level=%d:%d, "
3834 		     "Recheck\n", lockres->l_name, blocking,
3835 		     lockres->l_blocking, level, lockres->l_level);
3836 		goto recheck;
3837 	}
3838 
3839 downconvert:
3840 	ctl->requeue = 0;
3841 
3842 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3843 		if (lockres->l_level == DLM_LOCK_EX)
3844 			set_lvb = 1;
3845 
3846 		/*
3847 		 * We only set the lvb if the lock has been fully
3848 		 * refreshed - otherwise we risk setting stale
3849 		 * data. Otherwise, there's no need to actually clear
3850 		 * out the lvb here as it's value is still valid.
3851 		 */
3852 		if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3853 			lockres->l_ops->set_lvb(lockres);
3854 	}
3855 
3856 	gen = ocfs2_prepare_downconvert(lockres, new_level);
3857 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3858 	ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
3859 				     gen);
3860 
3861 leave:
3862 	if (ret)
3863 		mlog_errno(ret);
3864 	return ret;
3865 
3866 leave_requeue:
3867 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3868 	ctl->requeue = 1;
3869 
3870 	return 0;
3871 }
3872 
3873 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
3874 				     int blocking)
3875 {
3876 	struct inode *inode;
3877 	struct address_space *mapping;
3878 	struct ocfs2_inode_info *oi;
3879 
3880        	inode = ocfs2_lock_res_inode(lockres);
3881 	mapping = inode->i_mapping;
3882 
3883 	if (S_ISDIR(inode->i_mode)) {
3884 		oi = OCFS2_I(inode);
3885 		oi->ip_dir_lock_gen++;
3886 		mlog(0, "generation: %u\n", oi->ip_dir_lock_gen);
3887 		goto out;
3888 	}
3889 
3890 	if (!S_ISREG(inode->i_mode))
3891 		goto out;
3892 
3893 	/*
3894 	 * We need this before the filemap_fdatawrite() so that it can
3895 	 * transfer the dirty bit from the PTE to the
3896 	 * page. Unfortunately this means that even for EX->PR
3897 	 * downconverts, we'll lose our mappings and have to build
3898 	 * them up again.
3899 	 */
3900 	unmap_mapping_range(mapping, 0, 0, 0);
3901 
3902 	if (filemap_fdatawrite(mapping)) {
3903 		mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
3904 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
3905 	}
3906 	sync_mapping_buffers(mapping);
3907 	if (blocking == DLM_LOCK_EX) {
3908 		truncate_inode_pages(mapping, 0);
3909 	} else {
3910 		/* We only need to wait on the I/O if we're not also
3911 		 * truncating pages because truncate_inode_pages waits
3912 		 * for us above. We don't truncate pages if we're
3913 		 * blocking anything < EXMODE because we want to keep
3914 		 * them around in that case. */
3915 		filemap_fdatawait(mapping);
3916 	}
3917 
3918 	forget_all_cached_acls(inode);
3919 
3920 out:
3921 	return UNBLOCK_CONTINUE;
3922 }
3923 
3924 static int ocfs2_ci_checkpointed(struct ocfs2_caching_info *ci,
3925 				 struct ocfs2_lock_res *lockres,
3926 				 int new_level)
3927 {
3928 	int checkpointed = ocfs2_ci_fully_checkpointed(ci);
3929 
3930 	BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
3931 	BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
3932 
3933 	if (checkpointed)
3934 		return 1;
3935 
3936 	ocfs2_start_checkpoint(OCFS2_SB(ocfs2_metadata_cache_get_super(ci)));
3937 	return 0;
3938 }
3939 
3940 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
3941 					int new_level)
3942 {
3943 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3944 
3945 	return ocfs2_ci_checkpointed(INODE_CACHE(inode), lockres, new_level);
3946 }
3947 
3948 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
3949 {
3950 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3951 
3952 	__ocfs2_stuff_meta_lvb(inode);
3953 }
3954 
3955 /*
3956  * Does the final reference drop on our dentry lock. Right now this
3957  * happens in the downconvert thread, but we could choose to simplify the
3958  * dlmglue API and push these off to the ocfs2_wq in the future.
3959  */
3960 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
3961 				     struct ocfs2_lock_res *lockres)
3962 {
3963 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3964 	ocfs2_dentry_lock_put(osb, dl);
3965 }
3966 
3967 /*
3968  * d_delete() matching dentries before the lock downconvert.
3969  *
3970  * At this point, any process waiting to destroy the
3971  * dentry_lock due to last ref count is stopped by the
3972  * OCFS2_LOCK_QUEUED flag.
3973  *
3974  * We have two potential problems
3975  *
3976  * 1) If we do the last reference drop on our dentry_lock (via dput)
3977  *    we'll wind up in ocfs2_release_dentry_lock(), waiting on
3978  *    the downconvert to finish. Instead we take an elevated
3979  *    reference and push the drop until after we've completed our
3980  *    unblock processing.
3981  *
3982  * 2) There might be another process with a final reference,
3983  *    waiting on us to finish processing. If this is the case, we
3984  *    detect it and exit out - there's no more dentries anyway.
3985  */
3986 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
3987 				       int blocking)
3988 {
3989 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3990 	struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
3991 	struct dentry *dentry;
3992 	unsigned long flags;
3993 	int extra_ref = 0;
3994 
3995 	/*
3996 	 * This node is blocking another node from getting a read
3997 	 * lock. This happens when we've renamed within a
3998 	 * directory. We've forced the other nodes to d_delete(), but
3999 	 * we never actually dropped our lock because it's still
4000 	 * valid. The downconvert code will retain a PR for this node,
4001 	 * so there's no further work to do.
4002 	 */
4003 	if (blocking == DLM_LOCK_PR)
4004 		return UNBLOCK_CONTINUE;
4005 
4006 	/*
4007 	 * Mark this inode as potentially orphaned. The code in
4008 	 * ocfs2_delete_inode() will figure out whether it actually
4009 	 * needs to be freed or not.
4010 	 */
4011 	spin_lock(&oi->ip_lock);
4012 	oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
4013 	spin_unlock(&oi->ip_lock);
4014 
4015 	/*
4016 	 * Yuck. We need to make sure however that the check of
4017 	 * OCFS2_LOCK_FREEING and the extra reference are atomic with
4018 	 * respect to a reference decrement or the setting of that
4019 	 * flag.
4020 	 */
4021 	spin_lock_irqsave(&lockres->l_lock, flags);
4022 	spin_lock(&dentry_attach_lock);
4023 	if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
4024 	    && dl->dl_count) {
4025 		dl->dl_count++;
4026 		extra_ref = 1;
4027 	}
4028 	spin_unlock(&dentry_attach_lock);
4029 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4030 
4031 	mlog(0, "extra_ref = %d\n", extra_ref);
4032 
4033 	/*
4034 	 * We have a process waiting on us in ocfs2_dentry_iput(),
4035 	 * which means we can't have any more outstanding
4036 	 * aliases. There's no need to do any more work.
4037 	 */
4038 	if (!extra_ref)
4039 		return UNBLOCK_CONTINUE;
4040 
4041 	spin_lock(&dentry_attach_lock);
4042 	while (1) {
4043 		dentry = ocfs2_find_local_alias(dl->dl_inode,
4044 						dl->dl_parent_blkno, 1);
4045 		if (!dentry)
4046 			break;
4047 		spin_unlock(&dentry_attach_lock);
4048 
4049 		if (S_ISDIR(dl->dl_inode->i_mode))
4050 			shrink_dcache_parent(dentry);
4051 
4052 		mlog(0, "d_delete(%pd);\n", dentry);
4053 
4054 		/*
4055 		 * The following dcache calls may do an
4056 		 * iput(). Normally we don't want that from the
4057 		 * downconverting thread, but in this case it's ok
4058 		 * because the requesting node already has an
4059 		 * exclusive lock on the inode, so it can't be queued
4060 		 * for a downconvert.
4061 		 */
4062 		d_delete(dentry);
4063 		dput(dentry);
4064 
4065 		spin_lock(&dentry_attach_lock);
4066 	}
4067 	spin_unlock(&dentry_attach_lock);
4068 
4069 	/*
4070 	 * If we are the last holder of this dentry lock, there is no
4071 	 * reason to downconvert so skip straight to the unlock.
4072 	 */
4073 	if (dl->dl_count == 1)
4074 		return UNBLOCK_STOP_POST;
4075 
4076 	return UNBLOCK_CONTINUE_POST;
4077 }
4078 
4079 static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
4080 					    int new_level)
4081 {
4082 	struct ocfs2_refcount_tree *tree =
4083 				ocfs2_lock_res_refcount_tree(lockres);
4084 
4085 	return ocfs2_ci_checkpointed(&tree->rf_ci, lockres, new_level);
4086 }
4087 
4088 static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
4089 					 int blocking)
4090 {
4091 	struct ocfs2_refcount_tree *tree =
4092 				ocfs2_lock_res_refcount_tree(lockres);
4093 
4094 	ocfs2_metadata_cache_purge(&tree->rf_ci);
4095 
4096 	return UNBLOCK_CONTINUE;
4097 }
4098 
4099 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
4100 {
4101 	struct ocfs2_qinfo_lvb *lvb;
4102 	struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres);
4103 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
4104 					    oinfo->dqi_gi.dqi_type);
4105 
4106 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
4107 	lvb->lvb_version = OCFS2_QINFO_LVB_VERSION;
4108 	lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace);
4109 	lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace);
4110 	lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms);
4111 	lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks);
4112 	lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk);
4113 	lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry);
4114 }
4115 
4116 void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex)
4117 {
4118 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4119 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
4120 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4121 
4122 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
4123 		ocfs2_cluster_unlock(osb, lockres, level);
4124 }
4125 
4126 static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo)
4127 {
4128 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
4129 					    oinfo->dqi_gi.dqi_type);
4130 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4131 	struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
4132 	struct buffer_head *bh = NULL;
4133 	struct ocfs2_global_disk_dqinfo *gdinfo;
4134 	int status = 0;
4135 
4136 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
4137 	    lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) {
4138 		info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace);
4139 		info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace);
4140 		oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms);
4141 		oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks);
4142 		oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk);
4143 		oinfo->dqi_gi.dqi_free_entry =
4144 					be32_to_cpu(lvb->lvb_free_entry);
4145 	} else {
4146 		status = ocfs2_read_quota_phys_block(oinfo->dqi_gqinode,
4147 						     oinfo->dqi_giblk, &bh);
4148 		if (status) {
4149 			mlog_errno(status);
4150 			goto bail;
4151 		}
4152 		gdinfo = (struct ocfs2_global_disk_dqinfo *)
4153 					(bh->b_data + OCFS2_GLOBAL_INFO_OFF);
4154 		info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace);
4155 		info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace);
4156 		oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms);
4157 		oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks);
4158 		oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk);
4159 		oinfo->dqi_gi.dqi_free_entry =
4160 					le32_to_cpu(gdinfo->dqi_free_entry);
4161 		brelse(bh);
4162 		ocfs2_track_lock_refresh(lockres);
4163 	}
4164 
4165 bail:
4166 	return status;
4167 }
4168 
4169 /* Lock quota info, this function expects at least shared lock on the quota file
4170  * so that we can safely refresh quota info from disk. */
4171 int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex)
4172 {
4173 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4174 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
4175 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4176 	int status = 0;
4177 
4178 	/* On RO devices, locking really isn't needed... */
4179 	if (ocfs2_is_hard_readonly(osb)) {
4180 		if (ex)
4181 			status = -EROFS;
4182 		goto bail;
4183 	}
4184 	if (ocfs2_mount_local(osb))
4185 		goto bail;
4186 
4187 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
4188 	if (status < 0) {
4189 		mlog_errno(status);
4190 		goto bail;
4191 	}
4192 	if (!ocfs2_should_refresh_lock_res(lockres))
4193 		goto bail;
4194 	/* OK, we have the lock but we need to refresh the quota info */
4195 	status = ocfs2_refresh_qinfo(oinfo);
4196 	if (status)
4197 		ocfs2_qinfo_unlock(oinfo, ex);
4198 	ocfs2_complete_lock_res_refresh(lockres, status);
4199 bail:
4200 	return status;
4201 }
4202 
4203 int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex)
4204 {
4205 	int status;
4206 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4207 	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
4208 	struct ocfs2_super *osb = lockres->l_priv;
4209 
4210 
4211 	if (ocfs2_is_hard_readonly(osb))
4212 		return -EROFS;
4213 
4214 	if (ocfs2_mount_local(osb))
4215 		return 0;
4216 
4217 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
4218 	if (status < 0)
4219 		mlog_errno(status);
4220 
4221 	return status;
4222 }
4223 
4224 void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex)
4225 {
4226 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4227 	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
4228 	struct ocfs2_super *osb = lockres->l_priv;
4229 
4230 	if (!ocfs2_mount_local(osb))
4231 		ocfs2_cluster_unlock(osb, lockres, level);
4232 }
4233 
4234 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
4235 				       struct ocfs2_lock_res *lockres)
4236 {
4237 	int status;
4238 	struct ocfs2_unblock_ctl ctl = {0, 0,};
4239 	unsigned long flags;
4240 
4241 	/* Our reference to the lockres in this function can be
4242 	 * considered valid until we remove the OCFS2_LOCK_QUEUED
4243 	 * flag. */
4244 
4245 	BUG_ON(!lockres);
4246 	BUG_ON(!lockres->l_ops);
4247 
4248 	mlog(ML_BASTS, "lockres %s blocked\n", lockres->l_name);
4249 
4250 	/* Detect whether a lock has been marked as going away while
4251 	 * the downconvert thread was processing other things. A lock can
4252 	 * still be marked with OCFS2_LOCK_FREEING after this check,
4253 	 * but short circuiting here will still save us some
4254 	 * performance. */
4255 	spin_lock_irqsave(&lockres->l_lock, flags);
4256 	if (lockres->l_flags & OCFS2_LOCK_FREEING)
4257 		goto unqueue;
4258 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4259 
4260 	status = ocfs2_unblock_lock(osb, lockres, &ctl);
4261 	if (status < 0)
4262 		mlog_errno(status);
4263 
4264 	spin_lock_irqsave(&lockres->l_lock, flags);
4265 unqueue:
4266 	if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
4267 		lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
4268 	} else
4269 		ocfs2_schedule_blocked_lock(osb, lockres);
4270 
4271 	mlog(ML_BASTS, "lockres %s, requeue = %s.\n", lockres->l_name,
4272 	     ctl.requeue ? "yes" : "no");
4273 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4274 
4275 	if (ctl.unblock_action != UNBLOCK_CONTINUE
4276 	    && lockres->l_ops->post_unlock)
4277 		lockres->l_ops->post_unlock(osb, lockres);
4278 }
4279 
4280 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
4281 					struct ocfs2_lock_res *lockres)
4282 {
4283 	unsigned long flags;
4284 
4285 	assert_spin_locked(&lockres->l_lock);
4286 
4287 	if (lockres->l_flags & OCFS2_LOCK_FREEING) {
4288 		/* Do not schedule a lock for downconvert when it's on
4289 		 * the way to destruction - any nodes wanting access
4290 		 * to the resource will get it soon. */
4291 		mlog(ML_BASTS, "lockres %s won't be scheduled: flags 0x%lx\n",
4292 		     lockres->l_name, lockres->l_flags);
4293 		return;
4294 	}
4295 
4296 	lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
4297 
4298 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4299 	if (list_empty(&lockres->l_blocked_list)) {
4300 		list_add_tail(&lockres->l_blocked_list,
4301 			      &osb->blocked_lock_list);
4302 		osb->blocked_lock_count++;
4303 	}
4304 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4305 }
4306 
4307 static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
4308 {
4309 	unsigned long processed;
4310 	unsigned long flags;
4311 	struct ocfs2_lock_res *lockres;
4312 
4313 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4314 	/* grab this early so we know to try again if a state change and
4315 	 * wake happens part-way through our work  */
4316 	osb->dc_work_sequence = osb->dc_wake_sequence;
4317 
4318 	processed = osb->blocked_lock_count;
4319 	/*
4320 	 * blocked lock processing in this loop might call iput which can
4321 	 * remove items off osb->blocked_lock_list. Downconvert up to
4322 	 * 'processed' number of locks, but stop short if we had some
4323 	 * removed in ocfs2_mark_lockres_freeing when downconverting.
4324 	 */
4325 	while (processed && !list_empty(&osb->blocked_lock_list)) {
4326 		lockres = list_entry(osb->blocked_lock_list.next,
4327 				     struct ocfs2_lock_res, l_blocked_list);
4328 		list_del_init(&lockres->l_blocked_list);
4329 		osb->blocked_lock_count--;
4330 		spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4331 
4332 		BUG_ON(!processed);
4333 		processed--;
4334 
4335 		ocfs2_process_blocked_lock(osb, lockres);
4336 
4337 		spin_lock_irqsave(&osb->dc_task_lock, flags);
4338 	}
4339 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4340 }
4341 
4342 static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
4343 {
4344 	int empty = 0;
4345 	unsigned long flags;
4346 
4347 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4348 	if (list_empty(&osb->blocked_lock_list))
4349 		empty = 1;
4350 
4351 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4352 	return empty;
4353 }
4354 
4355 static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
4356 {
4357 	int should_wake = 0;
4358 	unsigned long flags;
4359 
4360 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4361 	if (osb->dc_work_sequence != osb->dc_wake_sequence)
4362 		should_wake = 1;
4363 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4364 
4365 	return should_wake;
4366 }
4367 
4368 static int ocfs2_downconvert_thread(void *arg)
4369 {
4370 	int status = 0;
4371 	struct ocfs2_super *osb = arg;
4372 
4373 	/* only quit once we've been asked to stop and there is no more
4374 	 * work available */
4375 	while (!(kthread_should_stop() &&
4376 		ocfs2_downconvert_thread_lists_empty(osb))) {
4377 
4378 		wait_event_interruptible(osb->dc_event,
4379 					 ocfs2_downconvert_thread_should_wake(osb) ||
4380 					 kthread_should_stop());
4381 
4382 		mlog(0, "downconvert_thread: awoken\n");
4383 
4384 		ocfs2_downconvert_thread_do_work(osb);
4385 	}
4386 
4387 	osb->dc_task = NULL;
4388 	return status;
4389 }
4390 
4391 void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
4392 {
4393 	unsigned long flags;
4394 
4395 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4396 	/* make sure the voting thread gets a swipe at whatever changes
4397 	 * the caller may have made to the voting state */
4398 	osb->dc_wake_sequence++;
4399 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4400 	wake_up(&osb->dc_event);
4401 }
4402