xref: /openbmc/linux/fs/ocfs2/dlmglue.c (revision fcbd8037)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* -*- mode: c; c-basic-offset: 8; -*-
3  * vim: noexpandtab sw=8 ts=8 sts=0:
4  *
5  * dlmglue.c
6  *
7  * Code which implements an OCFS2 specific interface to our DLM.
8  *
9  * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
10  */
11 
12 #include <linux/types.h>
13 #include <linux/slab.h>
14 #include <linux/highmem.h>
15 #include <linux/mm.h>
16 #include <linux/kthread.h>
17 #include <linux/pagemap.h>
18 #include <linux/debugfs.h>
19 #include <linux/seq_file.h>
20 #include <linux/time.h>
21 #include <linux/quotaops.h>
22 #include <linux/sched/signal.h>
23 
24 #define MLOG_MASK_PREFIX ML_DLM_GLUE
25 #include <cluster/masklog.h>
26 
27 #include "ocfs2.h"
28 #include "ocfs2_lockingver.h"
29 
30 #include "alloc.h"
31 #include "dcache.h"
32 #include "dlmglue.h"
33 #include "extent_map.h"
34 #include "file.h"
35 #include "heartbeat.h"
36 #include "inode.h"
37 #include "journal.h"
38 #include "stackglue.h"
39 #include "slot_map.h"
40 #include "super.h"
41 #include "uptodate.h"
42 #include "quota.h"
43 #include "refcounttree.h"
44 #include "acl.h"
45 
46 #include "buffer_head_io.h"
47 
48 struct ocfs2_mask_waiter {
49 	struct list_head	mw_item;
50 	int			mw_status;
51 	struct completion	mw_complete;
52 	unsigned long		mw_mask;
53 	unsigned long		mw_goal;
54 #ifdef CONFIG_OCFS2_FS_STATS
55 	ktime_t			mw_lock_start;
56 #endif
57 };
58 
59 static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres);
60 static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres);
61 static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres);
62 static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres);
63 
64 /*
65  * Return value from ->downconvert_worker functions.
66  *
67  * These control the precise actions of ocfs2_unblock_lock()
68  * and ocfs2_process_blocked_lock()
69  *
70  */
71 enum ocfs2_unblock_action {
72 	UNBLOCK_CONTINUE	= 0, /* Continue downconvert */
73 	UNBLOCK_CONTINUE_POST	= 1, /* Continue downconvert, fire
74 				      * ->post_unlock callback */
75 	UNBLOCK_STOP_POST	= 2, /* Do not downconvert, fire
76 				      * ->post_unlock() callback. */
77 };
78 
79 struct ocfs2_unblock_ctl {
80 	int requeue;
81 	enum ocfs2_unblock_action unblock_action;
82 };
83 
84 /* Lockdep class keys */
85 #ifdef CONFIG_DEBUG_LOCK_ALLOC
86 static struct lock_class_key lockdep_keys[OCFS2_NUM_LOCK_TYPES];
87 #endif
88 
89 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
90 					int new_level);
91 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);
92 
93 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
94 				     int blocking);
95 
96 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
97 				       int blocking);
98 
99 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
100 				     struct ocfs2_lock_res *lockres);
101 
102 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres);
103 
104 static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
105 					    int new_level);
106 static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
107 					 int blocking);
108 
109 #define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
110 
111 /* This aids in debugging situations where a bad LVB might be involved. */
112 static void ocfs2_dump_meta_lvb_info(u64 level,
113 				     const char *function,
114 				     unsigned int line,
115 				     struct ocfs2_lock_res *lockres)
116 {
117 	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
118 
119 	mlog(level, "LVB information for %s (called from %s:%u):\n",
120 	     lockres->l_name, function, line);
121 	mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
122 	     lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
123 	     be32_to_cpu(lvb->lvb_igeneration));
124 	mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
125 	     (unsigned long long)be64_to_cpu(lvb->lvb_isize),
126 	     be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
127 	     be16_to_cpu(lvb->lvb_imode));
128 	mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
129 	     "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
130 	     (long long)be64_to_cpu(lvb->lvb_iatime_packed),
131 	     (long long)be64_to_cpu(lvb->lvb_ictime_packed),
132 	     (long long)be64_to_cpu(lvb->lvb_imtime_packed),
133 	     be32_to_cpu(lvb->lvb_iattr));
134 }
135 
136 
137 /*
138  * OCFS2 Lock Resource Operations
139  *
140  * These fine tune the behavior of the generic dlmglue locking infrastructure.
141  *
142  * The most basic of lock types can point ->l_priv to their respective
143  * struct ocfs2_super and allow the default actions to manage things.
144  *
145  * Right now, each lock type also needs to implement an init function,
146  * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
147  * should be called when the lock is no longer needed (i.e., object
148  * destruction time).
149  */
150 struct ocfs2_lock_res_ops {
151 	/*
152 	 * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
153 	 * this callback if ->l_priv is not an ocfs2_super pointer
154 	 */
155 	struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
156 
157 	/*
158 	 * Optionally called in the downconvert thread after a
159 	 * successful downconvert. The lockres will not be referenced
160 	 * after this callback is called, so it is safe to free
161 	 * memory, etc.
162 	 *
163 	 * The exact semantics of when this is called are controlled
164 	 * by ->downconvert_worker()
165 	 */
166 	void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
167 
168 	/*
169 	 * Allow a lock type to add checks to determine whether it is
170 	 * safe to downconvert a lock. Return 0 to re-queue the
171 	 * downconvert at a later time, nonzero to continue.
172 	 *
173 	 * For most locks, the default checks that there are no
174 	 * incompatible holders are sufficient.
175 	 *
176 	 * Called with the lockres spinlock held.
177 	 */
178 	int (*check_downconvert)(struct ocfs2_lock_res *, int);
179 
180 	/*
181 	 * Allows a lock type to populate the lock value block. This
182 	 * is called on downconvert, and when we drop a lock.
183 	 *
184 	 * Locks that want to use this should set LOCK_TYPE_USES_LVB
185 	 * in the flags field.
186 	 *
187 	 * Called with the lockres spinlock held.
188 	 */
189 	void (*set_lvb)(struct ocfs2_lock_res *);
190 
191 	/*
192 	 * Called from the downconvert thread when it is determined
193 	 * that a lock will be downconverted. This is called without
194 	 * any locks held so the function can do work that might
195 	 * schedule (syncing out data, etc).
196 	 *
197 	 * This should return any one of the ocfs2_unblock_action
198 	 * values, depending on what it wants the thread to do.
199 	 */
200 	int (*downconvert_worker)(struct ocfs2_lock_res *, int);
201 
202 	/*
203 	 * LOCK_TYPE_* flags which describe the specific requirements
204 	 * of a lock type. Descriptions of each individual flag follow.
205 	 */
206 	int flags;
207 };
208 
209 /*
210  * Some locks want to "refresh" potentially stale data when a
211  * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
212  * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
213  * individual lockres l_flags member from the ast function. It is
214  * expected that the locking wrapper will clear the
215  * OCFS2_LOCK_NEEDS_REFRESH flag when done.
216  */
217 #define LOCK_TYPE_REQUIRES_REFRESH 0x1
218 
219 /*
220  * Indicate that a lock type makes use of the lock value block. The
221  * ->set_lvb lock type callback must be defined.
222  */
223 #define LOCK_TYPE_USES_LVB		0x2
224 
225 static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
226 	.get_osb	= ocfs2_get_inode_osb,
227 	.flags		= 0,
228 };
229 
230 static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = {
231 	.get_osb	= ocfs2_get_inode_osb,
232 	.check_downconvert = ocfs2_check_meta_downconvert,
233 	.set_lvb	= ocfs2_set_meta_lvb,
234 	.downconvert_worker = ocfs2_data_convert_worker,
235 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
236 };
237 
238 static struct ocfs2_lock_res_ops ocfs2_super_lops = {
239 	.flags		= LOCK_TYPE_REQUIRES_REFRESH,
240 };
241 
242 static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
243 	.flags		= 0,
244 };
245 
246 static struct ocfs2_lock_res_ops ocfs2_nfs_sync_lops = {
247 	.flags		= 0,
248 };
249 
250 static struct ocfs2_lock_res_ops ocfs2_trim_fs_lops = {
251 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
252 };
253 
254 static struct ocfs2_lock_res_ops ocfs2_orphan_scan_lops = {
255 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
256 };
257 
258 static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
259 	.get_osb	= ocfs2_get_dentry_osb,
260 	.post_unlock	= ocfs2_dentry_post_unlock,
261 	.downconvert_worker = ocfs2_dentry_convert_worker,
262 	.flags		= 0,
263 };
264 
265 static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
266 	.get_osb	= ocfs2_get_inode_osb,
267 	.flags		= 0,
268 };
269 
270 static struct ocfs2_lock_res_ops ocfs2_flock_lops = {
271 	.get_osb	= ocfs2_get_file_osb,
272 	.flags		= 0,
273 };
274 
275 static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = {
276 	.set_lvb	= ocfs2_set_qinfo_lvb,
277 	.get_osb	= ocfs2_get_qinfo_osb,
278 	.flags		= LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB,
279 };
280 
281 static struct ocfs2_lock_res_ops ocfs2_refcount_block_lops = {
282 	.check_downconvert = ocfs2_check_refcount_downconvert,
283 	.downconvert_worker = ocfs2_refcount_convert_worker,
284 	.flags		= 0,
285 };
286 
287 static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
288 {
289 	return lockres->l_type == OCFS2_LOCK_TYPE_META ||
290 		lockres->l_type == OCFS2_LOCK_TYPE_RW ||
291 		lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
292 }
293 
294 static inline struct ocfs2_lock_res *ocfs2_lksb_to_lock_res(struct ocfs2_dlm_lksb *lksb)
295 {
296 	return container_of(lksb, struct ocfs2_lock_res, l_lksb);
297 }
298 
299 static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
300 {
301 	BUG_ON(!ocfs2_is_inode_lock(lockres));
302 
303 	return (struct inode *) lockres->l_priv;
304 }
305 
306 static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
307 {
308 	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
309 
310 	return (struct ocfs2_dentry_lock *)lockres->l_priv;
311 }
312 
313 static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres)
314 {
315 	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO);
316 
317 	return (struct ocfs2_mem_dqinfo *)lockres->l_priv;
318 }
319 
320 static inline struct ocfs2_refcount_tree *
321 ocfs2_lock_res_refcount_tree(struct ocfs2_lock_res *res)
322 {
323 	return container_of(res, struct ocfs2_refcount_tree, rf_lockres);
324 }
325 
326 static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
327 {
328 	if (lockres->l_ops->get_osb)
329 		return lockres->l_ops->get_osb(lockres);
330 
331 	return (struct ocfs2_super *)lockres->l_priv;
332 }
333 
334 static int ocfs2_lock_create(struct ocfs2_super *osb,
335 			     struct ocfs2_lock_res *lockres,
336 			     int level,
337 			     u32 dlm_flags);
338 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
339 						     int wanted);
340 static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
341 				   struct ocfs2_lock_res *lockres,
342 				   int level, unsigned long caller_ip);
343 static inline void ocfs2_cluster_unlock(struct ocfs2_super *osb,
344 					struct ocfs2_lock_res *lockres,
345 					int level)
346 {
347 	__ocfs2_cluster_unlock(osb, lockres, level, _RET_IP_);
348 }
349 
350 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
351 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
352 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
353 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
354 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
355 					struct ocfs2_lock_res *lockres);
356 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
357 						int convert);
358 #define ocfs2_log_dlm_error(_func, _err, _lockres) do {					\
359 	if ((_lockres)->l_type != OCFS2_LOCK_TYPE_DENTRY)				\
360 		mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n",	\
361 		     _err, _func, _lockres->l_name);					\
362 	else										\
363 		mlog(ML_ERROR, "DLM error %d while calling %s on resource %.*s%08x\n",	\
364 		     _err, _func, OCFS2_DENTRY_LOCK_INO_START - 1, (_lockres)->l_name,	\
365 		     (unsigned int)ocfs2_get_dentry_lock_ino(_lockres));		\
366 } while (0)
367 static int ocfs2_downconvert_thread(void *arg);
368 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
369 					struct ocfs2_lock_res *lockres);
370 static int ocfs2_inode_lock_update(struct inode *inode,
371 				  struct buffer_head **bh);
372 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
373 static inline int ocfs2_highest_compat_lock_level(int level);
374 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
375 					      int new_level);
376 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
377 				  struct ocfs2_lock_res *lockres,
378 				  int new_level,
379 				  int lvb,
380 				  unsigned int generation);
381 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
382 				        struct ocfs2_lock_res *lockres);
383 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
384 				struct ocfs2_lock_res *lockres);
385 
386 
387 static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
388 				  u64 blkno,
389 				  u32 generation,
390 				  char *name)
391 {
392 	int len;
393 
394 	BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
395 
396 	len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
397 		       ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
398 		       (long long)blkno, generation);
399 
400 	BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
401 
402 	mlog(0, "built lock resource with name: %s\n", name);
403 }
404 
405 static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
406 
407 static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
408 				       struct ocfs2_dlm_debug *dlm_debug)
409 {
410 	mlog(0, "Add tracking for lockres %s\n", res->l_name);
411 
412 	spin_lock(&ocfs2_dlm_tracking_lock);
413 	list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
414 	spin_unlock(&ocfs2_dlm_tracking_lock);
415 }
416 
417 static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
418 {
419 	spin_lock(&ocfs2_dlm_tracking_lock);
420 	if (!list_empty(&res->l_debug_list))
421 		list_del_init(&res->l_debug_list);
422 	spin_unlock(&ocfs2_dlm_tracking_lock);
423 }
424 
425 #ifdef CONFIG_OCFS2_FS_STATS
426 static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
427 {
428 	res->l_lock_refresh = 0;
429 	res->l_lock_wait = 0;
430 	memset(&res->l_lock_prmode, 0, sizeof(struct ocfs2_lock_stats));
431 	memset(&res->l_lock_exmode, 0, sizeof(struct ocfs2_lock_stats));
432 }
433 
434 static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level,
435 				    struct ocfs2_mask_waiter *mw, int ret)
436 {
437 	u32 usec;
438 	ktime_t kt;
439 	struct ocfs2_lock_stats *stats;
440 
441 	if (level == LKM_PRMODE)
442 		stats = &res->l_lock_prmode;
443 	else if (level == LKM_EXMODE)
444 		stats = &res->l_lock_exmode;
445 	else
446 		return;
447 
448 	kt = ktime_sub(ktime_get(), mw->mw_lock_start);
449 	usec = ktime_to_us(kt);
450 
451 	stats->ls_gets++;
452 	stats->ls_total += ktime_to_ns(kt);
453 	/* overflow */
454 	if (unlikely(stats->ls_gets == 0)) {
455 		stats->ls_gets++;
456 		stats->ls_total = ktime_to_ns(kt);
457 	}
458 
459 	if (stats->ls_max < usec)
460 		stats->ls_max = usec;
461 
462 	if (ret)
463 		stats->ls_fail++;
464 
465 	stats->ls_last = ktime_to_us(ktime_get_real());
466 }
467 
468 static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
469 {
470 	lockres->l_lock_refresh++;
471 }
472 
473 static inline void ocfs2_track_lock_wait(struct ocfs2_lock_res *lockres)
474 {
475 	struct ocfs2_mask_waiter *mw;
476 
477 	if (list_empty(&lockres->l_mask_waiters)) {
478 		lockres->l_lock_wait = 0;
479 		return;
480 	}
481 
482 	mw = list_first_entry(&lockres->l_mask_waiters,
483 				struct ocfs2_mask_waiter, mw_item);
484 	lockres->l_lock_wait =
485 			ktime_to_us(ktime_mono_to_real(mw->mw_lock_start));
486 }
487 
488 static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
489 {
490 	mw->mw_lock_start = ktime_get();
491 }
492 #else
493 static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
494 {
495 }
496 static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res,
497 			   int level, struct ocfs2_mask_waiter *mw, int ret)
498 {
499 }
500 static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
501 {
502 }
503 static inline void ocfs2_track_lock_wait(struct ocfs2_lock_res *lockres)
504 {
505 }
506 static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
507 {
508 }
509 #endif
510 
511 static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
512 				       struct ocfs2_lock_res *res,
513 				       enum ocfs2_lock_type type,
514 				       struct ocfs2_lock_res_ops *ops,
515 				       void *priv)
516 {
517 	res->l_type          = type;
518 	res->l_ops           = ops;
519 	res->l_priv          = priv;
520 
521 	res->l_level         = DLM_LOCK_IV;
522 	res->l_requested     = DLM_LOCK_IV;
523 	res->l_blocking      = DLM_LOCK_IV;
524 	res->l_action        = OCFS2_AST_INVALID;
525 	res->l_unlock_action = OCFS2_UNLOCK_INVALID;
526 
527 	res->l_flags         = OCFS2_LOCK_INITIALIZED;
528 
529 	ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
530 
531 	ocfs2_init_lock_stats(res);
532 #ifdef CONFIG_DEBUG_LOCK_ALLOC
533 	if (type != OCFS2_LOCK_TYPE_OPEN)
534 		lockdep_init_map(&res->l_lockdep_map, ocfs2_lock_type_strings[type],
535 				 &lockdep_keys[type], 0);
536 	else
537 		res->l_lockdep_map.key = NULL;
538 #endif
539 }
540 
541 void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
542 {
543 	/* This also clears out the lock status block */
544 	memset(res, 0, sizeof(struct ocfs2_lock_res));
545 	spin_lock_init(&res->l_lock);
546 	init_waitqueue_head(&res->l_event);
547 	INIT_LIST_HEAD(&res->l_blocked_list);
548 	INIT_LIST_HEAD(&res->l_mask_waiters);
549 	INIT_LIST_HEAD(&res->l_holders);
550 }
551 
552 void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
553 			       enum ocfs2_lock_type type,
554 			       unsigned int generation,
555 			       struct inode *inode)
556 {
557 	struct ocfs2_lock_res_ops *ops;
558 
559 	switch(type) {
560 		case OCFS2_LOCK_TYPE_RW:
561 			ops = &ocfs2_inode_rw_lops;
562 			break;
563 		case OCFS2_LOCK_TYPE_META:
564 			ops = &ocfs2_inode_inode_lops;
565 			break;
566 		case OCFS2_LOCK_TYPE_OPEN:
567 			ops = &ocfs2_inode_open_lops;
568 			break;
569 		default:
570 			mlog_bug_on_msg(1, "type: %d\n", type);
571 			ops = NULL; /* thanks, gcc */
572 			break;
573 	};
574 
575 	ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
576 			      generation, res->l_name);
577 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
578 }
579 
580 static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
581 {
582 	struct inode *inode = ocfs2_lock_res_inode(lockres);
583 
584 	return OCFS2_SB(inode->i_sb);
585 }
586 
587 static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres)
588 {
589 	struct ocfs2_mem_dqinfo *info = lockres->l_priv;
590 
591 	return OCFS2_SB(info->dqi_gi.dqi_sb);
592 }
593 
594 static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres)
595 {
596 	struct ocfs2_file_private *fp = lockres->l_priv;
597 
598 	return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb);
599 }
600 
601 static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
602 {
603 	__be64 inode_blkno_be;
604 
605 	memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
606 	       sizeof(__be64));
607 
608 	return be64_to_cpu(inode_blkno_be);
609 }
610 
611 static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
612 {
613 	struct ocfs2_dentry_lock *dl = lockres->l_priv;
614 
615 	return OCFS2_SB(dl->dl_inode->i_sb);
616 }
617 
618 void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
619 				u64 parent, struct inode *inode)
620 {
621 	int len;
622 	u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
623 	__be64 inode_blkno_be = cpu_to_be64(inode_blkno);
624 	struct ocfs2_lock_res *lockres = &dl->dl_lockres;
625 
626 	ocfs2_lock_res_init_once(lockres);
627 
628 	/*
629 	 * Unfortunately, the standard lock naming scheme won't work
630 	 * here because we have two 16 byte values to use. Instead,
631 	 * we'll stuff the inode number as a binary value. We still
632 	 * want error prints to show something without garbling the
633 	 * display, so drop a null byte in there before the inode
634 	 * number. A future version of OCFS2 will likely use all
635 	 * binary lock names. The stringified names have been a
636 	 * tremendous aid in debugging, but now that the debugfs
637 	 * interface exists, we can mangle things there if need be.
638 	 *
639 	 * NOTE: We also drop the standard "pad" value (the total lock
640 	 * name size stays the same though - the last part is all
641 	 * zeros due to the memset in ocfs2_lock_res_init_once()
642 	 */
643 	len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
644 		       "%c%016llx",
645 		       ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
646 		       (long long)parent);
647 
648 	BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
649 
650 	memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
651 	       sizeof(__be64));
652 
653 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
654 				   OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
655 				   dl);
656 }
657 
658 static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
659 				      struct ocfs2_super *osb)
660 {
661 	/* Superblock lockres doesn't come from a slab so we call init
662 	 * once on it manually.  */
663 	ocfs2_lock_res_init_once(res);
664 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
665 			      0, res->l_name);
666 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
667 				   &ocfs2_super_lops, osb);
668 }
669 
670 static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
671 				       struct ocfs2_super *osb)
672 {
673 	/* Rename lockres doesn't come from a slab so we call init
674 	 * once on it manually.  */
675 	ocfs2_lock_res_init_once(res);
676 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
677 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
678 				   &ocfs2_rename_lops, osb);
679 }
680 
681 static void ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res *res,
682 					 struct ocfs2_super *osb)
683 {
684 	/* nfs_sync lockres doesn't come from a slab so we call init
685 	 * once on it manually.  */
686 	ocfs2_lock_res_init_once(res);
687 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_NFS_SYNC, 0, 0, res->l_name);
688 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_NFS_SYNC,
689 				   &ocfs2_nfs_sync_lops, osb);
690 }
691 
692 void ocfs2_trim_fs_lock_res_init(struct ocfs2_super *osb)
693 {
694 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
695 
696 	/* Only one trimfs thread are allowed to work at the same time. */
697 	mutex_lock(&osb->obs_trim_fs_mutex);
698 
699 	ocfs2_lock_res_init_once(lockres);
700 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_TRIM_FS, 0, 0, lockres->l_name);
701 	ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_TRIM_FS,
702 				   &ocfs2_trim_fs_lops, osb);
703 }
704 
705 void ocfs2_trim_fs_lock_res_uninit(struct ocfs2_super *osb)
706 {
707 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
708 
709 	ocfs2_simple_drop_lockres(osb, lockres);
710 	ocfs2_lock_res_free(lockres);
711 
712 	mutex_unlock(&osb->obs_trim_fs_mutex);
713 }
714 
715 static void ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res *res,
716 					    struct ocfs2_super *osb)
717 {
718 	ocfs2_lock_res_init_once(res);
719 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_ORPHAN_SCAN, 0, 0, res->l_name);
720 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_ORPHAN_SCAN,
721 				   &ocfs2_orphan_scan_lops, osb);
722 }
723 
724 void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
725 			      struct ocfs2_file_private *fp)
726 {
727 	struct inode *inode = fp->fp_file->f_mapping->host;
728 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
729 
730 	ocfs2_lock_res_init_once(lockres);
731 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
732 			      inode->i_generation, lockres->l_name);
733 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
734 				   OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
735 				   fp);
736 	lockres->l_flags |= OCFS2_LOCK_NOCACHE;
737 }
738 
739 void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
740 			       struct ocfs2_mem_dqinfo *info)
741 {
742 	ocfs2_lock_res_init_once(lockres);
743 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type,
744 			      0, lockres->l_name);
745 	ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres,
746 				   OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops,
747 				   info);
748 }
749 
750 void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres,
751 				  struct ocfs2_super *osb, u64 ref_blkno,
752 				  unsigned int generation)
753 {
754 	ocfs2_lock_res_init_once(lockres);
755 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_REFCOUNT, ref_blkno,
756 			      generation, lockres->l_name);
757 	ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_REFCOUNT,
758 				   &ocfs2_refcount_block_lops, osb);
759 }
760 
761 void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
762 {
763 	if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
764 		return;
765 
766 	ocfs2_remove_lockres_tracking(res);
767 
768 	mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
769 			"Lockres %s is on the blocked list\n",
770 			res->l_name);
771 	mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
772 			"Lockres %s has mask waiters pending\n",
773 			res->l_name);
774 	mlog_bug_on_msg(spin_is_locked(&res->l_lock),
775 			"Lockres %s is locked\n",
776 			res->l_name);
777 	mlog_bug_on_msg(res->l_ro_holders,
778 			"Lockres %s has %u ro holders\n",
779 			res->l_name, res->l_ro_holders);
780 	mlog_bug_on_msg(res->l_ex_holders,
781 			"Lockres %s has %u ex holders\n",
782 			res->l_name, res->l_ex_holders);
783 
784 	/* Need to clear out the lock status block for the dlm */
785 	memset(&res->l_lksb, 0, sizeof(res->l_lksb));
786 
787 	res->l_flags = 0UL;
788 }
789 
790 /*
791  * Keep a list of processes who have interest in a lockres.
792  * Note: this is now only uesed for check recursive cluster locking.
793  */
794 static inline void ocfs2_add_holder(struct ocfs2_lock_res *lockres,
795 				   struct ocfs2_lock_holder *oh)
796 {
797 	INIT_LIST_HEAD(&oh->oh_list);
798 	oh->oh_owner_pid = get_pid(task_pid(current));
799 
800 	spin_lock(&lockres->l_lock);
801 	list_add_tail(&oh->oh_list, &lockres->l_holders);
802 	spin_unlock(&lockres->l_lock);
803 }
804 
805 static struct ocfs2_lock_holder *
806 ocfs2_pid_holder(struct ocfs2_lock_res *lockres,
807 		struct pid *pid)
808 {
809 	struct ocfs2_lock_holder *oh;
810 
811 	spin_lock(&lockres->l_lock);
812 	list_for_each_entry(oh, &lockres->l_holders, oh_list) {
813 		if (oh->oh_owner_pid == pid) {
814 			spin_unlock(&lockres->l_lock);
815 			return oh;
816 		}
817 	}
818 	spin_unlock(&lockres->l_lock);
819 	return NULL;
820 }
821 
822 static inline void ocfs2_remove_holder(struct ocfs2_lock_res *lockres,
823 				       struct ocfs2_lock_holder *oh)
824 {
825 	spin_lock(&lockres->l_lock);
826 	list_del(&oh->oh_list);
827 	spin_unlock(&lockres->l_lock);
828 
829 	put_pid(oh->oh_owner_pid);
830 }
831 
832 
833 static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
834 				     int level)
835 {
836 	BUG_ON(!lockres);
837 
838 	switch(level) {
839 	case DLM_LOCK_EX:
840 		lockres->l_ex_holders++;
841 		break;
842 	case DLM_LOCK_PR:
843 		lockres->l_ro_holders++;
844 		break;
845 	default:
846 		BUG();
847 	}
848 }
849 
850 static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
851 				     int level)
852 {
853 	BUG_ON(!lockres);
854 
855 	switch(level) {
856 	case DLM_LOCK_EX:
857 		BUG_ON(!lockres->l_ex_holders);
858 		lockres->l_ex_holders--;
859 		break;
860 	case DLM_LOCK_PR:
861 		BUG_ON(!lockres->l_ro_holders);
862 		lockres->l_ro_holders--;
863 		break;
864 	default:
865 		BUG();
866 	}
867 }
868 
869 /* WARNING: This function lives in a world where the only three lock
870  * levels are EX, PR, and NL. It *will* have to be adjusted when more
871  * lock types are added. */
872 static inline int ocfs2_highest_compat_lock_level(int level)
873 {
874 	int new_level = DLM_LOCK_EX;
875 
876 	if (level == DLM_LOCK_EX)
877 		new_level = DLM_LOCK_NL;
878 	else if (level == DLM_LOCK_PR)
879 		new_level = DLM_LOCK_PR;
880 	return new_level;
881 }
882 
883 static void lockres_set_flags(struct ocfs2_lock_res *lockres,
884 			      unsigned long newflags)
885 {
886 	struct ocfs2_mask_waiter *mw, *tmp;
887 
888  	assert_spin_locked(&lockres->l_lock);
889 
890 	lockres->l_flags = newflags;
891 
892 	list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
893 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
894 			continue;
895 
896 		list_del_init(&mw->mw_item);
897 		mw->mw_status = 0;
898 		complete(&mw->mw_complete);
899 		ocfs2_track_lock_wait(lockres);
900 	}
901 }
902 static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
903 {
904 	lockres_set_flags(lockres, lockres->l_flags | or);
905 }
906 static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
907 				unsigned long clear)
908 {
909 	lockres_set_flags(lockres, lockres->l_flags & ~clear);
910 }
911 
912 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
913 {
914 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
915 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
916 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
917 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
918 
919 	lockres->l_level = lockres->l_requested;
920 	if (lockres->l_level <=
921 	    ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
922 		lockres->l_blocking = DLM_LOCK_NL;
923 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
924 	}
925 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
926 }
927 
928 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
929 {
930 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
931 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
932 
933 	/* Convert from RO to EX doesn't really need anything as our
934 	 * information is already up to data. Convert from NL to
935 	 * *anything* however should mark ourselves as needing an
936 	 * update */
937 	if (lockres->l_level == DLM_LOCK_NL &&
938 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
939 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
940 
941 	lockres->l_level = lockres->l_requested;
942 
943 	/*
944 	 * We set the OCFS2_LOCK_UPCONVERT_FINISHING flag before clearing
945 	 * the OCFS2_LOCK_BUSY flag to prevent the dc thread from
946 	 * downconverting the lock before the upconvert has fully completed.
947 	 * Do not prevent the dc thread from downconverting if NONBLOCK lock
948 	 * had already returned.
949 	 */
950 	if (!(lockres->l_flags & OCFS2_LOCK_NONBLOCK_FINISHED))
951 		lockres_or_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
952 	else
953 		lockres_clear_flags(lockres, OCFS2_LOCK_NONBLOCK_FINISHED);
954 
955 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
956 }
957 
958 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
959 {
960 	BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
961 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
962 
963 	if (lockres->l_requested > DLM_LOCK_NL &&
964 	    !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
965 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
966 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
967 
968 	lockres->l_level = lockres->l_requested;
969 	lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
970 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
971 }
972 
973 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
974 				     int level)
975 {
976 	int needs_downconvert = 0;
977 
978 	assert_spin_locked(&lockres->l_lock);
979 
980 	if (level > lockres->l_blocking) {
981 		/* only schedule a downconvert if we haven't already scheduled
982 		 * one that goes low enough to satisfy the level we're
983 		 * blocking.  this also catches the case where we get
984 		 * duplicate BASTs */
985 		if (ocfs2_highest_compat_lock_level(level) <
986 		    ocfs2_highest_compat_lock_level(lockres->l_blocking))
987 			needs_downconvert = 1;
988 
989 		lockres->l_blocking = level;
990 	}
991 
992 	mlog(ML_BASTS, "lockres %s, block %d, level %d, l_block %d, dwn %d\n",
993 	     lockres->l_name, level, lockres->l_level, lockres->l_blocking,
994 	     needs_downconvert);
995 
996 	if (needs_downconvert)
997 		lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
998 	mlog(0, "needs_downconvert = %d\n", needs_downconvert);
999 	return needs_downconvert;
1000 }
1001 
1002 /*
1003  * OCFS2_LOCK_PENDING and l_pending_gen.
1004  *
1005  * Why does OCFS2_LOCK_PENDING exist?  To close a race between setting
1006  * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock().  See ocfs2_unblock_lock()
1007  * for more details on the race.
1008  *
1009  * OCFS2_LOCK_PENDING closes the race quite nicely.  However, it introduces
1010  * a race on itself.  In o2dlm, we can get the ast before ocfs2_dlm_lock()
1011  * returns.  The ast clears OCFS2_LOCK_BUSY, and must therefore clear
1012  * OCFS2_LOCK_PENDING at the same time.  When ocfs2_dlm_lock() returns,
1013  * the caller is going to try to clear PENDING again.  If nothing else is
1014  * happening, __lockres_clear_pending() sees PENDING is unset and does
1015  * nothing.
1016  *
1017  * But what if another path (eg downconvert thread) has just started a
1018  * new locking action?  The other path has re-set PENDING.  Our path
1019  * cannot clear PENDING, because that will re-open the original race
1020  * window.
1021  *
1022  * [Example]
1023  *
1024  * ocfs2_meta_lock()
1025  *  ocfs2_cluster_lock()
1026  *   set BUSY
1027  *   set PENDING
1028  *   drop l_lock
1029  *   ocfs2_dlm_lock()
1030  *    ocfs2_locking_ast()		ocfs2_downconvert_thread()
1031  *     clear PENDING			 ocfs2_unblock_lock()
1032  *					  take_l_lock
1033  *					  !BUSY
1034  *					  ocfs2_prepare_downconvert()
1035  *					   set BUSY
1036  *					   set PENDING
1037  *					  drop l_lock
1038  *   take l_lock
1039  *   clear PENDING
1040  *   drop l_lock
1041  *			<window>
1042  *					  ocfs2_dlm_lock()
1043  *
1044  * So as you can see, we now have a window where l_lock is not held,
1045  * PENDING is not set, and ocfs2_dlm_lock() has not been called.
1046  *
1047  * The core problem is that ocfs2_cluster_lock() has cleared the PENDING
1048  * set by ocfs2_prepare_downconvert().  That wasn't nice.
1049  *
1050  * To solve this we introduce l_pending_gen.  A call to
1051  * lockres_clear_pending() will only do so when it is passed a generation
1052  * number that matches the lockres.  lockres_set_pending() will return the
1053  * current generation number.  When ocfs2_cluster_lock() goes to clear
1054  * PENDING, it passes the generation it got from set_pending().  In our
1055  * example above, the generation numbers will *not* match.  Thus,
1056  * ocfs2_cluster_lock() will not clear the PENDING set by
1057  * ocfs2_prepare_downconvert().
1058  */
1059 
1060 /* Unlocked version for ocfs2_locking_ast() */
1061 static void __lockres_clear_pending(struct ocfs2_lock_res *lockres,
1062 				    unsigned int generation,
1063 				    struct ocfs2_super *osb)
1064 {
1065 	assert_spin_locked(&lockres->l_lock);
1066 
1067 	/*
1068 	 * The ast and locking functions can race us here.  The winner
1069 	 * will clear pending, the loser will not.
1070 	 */
1071 	if (!(lockres->l_flags & OCFS2_LOCK_PENDING) ||
1072 	    (lockres->l_pending_gen != generation))
1073 		return;
1074 
1075 	lockres_clear_flags(lockres, OCFS2_LOCK_PENDING);
1076 	lockres->l_pending_gen++;
1077 
1078 	/*
1079 	 * The downconvert thread may have skipped us because we
1080 	 * were PENDING.  Wake it up.
1081 	 */
1082 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1083 		ocfs2_wake_downconvert_thread(osb);
1084 }
1085 
1086 /* Locked version for callers of ocfs2_dlm_lock() */
1087 static void lockres_clear_pending(struct ocfs2_lock_res *lockres,
1088 				  unsigned int generation,
1089 				  struct ocfs2_super *osb)
1090 {
1091 	unsigned long flags;
1092 
1093 	spin_lock_irqsave(&lockres->l_lock, flags);
1094 	__lockres_clear_pending(lockres, generation, osb);
1095 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1096 }
1097 
1098 static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres)
1099 {
1100 	assert_spin_locked(&lockres->l_lock);
1101 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
1102 
1103 	lockres_or_flags(lockres, OCFS2_LOCK_PENDING);
1104 
1105 	return lockres->l_pending_gen;
1106 }
1107 
1108 static void ocfs2_blocking_ast(struct ocfs2_dlm_lksb *lksb, int level)
1109 {
1110 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1111 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1112 	int needs_downconvert;
1113 	unsigned long flags;
1114 
1115 	BUG_ON(level <= DLM_LOCK_NL);
1116 
1117 	mlog(ML_BASTS, "BAST fired for lockres %s, blocking %d, level %d, "
1118 	     "type %s\n", lockres->l_name, level, lockres->l_level,
1119 	     ocfs2_lock_type_string(lockres->l_type));
1120 
1121 	/*
1122 	 * We can skip the bast for locks which don't enable caching -
1123 	 * they'll be dropped at the earliest possible time anyway.
1124 	 */
1125 	if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
1126 		return;
1127 
1128 	spin_lock_irqsave(&lockres->l_lock, flags);
1129 	needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
1130 	if (needs_downconvert)
1131 		ocfs2_schedule_blocked_lock(osb, lockres);
1132 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1133 
1134 	wake_up(&lockres->l_event);
1135 
1136 	ocfs2_wake_downconvert_thread(osb);
1137 }
1138 
1139 static void ocfs2_locking_ast(struct ocfs2_dlm_lksb *lksb)
1140 {
1141 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1142 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1143 	unsigned long flags;
1144 	int status;
1145 
1146 	spin_lock_irqsave(&lockres->l_lock, flags);
1147 
1148 	status = ocfs2_dlm_lock_status(&lockres->l_lksb);
1149 
1150 	if (status == -EAGAIN) {
1151 		lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1152 		goto out;
1153 	}
1154 
1155 	if (status) {
1156 		mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n",
1157 		     lockres->l_name, status);
1158 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1159 		return;
1160 	}
1161 
1162 	mlog(ML_BASTS, "AST fired for lockres %s, action %d, unlock %d, "
1163 	     "level %d => %d\n", lockres->l_name, lockres->l_action,
1164 	     lockres->l_unlock_action, lockres->l_level, lockres->l_requested);
1165 
1166 	switch(lockres->l_action) {
1167 	case OCFS2_AST_ATTACH:
1168 		ocfs2_generic_handle_attach_action(lockres);
1169 		lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
1170 		break;
1171 	case OCFS2_AST_CONVERT:
1172 		ocfs2_generic_handle_convert_action(lockres);
1173 		break;
1174 	case OCFS2_AST_DOWNCONVERT:
1175 		ocfs2_generic_handle_downconvert_action(lockres);
1176 		break;
1177 	default:
1178 		mlog(ML_ERROR, "lockres %s: AST fired with invalid action: %u, "
1179 		     "flags 0x%lx, unlock: %u\n",
1180 		     lockres->l_name, lockres->l_action, lockres->l_flags,
1181 		     lockres->l_unlock_action);
1182 		BUG();
1183 	}
1184 out:
1185 	/* set it to something invalid so if we get called again we
1186 	 * can catch it. */
1187 	lockres->l_action = OCFS2_AST_INVALID;
1188 
1189 	/* Did we try to cancel this lock?  Clear that state */
1190 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT)
1191 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1192 
1193 	/*
1194 	 * We may have beaten the locking functions here.  We certainly
1195 	 * know that dlm_lock() has been called :-)
1196 	 * Because we can't have two lock calls in flight at once, we
1197 	 * can use lockres->l_pending_gen.
1198 	 */
1199 	__lockres_clear_pending(lockres, lockres->l_pending_gen,  osb);
1200 
1201 	wake_up(&lockres->l_event);
1202 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1203 }
1204 
1205 static void ocfs2_unlock_ast(struct ocfs2_dlm_lksb *lksb, int error)
1206 {
1207 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1208 	unsigned long flags;
1209 
1210 	mlog(ML_BASTS, "UNLOCK AST fired for lockres %s, action = %d\n",
1211 	     lockres->l_name, lockres->l_unlock_action);
1212 
1213 	spin_lock_irqsave(&lockres->l_lock, flags);
1214 	if (error) {
1215 		mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
1216 		     "unlock_action %d\n", error, lockres->l_name,
1217 		     lockres->l_unlock_action);
1218 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1219 		return;
1220 	}
1221 
1222 	switch(lockres->l_unlock_action) {
1223 	case OCFS2_UNLOCK_CANCEL_CONVERT:
1224 		mlog(0, "Cancel convert success for %s\n", lockres->l_name);
1225 		lockres->l_action = OCFS2_AST_INVALID;
1226 		/* Downconvert thread may have requeued this lock, we
1227 		 * need to wake it. */
1228 		if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1229 			ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres));
1230 		break;
1231 	case OCFS2_UNLOCK_DROP_LOCK:
1232 		lockres->l_level = DLM_LOCK_IV;
1233 		break;
1234 	default:
1235 		BUG();
1236 	}
1237 
1238 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1239 	lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1240 	wake_up(&lockres->l_event);
1241 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1242 }
1243 
1244 /*
1245  * This is the filesystem locking protocol.  It provides the lock handling
1246  * hooks for the underlying DLM.  It has a maximum version number.
1247  * The version number allows interoperability with systems running at
1248  * the same major number and an equal or smaller minor number.
1249  *
1250  * Whenever the filesystem does new things with locks (adds or removes a
1251  * lock, orders them differently, does different things underneath a lock),
1252  * the version must be changed.  The protocol is negotiated when joining
1253  * the dlm domain.  A node may join the domain if its major version is
1254  * identical to all other nodes and its minor version is greater than
1255  * or equal to all other nodes.  When its minor version is greater than
1256  * the other nodes, it will run at the minor version specified by the
1257  * other nodes.
1258  *
1259  * If a locking change is made that will not be compatible with older
1260  * versions, the major number must be increased and the minor version set
1261  * to zero.  If a change merely adds a behavior that can be disabled when
1262  * speaking to older versions, the minor version must be increased.  If a
1263  * change adds a fully backwards compatible change (eg, LVB changes that
1264  * are just ignored by older versions), the version does not need to be
1265  * updated.
1266  */
1267 static struct ocfs2_locking_protocol lproto = {
1268 	.lp_max_version = {
1269 		.pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
1270 		.pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
1271 	},
1272 	.lp_lock_ast		= ocfs2_locking_ast,
1273 	.lp_blocking_ast	= ocfs2_blocking_ast,
1274 	.lp_unlock_ast		= ocfs2_unlock_ast,
1275 };
1276 
1277 void ocfs2_set_locking_protocol(void)
1278 {
1279 	ocfs2_stack_glue_set_max_proto_version(&lproto.lp_max_version);
1280 }
1281 
1282 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
1283 						int convert)
1284 {
1285 	unsigned long flags;
1286 
1287 	spin_lock_irqsave(&lockres->l_lock, flags);
1288 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1289 	lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1290 	if (convert)
1291 		lockres->l_action = OCFS2_AST_INVALID;
1292 	else
1293 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1294 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1295 
1296 	wake_up(&lockres->l_event);
1297 }
1298 
1299 /* Note: If we detect another process working on the lock (i.e.,
1300  * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
1301  * to do the right thing in that case.
1302  */
1303 static int ocfs2_lock_create(struct ocfs2_super *osb,
1304 			     struct ocfs2_lock_res *lockres,
1305 			     int level,
1306 			     u32 dlm_flags)
1307 {
1308 	int ret = 0;
1309 	unsigned long flags;
1310 	unsigned int gen;
1311 
1312 	mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
1313 	     dlm_flags);
1314 
1315 	spin_lock_irqsave(&lockres->l_lock, flags);
1316 	if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
1317 	    (lockres->l_flags & OCFS2_LOCK_BUSY)) {
1318 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1319 		goto bail;
1320 	}
1321 
1322 	lockres->l_action = OCFS2_AST_ATTACH;
1323 	lockres->l_requested = level;
1324 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1325 	gen = lockres_set_pending(lockres);
1326 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1327 
1328 	ret = ocfs2_dlm_lock(osb->cconn,
1329 			     level,
1330 			     &lockres->l_lksb,
1331 			     dlm_flags,
1332 			     lockres->l_name,
1333 			     OCFS2_LOCK_ID_MAX_LEN - 1);
1334 	lockres_clear_pending(lockres, gen, osb);
1335 	if (ret) {
1336 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
1337 		ocfs2_recover_from_dlm_error(lockres, 1);
1338 	}
1339 
1340 	mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
1341 
1342 bail:
1343 	return ret;
1344 }
1345 
1346 static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
1347 					int flag)
1348 {
1349 	unsigned long flags;
1350 	int ret;
1351 
1352 	spin_lock_irqsave(&lockres->l_lock, flags);
1353 	ret = lockres->l_flags & flag;
1354 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1355 
1356 	return ret;
1357 }
1358 
1359 static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
1360 
1361 {
1362 	wait_event(lockres->l_event,
1363 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
1364 }
1365 
1366 static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
1367 
1368 {
1369 	wait_event(lockres->l_event,
1370 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
1371 }
1372 
1373 /* predict what lock level we'll be dropping down to on behalf
1374  * of another node, and return true if the currently wanted
1375  * level will be compatible with it. */
1376 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
1377 						     int wanted)
1378 {
1379 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
1380 
1381 	return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
1382 }
1383 
1384 static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
1385 {
1386 	INIT_LIST_HEAD(&mw->mw_item);
1387 	init_completion(&mw->mw_complete);
1388 	ocfs2_init_start_time(mw);
1389 }
1390 
1391 static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
1392 {
1393 	wait_for_completion(&mw->mw_complete);
1394 	/* Re-arm the completion in case we want to wait on it again */
1395 	reinit_completion(&mw->mw_complete);
1396 	return mw->mw_status;
1397 }
1398 
1399 static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
1400 				    struct ocfs2_mask_waiter *mw,
1401 				    unsigned long mask,
1402 				    unsigned long goal)
1403 {
1404 	BUG_ON(!list_empty(&mw->mw_item));
1405 
1406 	assert_spin_locked(&lockres->l_lock);
1407 
1408 	list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
1409 	mw->mw_mask = mask;
1410 	mw->mw_goal = goal;
1411 	ocfs2_track_lock_wait(lockres);
1412 }
1413 
1414 /* returns 0 if the mw that was removed was already satisfied, -EBUSY
1415  * if the mask still hadn't reached its goal */
1416 static int __lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1417 				      struct ocfs2_mask_waiter *mw)
1418 {
1419 	int ret = 0;
1420 
1421 	assert_spin_locked(&lockres->l_lock);
1422 	if (!list_empty(&mw->mw_item)) {
1423 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
1424 			ret = -EBUSY;
1425 
1426 		list_del_init(&mw->mw_item);
1427 		init_completion(&mw->mw_complete);
1428 		ocfs2_track_lock_wait(lockres);
1429 	}
1430 
1431 	return ret;
1432 }
1433 
1434 static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1435 				      struct ocfs2_mask_waiter *mw)
1436 {
1437 	unsigned long flags;
1438 	int ret = 0;
1439 
1440 	spin_lock_irqsave(&lockres->l_lock, flags);
1441 	ret = __lockres_remove_mask_waiter(lockres, mw);
1442 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1443 
1444 	return ret;
1445 
1446 }
1447 
1448 static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
1449 					     struct ocfs2_lock_res *lockres)
1450 {
1451 	int ret;
1452 
1453 	ret = wait_for_completion_interruptible(&mw->mw_complete);
1454 	if (ret)
1455 		lockres_remove_mask_waiter(lockres, mw);
1456 	else
1457 		ret = mw->mw_status;
1458 	/* Re-arm the completion in case we want to wait on it again */
1459 	reinit_completion(&mw->mw_complete);
1460 	return ret;
1461 }
1462 
1463 static int __ocfs2_cluster_lock(struct ocfs2_super *osb,
1464 				struct ocfs2_lock_res *lockres,
1465 				int level,
1466 				u32 lkm_flags,
1467 				int arg_flags,
1468 				int l_subclass,
1469 				unsigned long caller_ip)
1470 {
1471 	struct ocfs2_mask_waiter mw;
1472 	int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
1473 	int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1474 	unsigned long flags;
1475 	unsigned int gen;
1476 	int noqueue_attempted = 0;
1477 	int dlm_locked = 0;
1478 	int kick_dc = 0;
1479 
1480 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED)) {
1481 		mlog_errno(-EINVAL);
1482 		return -EINVAL;
1483 	}
1484 
1485 	ocfs2_init_mask_waiter(&mw);
1486 
1487 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
1488 		lkm_flags |= DLM_LKF_VALBLK;
1489 
1490 again:
1491 	wait = 0;
1492 
1493 	spin_lock_irqsave(&lockres->l_lock, flags);
1494 
1495 	if (catch_signals && signal_pending(current)) {
1496 		ret = -ERESTARTSYS;
1497 		goto unlock;
1498 	}
1499 
1500 	mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
1501 			"Cluster lock called on freeing lockres %s! flags "
1502 			"0x%lx\n", lockres->l_name, lockres->l_flags);
1503 
1504 	/* We only compare against the currently granted level
1505 	 * here. If the lock is blocked waiting on a downconvert,
1506 	 * we'll get caught below. */
1507 	if (lockres->l_flags & OCFS2_LOCK_BUSY &&
1508 	    level > lockres->l_level) {
1509 		/* is someone sitting in dlm_lock? If so, wait on
1510 		 * them. */
1511 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1512 		wait = 1;
1513 		goto unlock;
1514 	}
1515 
1516 	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING) {
1517 		/*
1518 		 * We've upconverted. If the lock now has a level we can
1519 		 * work with, we take it. If, however, the lock is not at the
1520 		 * required level, we go thru the full cycle. One way this could
1521 		 * happen is if a process requesting an upconvert to PR is
1522 		 * closely followed by another requesting upconvert to an EX.
1523 		 * If the process requesting EX lands here, we want it to
1524 		 * continue attempting to upconvert and let the process
1525 		 * requesting PR take the lock.
1526 		 * If multiple processes request upconvert to PR, the first one
1527 		 * here will take the lock. The others will have to go thru the
1528 		 * OCFS2_LOCK_BLOCKED check to ensure that there is no pending
1529 		 * downconvert request.
1530 		 */
1531 		if (level <= lockres->l_level)
1532 			goto update_holders;
1533 	}
1534 
1535 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
1536 	    !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
1537 		/* is the lock is currently blocked on behalf of
1538 		 * another node */
1539 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
1540 		wait = 1;
1541 		goto unlock;
1542 	}
1543 
1544 	if (level > lockres->l_level) {
1545 		if (noqueue_attempted > 0) {
1546 			ret = -EAGAIN;
1547 			goto unlock;
1548 		}
1549 		if (lkm_flags & DLM_LKF_NOQUEUE)
1550 			noqueue_attempted = 1;
1551 
1552 		if (lockres->l_action != OCFS2_AST_INVALID)
1553 			mlog(ML_ERROR, "lockres %s has action %u pending\n",
1554 			     lockres->l_name, lockres->l_action);
1555 
1556 		if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1557 			lockres->l_action = OCFS2_AST_ATTACH;
1558 			lkm_flags &= ~DLM_LKF_CONVERT;
1559 		} else {
1560 			lockres->l_action = OCFS2_AST_CONVERT;
1561 			lkm_flags |= DLM_LKF_CONVERT;
1562 		}
1563 
1564 		lockres->l_requested = level;
1565 		lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1566 		gen = lockres_set_pending(lockres);
1567 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1568 
1569 		BUG_ON(level == DLM_LOCK_IV);
1570 		BUG_ON(level == DLM_LOCK_NL);
1571 
1572 		mlog(ML_BASTS, "lockres %s, convert from %d to %d\n",
1573 		     lockres->l_name, lockres->l_level, level);
1574 
1575 		/* call dlm_lock to upgrade lock now */
1576 		ret = ocfs2_dlm_lock(osb->cconn,
1577 				     level,
1578 				     &lockres->l_lksb,
1579 				     lkm_flags,
1580 				     lockres->l_name,
1581 				     OCFS2_LOCK_ID_MAX_LEN - 1);
1582 		lockres_clear_pending(lockres, gen, osb);
1583 		if (ret) {
1584 			if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
1585 			    (ret != -EAGAIN)) {
1586 				ocfs2_log_dlm_error("ocfs2_dlm_lock",
1587 						    ret, lockres);
1588 			}
1589 			ocfs2_recover_from_dlm_error(lockres, 1);
1590 			goto out;
1591 		}
1592 		dlm_locked = 1;
1593 
1594 		mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n",
1595 		     lockres->l_name);
1596 
1597 		/* At this point we've gone inside the dlm and need to
1598 		 * complete our work regardless. */
1599 		catch_signals = 0;
1600 
1601 		/* wait for busy to clear and carry on */
1602 		goto again;
1603 	}
1604 
1605 update_holders:
1606 	/* Ok, if we get here then we're good to go. */
1607 	ocfs2_inc_holders(lockres, level);
1608 
1609 	ret = 0;
1610 unlock:
1611 	lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1612 
1613 	/* ocfs2_unblock_lock reques on seeing OCFS2_LOCK_UPCONVERT_FINISHING */
1614 	kick_dc = (lockres->l_flags & OCFS2_LOCK_BLOCKED);
1615 
1616 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1617 	if (kick_dc)
1618 		ocfs2_wake_downconvert_thread(osb);
1619 out:
1620 	/*
1621 	 * This is helping work around a lock inversion between the page lock
1622 	 * and dlm locks.  One path holds the page lock while calling aops
1623 	 * which block acquiring dlm locks.  The voting thread holds dlm
1624 	 * locks while acquiring page locks while down converting data locks.
1625 	 * This block is helping an aop path notice the inversion and back
1626 	 * off to unlock its page lock before trying the dlm lock again.
1627 	 */
1628 	if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1629 	    mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1630 		wait = 0;
1631 		spin_lock_irqsave(&lockres->l_lock, flags);
1632 		if (__lockres_remove_mask_waiter(lockres, &mw)) {
1633 			if (dlm_locked)
1634 				lockres_or_flags(lockres,
1635 					OCFS2_LOCK_NONBLOCK_FINISHED);
1636 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1637 			ret = -EAGAIN;
1638 		} else {
1639 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1640 			goto again;
1641 		}
1642 	}
1643 	if (wait) {
1644 		ret = ocfs2_wait_for_mask(&mw);
1645 		if (ret == 0)
1646 			goto again;
1647 		mlog_errno(ret);
1648 	}
1649 	ocfs2_update_lock_stats(lockres, level, &mw, ret);
1650 
1651 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1652 	if (!ret && lockres->l_lockdep_map.key != NULL) {
1653 		if (level == DLM_LOCK_PR)
1654 			rwsem_acquire_read(&lockres->l_lockdep_map, l_subclass,
1655 				!!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1656 				caller_ip);
1657 		else
1658 			rwsem_acquire(&lockres->l_lockdep_map, l_subclass,
1659 				!!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1660 				caller_ip);
1661 	}
1662 #endif
1663 	return ret;
1664 }
1665 
1666 static inline int ocfs2_cluster_lock(struct ocfs2_super *osb,
1667 				     struct ocfs2_lock_res *lockres,
1668 				     int level,
1669 				     u32 lkm_flags,
1670 				     int arg_flags)
1671 {
1672 	return __ocfs2_cluster_lock(osb, lockres, level, lkm_flags, arg_flags,
1673 				    0, _RET_IP_);
1674 }
1675 
1676 
1677 static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
1678 				   struct ocfs2_lock_res *lockres,
1679 				   int level,
1680 				   unsigned long caller_ip)
1681 {
1682 	unsigned long flags;
1683 
1684 	spin_lock_irqsave(&lockres->l_lock, flags);
1685 	ocfs2_dec_holders(lockres, level);
1686 	ocfs2_downconvert_on_unlock(osb, lockres);
1687 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1688 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1689 	if (lockres->l_lockdep_map.key != NULL)
1690 		rwsem_release(&lockres->l_lockdep_map, 1, caller_ip);
1691 #endif
1692 }
1693 
1694 static int ocfs2_create_new_lock(struct ocfs2_super *osb,
1695 				 struct ocfs2_lock_res *lockres,
1696 				 int ex,
1697 				 int local)
1698 {
1699 	int level =  ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1700 	unsigned long flags;
1701 	u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
1702 
1703 	spin_lock_irqsave(&lockres->l_lock, flags);
1704 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1705 	lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1706 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1707 
1708 	return ocfs2_lock_create(osb, lockres, level, lkm_flags);
1709 }
1710 
1711 /* Grants us an EX lock on the data and metadata resources, skipping
1712  * the normal cluster directory lookup. Use this ONLY on newly created
1713  * inodes which other nodes can't possibly see, and which haven't been
1714  * hashed in the inode hash yet. This can give us a good performance
1715  * increase as it'll skip the network broadcast normally associated
1716  * with creating a new lock resource. */
1717 int ocfs2_create_new_inode_locks(struct inode *inode)
1718 {
1719 	int ret;
1720 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1721 
1722 	BUG_ON(!ocfs2_inode_is_new(inode));
1723 
1724 	mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
1725 
1726 	/* NOTE: That we don't increment any of the holder counts, nor
1727 	 * do we add anything to a journal handle. Since this is
1728 	 * supposed to be a new inode which the cluster doesn't know
1729 	 * about yet, there is no need to.  As far as the LVB handling
1730 	 * is concerned, this is basically like acquiring an EX lock
1731 	 * on a resource which has an invalid one -- we'll set it
1732 	 * valid when we release the EX. */
1733 
1734 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
1735 	if (ret) {
1736 		mlog_errno(ret);
1737 		goto bail;
1738 	}
1739 
1740 	/*
1741 	 * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
1742 	 * don't use a generation in their lock names.
1743 	 */
1744 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
1745 	if (ret) {
1746 		mlog_errno(ret);
1747 		goto bail;
1748 	}
1749 
1750 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
1751 	if (ret)
1752 		mlog_errno(ret);
1753 
1754 bail:
1755 	return ret;
1756 }
1757 
1758 int ocfs2_rw_lock(struct inode *inode, int write)
1759 {
1760 	int status, level;
1761 	struct ocfs2_lock_res *lockres;
1762 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1763 
1764 	mlog(0, "inode %llu take %s RW lock\n",
1765 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1766 	     write ? "EXMODE" : "PRMODE");
1767 
1768 	if (ocfs2_mount_local(osb))
1769 		return 0;
1770 
1771 	lockres = &OCFS2_I(inode)->ip_rw_lockres;
1772 
1773 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1774 
1775 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
1776 	if (status < 0)
1777 		mlog_errno(status);
1778 
1779 	return status;
1780 }
1781 
1782 int ocfs2_try_rw_lock(struct inode *inode, int write)
1783 {
1784 	int status, level;
1785 	struct ocfs2_lock_res *lockres;
1786 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1787 
1788 	mlog(0, "inode %llu try to take %s RW lock\n",
1789 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1790 	     write ? "EXMODE" : "PRMODE");
1791 
1792 	if (ocfs2_mount_local(osb))
1793 		return 0;
1794 
1795 	lockres = &OCFS2_I(inode)->ip_rw_lockres;
1796 
1797 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1798 
1799 	status = ocfs2_cluster_lock(osb, lockres, level, DLM_LKF_NOQUEUE, 0);
1800 	return status;
1801 }
1802 
1803 void ocfs2_rw_unlock(struct inode *inode, int write)
1804 {
1805 	int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1806 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
1807 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1808 
1809 	mlog(0, "inode %llu drop %s RW lock\n",
1810 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1811 	     write ? "EXMODE" : "PRMODE");
1812 
1813 	if (!ocfs2_mount_local(osb))
1814 		ocfs2_cluster_unlock(osb, lockres, level);
1815 }
1816 
1817 /*
1818  * ocfs2_open_lock always get PR mode lock.
1819  */
1820 int ocfs2_open_lock(struct inode *inode)
1821 {
1822 	int status = 0;
1823 	struct ocfs2_lock_res *lockres;
1824 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1825 
1826 	mlog(0, "inode %llu take PRMODE open lock\n",
1827 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1828 
1829 	if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1830 		goto out;
1831 
1832 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1833 
1834 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_PR, 0, 0);
1835 	if (status < 0)
1836 		mlog_errno(status);
1837 
1838 out:
1839 	return status;
1840 }
1841 
1842 int ocfs2_try_open_lock(struct inode *inode, int write)
1843 {
1844 	int status = 0, level;
1845 	struct ocfs2_lock_res *lockres;
1846 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1847 
1848 	mlog(0, "inode %llu try to take %s open lock\n",
1849 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1850 	     write ? "EXMODE" : "PRMODE");
1851 
1852 	if (ocfs2_is_hard_readonly(osb)) {
1853 		if (write)
1854 			status = -EROFS;
1855 		goto out;
1856 	}
1857 
1858 	if (ocfs2_mount_local(osb))
1859 		goto out;
1860 
1861 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1862 
1863 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1864 
1865 	/*
1866 	 * The file system may already holding a PRMODE/EXMODE open lock.
1867 	 * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
1868 	 * other nodes and the -EAGAIN will indicate to the caller that
1869 	 * this inode is still in use.
1870 	 */
1871 	status = ocfs2_cluster_lock(osb, lockres, level, DLM_LKF_NOQUEUE, 0);
1872 
1873 out:
1874 	return status;
1875 }
1876 
1877 /*
1878  * ocfs2_open_unlock unlock PR and EX mode open locks.
1879  */
1880 void ocfs2_open_unlock(struct inode *inode)
1881 {
1882 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
1883 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1884 
1885 	mlog(0, "inode %llu drop open lock\n",
1886 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1887 
1888 	if (ocfs2_mount_local(osb))
1889 		goto out;
1890 
1891 	if(lockres->l_ro_holders)
1892 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_PR);
1893 	if(lockres->l_ex_holders)
1894 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
1895 
1896 out:
1897 	return;
1898 }
1899 
1900 static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
1901 				     int level)
1902 {
1903 	int ret;
1904 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1905 	unsigned long flags;
1906 	struct ocfs2_mask_waiter mw;
1907 
1908 	ocfs2_init_mask_waiter(&mw);
1909 
1910 retry_cancel:
1911 	spin_lock_irqsave(&lockres->l_lock, flags);
1912 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
1913 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
1914 		if (ret) {
1915 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1916 			ret = ocfs2_cancel_convert(osb, lockres);
1917 			if (ret < 0) {
1918 				mlog_errno(ret);
1919 				goto out;
1920 			}
1921 			goto retry_cancel;
1922 		}
1923 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1924 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1925 
1926 		ocfs2_wait_for_mask(&mw);
1927 		goto retry_cancel;
1928 	}
1929 
1930 	ret = -ERESTARTSYS;
1931 	/*
1932 	 * We may still have gotten the lock, in which case there's no
1933 	 * point to restarting the syscall.
1934 	 */
1935 	if (lockres->l_level == level)
1936 		ret = 0;
1937 
1938 	mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
1939 	     lockres->l_flags, lockres->l_level, lockres->l_action);
1940 
1941 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1942 
1943 out:
1944 	return ret;
1945 }
1946 
1947 /*
1948  * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
1949  * flock() calls. The locking approach this requires is sufficiently
1950  * different from all other cluster lock types that we implement a
1951  * separate path to the "low-level" dlm calls. In particular:
1952  *
1953  * - No optimization of lock levels is done - we take at exactly
1954  *   what's been requested.
1955  *
1956  * - No lock caching is employed. We immediately downconvert to
1957  *   no-lock at unlock time. This also means flock locks never go on
1958  *   the blocking list).
1959  *
1960  * - Since userspace can trivially deadlock itself with flock, we make
1961  *   sure to allow cancellation of a misbehaving applications flock()
1962  *   request.
1963  *
1964  * - Access to any flock lockres doesn't require concurrency, so we
1965  *   can simplify the code by requiring the caller to guarantee
1966  *   serialization of dlmglue flock calls.
1967  */
1968 int ocfs2_file_lock(struct file *file, int ex, int trylock)
1969 {
1970 	int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1971 	unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0;
1972 	unsigned long flags;
1973 	struct ocfs2_file_private *fp = file->private_data;
1974 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
1975 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1976 	struct ocfs2_mask_waiter mw;
1977 
1978 	ocfs2_init_mask_waiter(&mw);
1979 
1980 	if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
1981 	    (lockres->l_level > DLM_LOCK_NL)) {
1982 		mlog(ML_ERROR,
1983 		     "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
1984 		     "level: %u\n", lockres->l_name, lockres->l_flags,
1985 		     lockres->l_level);
1986 		return -EINVAL;
1987 	}
1988 
1989 	spin_lock_irqsave(&lockres->l_lock, flags);
1990 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1991 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1992 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1993 
1994 		/*
1995 		 * Get the lock at NLMODE to start - that way we
1996 		 * can cancel the upconvert request if need be.
1997 		 */
1998 		ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0);
1999 		if (ret < 0) {
2000 			mlog_errno(ret);
2001 			goto out;
2002 		}
2003 
2004 		ret = ocfs2_wait_for_mask(&mw);
2005 		if (ret) {
2006 			mlog_errno(ret);
2007 			goto out;
2008 		}
2009 		spin_lock_irqsave(&lockres->l_lock, flags);
2010 	}
2011 
2012 	lockres->l_action = OCFS2_AST_CONVERT;
2013 	lkm_flags |= DLM_LKF_CONVERT;
2014 	lockres->l_requested = level;
2015 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2016 
2017 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
2018 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2019 
2020 	ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags,
2021 			     lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1);
2022 	if (ret) {
2023 		if (!trylock || (ret != -EAGAIN)) {
2024 			ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
2025 			ret = -EINVAL;
2026 		}
2027 
2028 		ocfs2_recover_from_dlm_error(lockres, 1);
2029 		lockres_remove_mask_waiter(lockres, &mw);
2030 		goto out;
2031 	}
2032 
2033 	ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
2034 	if (ret == -ERESTARTSYS) {
2035 		/*
2036 		 * Userspace can cause deadlock itself with
2037 		 * flock(). Current behavior locally is to allow the
2038 		 * deadlock, but abort the system call if a signal is
2039 		 * received. We follow this example, otherwise a
2040 		 * poorly written program could sit in kernel until
2041 		 * reboot.
2042 		 *
2043 		 * Handling this is a bit more complicated for Ocfs2
2044 		 * though. We can't exit this function with an
2045 		 * outstanding lock request, so a cancel convert is
2046 		 * required. We intentionally overwrite 'ret' - if the
2047 		 * cancel fails and the lock was granted, it's easier
2048 		 * to just bubble success back up to the user.
2049 		 */
2050 		ret = ocfs2_flock_handle_signal(lockres, level);
2051 	} else if (!ret && (level > lockres->l_level)) {
2052 		/* Trylock failed asynchronously */
2053 		BUG_ON(!trylock);
2054 		ret = -EAGAIN;
2055 	}
2056 
2057 out:
2058 
2059 	mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
2060 	     lockres->l_name, ex, trylock, ret);
2061 	return ret;
2062 }
2063 
2064 void ocfs2_file_unlock(struct file *file)
2065 {
2066 	int ret;
2067 	unsigned int gen;
2068 	unsigned long flags;
2069 	struct ocfs2_file_private *fp = file->private_data;
2070 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
2071 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
2072 	struct ocfs2_mask_waiter mw;
2073 
2074 	ocfs2_init_mask_waiter(&mw);
2075 
2076 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
2077 		return;
2078 
2079 	if (lockres->l_level == DLM_LOCK_NL)
2080 		return;
2081 
2082 	mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
2083 	     lockres->l_name, lockres->l_flags, lockres->l_level,
2084 	     lockres->l_action);
2085 
2086 	spin_lock_irqsave(&lockres->l_lock, flags);
2087 	/*
2088 	 * Fake a blocking ast for the downconvert code.
2089 	 */
2090 	lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
2091 	lockres->l_blocking = DLM_LOCK_EX;
2092 
2093 	gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL);
2094 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
2095 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2096 
2097 	ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen);
2098 	if (ret) {
2099 		mlog_errno(ret);
2100 		return;
2101 	}
2102 
2103 	ret = ocfs2_wait_for_mask(&mw);
2104 	if (ret)
2105 		mlog_errno(ret);
2106 }
2107 
2108 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
2109 					struct ocfs2_lock_res *lockres)
2110 {
2111 	int kick = 0;
2112 
2113 	/* If we know that another node is waiting on our lock, kick
2114 	 * the downconvert thread * pre-emptively when we reach a release
2115 	 * condition. */
2116 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
2117 		switch(lockres->l_blocking) {
2118 		case DLM_LOCK_EX:
2119 			if (!lockres->l_ex_holders && !lockres->l_ro_holders)
2120 				kick = 1;
2121 			break;
2122 		case DLM_LOCK_PR:
2123 			if (!lockres->l_ex_holders)
2124 				kick = 1;
2125 			break;
2126 		default:
2127 			BUG();
2128 		}
2129 	}
2130 
2131 	if (kick)
2132 		ocfs2_wake_downconvert_thread(osb);
2133 }
2134 
2135 #define OCFS2_SEC_BITS   34
2136 #define OCFS2_SEC_SHIFT  (64 - 34)
2137 #define OCFS2_NSEC_MASK  ((1ULL << OCFS2_SEC_SHIFT) - 1)
2138 
2139 /* LVB only has room for 64 bits of time here so we pack it for
2140  * now. */
2141 static u64 ocfs2_pack_timespec(struct timespec64 *spec)
2142 {
2143 	u64 res;
2144 	u64 sec = clamp_t(time64_t, spec->tv_sec, 0, 0x3ffffffffull);
2145 	u32 nsec = spec->tv_nsec;
2146 
2147 	res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
2148 
2149 	return res;
2150 }
2151 
2152 /* Call this with the lockres locked. I am reasonably sure we don't
2153  * need ip_lock in this function as anyone who would be changing those
2154  * values is supposed to be blocked in ocfs2_inode_lock right now. */
2155 static void __ocfs2_stuff_meta_lvb(struct inode *inode)
2156 {
2157 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2158 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2159 	struct ocfs2_meta_lvb *lvb;
2160 
2161 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2162 
2163 	/*
2164 	 * Invalidate the LVB of a deleted inode - this way other
2165 	 * nodes are forced to go to disk and discover the new inode
2166 	 * status.
2167 	 */
2168 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
2169 		lvb->lvb_version = 0;
2170 		goto out;
2171 	}
2172 
2173 	lvb->lvb_version   = OCFS2_LVB_VERSION;
2174 	lvb->lvb_isize	   = cpu_to_be64(i_size_read(inode));
2175 	lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
2176 	lvb->lvb_iuid      = cpu_to_be32(i_uid_read(inode));
2177 	lvb->lvb_igid      = cpu_to_be32(i_gid_read(inode));
2178 	lvb->lvb_imode     = cpu_to_be16(inode->i_mode);
2179 	lvb->lvb_inlink    = cpu_to_be16(inode->i_nlink);
2180 	lvb->lvb_iatime_packed  =
2181 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
2182 	lvb->lvb_ictime_packed =
2183 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
2184 	lvb->lvb_imtime_packed =
2185 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
2186 	lvb->lvb_iattr    = cpu_to_be32(oi->ip_attr);
2187 	lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
2188 	lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
2189 
2190 out:
2191 	mlog_meta_lvb(0, lockres);
2192 }
2193 
2194 static void ocfs2_unpack_timespec(struct timespec64 *spec,
2195 				  u64 packed_time)
2196 {
2197 	spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
2198 	spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
2199 }
2200 
2201 static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
2202 {
2203 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2204 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2205 	struct ocfs2_meta_lvb *lvb;
2206 
2207 	mlog_meta_lvb(0, lockres);
2208 
2209 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2210 
2211 	/* We're safe here without the lockres lock... */
2212 	spin_lock(&oi->ip_lock);
2213 	oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
2214 	i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
2215 
2216 	oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
2217 	oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
2218 	ocfs2_set_inode_flags(inode);
2219 
2220 	/* fast-symlinks are a special case */
2221 	if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
2222 		inode->i_blocks = 0;
2223 	else
2224 		inode->i_blocks = ocfs2_inode_sector_count(inode);
2225 
2226 	i_uid_write(inode, be32_to_cpu(lvb->lvb_iuid));
2227 	i_gid_write(inode, be32_to_cpu(lvb->lvb_igid));
2228 	inode->i_mode    = be16_to_cpu(lvb->lvb_imode);
2229 	set_nlink(inode, be16_to_cpu(lvb->lvb_inlink));
2230 	ocfs2_unpack_timespec(&inode->i_atime,
2231 			      be64_to_cpu(lvb->lvb_iatime_packed));
2232 	ocfs2_unpack_timespec(&inode->i_mtime,
2233 			      be64_to_cpu(lvb->lvb_imtime_packed));
2234 	ocfs2_unpack_timespec(&inode->i_ctime,
2235 			      be64_to_cpu(lvb->lvb_ictime_packed));
2236 	spin_unlock(&oi->ip_lock);
2237 }
2238 
2239 static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
2240 					      struct ocfs2_lock_res *lockres)
2241 {
2242 	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2243 
2244 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)
2245 	    && lvb->lvb_version == OCFS2_LVB_VERSION
2246 	    && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
2247 		return 1;
2248 	return 0;
2249 }
2250 
2251 /* Determine whether a lock resource needs to be refreshed, and
2252  * arbitrate who gets to refresh it.
2253  *
2254  *   0 means no refresh needed.
2255  *
2256  *   > 0 means you need to refresh this and you MUST call
2257  *   ocfs2_complete_lock_res_refresh afterwards. */
2258 static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
2259 {
2260 	unsigned long flags;
2261 	int status = 0;
2262 
2263 refresh_check:
2264 	spin_lock_irqsave(&lockres->l_lock, flags);
2265 	if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
2266 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2267 		goto bail;
2268 	}
2269 
2270 	if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
2271 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2272 
2273 		ocfs2_wait_on_refreshing_lock(lockres);
2274 		goto refresh_check;
2275 	}
2276 
2277 	/* Ok, I'll be the one to refresh this lock. */
2278 	lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
2279 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2280 
2281 	status = 1;
2282 bail:
2283 	mlog(0, "status %d\n", status);
2284 	return status;
2285 }
2286 
2287 /* If status is non zero, I'll mark it as not being in refresh
2288  * anymroe, but i won't clear the needs refresh flag. */
2289 static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
2290 						   int status)
2291 {
2292 	unsigned long flags;
2293 
2294 	spin_lock_irqsave(&lockres->l_lock, flags);
2295 	lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
2296 	if (!status)
2297 		lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
2298 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2299 
2300 	wake_up(&lockres->l_event);
2301 }
2302 
2303 /* may or may not return a bh if it went to disk. */
2304 static int ocfs2_inode_lock_update(struct inode *inode,
2305 				  struct buffer_head **bh)
2306 {
2307 	int status = 0;
2308 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2309 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2310 	struct ocfs2_dinode *fe;
2311 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2312 
2313 	if (ocfs2_mount_local(osb))
2314 		goto bail;
2315 
2316 	spin_lock(&oi->ip_lock);
2317 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
2318 		mlog(0, "Orphaned inode %llu was deleted while we "
2319 		     "were waiting on a lock. ip_flags = 0x%x\n",
2320 		     (unsigned long long)oi->ip_blkno, oi->ip_flags);
2321 		spin_unlock(&oi->ip_lock);
2322 		status = -ENOENT;
2323 		goto bail;
2324 	}
2325 	spin_unlock(&oi->ip_lock);
2326 
2327 	if (!ocfs2_should_refresh_lock_res(lockres))
2328 		goto bail;
2329 
2330 	/* This will discard any caching information we might have had
2331 	 * for the inode metadata. */
2332 	ocfs2_metadata_cache_purge(INODE_CACHE(inode));
2333 
2334 	ocfs2_extent_map_trunc(inode, 0);
2335 
2336 	if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
2337 		mlog(0, "Trusting LVB on inode %llu\n",
2338 		     (unsigned long long)oi->ip_blkno);
2339 		ocfs2_refresh_inode_from_lvb(inode);
2340 	} else {
2341 		/* Boo, we have to go to disk. */
2342 		/* read bh, cast, ocfs2_refresh_inode */
2343 		status = ocfs2_read_inode_block(inode, bh);
2344 		if (status < 0) {
2345 			mlog_errno(status);
2346 			goto bail_refresh;
2347 		}
2348 		fe = (struct ocfs2_dinode *) (*bh)->b_data;
2349 
2350 		/* This is a good chance to make sure we're not
2351 		 * locking an invalid object.  ocfs2_read_inode_block()
2352 		 * already checked that the inode block is sane.
2353 		 *
2354 		 * We bug on a stale inode here because we checked
2355 		 * above whether it was wiped from disk. The wiping
2356 		 * node provides a guarantee that we receive that
2357 		 * message and can mark the inode before dropping any
2358 		 * locks associated with it. */
2359 		mlog_bug_on_msg(inode->i_generation !=
2360 				le32_to_cpu(fe->i_generation),
2361 				"Invalid dinode %llu disk generation: %u "
2362 				"inode->i_generation: %u\n",
2363 				(unsigned long long)oi->ip_blkno,
2364 				le32_to_cpu(fe->i_generation),
2365 				inode->i_generation);
2366 		mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
2367 				!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
2368 				"Stale dinode %llu dtime: %llu flags: 0x%x\n",
2369 				(unsigned long long)oi->ip_blkno,
2370 				(unsigned long long)le64_to_cpu(fe->i_dtime),
2371 				le32_to_cpu(fe->i_flags));
2372 
2373 		ocfs2_refresh_inode(inode, fe);
2374 		ocfs2_track_lock_refresh(lockres);
2375 	}
2376 
2377 	status = 0;
2378 bail_refresh:
2379 	ocfs2_complete_lock_res_refresh(lockres, status);
2380 bail:
2381 	return status;
2382 }
2383 
2384 static int ocfs2_assign_bh(struct inode *inode,
2385 			   struct buffer_head **ret_bh,
2386 			   struct buffer_head *passed_bh)
2387 {
2388 	int status;
2389 
2390 	if (passed_bh) {
2391 		/* Ok, the update went to disk for us, use the
2392 		 * returned bh. */
2393 		*ret_bh = passed_bh;
2394 		get_bh(*ret_bh);
2395 
2396 		return 0;
2397 	}
2398 
2399 	status = ocfs2_read_inode_block(inode, ret_bh);
2400 	if (status < 0)
2401 		mlog_errno(status);
2402 
2403 	return status;
2404 }
2405 
2406 /*
2407  * returns < 0 error if the callback will never be called, otherwise
2408  * the result of the lock will be communicated via the callback.
2409  */
2410 int ocfs2_inode_lock_full_nested(struct inode *inode,
2411 				 struct buffer_head **ret_bh,
2412 				 int ex,
2413 				 int arg_flags,
2414 				 int subclass)
2415 {
2416 	int status, level, acquired;
2417 	u32 dlm_flags;
2418 	struct ocfs2_lock_res *lockres = NULL;
2419 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2420 	struct buffer_head *local_bh = NULL;
2421 
2422 	mlog(0, "inode %llu, take %s META lock\n",
2423 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2424 	     ex ? "EXMODE" : "PRMODE");
2425 
2426 	status = 0;
2427 	acquired = 0;
2428 	/* We'll allow faking a readonly metadata lock for
2429 	 * rodevices. */
2430 	if (ocfs2_is_hard_readonly(osb)) {
2431 		if (ex)
2432 			status = -EROFS;
2433 		goto getbh;
2434 	}
2435 
2436 	if ((arg_flags & OCFS2_META_LOCK_GETBH) ||
2437 	    ocfs2_mount_local(osb))
2438 		goto update;
2439 
2440 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2441 		ocfs2_wait_for_recovery(osb);
2442 
2443 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2444 	level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2445 	dlm_flags = 0;
2446 	if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
2447 		dlm_flags |= DLM_LKF_NOQUEUE;
2448 
2449 	status = __ocfs2_cluster_lock(osb, lockres, level, dlm_flags,
2450 				      arg_flags, subclass, _RET_IP_);
2451 	if (status < 0) {
2452 		if (status != -EAGAIN)
2453 			mlog_errno(status);
2454 		goto bail;
2455 	}
2456 
2457 	/* Notify the error cleanup path to drop the cluster lock. */
2458 	acquired = 1;
2459 
2460 	/* We wait twice because a node may have died while we were in
2461 	 * the lower dlm layers. The second time though, we've
2462 	 * committed to owning this lock so we don't allow signals to
2463 	 * abort the operation. */
2464 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2465 		ocfs2_wait_for_recovery(osb);
2466 
2467 update:
2468 	/*
2469 	 * We only see this flag if we're being called from
2470 	 * ocfs2_read_locked_inode(). It means we're locking an inode
2471 	 * which hasn't been populated yet, so clear the refresh flag
2472 	 * and let the caller handle it.
2473 	 */
2474 	if (inode->i_state & I_NEW) {
2475 		status = 0;
2476 		if (lockres)
2477 			ocfs2_complete_lock_res_refresh(lockres, 0);
2478 		goto bail;
2479 	}
2480 
2481 	/* This is fun. The caller may want a bh back, or it may
2482 	 * not. ocfs2_inode_lock_update definitely wants one in, but
2483 	 * may or may not read one, depending on what's in the
2484 	 * LVB. The result of all of this is that we've *only* gone to
2485 	 * disk if we have to, so the complexity is worthwhile. */
2486 	status = ocfs2_inode_lock_update(inode, &local_bh);
2487 	if (status < 0) {
2488 		if (status != -ENOENT)
2489 			mlog_errno(status);
2490 		goto bail;
2491 	}
2492 getbh:
2493 	if (ret_bh) {
2494 		status = ocfs2_assign_bh(inode, ret_bh, local_bh);
2495 		if (status < 0) {
2496 			mlog_errno(status);
2497 			goto bail;
2498 		}
2499 	}
2500 
2501 bail:
2502 	if (status < 0) {
2503 		if (ret_bh && (*ret_bh)) {
2504 			brelse(*ret_bh);
2505 			*ret_bh = NULL;
2506 		}
2507 		if (acquired)
2508 			ocfs2_inode_unlock(inode, ex);
2509 	}
2510 
2511 	brelse(local_bh);
2512 	return status;
2513 }
2514 
2515 /*
2516  * This is working around a lock inversion between tasks acquiring DLM
2517  * locks while holding a page lock and the downconvert thread which
2518  * blocks dlm lock acquiry while acquiring page locks.
2519  *
2520  * ** These _with_page variantes are only intended to be called from aop
2521  * methods that hold page locks and return a very specific *positive* error
2522  * code that aop methods pass up to the VFS -- test for errors with != 0. **
2523  *
2524  * The DLM is called such that it returns -EAGAIN if it would have
2525  * blocked waiting for the downconvert thread.  In that case we unlock
2526  * our page so the downconvert thread can make progress.  Once we've
2527  * done this we have to return AOP_TRUNCATED_PAGE so the aop method
2528  * that called us can bubble that back up into the VFS who will then
2529  * immediately retry the aop call.
2530  */
2531 int ocfs2_inode_lock_with_page(struct inode *inode,
2532 			      struct buffer_head **ret_bh,
2533 			      int ex,
2534 			      struct page *page)
2535 {
2536 	int ret;
2537 
2538 	ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
2539 	if (ret == -EAGAIN) {
2540 		unlock_page(page);
2541 		/*
2542 		 * If we can't get inode lock immediately, we should not return
2543 		 * directly here, since this will lead to a softlockup problem.
2544 		 * The method is to get a blocking lock and immediately unlock
2545 		 * before returning, this can avoid CPU resource waste due to
2546 		 * lots of retries, and benefits fairness in getting lock.
2547 		 */
2548 		if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
2549 			ocfs2_inode_unlock(inode, ex);
2550 		ret = AOP_TRUNCATED_PAGE;
2551 	}
2552 
2553 	return ret;
2554 }
2555 
2556 int ocfs2_inode_lock_atime(struct inode *inode,
2557 			  struct vfsmount *vfsmnt,
2558 			  int *level, int wait)
2559 {
2560 	int ret;
2561 
2562 	if (wait)
2563 		ret = ocfs2_inode_lock(inode, NULL, 0);
2564 	else
2565 		ret = ocfs2_try_inode_lock(inode, NULL, 0);
2566 
2567 	if (ret < 0) {
2568 		if (ret != -EAGAIN)
2569 			mlog_errno(ret);
2570 		return ret;
2571 	}
2572 
2573 	/*
2574 	 * If we should update atime, we will get EX lock,
2575 	 * otherwise we just get PR lock.
2576 	 */
2577 	if (ocfs2_should_update_atime(inode, vfsmnt)) {
2578 		struct buffer_head *bh = NULL;
2579 
2580 		ocfs2_inode_unlock(inode, 0);
2581 		if (wait)
2582 			ret = ocfs2_inode_lock(inode, &bh, 1);
2583 		else
2584 			ret = ocfs2_try_inode_lock(inode, &bh, 1);
2585 
2586 		if (ret < 0) {
2587 			if (ret != -EAGAIN)
2588 				mlog_errno(ret);
2589 			return ret;
2590 		}
2591 		*level = 1;
2592 		if (ocfs2_should_update_atime(inode, vfsmnt))
2593 			ocfs2_update_inode_atime(inode, bh);
2594 		brelse(bh);
2595 	} else
2596 		*level = 0;
2597 
2598 	return ret;
2599 }
2600 
2601 void ocfs2_inode_unlock(struct inode *inode,
2602 		       int ex)
2603 {
2604 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2605 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
2606 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2607 
2608 	mlog(0, "inode %llu drop %s META lock\n",
2609 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2610 	     ex ? "EXMODE" : "PRMODE");
2611 
2612 	if (!ocfs2_is_hard_readonly(osb) &&
2613 	    !ocfs2_mount_local(osb))
2614 		ocfs2_cluster_unlock(osb, lockres, level);
2615 }
2616 
2617 /*
2618  * This _tracker variantes are introduced to deal with the recursive cluster
2619  * locking issue. The idea is to keep track of a lock holder on the stack of
2620  * the current process. If there's a lock holder on the stack, we know the
2621  * task context is already protected by cluster locking. Currently, they're
2622  * used in some VFS entry routines.
2623  *
2624  * return < 0 on error, return == 0 if there's no lock holder on the stack
2625  * before this call, return == 1 if this call would be a recursive locking.
2626  * return == -1 if this lock attempt will cause an upgrade which is forbidden.
2627  *
2628  * When taking lock levels into account,we face some different situations.
2629  *
2630  * 1. no lock is held
2631  *    In this case, just lock the inode as requested and return 0
2632  *
2633  * 2. We are holding a lock
2634  *    For this situation, things diverges into several cases
2635  *
2636  *    wanted     holding	     what to do
2637  *    ex		ex	    see 2.1 below
2638  *    ex		pr	    see 2.2 below
2639  *    pr		ex	    see 2.1 below
2640  *    pr		pr	    see 2.1 below
2641  *
2642  *    2.1 lock level that is been held is compatible
2643  *    with the wanted level, so no lock action will be tacken.
2644  *
2645  *    2.2 Otherwise, an upgrade is needed, but it is forbidden.
2646  *
2647  * Reason why upgrade within a process is forbidden is that
2648  * lock upgrade may cause dead lock. The following illustrates
2649  * how it happens.
2650  *
2651  *         thread on node1                             thread on node2
2652  * ocfs2_inode_lock_tracker(ex=0)
2653  *
2654  *                                <======   ocfs2_inode_lock_tracker(ex=1)
2655  *
2656  * ocfs2_inode_lock_tracker(ex=1)
2657  */
2658 int ocfs2_inode_lock_tracker(struct inode *inode,
2659 			     struct buffer_head **ret_bh,
2660 			     int ex,
2661 			     struct ocfs2_lock_holder *oh)
2662 {
2663 	int status = 0;
2664 	struct ocfs2_lock_res *lockres;
2665 	struct ocfs2_lock_holder *tmp_oh;
2666 	struct pid *pid = task_pid(current);
2667 
2668 
2669 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2670 	tmp_oh = ocfs2_pid_holder(lockres, pid);
2671 
2672 	if (!tmp_oh) {
2673 		/*
2674 		 * This corresponds to the case 1.
2675 		 * We haven't got any lock before.
2676 		 */
2677 		status = ocfs2_inode_lock_full(inode, ret_bh, ex, 0);
2678 		if (status < 0) {
2679 			if (status != -ENOENT)
2680 				mlog_errno(status);
2681 			return status;
2682 		}
2683 
2684 		oh->oh_ex = ex;
2685 		ocfs2_add_holder(lockres, oh);
2686 		return 0;
2687 	}
2688 
2689 	if (unlikely(ex && !tmp_oh->oh_ex)) {
2690 		/*
2691 		 * case 2.2 upgrade may cause dead lock, forbid it.
2692 		 */
2693 		mlog(ML_ERROR, "Recursive locking is not permitted to "
2694 		     "upgrade to EX level from PR level.\n");
2695 		dump_stack();
2696 		return -EINVAL;
2697 	}
2698 
2699 	/*
2700 	 *  case 2.1 OCFS2_META_LOCK_GETBH flag make ocfs2_inode_lock_full.
2701 	 *  ignore the lock level and just update it.
2702 	 */
2703 	if (ret_bh) {
2704 		status = ocfs2_inode_lock_full(inode, ret_bh, ex,
2705 					       OCFS2_META_LOCK_GETBH);
2706 		if (status < 0) {
2707 			if (status != -ENOENT)
2708 				mlog_errno(status);
2709 			return status;
2710 		}
2711 	}
2712 	return tmp_oh ? 1 : 0;
2713 }
2714 
2715 void ocfs2_inode_unlock_tracker(struct inode *inode,
2716 				int ex,
2717 				struct ocfs2_lock_holder *oh,
2718 				int had_lock)
2719 {
2720 	struct ocfs2_lock_res *lockres;
2721 
2722 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2723 	/* had_lock means that the currect process already takes the cluster
2724 	 * lock previously.
2725 	 * If had_lock is 1, we have nothing to do here.
2726 	 * If had_lock is 0, we will release the lock.
2727 	 */
2728 	if (!had_lock) {
2729 		ocfs2_inode_unlock(inode, oh->oh_ex);
2730 		ocfs2_remove_holder(lockres, oh);
2731 	}
2732 }
2733 
2734 int ocfs2_orphan_scan_lock(struct ocfs2_super *osb, u32 *seqno)
2735 {
2736 	struct ocfs2_lock_res *lockres;
2737 	struct ocfs2_orphan_scan_lvb *lvb;
2738 	int status = 0;
2739 
2740 	if (ocfs2_is_hard_readonly(osb))
2741 		return -EROFS;
2742 
2743 	if (ocfs2_mount_local(osb))
2744 		return 0;
2745 
2746 	lockres = &osb->osb_orphan_scan.os_lockres;
2747 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2748 	if (status < 0)
2749 		return status;
2750 
2751 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2752 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2753 	    lvb->lvb_version == OCFS2_ORPHAN_LVB_VERSION)
2754 		*seqno = be32_to_cpu(lvb->lvb_os_seqno);
2755 	else
2756 		*seqno = osb->osb_orphan_scan.os_seqno + 1;
2757 
2758 	return status;
2759 }
2760 
2761 void ocfs2_orphan_scan_unlock(struct ocfs2_super *osb, u32 seqno)
2762 {
2763 	struct ocfs2_lock_res *lockres;
2764 	struct ocfs2_orphan_scan_lvb *lvb;
2765 
2766 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) {
2767 		lockres = &osb->osb_orphan_scan.os_lockres;
2768 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2769 		lvb->lvb_version = OCFS2_ORPHAN_LVB_VERSION;
2770 		lvb->lvb_os_seqno = cpu_to_be32(seqno);
2771 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2772 	}
2773 }
2774 
2775 int ocfs2_super_lock(struct ocfs2_super *osb,
2776 		     int ex)
2777 {
2778 	int status = 0;
2779 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2780 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2781 
2782 	if (ocfs2_is_hard_readonly(osb))
2783 		return -EROFS;
2784 
2785 	if (ocfs2_mount_local(osb))
2786 		goto bail;
2787 
2788 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
2789 	if (status < 0) {
2790 		mlog_errno(status);
2791 		goto bail;
2792 	}
2793 
2794 	/* The super block lock path is really in the best position to
2795 	 * know when resources covered by the lock need to be
2796 	 * refreshed, so we do it here. Of course, making sense of
2797 	 * everything is up to the caller :) */
2798 	status = ocfs2_should_refresh_lock_res(lockres);
2799 	if (status) {
2800 		status = ocfs2_refresh_slot_info(osb);
2801 
2802 		ocfs2_complete_lock_res_refresh(lockres, status);
2803 
2804 		if (status < 0) {
2805 			ocfs2_cluster_unlock(osb, lockres, level);
2806 			mlog_errno(status);
2807 		}
2808 		ocfs2_track_lock_refresh(lockres);
2809 	}
2810 bail:
2811 	return status;
2812 }
2813 
2814 void ocfs2_super_unlock(struct ocfs2_super *osb,
2815 			int ex)
2816 {
2817 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2818 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2819 
2820 	if (!ocfs2_mount_local(osb))
2821 		ocfs2_cluster_unlock(osb, lockres, level);
2822 }
2823 
2824 int ocfs2_rename_lock(struct ocfs2_super *osb)
2825 {
2826 	int status;
2827 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2828 
2829 	if (ocfs2_is_hard_readonly(osb))
2830 		return -EROFS;
2831 
2832 	if (ocfs2_mount_local(osb))
2833 		return 0;
2834 
2835 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2836 	if (status < 0)
2837 		mlog_errno(status);
2838 
2839 	return status;
2840 }
2841 
2842 void ocfs2_rename_unlock(struct ocfs2_super *osb)
2843 {
2844 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2845 
2846 	if (!ocfs2_mount_local(osb))
2847 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2848 }
2849 
2850 int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex)
2851 {
2852 	int status;
2853 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2854 
2855 	if (ocfs2_is_hard_readonly(osb))
2856 		return -EROFS;
2857 
2858 	if (ocfs2_mount_local(osb))
2859 		return 0;
2860 
2861 	status = ocfs2_cluster_lock(osb, lockres, ex ? LKM_EXMODE : LKM_PRMODE,
2862 				    0, 0);
2863 	if (status < 0)
2864 		mlog(ML_ERROR, "lock on nfs sync lock failed %d\n", status);
2865 
2866 	return status;
2867 }
2868 
2869 void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex)
2870 {
2871 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2872 
2873 	if (!ocfs2_mount_local(osb))
2874 		ocfs2_cluster_unlock(osb, lockres,
2875 				     ex ? LKM_EXMODE : LKM_PRMODE);
2876 }
2877 
2878 int ocfs2_trim_fs_lock(struct ocfs2_super *osb,
2879 		       struct ocfs2_trim_fs_info *info, int trylock)
2880 {
2881 	int status;
2882 	struct ocfs2_trim_fs_lvb *lvb;
2883 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
2884 
2885 	if (info)
2886 		info->tf_valid = 0;
2887 
2888 	if (ocfs2_is_hard_readonly(osb))
2889 		return -EROFS;
2890 
2891 	if (ocfs2_mount_local(osb))
2892 		return 0;
2893 
2894 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX,
2895 				    trylock ? DLM_LKF_NOQUEUE : 0, 0);
2896 	if (status < 0) {
2897 		if (status != -EAGAIN)
2898 			mlog_errno(status);
2899 		return status;
2900 	}
2901 
2902 	if (info) {
2903 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2904 		if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2905 		    lvb->lvb_version == OCFS2_TRIMFS_LVB_VERSION) {
2906 			info->tf_valid = 1;
2907 			info->tf_success = lvb->lvb_success;
2908 			info->tf_nodenum = be32_to_cpu(lvb->lvb_nodenum);
2909 			info->tf_start = be64_to_cpu(lvb->lvb_start);
2910 			info->tf_len = be64_to_cpu(lvb->lvb_len);
2911 			info->tf_minlen = be64_to_cpu(lvb->lvb_minlen);
2912 			info->tf_trimlen = be64_to_cpu(lvb->lvb_trimlen);
2913 		}
2914 	}
2915 
2916 	return status;
2917 }
2918 
2919 void ocfs2_trim_fs_unlock(struct ocfs2_super *osb,
2920 			  struct ocfs2_trim_fs_info *info)
2921 {
2922 	struct ocfs2_trim_fs_lvb *lvb;
2923 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
2924 
2925 	if (ocfs2_mount_local(osb))
2926 		return;
2927 
2928 	if (info) {
2929 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2930 		lvb->lvb_version = OCFS2_TRIMFS_LVB_VERSION;
2931 		lvb->lvb_success = info->tf_success;
2932 		lvb->lvb_nodenum = cpu_to_be32(info->tf_nodenum);
2933 		lvb->lvb_start = cpu_to_be64(info->tf_start);
2934 		lvb->lvb_len = cpu_to_be64(info->tf_len);
2935 		lvb->lvb_minlen = cpu_to_be64(info->tf_minlen);
2936 		lvb->lvb_trimlen = cpu_to_be64(info->tf_trimlen);
2937 	}
2938 
2939 	ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2940 }
2941 
2942 int ocfs2_dentry_lock(struct dentry *dentry, int ex)
2943 {
2944 	int ret;
2945 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2946 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2947 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2948 
2949 	BUG_ON(!dl);
2950 
2951 	if (ocfs2_is_hard_readonly(osb)) {
2952 		if (ex)
2953 			return -EROFS;
2954 		return 0;
2955 	}
2956 
2957 	if (ocfs2_mount_local(osb))
2958 		return 0;
2959 
2960 	ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2961 	if (ret < 0)
2962 		mlog_errno(ret);
2963 
2964 	return ret;
2965 }
2966 
2967 void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2968 {
2969 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2970 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2971 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2972 
2973 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
2974 		ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
2975 }
2976 
2977 /* Reference counting of the dlm debug structure. We want this because
2978  * open references on the debug inodes can live on after a mount, so
2979  * we can't rely on the ocfs2_super to always exist. */
2980 static void ocfs2_dlm_debug_free(struct kref *kref)
2981 {
2982 	struct ocfs2_dlm_debug *dlm_debug;
2983 
2984 	dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
2985 
2986 	kfree(dlm_debug);
2987 }
2988 
2989 void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
2990 {
2991 	if (dlm_debug)
2992 		kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
2993 }
2994 
2995 static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
2996 {
2997 	kref_get(&debug->d_refcnt);
2998 }
2999 
3000 struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
3001 {
3002 	struct ocfs2_dlm_debug *dlm_debug;
3003 
3004 	dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
3005 	if (!dlm_debug) {
3006 		mlog_errno(-ENOMEM);
3007 		goto out;
3008 	}
3009 
3010 	kref_init(&dlm_debug->d_refcnt);
3011 	INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
3012 	dlm_debug->d_filter_secs = 0;
3013 out:
3014 	return dlm_debug;
3015 }
3016 
3017 /* Access to this is arbitrated for us via seq_file->sem. */
3018 struct ocfs2_dlm_seq_priv {
3019 	struct ocfs2_dlm_debug *p_dlm_debug;
3020 	struct ocfs2_lock_res p_iter_res;
3021 	struct ocfs2_lock_res p_tmp_res;
3022 };
3023 
3024 static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
3025 						 struct ocfs2_dlm_seq_priv *priv)
3026 {
3027 	struct ocfs2_lock_res *iter, *ret = NULL;
3028 	struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
3029 
3030 	assert_spin_locked(&ocfs2_dlm_tracking_lock);
3031 
3032 	list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
3033 		/* discover the head of the list */
3034 		if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
3035 			mlog(0, "End of list found, %p\n", ret);
3036 			break;
3037 		}
3038 
3039 		/* We track our "dummy" iteration lockres' by a NULL
3040 		 * l_ops field. */
3041 		if (iter->l_ops != NULL) {
3042 			ret = iter;
3043 			break;
3044 		}
3045 	}
3046 
3047 	return ret;
3048 }
3049 
3050 static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
3051 {
3052 	struct ocfs2_dlm_seq_priv *priv = m->private;
3053 	struct ocfs2_lock_res *iter;
3054 
3055 	spin_lock(&ocfs2_dlm_tracking_lock);
3056 	iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
3057 	if (iter) {
3058 		/* Since lockres' have the lifetime of their container
3059 		 * (which can be inodes, ocfs2_supers, etc) we want to
3060 		 * copy this out to a temporary lockres while still
3061 		 * under the spinlock. Obviously after this we can't
3062 		 * trust any pointers on the copy returned, but that's
3063 		 * ok as the information we want isn't typically held
3064 		 * in them. */
3065 		priv->p_tmp_res = *iter;
3066 		iter = &priv->p_tmp_res;
3067 	}
3068 	spin_unlock(&ocfs2_dlm_tracking_lock);
3069 
3070 	return iter;
3071 }
3072 
3073 static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
3074 {
3075 }
3076 
3077 static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
3078 {
3079 	struct ocfs2_dlm_seq_priv *priv = m->private;
3080 	struct ocfs2_lock_res *iter = v;
3081 	struct ocfs2_lock_res *dummy = &priv->p_iter_res;
3082 
3083 	spin_lock(&ocfs2_dlm_tracking_lock);
3084 	iter = ocfs2_dlm_next_res(iter, priv);
3085 	list_del_init(&dummy->l_debug_list);
3086 	if (iter) {
3087 		list_add(&dummy->l_debug_list, &iter->l_debug_list);
3088 		priv->p_tmp_res = *iter;
3089 		iter = &priv->p_tmp_res;
3090 	}
3091 	spin_unlock(&ocfs2_dlm_tracking_lock);
3092 
3093 	return iter;
3094 }
3095 
3096 /*
3097  * Version is used by debugfs.ocfs2 to determine the format being used
3098  *
3099  * New in version 2
3100  *	- Lock stats printed
3101  * New in version 3
3102  *	- Max time in lock stats is in usecs (instead of nsecs)
3103  * New in version 4
3104  *	- Add last pr/ex unlock times and first lock wait time in usecs
3105  */
3106 #define OCFS2_DLM_DEBUG_STR_VERSION 4
3107 static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
3108 {
3109 	int i;
3110 	char *lvb;
3111 	struct ocfs2_lock_res *lockres = v;
3112 #ifdef CONFIG_OCFS2_FS_STATS
3113 	u64 now, last;
3114 	struct ocfs2_dlm_debug *dlm_debug =
3115 			((struct ocfs2_dlm_seq_priv *)m->private)->p_dlm_debug;
3116 #endif
3117 
3118 	if (!lockres)
3119 		return -EINVAL;
3120 
3121 #ifdef CONFIG_OCFS2_FS_STATS
3122 	if (!lockres->l_lock_wait && dlm_debug->d_filter_secs) {
3123 		now = ktime_to_us(ktime_get_real());
3124 		if (lockres->l_lock_prmode.ls_last >
3125 		    lockres->l_lock_exmode.ls_last)
3126 			last = lockres->l_lock_prmode.ls_last;
3127 		else
3128 			last = lockres->l_lock_exmode.ls_last;
3129 		/*
3130 		 * Use d_filter_secs field to filter lock resources dump,
3131 		 * the default d_filter_secs(0) value filters nothing,
3132 		 * otherwise, only dump the last N seconds active lock
3133 		 * resources.
3134 		 */
3135 		if (div_u64(now - last, 1000000) > dlm_debug->d_filter_secs)
3136 			return 0;
3137 	}
3138 #endif
3139 
3140 	seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
3141 
3142 	if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
3143 		seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
3144 			   lockres->l_name,
3145 			   (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
3146 	else
3147 		seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
3148 
3149 	seq_printf(m, "%d\t"
3150 		   "0x%lx\t"
3151 		   "0x%x\t"
3152 		   "0x%x\t"
3153 		   "%u\t"
3154 		   "%u\t"
3155 		   "%d\t"
3156 		   "%d\t",
3157 		   lockres->l_level,
3158 		   lockres->l_flags,
3159 		   lockres->l_action,
3160 		   lockres->l_unlock_action,
3161 		   lockres->l_ro_holders,
3162 		   lockres->l_ex_holders,
3163 		   lockres->l_requested,
3164 		   lockres->l_blocking);
3165 
3166 	/* Dump the raw LVB */
3167 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
3168 	for(i = 0; i < DLM_LVB_LEN; i++)
3169 		seq_printf(m, "0x%x\t", lvb[i]);
3170 
3171 #ifdef CONFIG_OCFS2_FS_STATS
3172 # define lock_num_prmode(_l)		((_l)->l_lock_prmode.ls_gets)
3173 # define lock_num_exmode(_l)		((_l)->l_lock_exmode.ls_gets)
3174 # define lock_num_prmode_failed(_l)	((_l)->l_lock_prmode.ls_fail)
3175 # define lock_num_exmode_failed(_l)	((_l)->l_lock_exmode.ls_fail)
3176 # define lock_total_prmode(_l)		((_l)->l_lock_prmode.ls_total)
3177 # define lock_total_exmode(_l)		((_l)->l_lock_exmode.ls_total)
3178 # define lock_max_prmode(_l)		((_l)->l_lock_prmode.ls_max)
3179 # define lock_max_exmode(_l)		((_l)->l_lock_exmode.ls_max)
3180 # define lock_refresh(_l)		((_l)->l_lock_refresh)
3181 # define lock_last_prmode(_l)		((_l)->l_lock_prmode.ls_last)
3182 # define lock_last_exmode(_l)		((_l)->l_lock_exmode.ls_last)
3183 # define lock_wait(_l)			((_l)->l_lock_wait)
3184 #else
3185 # define lock_num_prmode(_l)		(0)
3186 # define lock_num_exmode(_l)		(0)
3187 # define lock_num_prmode_failed(_l)	(0)
3188 # define lock_num_exmode_failed(_l)	(0)
3189 # define lock_total_prmode(_l)		(0ULL)
3190 # define lock_total_exmode(_l)		(0ULL)
3191 # define lock_max_prmode(_l)		(0)
3192 # define lock_max_exmode(_l)		(0)
3193 # define lock_refresh(_l)		(0)
3194 # define lock_last_prmode(_l)		(0ULL)
3195 # define lock_last_exmode(_l)		(0ULL)
3196 # define lock_wait(_l)			(0ULL)
3197 #endif
3198 	/* The following seq_print was added in version 2 of this output */
3199 	seq_printf(m, "%u\t"
3200 		   "%u\t"
3201 		   "%u\t"
3202 		   "%u\t"
3203 		   "%llu\t"
3204 		   "%llu\t"
3205 		   "%u\t"
3206 		   "%u\t"
3207 		   "%u\t"
3208 		   "%llu\t"
3209 		   "%llu\t"
3210 		   "%llu\t",
3211 		   lock_num_prmode(lockres),
3212 		   lock_num_exmode(lockres),
3213 		   lock_num_prmode_failed(lockres),
3214 		   lock_num_exmode_failed(lockres),
3215 		   lock_total_prmode(lockres),
3216 		   lock_total_exmode(lockres),
3217 		   lock_max_prmode(lockres),
3218 		   lock_max_exmode(lockres),
3219 		   lock_refresh(lockres),
3220 		   lock_last_prmode(lockres),
3221 		   lock_last_exmode(lockres),
3222 		   lock_wait(lockres));
3223 
3224 	/* End the line */
3225 	seq_printf(m, "\n");
3226 	return 0;
3227 }
3228 
3229 static const struct seq_operations ocfs2_dlm_seq_ops = {
3230 	.start =	ocfs2_dlm_seq_start,
3231 	.stop =		ocfs2_dlm_seq_stop,
3232 	.next =		ocfs2_dlm_seq_next,
3233 	.show =		ocfs2_dlm_seq_show,
3234 };
3235 
3236 static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
3237 {
3238 	struct seq_file *seq = file->private_data;
3239 	struct ocfs2_dlm_seq_priv *priv = seq->private;
3240 	struct ocfs2_lock_res *res = &priv->p_iter_res;
3241 
3242 	ocfs2_remove_lockres_tracking(res);
3243 	ocfs2_put_dlm_debug(priv->p_dlm_debug);
3244 	return seq_release_private(inode, file);
3245 }
3246 
3247 static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
3248 {
3249 	struct ocfs2_dlm_seq_priv *priv;
3250 	struct ocfs2_super *osb;
3251 
3252 	priv = __seq_open_private(file, &ocfs2_dlm_seq_ops, sizeof(*priv));
3253 	if (!priv) {
3254 		mlog_errno(-ENOMEM);
3255 		return -ENOMEM;
3256 	}
3257 
3258 	osb = inode->i_private;
3259 	ocfs2_get_dlm_debug(osb->osb_dlm_debug);
3260 	priv->p_dlm_debug = osb->osb_dlm_debug;
3261 	INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
3262 
3263 	ocfs2_add_lockres_tracking(&priv->p_iter_res,
3264 				   priv->p_dlm_debug);
3265 
3266 	return 0;
3267 }
3268 
3269 static const struct file_operations ocfs2_dlm_debug_fops = {
3270 	.open =		ocfs2_dlm_debug_open,
3271 	.release =	ocfs2_dlm_debug_release,
3272 	.read =		seq_read,
3273 	.llseek =	seq_lseek,
3274 };
3275 
3276 static void ocfs2_dlm_init_debug(struct ocfs2_super *osb)
3277 {
3278 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
3279 
3280 	debugfs_create_file("locking_state", S_IFREG|S_IRUSR,
3281 			    osb->osb_debug_root, osb, &ocfs2_dlm_debug_fops);
3282 
3283 	debugfs_create_u32("locking_filter", 0600, osb->osb_debug_root,
3284 			   &dlm_debug->d_filter_secs);
3285 }
3286 
3287 static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
3288 {
3289 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
3290 
3291 	if (dlm_debug)
3292 		ocfs2_put_dlm_debug(dlm_debug);
3293 }
3294 
3295 int ocfs2_dlm_init(struct ocfs2_super *osb)
3296 {
3297 	int status = 0;
3298 	struct ocfs2_cluster_connection *conn = NULL;
3299 
3300 	if (ocfs2_mount_local(osb)) {
3301 		osb->node_num = 0;
3302 		goto local;
3303 	}
3304 
3305 	ocfs2_dlm_init_debug(osb);
3306 
3307 	/* launch downconvert thread */
3308 	osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc-%s",
3309 			osb->uuid_str);
3310 	if (IS_ERR(osb->dc_task)) {
3311 		status = PTR_ERR(osb->dc_task);
3312 		osb->dc_task = NULL;
3313 		mlog_errno(status);
3314 		goto bail;
3315 	}
3316 
3317 	/* for now, uuid == domain */
3318 	status = ocfs2_cluster_connect(osb->osb_cluster_stack,
3319 				       osb->osb_cluster_name,
3320 				       strlen(osb->osb_cluster_name),
3321 				       osb->uuid_str,
3322 				       strlen(osb->uuid_str),
3323 				       &lproto, ocfs2_do_node_down, osb,
3324 				       &conn);
3325 	if (status) {
3326 		mlog_errno(status);
3327 		goto bail;
3328 	}
3329 
3330 	status = ocfs2_cluster_this_node(conn, &osb->node_num);
3331 	if (status < 0) {
3332 		mlog_errno(status);
3333 		mlog(ML_ERROR,
3334 		     "could not find this host's node number\n");
3335 		ocfs2_cluster_disconnect(conn, 0);
3336 		goto bail;
3337 	}
3338 
3339 local:
3340 	ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
3341 	ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
3342 	ocfs2_nfs_sync_lock_res_init(&osb->osb_nfs_sync_lockres, osb);
3343 	ocfs2_orphan_scan_lock_res_init(&osb->osb_orphan_scan.os_lockres, osb);
3344 
3345 	osb->cconn = conn;
3346 bail:
3347 	if (status < 0) {
3348 		ocfs2_dlm_shutdown_debug(osb);
3349 		if (osb->dc_task)
3350 			kthread_stop(osb->dc_task);
3351 	}
3352 
3353 	return status;
3354 }
3355 
3356 void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
3357 			int hangup_pending)
3358 {
3359 	ocfs2_drop_osb_locks(osb);
3360 
3361 	/*
3362 	 * Now that we have dropped all locks and ocfs2_dismount_volume()
3363 	 * has disabled recovery, the DLM won't be talking to us.  It's
3364 	 * safe to tear things down before disconnecting the cluster.
3365 	 */
3366 
3367 	if (osb->dc_task) {
3368 		kthread_stop(osb->dc_task);
3369 		osb->dc_task = NULL;
3370 	}
3371 
3372 	ocfs2_lock_res_free(&osb->osb_super_lockres);
3373 	ocfs2_lock_res_free(&osb->osb_rename_lockres);
3374 	ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres);
3375 	ocfs2_lock_res_free(&osb->osb_orphan_scan.os_lockres);
3376 
3377 	ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
3378 	osb->cconn = NULL;
3379 
3380 	ocfs2_dlm_shutdown_debug(osb);
3381 }
3382 
3383 static int ocfs2_drop_lock(struct ocfs2_super *osb,
3384 			   struct ocfs2_lock_res *lockres)
3385 {
3386 	int ret;
3387 	unsigned long flags;
3388 	u32 lkm_flags = 0;
3389 
3390 	/* We didn't get anywhere near actually using this lockres. */
3391 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
3392 		goto out;
3393 
3394 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
3395 		lkm_flags |= DLM_LKF_VALBLK;
3396 
3397 	spin_lock_irqsave(&lockres->l_lock, flags);
3398 
3399 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
3400 			"lockres %s, flags 0x%lx\n",
3401 			lockres->l_name, lockres->l_flags);
3402 
3403 	while (lockres->l_flags & OCFS2_LOCK_BUSY) {
3404 		mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
3405 		     "%u, unlock_action = %u\n",
3406 		     lockres->l_name, lockres->l_flags, lockres->l_action,
3407 		     lockres->l_unlock_action);
3408 
3409 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3410 
3411 		/* XXX: Today we just wait on any busy
3412 		 * locks... Perhaps we need to cancel converts in the
3413 		 * future? */
3414 		ocfs2_wait_on_busy_lock(lockres);
3415 
3416 		spin_lock_irqsave(&lockres->l_lock, flags);
3417 	}
3418 
3419 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3420 		if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
3421 		    lockres->l_level == DLM_LOCK_EX &&
3422 		    !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3423 			lockres->l_ops->set_lvb(lockres);
3424 	}
3425 
3426 	if (lockres->l_flags & OCFS2_LOCK_BUSY)
3427 		mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
3428 		     lockres->l_name);
3429 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
3430 		mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
3431 
3432 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
3433 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3434 		goto out;
3435 	}
3436 
3437 	lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
3438 
3439 	/* make sure we never get here while waiting for an ast to
3440 	 * fire. */
3441 	BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
3442 
3443 	/* is this necessary? */
3444 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3445 	lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
3446 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3447 
3448 	mlog(0, "lock %s\n", lockres->l_name);
3449 
3450 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags);
3451 	if (ret) {
3452 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3453 		mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
3454 		ocfs2_dlm_dump_lksb(&lockres->l_lksb);
3455 		BUG();
3456 	}
3457 	mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n",
3458 	     lockres->l_name);
3459 
3460 	ocfs2_wait_on_busy_lock(lockres);
3461 out:
3462 	return 0;
3463 }
3464 
3465 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3466 				       struct ocfs2_lock_res *lockres);
3467 
3468 /* Mark the lockres as being dropped. It will no longer be
3469  * queued if blocking, but we still may have to wait on it
3470  * being dequeued from the downconvert thread before we can consider
3471  * it safe to drop.
3472  *
3473  * You can *not* attempt to call cluster_lock on this lockres anymore. */
3474 void ocfs2_mark_lockres_freeing(struct ocfs2_super *osb,
3475 				struct ocfs2_lock_res *lockres)
3476 {
3477 	int status;
3478 	struct ocfs2_mask_waiter mw;
3479 	unsigned long flags, flags2;
3480 
3481 	ocfs2_init_mask_waiter(&mw);
3482 
3483 	spin_lock_irqsave(&lockres->l_lock, flags);
3484 	lockres->l_flags |= OCFS2_LOCK_FREEING;
3485 	if (lockres->l_flags & OCFS2_LOCK_QUEUED && current == osb->dc_task) {
3486 		/*
3487 		 * We know the downconvert is queued but not in progress
3488 		 * because we are the downconvert thread and processing
3489 		 * different lock. So we can just remove the lock from the
3490 		 * queue. This is not only an optimization but also a way
3491 		 * to avoid the following deadlock:
3492 		 *   ocfs2_dentry_post_unlock()
3493 		 *     ocfs2_dentry_lock_put()
3494 		 *       ocfs2_drop_dentry_lock()
3495 		 *         iput()
3496 		 *           ocfs2_evict_inode()
3497 		 *             ocfs2_clear_inode()
3498 		 *               ocfs2_mark_lockres_freeing()
3499 		 *                 ... blocks waiting for OCFS2_LOCK_QUEUED
3500 		 *                 since we are the downconvert thread which
3501 		 *                 should clear the flag.
3502 		 */
3503 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3504 		spin_lock_irqsave(&osb->dc_task_lock, flags2);
3505 		list_del_init(&lockres->l_blocked_list);
3506 		osb->blocked_lock_count--;
3507 		spin_unlock_irqrestore(&osb->dc_task_lock, flags2);
3508 		/*
3509 		 * Warn if we recurse into another post_unlock call.  Strictly
3510 		 * speaking it isn't a problem but we need to be careful if
3511 		 * that happens (stack overflow, deadlocks, ...) so warn if
3512 		 * ocfs2 grows a path for which this can happen.
3513 		 */
3514 		WARN_ON_ONCE(lockres->l_ops->post_unlock);
3515 		/* Since the lock is freeing we don't do much in the fn below */
3516 		ocfs2_process_blocked_lock(osb, lockres);
3517 		return;
3518 	}
3519 	while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
3520 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
3521 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3522 
3523 		mlog(0, "Waiting on lockres %s\n", lockres->l_name);
3524 
3525 		status = ocfs2_wait_for_mask(&mw);
3526 		if (status)
3527 			mlog_errno(status);
3528 
3529 		spin_lock_irqsave(&lockres->l_lock, flags);
3530 	}
3531 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3532 }
3533 
3534 void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
3535 			       struct ocfs2_lock_res *lockres)
3536 {
3537 	int ret;
3538 
3539 	ocfs2_mark_lockres_freeing(osb, lockres);
3540 	ret = ocfs2_drop_lock(osb, lockres);
3541 	if (ret)
3542 		mlog_errno(ret);
3543 }
3544 
3545 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
3546 {
3547 	ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
3548 	ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
3549 	ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres);
3550 	ocfs2_simple_drop_lockres(osb, &osb->osb_orphan_scan.os_lockres);
3551 }
3552 
3553 int ocfs2_drop_inode_locks(struct inode *inode)
3554 {
3555 	int status, err;
3556 
3557 	/* No need to call ocfs2_mark_lockres_freeing here -
3558 	 * ocfs2_clear_inode has done it for us. */
3559 
3560 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3561 			      &OCFS2_I(inode)->ip_open_lockres);
3562 	if (err < 0)
3563 		mlog_errno(err);
3564 
3565 	status = err;
3566 
3567 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3568 			      &OCFS2_I(inode)->ip_inode_lockres);
3569 	if (err < 0)
3570 		mlog_errno(err);
3571 	if (err < 0 && !status)
3572 		status = err;
3573 
3574 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3575 			      &OCFS2_I(inode)->ip_rw_lockres);
3576 	if (err < 0)
3577 		mlog_errno(err);
3578 	if (err < 0 && !status)
3579 		status = err;
3580 
3581 	return status;
3582 }
3583 
3584 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
3585 					      int new_level)
3586 {
3587 	assert_spin_locked(&lockres->l_lock);
3588 
3589 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
3590 
3591 	if (lockres->l_level <= new_level) {
3592 		mlog(ML_ERROR, "lockres %s, lvl %d <= %d, blcklst %d, mask %d, "
3593 		     "type %d, flags 0x%lx, hold %d %d, act %d %d, req %d, "
3594 		     "block %d, pgen %d\n", lockres->l_name, lockres->l_level,
3595 		     new_level, list_empty(&lockres->l_blocked_list),
3596 		     list_empty(&lockres->l_mask_waiters), lockres->l_type,
3597 		     lockres->l_flags, lockres->l_ro_holders,
3598 		     lockres->l_ex_holders, lockres->l_action,
3599 		     lockres->l_unlock_action, lockres->l_requested,
3600 		     lockres->l_blocking, lockres->l_pending_gen);
3601 		BUG();
3602 	}
3603 
3604 	mlog(ML_BASTS, "lockres %s, level %d => %d, blocking %d\n",
3605 	     lockres->l_name, lockres->l_level, new_level, lockres->l_blocking);
3606 
3607 	lockres->l_action = OCFS2_AST_DOWNCONVERT;
3608 	lockres->l_requested = new_level;
3609 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3610 	return lockres_set_pending(lockres);
3611 }
3612 
3613 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
3614 				  struct ocfs2_lock_res *lockres,
3615 				  int new_level,
3616 				  int lvb,
3617 				  unsigned int generation)
3618 {
3619 	int ret;
3620 	u32 dlm_flags = DLM_LKF_CONVERT;
3621 
3622 	mlog(ML_BASTS, "lockres %s, level %d => %d\n", lockres->l_name,
3623 	     lockres->l_level, new_level);
3624 
3625 	/*
3626 	 * On DLM_LKF_VALBLK, fsdlm behaves differently with o2cb. It always
3627 	 * expects DLM_LKF_VALBLK being set if the LKB has LVB, so that
3628 	 * we can recover correctly from node failure. Otherwise, we may get
3629 	 * invalid LVB in LKB, but without DLM_SBF_VALNOTVALID being set.
3630 	 */
3631 	if (ocfs2_userspace_stack(osb) &&
3632 	    lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
3633 		lvb = 1;
3634 
3635 	if (lvb)
3636 		dlm_flags |= DLM_LKF_VALBLK;
3637 
3638 	ret = ocfs2_dlm_lock(osb->cconn,
3639 			     new_level,
3640 			     &lockres->l_lksb,
3641 			     dlm_flags,
3642 			     lockres->l_name,
3643 			     OCFS2_LOCK_ID_MAX_LEN - 1);
3644 	lockres_clear_pending(lockres, generation, osb);
3645 	if (ret) {
3646 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
3647 		ocfs2_recover_from_dlm_error(lockres, 1);
3648 		goto bail;
3649 	}
3650 
3651 	ret = 0;
3652 bail:
3653 	return ret;
3654 }
3655 
3656 /* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
3657 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
3658 				        struct ocfs2_lock_res *lockres)
3659 {
3660 	assert_spin_locked(&lockres->l_lock);
3661 
3662 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
3663 		/* If we're already trying to cancel a lock conversion
3664 		 * then just drop the spinlock and allow the caller to
3665 		 * requeue this lock. */
3666 		mlog(ML_BASTS, "lockres %s, skip convert\n", lockres->l_name);
3667 		return 0;
3668 	}
3669 
3670 	/* were we in a convert when we got the bast fire? */
3671 	BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
3672 	       lockres->l_action != OCFS2_AST_DOWNCONVERT);
3673 	/* set things up for the unlockast to know to just
3674 	 * clear out the ast_action and unset busy, etc. */
3675 	lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
3676 
3677 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
3678 			"lock %s, invalid flags: 0x%lx\n",
3679 			lockres->l_name, lockres->l_flags);
3680 
3681 	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3682 
3683 	return 1;
3684 }
3685 
3686 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
3687 				struct ocfs2_lock_res *lockres)
3688 {
3689 	int ret;
3690 
3691 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
3692 			       DLM_LKF_CANCEL);
3693 	if (ret) {
3694 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3695 		ocfs2_recover_from_dlm_error(lockres, 0);
3696 	}
3697 
3698 	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3699 
3700 	return ret;
3701 }
3702 
3703 static int ocfs2_unblock_lock(struct ocfs2_super *osb,
3704 			      struct ocfs2_lock_res *lockres,
3705 			      struct ocfs2_unblock_ctl *ctl)
3706 {
3707 	unsigned long flags;
3708 	int blocking;
3709 	int new_level;
3710 	int level;
3711 	int ret = 0;
3712 	int set_lvb = 0;
3713 	unsigned int gen;
3714 
3715 	spin_lock_irqsave(&lockres->l_lock, flags);
3716 
3717 recheck:
3718 	/*
3719 	 * Is it still blocking? If not, we have no more work to do.
3720 	 */
3721 	if (!(lockres->l_flags & OCFS2_LOCK_BLOCKED)) {
3722 		BUG_ON(lockres->l_blocking != DLM_LOCK_NL);
3723 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3724 		ret = 0;
3725 		goto leave;
3726 	}
3727 
3728 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
3729 		/* XXX
3730 		 * This is a *big* race.  The OCFS2_LOCK_PENDING flag
3731 		 * exists entirely for one reason - another thread has set
3732 		 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
3733 		 *
3734 		 * If we do ocfs2_cancel_convert() before the other thread
3735 		 * calls dlm_lock(), our cancel will do nothing.  We will
3736 		 * get no ast, and we will have no way of knowing the
3737 		 * cancel failed.  Meanwhile, the other thread will call
3738 		 * into dlm_lock() and wait...forever.
3739 		 *
3740 		 * Why forever?  Because another node has asked for the
3741 		 * lock first; that's why we're here in unblock_lock().
3742 		 *
3743 		 * The solution is OCFS2_LOCK_PENDING.  When PENDING is
3744 		 * set, we just requeue the unblock.  Only when the other
3745 		 * thread has called dlm_lock() and cleared PENDING will
3746 		 * we then cancel their request.
3747 		 *
3748 		 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
3749 		 * at the same time they set OCFS2_DLM_BUSY.  They must
3750 		 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
3751 		 */
3752 		if (lockres->l_flags & OCFS2_LOCK_PENDING) {
3753 			mlog(ML_BASTS, "lockres %s, ReQ: Pending\n",
3754 			     lockres->l_name);
3755 			goto leave_requeue;
3756 		}
3757 
3758 		ctl->requeue = 1;
3759 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
3760 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3761 		if (ret) {
3762 			ret = ocfs2_cancel_convert(osb, lockres);
3763 			if (ret < 0)
3764 				mlog_errno(ret);
3765 		}
3766 		goto leave;
3767 	}
3768 
3769 	/*
3770 	 * This prevents livelocks. OCFS2_LOCK_UPCONVERT_FINISHING flag is
3771 	 * set when the ast is received for an upconvert just before the
3772 	 * OCFS2_LOCK_BUSY flag is cleared. Now if the fs received a bast
3773 	 * on the heels of the ast, we want to delay the downconvert just
3774 	 * enough to allow the up requestor to do its task. Because this
3775 	 * lock is in the blocked queue, the lock will be downconverted
3776 	 * as soon as the requestor is done with the lock.
3777 	 */
3778 	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING)
3779 		goto leave_requeue;
3780 
3781 	/*
3782 	 * How can we block and yet be at NL?  We were trying to upconvert
3783 	 * from NL and got canceled.  The code comes back here, and now
3784 	 * we notice and clear BLOCKING.
3785 	 */
3786 	if (lockres->l_level == DLM_LOCK_NL) {
3787 		BUG_ON(lockres->l_ex_holders || lockres->l_ro_holders);
3788 		mlog(ML_BASTS, "lockres %s, Aborting dc\n", lockres->l_name);
3789 		lockres->l_blocking = DLM_LOCK_NL;
3790 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
3791 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3792 		goto leave;
3793 	}
3794 
3795 	/* if we're blocking an exclusive and we have *any* holders,
3796 	 * then requeue. */
3797 	if ((lockres->l_blocking == DLM_LOCK_EX)
3798 	    && (lockres->l_ex_holders || lockres->l_ro_holders)) {
3799 		mlog(ML_BASTS, "lockres %s, ReQ: EX/PR Holders %u,%u\n",
3800 		     lockres->l_name, lockres->l_ex_holders,
3801 		     lockres->l_ro_holders);
3802 		goto leave_requeue;
3803 	}
3804 
3805 	/* If it's a PR we're blocking, then only
3806 	 * requeue if we've got any EX holders */
3807 	if (lockres->l_blocking == DLM_LOCK_PR &&
3808 	    lockres->l_ex_holders) {
3809 		mlog(ML_BASTS, "lockres %s, ReQ: EX Holders %u\n",
3810 		     lockres->l_name, lockres->l_ex_holders);
3811 		goto leave_requeue;
3812 	}
3813 
3814 	/*
3815 	 * Can we get a lock in this state if the holder counts are
3816 	 * zero? The meta data unblock code used to check this.
3817 	 */
3818 	if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
3819 	    && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) {
3820 		mlog(ML_BASTS, "lockres %s, ReQ: Lock Refreshing\n",
3821 		     lockres->l_name);
3822 		goto leave_requeue;
3823 	}
3824 
3825 	new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
3826 
3827 	if (lockres->l_ops->check_downconvert
3828 	    && !lockres->l_ops->check_downconvert(lockres, new_level)) {
3829 		mlog(ML_BASTS, "lockres %s, ReQ: Checkpointing\n",
3830 		     lockres->l_name);
3831 		goto leave_requeue;
3832 	}
3833 
3834 	/* If we get here, then we know that there are no more
3835 	 * incompatible holders (and anyone asking for an incompatible
3836 	 * lock is blocked). We can now downconvert the lock */
3837 	if (!lockres->l_ops->downconvert_worker)
3838 		goto downconvert;
3839 
3840 	/* Some lockres types want to do a bit of work before
3841 	 * downconverting a lock. Allow that here. The worker function
3842 	 * may sleep, so we save off a copy of what we're blocking as
3843 	 * it may change while we're not holding the spin lock. */
3844 	blocking = lockres->l_blocking;
3845 	level = lockres->l_level;
3846 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3847 
3848 	ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
3849 
3850 	if (ctl->unblock_action == UNBLOCK_STOP_POST) {
3851 		mlog(ML_BASTS, "lockres %s, UNBLOCK_STOP_POST\n",
3852 		     lockres->l_name);
3853 		goto leave;
3854 	}
3855 
3856 	spin_lock_irqsave(&lockres->l_lock, flags);
3857 	if ((blocking != lockres->l_blocking) || (level != lockres->l_level)) {
3858 		/* If this changed underneath us, then we can't drop
3859 		 * it just yet. */
3860 		mlog(ML_BASTS, "lockres %s, block=%d:%d, level=%d:%d, "
3861 		     "Recheck\n", lockres->l_name, blocking,
3862 		     lockres->l_blocking, level, lockres->l_level);
3863 		goto recheck;
3864 	}
3865 
3866 downconvert:
3867 	ctl->requeue = 0;
3868 
3869 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3870 		if (lockres->l_level == DLM_LOCK_EX)
3871 			set_lvb = 1;
3872 
3873 		/*
3874 		 * We only set the lvb if the lock has been fully
3875 		 * refreshed - otherwise we risk setting stale
3876 		 * data. Otherwise, there's no need to actually clear
3877 		 * out the lvb here as it's value is still valid.
3878 		 */
3879 		if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3880 			lockres->l_ops->set_lvb(lockres);
3881 	}
3882 
3883 	gen = ocfs2_prepare_downconvert(lockres, new_level);
3884 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3885 	ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
3886 				     gen);
3887 
3888 leave:
3889 	if (ret)
3890 		mlog_errno(ret);
3891 	return ret;
3892 
3893 leave_requeue:
3894 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3895 	ctl->requeue = 1;
3896 
3897 	return 0;
3898 }
3899 
3900 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
3901 				     int blocking)
3902 {
3903 	struct inode *inode;
3904 	struct address_space *mapping;
3905 	struct ocfs2_inode_info *oi;
3906 
3907        	inode = ocfs2_lock_res_inode(lockres);
3908 	mapping = inode->i_mapping;
3909 
3910 	if (S_ISDIR(inode->i_mode)) {
3911 		oi = OCFS2_I(inode);
3912 		oi->ip_dir_lock_gen++;
3913 		mlog(0, "generation: %u\n", oi->ip_dir_lock_gen);
3914 		goto out;
3915 	}
3916 
3917 	if (!S_ISREG(inode->i_mode))
3918 		goto out;
3919 
3920 	/*
3921 	 * We need this before the filemap_fdatawrite() so that it can
3922 	 * transfer the dirty bit from the PTE to the
3923 	 * page. Unfortunately this means that even for EX->PR
3924 	 * downconverts, we'll lose our mappings and have to build
3925 	 * them up again.
3926 	 */
3927 	unmap_mapping_range(mapping, 0, 0, 0);
3928 
3929 	if (filemap_fdatawrite(mapping)) {
3930 		mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
3931 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
3932 	}
3933 	sync_mapping_buffers(mapping);
3934 	if (blocking == DLM_LOCK_EX) {
3935 		truncate_inode_pages(mapping, 0);
3936 	} else {
3937 		/* We only need to wait on the I/O if we're not also
3938 		 * truncating pages because truncate_inode_pages waits
3939 		 * for us above. We don't truncate pages if we're
3940 		 * blocking anything < EXMODE because we want to keep
3941 		 * them around in that case. */
3942 		filemap_fdatawait(mapping);
3943 	}
3944 
3945 	forget_all_cached_acls(inode);
3946 
3947 out:
3948 	return UNBLOCK_CONTINUE;
3949 }
3950 
3951 static int ocfs2_ci_checkpointed(struct ocfs2_caching_info *ci,
3952 				 struct ocfs2_lock_res *lockres,
3953 				 int new_level)
3954 {
3955 	int checkpointed = ocfs2_ci_fully_checkpointed(ci);
3956 
3957 	BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
3958 	BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
3959 
3960 	if (checkpointed)
3961 		return 1;
3962 
3963 	ocfs2_start_checkpoint(OCFS2_SB(ocfs2_metadata_cache_get_super(ci)));
3964 	return 0;
3965 }
3966 
3967 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
3968 					int new_level)
3969 {
3970 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3971 
3972 	return ocfs2_ci_checkpointed(INODE_CACHE(inode), lockres, new_level);
3973 }
3974 
3975 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
3976 {
3977 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3978 
3979 	__ocfs2_stuff_meta_lvb(inode);
3980 }
3981 
3982 /*
3983  * Does the final reference drop on our dentry lock. Right now this
3984  * happens in the downconvert thread, but we could choose to simplify the
3985  * dlmglue API and push these off to the ocfs2_wq in the future.
3986  */
3987 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
3988 				     struct ocfs2_lock_res *lockres)
3989 {
3990 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3991 	ocfs2_dentry_lock_put(osb, dl);
3992 }
3993 
3994 /*
3995  * d_delete() matching dentries before the lock downconvert.
3996  *
3997  * At this point, any process waiting to destroy the
3998  * dentry_lock due to last ref count is stopped by the
3999  * OCFS2_LOCK_QUEUED flag.
4000  *
4001  * We have two potential problems
4002  *
4003  * 1) If we do the last reference drop on our dentry_lock (via dput)
4004  *    we'll wind up in ocfs2_release_dentry_lock(), waiting on
4005  *    the downconvert to finish. Instead we take an elevated
4006  *    reference and push the drop until after we've completed our
4007  *    unblock processing.
4008  *
4009  * 2) There might be another process with a final reference,
4010  *    waiting on us to finish processing. If this is the case, we
4011  *    detect it and exit out - there's no more dentries anyway.
4012  */
4013 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
4014 				       int blocking)
4015 {
4016 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
4017 	struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
4018 	struct dentry *dentry;
4019 	unsigned long flags;
4020 	int extra_ref = 0;
4021 
4022 	/*
4023 	 * This node is blocking another node from getting a read
4024 	 * lock. This happens when we've renamed within a
4025 	 * directory. We've forced the other nodes to d_delete(), but
4026 	 * we never actually dropped our lock because it's still
4027 	 * valid. The downconvert code will retain a PR for this node,
4028 	 * so there's no further work to do.
4029 	 */
4030 	if (blocking == DLM_LOCK_PR)
4031 		return UNBLOCK_CONTINUE;
4032 
4033 	/*
4034 	 * Mark this inode as potentially orphaned. The code in
4035 	 * ocfs2_delete_inode() will figure out whether it actually
4036 	 * needs to be freed or not.
4037 	 */
4038 	spin_lock(&oi->ip_lock);
4039 	oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
4040 	spin_unlock(&oi->ip_lock);
4041 
4042 	/*
4043 	 * Yuck. We need to make sure however that the check of
4044 	 * OCFS2_LOCK_FREEING and the extra reference are atomic with
4045 	 * respect to a reference decrement or the setting of that
4046 	 * flag.
4047 	 */
4048 	spin_lock_irqsave(&lockres->l_lock, flags);
4049 	spin_lock(&dentry_attach_lock);
4050 	if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
4051 	    && dl->dl_count) {
4052 		dl->dl_count++;
4053 		extra_ref = 1;
4054 	}
4055 	spin_unlock(&dentry_attach_lock);
4056 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4057 
4058 	mlog(0, "extra_ref = %d\n", extra_ref);
4059 
4060 	/*
4061 	 * We have a process waiting on us in ocfs2_dentry_iput(),
4062 	 * which means we can't have any more outstanding
4063 	 * aliases. There's no need to do any more work.
4064 	 */
4065 	if (!extra_ref)
4066 		return UNBLOCK_CONTINUE;
4067 
4068 	spin_lock(&dentry_attach_lock);
4069 	while (1) {
4070 		dentry = ocfs2_find_local_alias(dl->dl_inode,
4071 						dl->dl_parent_blkno, 1);
4072 		if (!dentry)
4073 			break;
4074 		spin_unlock(&dentry_attach_lock);
4075 
4076 		if (S_ISDIR(dl->dl_inode->i_mode))
4077 			shrink_dcache_parent(dentry);
4078 
4079 		mlog(0, "d_delete(%pd);\n", dentry);
4080 
4081 		/*
4082 		 * The following dcache calls may do an
4083 		 * iput(). Normally we don't want that from the
4084 		 * downconverting thread, but in this case it's ok
4085 		 * because the requesting node already has an
4086 		 * exclusive lock on the inode, so it can't be queued
4087 		 * for a downconvert.
4088 		 */
4089 		d_delete(dentry);
4090 		dput(dentry);
4091 
4092 		spin_lock(&dentry_attach_lock);
4093 	}
4094 	spin_unlock(&dentry_attach_lock);
4095 
4096 	/*
4097 	 * If we are the last holder of this dentry lock, there is no
4098 	 * reason to downconvert so skip straight to the unlock.
4099 	 */
4100 	if (dl->dl_count == 1)
4101 		return UNBLOCK_STOP_POST;
4102 
4103 	return UNBLOCK_CONTINUE_POST;
4104 }
4105 
4106 static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
4107 					    int new_level)
4108 {
4109 	struct ocfs2_refcount_tree *tree =
4110 				ocfs2_lock_res_refcount_tree(lockres);
4111 
4112 	return ocfs2_ci_checkpointed(&tree->rf_ci, lockres, new_level);
4113 }
4114 
4115 static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
4116 					 int blocking)
4117 {
4118 	struct ocfs2_refcount_tree *tree =
4119 				ocfs2_lock_res_refcount_tree(lockres);
4120 
4121 	ocfs2_metadata_cache_purge(&tree->rf_ci);
4122 
4123 	return UNBLOCK_CONTINUE;
4124 }
4125 
4126 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
4127 {
4128 	struct ocfs2_qinfo_lvb *lvb;
4129 	struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres);
4130 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
4131 					    oinfo->dqi_gi.dqi_type);
4132 
4133 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
4134 	lvb->lvb_version = OCFS2_QINFO_LVB_VERSION;
4135 	lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace);
4136 	lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace);
4137 	lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms);
4138 	lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks);
4139 	lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk);
4140 	lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry);
4141 }
4142 
4143 void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex)
4144 {
4145 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4146 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
4147 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4148 
4149 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
4150 		ocfs2_cluster_unlock(osb, lockres, level);
4151 }
4152 
4153 static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo)
4154 {
4155 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
4156 					    oinfo->dqi_gi.dqi_type);
4157 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4158 	struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
4159 	struct buffer_head *bh = NULL;
4160 	struct ocfs2_global_disk_dqinfo *gdinfo;
4161 	int status = 0;
4162 
4163 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
4164 	    lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) {
4165 		info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace);
4166 		info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace);
4167 		oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms);
4168 		oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks);
4169 		oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk);
4170 		oinfo->dqi_gi.dqi_free_entry =
4171 					be32_to_cpu(lvb->lvb_free_entry);
4172 	} else {
4173 		status = ocfs2_read_quota_phys_block(oinfo->dqi_gqinode,
4174 						     oinfo->dqi_giblk, &bh);
4175 		if (status) {
4176 			mlog_errno(status);
4177 			goto bail;
4178 		}
4179 		gdinfo = (struct ocfs2_global_disk_dqinfo *)
4180 					(bh->b_data + OCFS2_GLOBAL_INFO_OFF);
4181 		info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace);
4182 		info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace);
4183 		oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms);
4184 		oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks);
4185 		oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk);
4186 		oinfo->dqi_gi.dqi_free_entry =
4187 					le32_to_cpu(gdinfo->dqi_free_entry);
4188 		brelse(bh);
4189 		ocfs2_track_lock_refresh(lockres);
4190 	}
4191 
4192 bail:
4193 	return status;
4194 }
4195 
4196 /* Lock quota info, this function expects at least shared lock on the quota file
4197  * so that we can safely refresh quota info from disk. */
4198 int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex)
4199 {
4200 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4201 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
4202 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4203 	int status = 0;
4204 
4205 	/* On RO devices, locking really isn't needed... */
4206 	if (ocfs2_is_hard_readonly(osb)) {
4207 		if (ex)
4208 			status = -EROFS;
4209 		goto bail;
4210 	}
4211 	if (ocfs2_mount_local(osb))
4212 		goto bail;
4213 
4214 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
4215 	if (status < 0) {
4216 		mlog_errno(status);
4217 		goto bail;
4218 	}
4219 	if (!ocfs2_should_refresh_lock_res(lockres))
4220 		goto bail;
4221 	/* OK, we have the lock but we need to refresh the quota info */
4222 	status = ocfs2_refresh_qinfo(oinfo);
4223 	if (status)
4224 		ocfs2_qinfo_unlock(oinfo, ex);
4225 	ocfs2_complete_lock_res_refresh(lockres, status);
4226 bail:
4227 	return status;
4228 }
4229 
4230 int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex)
4231 {
4232 	int status;
4233 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4234 	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
4235 	struct ocfs2_super *osb = lockres->l_priv;
4236 
4237 
4238 	if (ocfs2_is_hard_readonly(osb))
4239 		return -EROFS;
4240 
4241 	if (ocfs2_mount_local(osb))
4242 		return 0;
4243 
4244 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
4245 	if (status < 0)
4246 		mlog_errno(status);
4247 
4248 	return status;
4249 }
4250 
4251 void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex)
4252 {
4253 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4254 	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
4255 	struct ocfs2_super *osb = lockres->l_priv;
4256 
4257 	if (!ocfs2_mount_local(osb))
4258 		ocfs2_cluster_unlock(osb, lockres, level);
4259 }
4260 
4261 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
4262 				       struct ocfs2_lock_res *lockres)
4263 {
4264 	int status;
4265 	struct ocfs2_unblock_ctl ctl = {0, 0,};
4266 	unsigned long flags;
4267 
4268 	/* Our reference to the lockres in this function can be
4269 	 * considered valid until we remove the OCFS2_LOCK_QUEUED
4270 	 * flag. */
4271 
4272 	BUG_ON(!lockres);
4273 	BUG_ON(!lockres->l_ops);
4274 
4275 	mlog(ML_BASTS, "lockres %s blocked\n", lockres->l_name);
4276 
4277 	/* Detect whether a lock has been marked as going away while
4278 	 * the downconvert thread was processing other things. A lock can
4279 	 * still be marked with OCFS2_LOCK_FREEING after this check,
4280 	 * but short circuiting here will still save us some
4281 	 * performance. */
4282 	spin_lock_irqsave(&lockres->l_lock, flags);
4283 	if (lockres->l_flags & OCFS2_LOCK_FREEING)
4284 		goto unqueue;
4285 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4286 
4287 	status = ocfs2_unblock_lock(osb, lockres, &ctl);
4288 	if (status < 0)
4289 		mlog_errno(status);
4290 
4291 	spin_lock_irqsave(&lockres->l_lock, flags);
4292 unqueue:
4293 	if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
4294 		lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
4295 	} else
4296 		ocfs2_schedule_blocked_lock(osb, lockres);
4297 
4298 	mlog(ML_BASTS, "lockres %s, requeue = %s.\n", lockres->l_name,
4299 	     ctl.requeue ? "yes" : "no");
4300 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4301 
4302 	if (ctl.unblock_action != UNBLOCK_CONTINUE
4303 	    && lockres->l_ops->post_unlock)
4304 		lockres->l_ops->post_unlock(osb, lockres);
4305 }
4306 
4307 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
4308 					struct ocfs2_lock_res *lockres)
4309 {
4310 	unsigned long flags;
4311 
4312 	assert_spin_locked(&lockres->l_lock);
4313 
4314 	if (lockres->l_flags & OCFS2_LOCK_FREEING) {
4315 		/* Do not schedule a lock for downconvert when it's on
4316 		 * the way to destruction - any nodes wanting access
4317 		 * to the resource will get it soon. */
4318 		mlog(ML_BASTS, "lockres %s won't be scheduled: flags 0x%lx\n",
4319 		     lockres->l_name, lockres->l_flags);
4320 		return;
4321 	}
4322 
4323 	lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
4324 
4325 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4326 	if (list_empty(&lockres->l_blocked_list)) {
4327 		list_add_tail(&lockres->l_blocked_list,
4328 			      &osb->blocked_lock_list);
4329 		osb->blocked_lock_count++;
4330 	}
4331 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4332 }
4333 
4334 static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
4335 {
4336 	unsigned long processed;
4337 	unsigned long flags;
4338 	struct ocfs2_lock_res *lockres;
4339 
4340 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4341 	/* grab this early so we know to try again if a state change and
4342 	 * wake happens part-way through our work  */
4343 	osb->dc_work_sequence = osb->dc_wake_sequence;
4344 
4345 	processed = osb->blocked_lock_count;
4346 	/*
4347 	 * blocked lock processing in this loop might call iput which can
4348 	 * remove items off osb->blocked_lock_list. Downconvert up to
4349 	 * 'processed' number of locks, but stop short if we had some
4350 	 * removed in ocfs2_mark_lockres_freeing when downconverting.
4351 	 */
4352 	while (processed && !list_empty(&osb->blocked_lock_list)) {
4353 		lockres = list_entry(osb->blocked_lock_list.next,
4354 				     struct ocfs2_lock_res, l_blocked_list);
4355 		list_del_init(&lockres->l_blocked_list);
4356 		osb->blocked_lock_count--;
4357 		spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4358 
4359 		BUG_ON(!processed);
4360 		processed--;
4361 
4362 		ocfs2_process_blocked_lock(osb, lockres);
4363 
4364 		spin_lock_irqsave(&osb->dc_task_lock, flags);
4365 	}
4366 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4367 }
4368 
4369 static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
4370 {
4371 	int empty = 0;
4372 	unsigned long flags;
4373 
4374 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4375 	if (list_empty(&osb->blocked_lock_list))
4376 		empty = 1;
4377 
4378 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4379 	return empty;
4380 }
4381 
4382 static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
4383 {
4384 	int should_wake = 0;
4385 	unsigned long flags;
4386 
4387 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4388 	if (osb->dc_work_sequence != osb->dc_wake_sequence)
4389 		should_wake = 1;
4390 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4391 
4392 	return should_wake;
4393 }
4394 
4395 static int ocfs2_downconvert_thread(void *arg)
4396 {
4397 	struct ocfs2_super *osb = arg;
4398 
4399 	/* only quit once we've been asked to stop and there is no more
4400 	 * work available */
4401 	while (!(kthread_should_stop() &&
4402 		ocfs2_downconvert_thread_lists_empty(osb))) {
4403 
4404 		wait_event_interruptible(osb->dc_event,
4405 					 ocfs2_downconvert_thread_should_wake(osb) ||
4406 					 kthread_should_stop());
4407 
4408 		mlog(0, "downconvert_thread: awoken\n");
4409 
4410 		ocfs2_downconvert_thread_do_work(osb);
4411 	}
4412 
4413 	osb->dc_task = NULL;
4414 	return 0;
4415 }
4416 
4417 void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
4418 {
4419 	unsigned long flags;
4420 
4421 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4422 	/* make sure the voting thread gets a swipe at whatever changes
4423 	 * the caller may have made to the voting state */
4424 	osb->dc_wake_sequence++;
4425 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4426 	wake_up(&osb->dc_event);
4427 }
4428