xref: /openbmc/linux/fs/ocfs2/dlmglue.c (revision 59b4412f)
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 static void ocfs2_nfs_sync_lock_init(struct ocfs2_super *osb)
693 {
694 	ocfs2_nfs_sync_lock_res_init(&osb->osb_nfs_sync_lockres, osb);
695 	init_rwsem(&osb->nfs_sync_rwlock);
696 }
697 
698 void ocfs2_trim_fs_lock_res_init(struct ocfs2_super *osb)
699 {
700 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
701 
702 	/* Only one trimfs thread are allowed to work at the same time. */
703 	mutex_lock(&osb->obs_trim_fs_mutex);
704 
705 	ocfs2_lock_res_init_once(lockres);
706 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_TRIM_FS, 0, 0, lockres->l_name);
707 	ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_TRIM_FS,
708 				   &ocfs2_trim_fs_lops, osb);
709 }
710 
711 void ocfs2_trim_fs_lock_res_uninit(struct ocfs2_super *osb)
712 {
713 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
714 
715 	ocfs2_simple_drop_lockres(osb, lockres);
716 	ocfs2_lock_res_free(lockres);
717 
718 	mutex_unlock(&osb->obs_trim_fs_mutex);
719 }
720 
721 static void ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res *res,
722 					    struct ocfs2_super *osb)
723 {
724 	ocfs2_lock_res_init_once(res);
725 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_ORPHAN_SCAN, 0, 0, res->l_name);
726 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_ORPHAN_SCAN,
727 				   &ocfs2_orphan_scan_lops, osb);
728 }
729 
730 void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
731 			      struct ocfs2_file_private *fp)
732 {
733 	struct inode *inode = fp->fp_file->f_mapping->host;
734 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
735 
736 	ocfs2_lock_res_init_once(lockres);
737 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
738 			      inode->i_generation, lockres->l_name);
739 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
740 				   OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
741 				   fp);
742 	lockres->l_flags |= OCFS2_LOCK_NOCACHE;
743 }
744 
745 void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
746 			       struct ocfs2_mem_dqinfo *info)
747 {
748 	ocfs2_lock_res_init_once(lockres);
749 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type,
750 			      0, lockres->l_name);
751 	ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres,
752 				   OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops,
753 				   info);
754 }
755 
756 void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres,
757 				  struct ocfs2_super *osb, u64 ref_blkno,
758 				  unsigned int generation)
759 {
760 	ocfs2_lock_res_init_once(lockres);
761 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_REFCOUNT, ref_blkno,
762 			      generation, lockres->l_name);
763 	ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_REFCOUNT,
764 				   &ocfs2_refcount_block_lops, osb);
765 }
766 
767 void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
768 {
769 	if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
770 		return;
771 
772 	ocfs2_remove_lockres_tracking(res);
773 
774 	mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
775 			"Lockres %s is on the blocked list\n",
776 			res->l_name);
777 	mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
778 			"Lockres %s has mask waiters pending\n",
779 			res->l_name);
780 	mlog_bug_on_msg(spin_is_locked(&res->l_lock),
781 			"Lockres %s is locked\n",
782 			res->l_name);
783 	mlog_bug_on_msg(res->l_ro_holders,
784 			"Lockres %s has %u ro holders\n",
785 			res->l_name, res->l_ro_holders);
786 	mlog_bug_on_msg(res->l_ex_holders,
787 			"Lockres %s has %u ex holders\n",
788 			res->l_name, res->l_ex_holders);
789 
790 	/* Need to clear out the lock status block for the dlm */
791 	memset(&res->l_lksb, 0, sizeof(res->l_lksb));
792 
793 	res->l_flags = 0UL;
794 }
795 
796 /*
797  * Keep a list of processes who have interest in a lockres.
798  * Note: this is now only uesed for check recursive cluster locking.
799  */
800 static inline void ocfs2_add_holder(struct ocfs2_lock_res *lockres,
801 				   struct ocfs2_lock_holder *oh)
802 {
803 	INIT_LIST_HEAD(&oh->oh_list);
804 	oh->oh_owner_pid = get_pid(task_pid(current));
805 
806 	spin_lock(&lockres->l_lock);
807 	list_add_tail(&oh->oh_list, &lockres->l_holders);
808 	spin_unlock(&lockres->l_lock);
809 }
810 
811 static struct ocfs2_lock_holder *
812 ocfs2_pid_holder(struct ocfs2_lock_res *lockres,
813 		struct pid *pid)
814 {
815 	struct ocfs2_lock_holder *oh;
816 
817 	spin_lock(&lockres->l_lock);
818 	list_for_each_entry(oh, &lockres->l_holders, oh_list) {
819 		if (oh->oh_owner_pid == pid) {
820 			spin_unlock(&lockres->l_lock);
821 			return oh;
822 		}
823 	}
824 	spin_unlock(&lockres->l_lock);
825 	return NULL;
826 }
827 
828 static inline void ocfs2_remove_holder(struct ocfs2_lock_res *lockres,
829 				       struct ocfs2_lock_holder *oh)
830 {
831 	spin_lock(&lockres->l_lock);
832 	list_del(&oh->oh_list);
833 	spin_unlock(&lockres->l_lock);
834 
835 	put_pid(oh->oh_owner_pid);
836 }
837 
838 
839 static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
840 				     int level)
841 {
842 	BUG_ON(!lockres);
843 
844 	switch(level) {
845 	case DLM_LOCK_EX:
846 		lockres->l_ex_holders++;
847 		break;
848 	case DLM_LOCK_PR:
849 		lockres->l_ro_holders++;
850 		break;
851 	default:
852 		BUG();
853 	}
854 }
855 
856 static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
857 				     int level)
858 {
859 	BUG_ON(!lockres);
860 
861 	switch(level) {
862 	case DLM_LOCK_EX:
863 		BUG_ON(!lockres->l_ex_holders);
864 		lockres->l_ex_holders--;
865 		break;
866 	case DLM_LOCK_PR:
867 		BUG_ON(!lockres->l_ro_holders);
868 		lockres->l_ro_holders--;
869 		break;
870 	default:
871 		BUG();
872 	}
873 }
874 
875 /* WARNING: This function lives in a world where the only three lock
876  * levels are EX, PR, and NL. It *will* have to be adjusted when more
877  * lock types are added. */
878 static inline int ocfs2_highest_compat_lock_level(int level)
879 {
880 	int new_level = DLM_LOCK_EX;
881 
882 	if (level == DLM_LOCK_EX)
883 		new_level = DLM_LOCK_NL;
884 	else if (level == DLM_LOCK_PR)
885 		new_level = DLM_LOCK_PR;
886 	return new_level;
887 }
888 
889 static void lockres_set_flags(struct ocfs2_lock_res *lockres,
890 			      unsigned long newflags)
891 {
892 	struct ocfs2_mask_waiter *mw, *tmp;
893 
894  	assert_spin_locked(&lockres->l_lock);
895 
896 	lockres->l_flags = newflags;
897 
898 	list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
899 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
900 			continue;
901 
902 		list_del_init(&mw->mw_item);
903 		mw->mw_status = 0;
904 		complete(&mw->mw_complete);
905 		ocfs2_track_lock_wait(lockres);
906 	}
907 }
908 static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
909 {
910 	lockres_set_flags(lockres, lockres->l_flags | or);
911 }
912 static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
913 				unsigned long clear)
914 {
915 	lockres_set_flags(lockres, lockres->l_flags & ~clear);
916 }
917 
918 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
919 {
920 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
921 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
922 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
923 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
924 
925 	lockres->l_level = lockres->l_requested;
926 	if (lockres->l_level <=
927 	    ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
928 		lockres->l_blocking = DLM_LOCK_NL;
929 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
930 	}
931 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
932 }
933 
934 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
935 {
936 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
937 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
938 
939 	/* Convert from RO to EX doesn't really need anything as our
940 	 * information is already up to data. Convert from NL to
941 	 * *anything* however should mark ourselves as needing an
942 	 * update */
943 	if (lockres->l_level == DLM_LOCK_NL &&
944 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
945 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
946 
947 	lockres->l_level = lockres->l_requested;
948 
949 	/*
950 	 * We set the OCFS2_LOCK_UPCONVERT_FINISHING flag before clearing
951 	 * the OCFS2_LOCK_BUSY flag to prevent the dc thread from
952 	 * downconverting the lock before the upconvert has fully completed.
953 	 * Do not prevent the dc thread from downconverting if NONBLOCK lock
954 	 * had already returned.
955 	 */
956 	if (!(lockres->l_flags & OCFS2_LOCK_NONBLOCK_FINISHED))
957 		lockres_or_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
958 	else
959 		lockres_clear_flags(lockres, OCFS2_LOCK_NONBLOCK_FINISHED);
960 
961 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
962 }
963 
964 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
965 {
966 	BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
967 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
968 
969 	if (lockres->l_requested > DLM_LOCK_NL &&
970 	    !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
971 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
972 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
973 
974 	lockres->l_level = lockres->l_requested;
975 	lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
976 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
977 }
978 
979 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
980 				     int level)
981 {
982 	int needs_downconvert = 0;
983 
984 	assert_spin_locked(&lockres->l_lock);
985 
986 	if (level > lockres->l_blocking) {
987 		/* only schedule a downconvert if we haven't already scheduled
988 		 * one that goes low enough to satisfy the level we're
989 		 * blocking.  this also catches the case where we get
990 		 * duplicate BASTs */
991 		if (ocfs2_highest_compat_lock_level(level) <
992 		    ocfs2_highest_compat_lock_level(lockres->l_blocking))
993 			needs_downconvert = 1;
994 
995 		lockres->l_blocking = level;
996 	}
997 
998 	mlog(ML_BASTS, "lockres %s, block %d, level %d, l_block %d, dwn %d\n",
999 	     lockres->l_name, level, lockres->l_level, lockres->l_blocking,
1000 	     needs_downconvert);
1001 
1002 	if (needs_downconvert)
1003 		lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
1004 	mlog(0, "needs_downconvert = %d\n", needs_downconvert);
1005 	return needs_downconvert;
1006 }
1007 
1008 /*
1009  * OCFS2_LOCK_PENDING and l_pending_gen.
1010  *
1011  * Why does OCFS2_LOCK_PENDING exist?  To close a race between setting
1012  * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock().  See ocfs2_unblock_lock()
1013  * for more details on the race.
1014  *
1015  * OCFS2_LOCK_PENDING closes the race quite nicely.  However, it introduces
1016  * a race on itself.  In o2dlm, we can get the ast before ocfs2_dlm_lock()
1017  * returns.  The ast clears OCFS2_LOCK_BUSY, and must therefore clear
1018  * OCFS2_LOCK_PENDING at the same time.  When ocfs2_dlm_lock() returns,
1019  * the caller is going to try to clear PENDING again.  If nothing else is
1020  * happening, __lockres_clear_pending() sees PENDING is unset and does
1021  * nothing.
1022  *
1023  * But what if another path (eg downconvert thread) has just started a
1024  * new locking action?  The other path has re-set PENDING.  Our path
1025  * cannot clear PENDING, because that will re-open the original race
1026  * window.
1027  *
1028  * [Example]
1029  *
1030  * ocfs2_meta_lock()
1031  *  ocfs2_cluster_lock()
1032  *   set BUSY
1033  *   set PENDING
1034  *   drop l_lock
1035  *   ocfs2_dlm_lock()
1036  *    ocfs2_locking_ast()		ocfs2_downconvert_thread()
1037  *     clear PENDING			 ocfs2_unblock_lock()
1038  *					  take_l_lock
1039  *					  !BUSY
1040  *					  ocfs2_prepare_downconvert()
1041  *					   set BUSY
1042  *					   set PENDING
1043  *					  drop l_lock
1044  *   take l_lock
1045  *   clear PENDING
1046  *   drop l_lock
1047  *			<window>
1048  *					  ocfs2_dlm_lock()
1049  *
1050  * So as you can see, we now have a window where l_lock is not held,
1051  * PENDING is not set, and ocfs2_dlm_lock() has not been called.
1052  *
1053  * The core problem is that ocfs2_cluster_lock() has cleared the PENDING
1054  * set by ocfs2_prepare_downconvert().  That wasn't nice.
1055  *
1056  * To solve this we introduce l_pending_gen.  A call to
1057  * lockres_clear_pending() will only do so when it is passed a generation
1058  * number that matches the lockres.  lockres_set_pending() will return the
1059  * current generation number.  When ocfs2_cluster_lock() goes to clear
1060  * PENDING, it passes the generation it got from set_pending().  In our
1061  * example above, the generation numbers will *not* match.  Thus,
1062  * ocfs2_cluster_lock() will not clear the PENDING set by
1063  * ocfs2_prepare_downconvert().
1064  */
1065 
1066 /* Unlocked version for ocfs2_locking_ast() */
1067 static void __lockres_clear_pending(struct ocfs2_lock_res *lockres,
1068 				    unsigned int generation,
1069 				    struct ocfs2_super *osb)
1070 {
1071 	assert_spin_locked(&lockres->l_lock);
1072 
1073 	/*
1074 	 * The ast and locking functions can race us here.  The winner
1075 	 * will clear pending, the loser will not.
1076 	 */
1077 	if (!(lockres->l_flags & OCFS2_LOCK_PENDING) ||
1078 	    (lockres->l_pending_gen != generation))
1079 		return;
1080 
1081 	lockres_clear_flags(lockres, OCFS2_LOCK_PENDING);
1082 	lockres->l_pending_gen++;
1083 
1084 	/*
1085 	 * The downconvert thread may have skipped us because we
1086 	 * were PENDING.  Wake it up.
1087 	 */
1088 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1089 		ocfs2_wake_downconvert_thread(osb);
1090 }
1091 
1092 /* Locked version for callers of ocfs2_dlm_lock() */
1093 static void lockres_clear_pending(struct ocfs2_lock_res *lockres,
1094 				  unsigned int generation,
1095 				  struct ocfs2_super *osb)
1096 {
1097 	unsigned long flags;
1098 
1099 	spin_lock_irqsave(&lockres->l_lock, flags);
1100 	__lockres_clear_pending(lockres, generation, osb);
1101 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1102 }
1103 
1104 static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres)
1105 {
1106 	assert_spin_locked(&lockres->l_lock);
1107 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
1108 
1109 	lockres_or_flags(lockres, OCFS2_LOCK_PENDING);
1110 
1111 	return lockres->l_pending_gen;
1112 }
1113 
1114 static void ocfs2_blocking_ast(struct ocfs2_dlm_lksb *lksb, int level)
1115 {
1116 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1117 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1118 	int needs_downconvert;
1119 	unsigned long flags;
1120 
1121 	BUG_ON(level <= DLM_LOCK_NL);
1122 
1123 	mlog(ML_BASTS, "BAST fired for lockres %s, blocking %d, level %d, "
1124 	     "type %s\n", lockres->l_name, level, lockres->l_level,
1125 	     ocfs2_lock_type_string(lockres->l_type));
1126 
1127 	/*
1128 	 * We can skip the bast for locks which don't enable caching -
1129 	 * they'll be dropped at the earliest possible time anyway.
1130 	 */
1131 	if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
1132 		return;
1133 
1134 	spin_lock_irqsave(&lockres->l_lock, flags);
1135 	needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
1136 	if (needs_downconvert)
1137 		ocfs2_schedule_blocked_lock(osb, lockres);
1138 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1139 
1140 	wake_up(&lockres->l_event);
1141 
1142 	ocfs2_wake_downconvert_thread(osb);
1143 }
1144 
1145 static void ocfs2_locking_ast(struct ocfs2_dlm_lksb *lksb)
1146 {
1147 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1148 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1149 	unsigned long flags;
1150 	int status;
1151 
1152 	spin_lock_irqsave(&lockres->l_lock, flags);
1153 
1154 	status = ocfs2_dlm_lock_status(&lockres->l_lksb);
1155 
1156 	if (status == -EAGAIN) {
1157 		lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1158 		goto out;
1159 	}
1160 
1161 	if (status) {
1162 		mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n",
1163 		     lockres->l_name, status);
1164 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1165 		return;
1166 	}
1167 
1168 	mlog(ML_BASTS, "AST fired for lockres %s, action %d, unlock %d, "
1169 	     "level %d => %d\n", lockres->l_name, lockres->l_action,
1170 	     lockres->l_unlock_action, lockres->l_level, lockres->l_requested);
1171 
1172 	switch(lockres->l_action) {
1173 	case OCFS2_AST_ATTACH:
1174 		ocfs2_generic_handle_attach_action(lockres);
1175 		lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
1176 		break;
1177 	case OCFS2_AST_CONVERT:
1178 		ocfs2_generic_handle_convert_action(lockres);
1179 		break;
1180 	case OCFS2_AST_DOWNCONVERT:
1181 		ocfs2_generic_handle_downconvert_action(lockres);
1182 		break;
1183 	default:
1184 		mlog(ML_ERROR, "lockres %s: AST fired with invalid action: %u, "
1185 		     "flags 0x%lx, unlock: %u\n",
1186 		     lockres->l_name, lockres->l_action, lockres->l_flags,
1187 		     lockres->l_unlock_action);
1188 		BUG();
1189 	}
1190 out:
1191 	/* set it to something invalid so if we get called again we
1192 	 * can catch it. */
1193 	lockres->l_action = OCFS2_AST_INVALID;
1194 
1195 	/* Did we try to cancel this lock?  Clear that state */
1196 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT)
1197 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1198 
1199 	/*
1200 	 * We may have beaten the locking functions here.  We certainly
1201 	 * know that dlm_lock() has been called :-)
1202 	 * Because we can't have two lock calls in flight at once, we
1203 	 * can use lockres->l_pending_gen.
1204 	 */
1205 	__lockres_clear_pending(lockres, lockres->l_pending_gen,  osb);
1206 
1207 	wake_up(&lockres->l_event);
1208 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1209 }
1210 
1211 static void ocfs2_unlock_ast(struct ocfs2_dlm_lksb *lksb, int error)
1212 {
1213 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1214 	unsigned long flags;
1215 
1216 	mlog(ML_BASTS, "UNLOCK AST fired for lockres %s, action = %d\n",
1217 	     lockres->l_name, lockres->l_unlock_action);
1218 
1219 	spin_lock_irqsave(&lockres->l_lock, flags);
1220 	if (error) {
1221 		mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
1222 		     "unlock_action %d\n", error, lockres->l_name,
1223 		     lockres->l_unlock_action);
1224 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1225 		return;
1226 	}
1227 
1228 	switch(lockres->l_unlock_action) {
1229 	case OCFS2_UNLOCK_CANCEL_CONVERT:
1230 		mlog(0, "Cancel convert success for %s\n", lockres->l_name);
1231 		lockres->l_action = OCFS2_AST_INVALID;
1232 		/* Downconvert thread may have requeued this lock, we
1233 		 * need to wake it. */
1234 		if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1235 			ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres));
1236 		break;
1237 	case OCFS2_UNLOCK_DROP_LOCK:
1238 		lockres->l_level = DLM_LOCK_IV;
1239 		break;
1240 	default:
1241 		BUG();
1242 	}
1243 
1244 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1245 	lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1246 	wake_up(&lockres->l_event);
1247 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1248 }
1249 
1250 /*
1251  * This is the filesystem locking protocol.  It provides the lock handling
1252  * hooks for the underlying DLM.  It has a maximum version number.
1253  * The version number allows interoperability with systems running at
1254  * the same major number and an equal or smaller minor number.
1255  *
1256  * Whenever the filesystem does new things with locks (adds or removes a
1257  * lock, orders them differently, does different things underneath a lock),
1258  * the version must be changed.  The protocol is negotiated when joining
1259  * the dlm domain.  A node may join the domain if its major version is
1260  * identical to all other nodes and its minor version is greater than
1261  * or equal to all other nodes.  When its minor version is greater than
1262  * the other nodes, it will run at the minor version specified by the
1263  * other nodes.
1264  *
1265  * If a locking change is made that will not be compatible with older
1266  * versions, the major number must be increased and the minor version set
1267  * to zero.  If a change merely adds a behavior that can be disabled when
1268  * speaking to older versions, the minor version must be increased.  If a
1269  * change adds a fully backwards compatible change (eg, LVB changes that
1270  * are just ignored by older versions), the version does not need to be
1271  * updated.
1272  */
1273 static struct ocfs2_locking_protocol lproto = {
1274 	.lp_max_version = {
1275 		.pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
1276 		.pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
1277 	},
1278 	.lp_lock_ast		= ocfs2_locking_ast,
1279 	.lp_blocking_ast	= ocfs2_blocking_ast,
1280 	.lp_unlock_ast		= ocfs2_unlock_ast,
1281 };
1282 
1283 void ocfs2_set_locking_protocol(void)
1284 {
1285 	ocfs2_stack_glue_set_max_proto_version(&lproto.lp_max_version);
1286 }
1287 
1288 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
1289 						int convert)
1290 {
1291 	unsigned long flags;
1292 
1293 	spin_lock_irqsave(&lockres->l_lock, flags);
1294 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1295 	lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1296 	if (convert)
1297 		lockres->l_action = OCFS2_AST_INVALID;
1298 	else
1299 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1300 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1301 
1302 	wake_up(&lockres->l_event);
1303 }
1304 
1305 /* Note: If we detect another process working on the lock (i.e.,
1306  * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
1307  * to do the right thing in that case.
1308  */
1309 static int ocfs2_lock_create(struct ocfs2_super *osb,
1310 			     struct ocfs2_lock_res *lockres,
1311 			     int level,
1312 			     u32 dlm_flags)
1313 {
1314 	int ret = 0;
1315 	unsigned long flags;
1316 	unsigned int gen;
1317 
1318 	mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
1319 	     dlm_flags);
1320 
1321 	spin_lock_irqsave(&lockres->l_lock, flags);
1322 	if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
1323 	    (lockres->l_flags & OCFS2_LOCK_BUSY)) {
1324 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1325 		goto bail;
1326 	}
1327 
1328 	lockres->l_action = OCFS2_AST_ATTACH;
1329 	lockres->l_requested = level;
1330 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1331 	gen = lockres_set_pending(lockres);
1332 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1333 
1334 	ret = ocfs2_dlm_lock(osb->cconn,
1335 			     level,
1336 			     &lockres->l_lksb,
1337 			     dlm_flags,
1338 			     lockres->l_name,
1339 			     OCFS2_LOCK_ID_MAX_LEN - 1);
1340 	lockres_clear_pending(lockres, gen, osb);
1341 	if (ret) {
1342 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
1343 		ocfs2_recover_from_dlm_error(lockres, 1);
1344 	}
1345 
1346 	mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
1347 
1348 bail:
1349 	return ret;
1350 }
1351 
1352 static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
1353 					int flag)
1354 {
1355 	unsigned long flags;
1356 	int ret;
1357 
1358 	spin_lock_irqsave(&lockres->l_lock, flags);
1359 	ret = lockres->l_flags & flag;
1360 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1361 
1362 	return ret;
1363 }
1364 
1365 static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
1366 
1367 {
1368 	wait_event(lockres->l_event,
1369 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
1370 }
1371 
1372 static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
1373 
1374 {
1375 	wait_event(lockres->l_event,
1376 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
1377 }
1378 
1379 /* predict what lock level we'll be dropping down to on behalf
1380  * of another node, and return true if the currently wanted
1381  * level will be compatible with it. */
1382 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
1383 						     int wanted)
1384 {
1385 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
1386 
1387 	return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
1388 }
1389 
1390 static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
1391 {
1392 	INIT_LIST_HEAD(&mw->mw_item);
1393 	init_completion(&mw->mw_complete);
1394 	ocfs2_init_start_time(mw);
1395 }
1396 
1397 static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
1398 {
1399 	wait_for_completion(&mw->mw_complete);
1400 	/* Re-arm the completion in case we want to wait on it again */
1401 	reinit_completion(&mw->mw_complete);
1402 	return mw->mw_status;
1403 }
1404 
1405 static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
1406 				    struct ocfs2_mask_waiter *mw,
1407 				    unsigned long mask,
1408 				    unsigned long goal)
1409 {
1410 	BUG_ON(!list_empty(&mw->mw_item));
1411 
1412 	assert_spin_locked(&lockres->l_lock);
1413 
1414 	list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
1415 	mw->mw_mask = mask;
1416 	mw->mw_goal = goal;
1417 	ocfs2_track_lock_wait(lockres);
1418 }
1419 
1420 /* returns 0 if the mw that was removed was already satisfied, -EBUSY
1421  * if the mask still hadn't reached its goal */
1422 static int __lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1423 				      struct ocfs2_mask_waiter *mw)
1424 {
1425 	int ret = 0;
1426 
1427 	assert_spin_locked(&lockres->l_lock);
1428 	if (!list_empty(&mw->mw_item)) {
1429 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
1430 			ret = -EBUSY;
1431 
1432 		list_del_init(&mw->mw_item);
1433 		init_completion(&mw->mw_complete);
1434 		ocfs2_track_lock_wait(lockres);
1435 	}
1436 
1437 	return ret;
1438 }
1439 
1440 static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1441 				      struct ocfs2_mask_waiter *mw)
1442 {
1443 	unsigned long flags;
1444 	int ret = 0;
1445 
1446 	spin_lock_irqsave(&lockres->l_lock, flags);
1447 	ret = __lockres_remove_mask_waiter(lockres, mw);
1448 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1449 
1450 	return ret;
1451 
1452 }
1453 
1454 static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
1455 					     struct ocfs2_lock_res *lockres)
1456 {
1457 	int ret;
1458 
1459 	ret = wait_for_completion_interruptible(&mw->mw_complete);
1460 	if (ret)
1461 		lockres_remove_mask_waiter(lockres, mw);
1462 	else
1463 		ret = mw->mw_status;
1464 	/* Re-arm the completion in case we want to wait on it again */
1465 	reinit_completion(&mw->mw_complete);
1466 	return ret;
1467 }
1468 
1469 static int __ocfs2_cluster_lock(struct ocfs2_super *osb,
1470 				struct ocfs2_lock_res *lockres,
1471 				int level,
1472 				u32 lkm_flags,
1473 				int arg_flags,
1474 				int l_subclass,
1475 				unsigned long caller_ip)
1476 {
1477 	struct ocfs2_mask_waiter mw;
1478 	int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
1479 	int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1480 	unsigned long flags;
1481 	unsigned int gen;
1482 	int noqueue_attempted = 0;
1483 	int dlm_locked = 0;
1484 	int kick_dc = 0;
1485 
1486 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED)) {
1487 		mlog_errno(-EINVAL);
1488 		return -EINVAL;
1489 	}
1490 
1491 	ocfs2_init_mask_waiter(&mw);
1492 
1493 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
1494 		lkm_flags |= DLM_LKF_VALBLK;
1495 
1496 again:
1497 	wait = 0;
1498 
1499 	spin_lock_irqsave(&lockres->l_lock, flags);
1500 
1501 	if (catch_signals && signal_pending(current)) {
1502 		ret = -ERESTARTSYS;
1503 		goto unlock;
1504 	}
1505 
1506 	mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
1507 			"Cluster lock called on freeing lockres %s! flags "
1508 			"0x%lx\n", lockres->l_name, lockres->l_flags);
1509 
1510 	/* We only compare against the currently granted level
1511 	 * here. If the lock is blocked waiting on a downconvert,
1512 	 * we'll get caught below. */
1513 	if (lockres->l_flags & OCFS2_LOCK_BUSY &&
1514 	    level > lockres->l_level) {
1515 		/* is someone sitting in dlm_lock? If so, wait on
1516 		 * them. */
1517 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1518 		wait = 1;
1519 		goto unlock;
1520 	}
1521 
1522 	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING) {
1523 		/*
1524 		 * We've upconverted. If the lock now has a level we can
1525 		 * work with, we take it. If, however, the lock is not at the
1526 		 * required level, we go thru the full cycle. One way this could
1527 		 * happen is if a process requesting an upconvert to PR is
1528 		 * closely followed by another requesting upconvert to an EX.
1529 		 * If the process requesting EX lands here, we want it to
1530 		 * continue attempting to upconvert and let the process
1531 		 * requesting PR take the lock.
1532 		 * If multiple processes request upconvert to PR, the first one
1533 		 * here will take the lock. The others will have to go thru the
1534 		 * OCFS2_LOCK_BLOCKED check to ensure that there is no pending
1535 		 * downconvert request.
1536 		 */
1537 		if (level <= lockres->l_level)
1538 			goto update_holders;
1539 	}
1540 
1541 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
1542 	    !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
1543 		/* is the lock is currently blocked on behalf of
1544 		 * another node */
1545 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
1546 		wait = 1;
1547 		goto unlock;
1548 	}
1549 
1550 	if (level > lockres->l_level) {
1551 		if (noqueue_attempted > 0) {
1552 			ret = -EAGAIN;
1553 			goto unlock;
1554 		}
1555 		if (lkm_flags & DLM_LKF_NOQUEUE)
1556 			noqueue_attempted = 1;
1557 
1558 		if (lockres->l_action != OCFS2_AST_INVALID)
1559 			mlog(ML_ERROR, "lockres %s has action %u pending\n",
1560 			     lockres->l_name, lockres->l_action);
1561 
1562 		if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1563 			lockres->l_action = OCFS2_AST_ATTACH;
1564 			lkm_flags &= ~DLM_LKF_CONVERT;
1565 		} else {
1566 			lockres->l_action = OCFS2_AST_CONVERT;
1567 			lkm_flags |= DLM_LKF_CONVERT;
1568 		}
1569 
1570 		lockres->l_requested = level;
1571 		lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1572 		gen = lockres_set_pending(lockres);
1573 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1574 
1575 		BUG_ON(level == DLM_LOCK_IV);
1576 		BUG_ON(level == DLM_LOCK_NL);
1577 
1578 		mlog(ML_BASTS, "lockres %s, convert from %d to %d\n",
1579 		     lockres->l_name, lockres->l_level, level);
1580 
1581 		/* call dlm_lock to upgrade lock now */
1582 		ret = ocfs2_dlm_lock(osb->cconn,
1583 				     level,
1584 				     &lockres->l_lksb,
1585 				     lkm_flags,
1586 				     lockres->l_name,
1587 				     OCFS2_LOCK_ID_MAX_LEN - 1);
1588 		lockres_clear_pending(lockres, gen, osb);
1589 		if (ret) {
1590 			if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
1591 			    (ret != -EAGAIN)) {
1592 				ocfs2_log_dlm_error("ocfs2_dlm_lock",
1593 						    ret, lockres);
1594 			}
1595 			ocfs2_recover_from_dlm_error(lockres, 1);
1596 			goto out;
1597 		}
1598 		dlm_locked = 1;
1599 
1600 		mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n",
1601 		     lockres->l_name);
1602 
1603 		/* At this point we've gone inside the dlm and need to
1604 		 * complete our work regardless. */
1605 		catch_signals = 0;
1606 
1607 		/* wait for busy to clear and carry on */
1608 		goto again;
1609 	}
1610 
1611 update_holders:
1612 	/* Ok, if we get here then we're good to go. */
1613 	ocfs2_inc_holders(lockres, level);
1614 
1615 	ret = 0;
1616 unlock:
1617 	lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1618 
1619 	/* ocfs2_unblock_lock reques on seeing OCFS2_LOCK_UPCONVERT_FINISHING */
1620 	kick_dc = (lockres->l_flags & OCFS2_LOCK_BLOCKED);
1621 
1622 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1623 	if (kick_dc)
1624 		ocfs2_wake_downconvert_thread(osb);
1625 out:
1626 	/*
1627 	 * This is helping work around a lock inversion between the page lock
1628 	 * and dlm locks.  One path holds the page lock while calling aops
1629 	 * which block acquiring dlm locks.  The voting thread holds dlm
1630 	 * locks while acquiring page locks while down converting data locks.
1631 	 * This block is helping an aop path notice the inversion and back
1632 	 * off to unlock its page lock before trying the dlm lock again.
1633 	 */
1634 	if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1635 	    mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1636 		wait = 0;
1637 		spin_lock_irqsave(&lockres->l_lock, flags);
1638 		if (__lockres_remove_mask_waiter(lockres, &mw)) {
1639 			if (dlm_locked)
1640 				lockres_or_flags(lockres,
1641 					OCFS2_LOCK_NONBLOCK_FINISHED);
1642 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1643 			ret = -EAGAIN;
1644 		} else {
1645 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1646 			goto again;
1647 		}
1648 	}
1649 	if (wait) {
1650 		ret = ocfs2_wait_for_mask(&mw);
1651 		if (ret == 0)
1652 			goto again;
1653 		mlog_errno(ret);
1654 	}
1655 	ocfs2_update_lock_stats(lockres, level, &mw, ret);
1656 
1657 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1658 	if (!ret && lockres->l_lockdep_map.key != NULL) {
1659 		if (level == DLM_LOCK_PR)
1660 			rwsem_acquire_read(&lockres->l_lockdep_map, l_subclass,
1661 				!!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1662 				caller_ip);
1663 		else
1664 			rwsem_acquire(&lockres->l_lockdep_map, l_subclass,
1665 				!!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1666 				caller_ip);
1667 	}
1668 #endif
1669 	return ret;
1670 }
1671 
1672 static inline int ocfs2_cluster_lock(struct ocfs2_super *osb,
1673 				     struct ocfs2_lock_res *lockres,
1674 				     int level,
1675 				     u32 lkm_flags,
1676 				     int arg_flags)
1677 {
1678 	return __ocfs2_cluster_lock(osb, lockres, level, lkm_flags, arg_flags,
1679 				    0, _RET_IP_);
1680 }
1681 
1682 
1683 static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
1684 				   struct ocfs2_lock_res *lockres,
1685 				   int level,
1686 				   unsigned long caller_ip)
1687 {
1688 	unsigned long flags;
1689 
1690 	spin_lock_irqsave(&lockres->l_lock, flags);
1691 	ocfs2_dec_holders(lockres, level);
1692 	ocfs2_downconvert_on_unlock(osb, lockres);
1693 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1694 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1695 	if (lockres->l_lockdep_map.key != NULL)
1696 		rwsem_release(&lockres->l_lockdep_map, caller_ip);
1697 #endif
1698 }
1699 
1700 static int ocfs2_create_new_lock(struct ocfs2_super *osb,
1701 				 struct ocfs2_lock_res *lockres,
1702 				 int ex,
1703 				 int local)
1704 {
1705 	int level =  ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1706 	unsigned long flags;
1707 	u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
1708 
1709 	spin_lock_irqsave(&lockres->l_lock, flags);
1710 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1711 	lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1712 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1713 
1714 	return ocfs2_lock_create(osb, lockres, level, lkm_flags);
1715 }
1716 
1717 /* Grants us an EX lock on the data and metadata resources, skipping
1718  * the normal cluster directory lookup. Use this ONLY on newly created
1719  * inodes which other nodes can't possibly see, and which haven't been
1720  * hashed in the inode hash yet. This can give us a good performance
1721  * increase as it'll skip the network broadcast normally associated
1722  * with creating a new lock resource. */
1723 int ocfs2_create_new_inode_locks(struct inode *inode)
1724 {
1725 	int ret;
1726 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1727 
1728 	BUG_ON(!ocfs2_inode_is_new(inode));
1729 
1730 	mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
1731 
1732 	/* NOTE: That we don't increment any of the holder counts, nor
1733 	 * do we add anything to a journal handle. Since this is
1734 	 * supposed to be a new inode which the cluster doesn't know
1735 	 * about yet, there is no need to.  As far as the LVB handling
1736 	 * is concerned, this is basically like acquiring an EX lock
1737 	 * on a resource which has an invalid one -- we'll set it
1738 	 * valid when we release the EX. */
1739 
1740 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
1741 	if (ret) {
1742 		mlog_errno(ret);
1743 		goto bail;
1744 	}
1745 
1746 	/*
1747 	 * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
1748 	 * don't use a generation in their lock names.
1749 	 */
1750 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
1751 	if (ret) {
1752 		mlog_errno(ret);
1753 		goto bail;
1754 	}
1755 
1756 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
1757 	if (ret)
1758 		mlog_errno(ret);
1759 
1760 bail:
1761 	return ret;
1762 }
1763 
1764 int ocfs2_rw_lock(struct inode *inode, int write)
1765 {
1766 	int status, level;
1767 	struct ocfs2_lock_res *lockres;
1768 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1769 
1770 	mlog(0, "inode %llu take %s RW lock\n",
1771 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1772 	     write ? "EXMODE" : "PRMODE");
1773 
1774 	if (ocfs2_mount_local(osb))
1775 		return 0;
1776 
1777 	lockres = &OCFS2_I(inode)->ip_rw_lockres;
1778 
1779 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1780 
1781 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
1782 	if (status < 0)
1783 		mlog_errno(status);
1784 
1785 	return status;
1786 }
1787 
1788 int ocfs2_try_rw_lock(struct inode *inode, int write)
1789 {
1790 	int status, level;
1791 	struct ocfs2_lock_res *lockres;
1792 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1793 
1794 	mlog(0, "inode %llu try to take %s RW lock\n",
1795 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1796 	     write ? "EXMODE" : "PRMODE");
1797 
1798 	if (ocfs2_mount_local(osb))
1799 		return 0;
1800 
1801 	lockres = &OCFS2_I(inode)->ip_rw_lockres;
1802 
1803 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1804 
1805 	status = ocfs2_cluster_lock(osb, lockres, level, DLM_LKF_NOQUEUE, 0);
1806 	return status;
1807 }
1808 
1809 void ocfs2_rw_unlock(struct inode *inode, int write)
1810 {
1811 	int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1812 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
1813 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1814 
1815 	mlog(0, "inode %llu drop %s RW lock\n",
1816 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1817 	     write ? "EXMODE" : "PRMODE");
1818 
1819 	if (!ocfs2_mount_local(osb))
1820 		ocfs2_cluster_unlock(osb, lockres, level);
1821 }
1822 
1823 /*
1824  * ocfs2_open_lock always get PR mode lock.
1825  */
1826 int ocfs2_open_lock(struct inode *inode)
1827 {
1828 	int status = 0;
1829 	struct ocfs2_lock_res *lockres;
1830 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1831 
1832 	mlog(0, "inode %llu take PRMODE open lock\n",
1833 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1834 
1835 	if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1836 		goto out;
1837 
1838 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1839 
1840 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_PR, 0, 0);
1841 	if (status < 0)
1842 		mlog_errno(status);
1843 
1844 out:
1845 	return status;
1846 }
1847 
1848 int ocfs2_try_open_lock(struct inode *inode, int write)
1849 {
1850 	int status = 0, level;
1851 	struct ocfs2_lock_res *lockres;
1852 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1853 
1854 	mlog(0, "inode %llu try to take %s open lock\n",
1855 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1856 	     write ? "EXMODE" : "PRMODE");
1857 
1858 	if (ocfs2_is_hard_readonly(osb)) {
1859 		if (write)
1860 			status = -EROFS;
1861 		goto out;
1862 	}
1863 
1864 	if (ocfs2_mount_local(osb))
1865 		goto out;
1866 
1867 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1868 
1869 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1870 
1871 	/*
1872 	 * The file system may already holding a PRMODE/EXMODE open lock.
1873 	 * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
1874 	 * other nodes and the -EAGAIN will indicate to the caller that
1875 	 * this inode is still in use.
1876 	 */
1877 	status = ocfs2_cluster_lock(osb, lockres, level, DLM_LKF_NOQUEUE, 0);
1878 
1879 out:
1880 	return status;
1881 }
1882 
1883 /*
1884  * ocfs2_open_unlock unlock PR and EX mode open locks.
1885  */
1886 void ocfs2_open_unlock(struct inode *inode)
1887 {
1888 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
1889 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1890 
1891 	mlog(0, "inode %llu drop open lock\n",
1892 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1893 
1894 	if (ocfs2_mount_local(osb))
1895 		goto out;
1896 
1897 	if(lockres->l_ro_holders)
1898 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_PR);
1899 	if(lockres->l_ex_holders)
1900 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
1901 
1902 out:
1903 	return;
1904 }
1905 
1906 static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
1907 				     int level)
1908 {
1909 	int ret;
1910 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1911 	unsigned long flags;
1912 	struct ocfs2_mask_waiter mw;
1913 
1914 	ocfs2_init_mask_waiter(&mw);
1915 
1916 retry_cancel:
1917 	spin_lock_irqsave(&lockres->l_lock, flags);
1918 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
1919 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
1920 		if (ret) {
1921 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1922 			ret = ocfs2_cancel_convert(osb, lockres);
1923 			if (ret < 0) {
1924 				mlog_errno(ret);
1925 				goto out;
1926 			}
1927 			goto retry_cancel;
1928 		}
1929 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1930 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1931 
1932 		ocfs2_wait_for_mask(&mw);
1933 		goto retry_cancel;
1934 	}
1935 
1936 	ret = -ERESTARTSYS;
1937 	/*
1938 	 * We may still have gotten the lock, in which case there's no
1939 	 * point to restarting the syscall.
1940 	 */
1941 	if (lockres->l_level == level)
1942 		ret = 0;
1943 
1944 	mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
1945 	     lockres->l_flags, lockres->l_level, lockres->l_action);
1946 
1947 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1948 
1949 out:
1950 	return ret;
1951 }
1952 
1953 /*
1954  * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
1955  * flock() calls. The locking approach this requires is sufficiently
1956  * different from all other cluster lock types that we implement a
1957  * separate path to the "low-level" dlm calls. In particular:
1958  *
1959  * - No optimization of lock levels is done - we take at exactly
1960  *   what's been requested.
1961  *
1962  * - No lock caching is employed. We immediately downconvert to
1963  *   no-lock at unlock time. This also means flock locks never go on
1964  *   the blocking list).
1965  *
1966  * - Since userspace can trivially deadlock itself with flock, we make
1967  *   sure to allow cancellation of a misbehaving applications flock()
1968  *   request.
1969  *
1970  * - Access to any flock lockres doesn't require concurrency, so we
1971  *   can simplify the code by requiring the caller to guarantee
1972  *   serialization of dlmglue flock calls.
1973  */
1974 int ocfs2_file_lock(struct file *file, int ex, int trylock)
1975 {
1976 	int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1977 	unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0;
1978 	unsigned long flags;
1979 	struct ocfs2_file_private *fp = file->private_data;
1980 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
1981 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1982 	struct ocfs2_mask_waiter mw;
1983 
1984 	ocfs2_init_mask_waiter(&mw);
1985 
1986 	if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
1987 	    (lockres->l_level > DLM_LOCK_NL)) {
1988 		mlog(ML_ERROR,
1989 		     "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
1990 		     "level: %u\n", lockres->l_name, lockres->l_flags,
1991 		     lockres->l_level);
1992 		return -EINVAL;
1993 	}
1994 
1995 	spin_lock_irqsave(&lockres->l_lock, flags);
1996 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1997 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1998 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1999 
2000 		/*
2001 		 * Get the lock at NLMODE to start - that way we
2002 		 * can cancel the upconvert request if need be.
2003 		 */
2004 		ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0);
2005 		if (ret < 0) {
2006 			mlog_errno(ret);
2007 			goto out;
2008 		}
2009 
2010 		ret = ocfs2_wait_for_mask(&mw);
2011 		if (ret) {
2012 			mlog_errno(ret);
2013 			goto out;
2014 		}
2015 		spin_lock_irqsave(&lockres->l_lock, flags);
2016 	}
2017 
2018 	lockres->l_action = OCFS2_AST_CONVERT;
2019 	lkm_flags |= DLM_LKF_CONVERT;
2020 	lockres->l_requested = level;
2021 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2022 
2023 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
2024 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2025 
2026 	ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags,
2027 			     lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1);
2028 	if (ret) {
2029 		if (!trylock || (ret != -EAGAIN)) {
2030 			ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
2031 			ret = -EINVAL;
2032 		}
2033 
2034 		ocfs2_recover_from_dlm_error(lockres, 1);
2035 		lockres_remove_mask_waiter(lockres, &mw);
2036 		goto out;
2037 	}
2038 
2039 	ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
2040 	if (ret == -ERESTARTSYS) {
2041 		/*
2042 		 * Userspace can cause deadlock itself with
2043 		 * flock(). Current behavior locally is to allow the
2044 		 * deadlock, but abort the system call if a signal is
2045 		 * received. We follow this example, otherwise a
2046 		 * poorly written program could sit in kernel until
2047 		 * reboot.
2048 		 *
2049 		 * Handling this is a bit more complicated for Ocfs2
2050 		 * though. We can't exit this function with an
2051 		 * outstanding lock request, so a cancel convert is
2052 		 * required. We intentionally overwrite 'ret' - if the
2053 		 * cancel fails and the lock was granted, it's easier
2054 		 * to just bubble success back up to the user.
2055 		 */
2056 		ret = ocfs2_flock_handle_signal(lockres, level);
2057 	} else if (!ret && (level > lockres->l_level)) {
2058 		/* Trylock failed asynchronously */
2059 		BUG_ON(!trylock);
2060 		ret = -EAGAIN;
2061 	}
2062 
2063 out:
2064 
2065 	mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
2066 	     lockres->l_name, ex, trylock, ret);
2067 	return ret;
2068 }
2069 
2070 void ocfs2_file_unlock(struct file *file)
2071 {
2072 	int ret;
2073 	unsigned int gen;
2074 	unsigned long flags;
2075 	struct ocfs2_file_private *fp = file->private_data;
2076 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
2077 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
2078 	struct ocfs2_mask_waiter mw;
2079 
2080 	ocfs2_init_mask_waiter(&mw);
2081 
2082 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
2083 		return;
2084 
2085 	if (lockres->l_level == DLM_LOCK_NL)
2086 		return;
2087 
2088 	mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
2089 	     lockres->l_name, lockres->l_flags, lockres->l_level,
2090 	     lockres->l_action);
2091 
2092 	spin_lock_irqsave(&lockres->l_lock, flags);
2093 	/*
2094 	 * Fake a blocking ast for the downconvert code.
2095 	 */
2096 	lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
2097 	lockres->l_blocking = DLM_LOCK_EX;
2098 
2099 	gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL);
2100 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
2101 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2102 
2103 	ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen);
2104 	if (ret) {
2105 		mlog_errno(ret);
2106 		return;
2107 	}
2108 
2109 	ret = ocfs2_wait_for_mask(&mw);
2110 	if (ret)
2111 		mlog_errno(ret);
2112 }
2113 
2114 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
2115 					struct ocfs2_lock_res *lockres)
2116 {
2117 	int kick = 0;
2118 
2119 	/* If we know that another node is waiting on our lock, kick
2120 	 * the downconvert thread * pre-emptively when we reach a release
2121 	 * condition. */
2122 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
2123 		switch(lockres->l_blocking) {
2124 		case DLM_LOCK_EX:
2125 			if (!lockres->l_ex_holders && !lockres->l_ro_holders)
2126 				kick = 1;
2127 			break;
2128 		case DLM_LOCK_PR:
2129 			if (!lockres->l_ex_holders)
2130 				kick = 1;
2131 			break;
2132 		default:
2133 			BUG();
2134 		}
2135 	}
2136 
2137 	if (kick)
2138 		ocfs2_wake_downconvert_thread(osb);
2139 }
2140 
2141 #define OCFS2_SEC_BITS   34
2142 #define OCFS2_SEC_SHIFT  (64 - OCFS2_SEC_BITS)
2143 #define OCFS2_NSEC_MASK  ((1ULL << OCFS2_SEC_SHIFT) - 1)
2144 
2145 /* LVB only has room for 64 bits of time here so we pack it for
2146  * now. */
2147 static u64 ocfs2_pack_timespec(struct timespec64 *spec)
2148 {
2149 	u64 res;
2150 	u64 sec = clamp_t(time64_t, spec->tv_sec, 0, 0x3ffffffffull);
2151 	u32 nsec = spec->tv_nsec;
2152 
2153 	res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
2154 
2155 	return res;
2156 }
2157 
2158 /* Call this with the lockres locked. I am reasonably sure we don't
2159  * need ip_lock in this function as anyone who would be changing those
2160  * values is supposed to be blocked in ocfs2_inode_lock right now. */
2161 static void __ocfs2_stuff_meta_lvb(struct inode *inode)
2162 {
2163 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2164 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2165 	struct ocfs2_meta_lvb *lvb;
2166 
2167 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2168 
2169 	/*
2170 	 * Invalidate the LVB of a deleted inode - this way other
2171 	 * nodes are forced to go to disk and discover the new inode
2172 	 * status.
2173 	 */
2174 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
2175 		lvb->lvb_version = 0;
2176 		goto out;
2177 	}
2178 
2179 	lvb->lvb_version   = OCFS2_LVB_VERSION;
2180 	lvb->lvb_isize	   = cpu_to_be64(i_size_read(inode));
2181 	lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
2182 	lvb->lvb_iuid      = cpu_to_be32(i_uid_read(inode));
2183 	lvb->lvb_igid      = cpu_to_be32(i_gid_read(inode));
2184 	lvb->lvb_imode     = cpu_to_be16(inode->i_mode);
2185 	lvb->lvb_inlink    = cpu_to_be16(inode->i_nlink);
2186 	lvb->lvb_iatime_packed  =
2187 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
2188 	lvb->lvb_ictime_packed =
2189 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
2190 	lvb->lvb_imtime_packed =
2191 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
2192 	lvb->lvb_iattr    = cpu_to_be32(oi->ip_attr);
2193 	lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
2194 	lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
2195 
2196 out:
2197 	mlog_meta_lvb(0, lockres);
2198 }
2199 
2200 static void ocfs2_unpack_timespec(struct timespec64 *spec,
2201 				  u64 packed_time)
2202 {
2203 	spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
2204 	spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
2205 }
2206 
2207 static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
2208 {
2209 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2210 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2211 	struct ocfs2_meta_lvb *lvb;
2212 
2213 	mlog_meta_lvb(0, lockres);
2214 
2215 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2216 
2217 	/* We're safe here without the lockres lock... */
2218 	spin_lock(&oi->ip_lock);
2219 	oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
2220 	i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
2221 
2222 	oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
2223 	oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
2224 	ocfs2_set_inode_flags(inode);
2225 
2226 	/* fast-symlinks are a special case */
2227 	if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
2228 		inode->i_blocks = 0;
2229 	else
2230 		inode->i_blocks = ocfs2_inode_sector_count(inode);
2231 
2232 	i_uid_write(inode, be32_to_cpu(lvb->lvb_iuid));
2233 	i_gid_write(inode, be32_to_cpu(lvb->lvb_igid));
2234 	inode->i_mode    = be16_to_cpu(lvb->lvb_imode);
2235 	set_nlink(inode, be16_to_cpu(lvb->lvb_inlink));
2236 	ocfs2_unpack_timespec(&inode->i_atime,
2237 			      be64_to_cpu(lvb->lvb_iatime_packed));
2238 	ocfs2_unpack_timespec(&inode->i_mtime,
2239 			      be64_to_cpu(lvb->lvb_imtime_packed));
2240 	ocfs2_unpack_timespec(&inode->i_ctime,
2241 			      be64_to_cpu(lvb->lvb_ictime_packed));
2242 	spin_unlock(&oi->ip_lock);
2243 }
2244 
2245 static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
2246 					      struct ocfs2_lock_res *lockres)
2247 {
2248 	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2249 
2250 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)
2251 	    && lvb->lvb_version == OCFS2_LVB_VERSION
2252 	    && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
2253 		return 1;
2254 	return 0;
2255 }
2256 
2257 /* Determine whether a lock resource needs to be refreshed, and
2258  * arbitrate who gets to refresh it.
2259  *
2260  *   0 means no refresh needed.
2261  *
2262  *   > 0 means you need to refresh this and you MUST call
2263  *   ocfs2_complete_lock_res_refresh afterwards. */
2264 static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
2265 {
2266 	unsigned long flags;
2267 	int status = 0;
2268 
2269 refresh_check:
2270 	spin_lock_irqsave(&lockres->l_lock, flags);
2271 	if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
2272 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2273 		goto bail;
2274 	}
2275 
2276 	if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
2277 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2278 
2279 		ocfs2_wait_on_refreshing_lock(lockres);
2280 		goto refresh_check;
2281 	}
2282 
2283 	/* Ok, I'll be the one to refresh this lock. */
2284 	lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
2285 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2286 
2287 	status = 1;
2288 bail:
2289 	mlog(0, "status %d\n", status);
2290 	return status;
2291 }
2292 
2293 /* If status is non zero, I'll mark it as not being in refresh
2294  * anymroe, but i won't clear the needs refresh flag. */
2295 static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
2296 						   int status)
2297 {
2298 	unsigned long flags;
2299 
2300 	spin_lock_irqsave(&lockres->l_lock, flags);
2301 	lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
2302 	if (!status)
2303 		lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
2304 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2305 
2306 	wake_up(&lockres->l_event);
2307 }
2308 
2309 /* may or may not return a bh if it went to disk. */
2310 static int ocfs2_inode_lock_update(struct inode *inode,
2311 				  struct buffer_head **bh)
2312 {
2313 	int status = 0;
2314 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2315 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2316 	struct ocfs2_dinode *fe;
2317 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2318 
2319 	if (ocfs2_mount_local(osb))
2320 		goto bail;
2321 
2322 	spin_lock(&oi->ip_lock);
2323 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
2324 		mlog(0, "Orphaned inode %llu was deleted while we "
2325 		     "were waiting on a lock. ip_flags = 0x%x\n",
2326 		     (unsigned long long)oi->ip_blkno, oi->ip_flags);
2327 		spin_unlock(&oi->ip_lock);
2328 		status = -ENOENT;
2329 		goto bail;
2330 	}
2331 	spin_unlock(&oi->ip_lock);
2332 
2333 	if (!ocfs2_should_refresh_lock_res(lockres))
2334 		goto bail;
2335 
2336 	/* This will discard any caching information we might have had
2337 	 * for the inode metadata. */
2338 	ocfs2_metadata_cache_purge(INODE_CACHE(inode));
2339 
2340 	ocfs2_extent_map_trunc(inode, 0);
2341 
2342 	if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
2343 		mlog(0, "Trusting LVB on inode %llu\n",
2344 		     (unsigned long long)oi->ip_blkno);
2345 		ocfs2_refresh_inode_from_lvb(inode);
2346 	} else {
2347 		/* Boo, we have to go to disk. */
2348 		/* read bh, cast, ocfs2_refresh_inode */
2349 		status = ocfs2_read_inode_block(inode, bh);
2350 		if (status < 0) {
2351 			mlog_errno(status);
2352 			goto bail_refresh;
2353 		}
2354 		fe = (struct ocfs2_dinode *) (*bh)->b_data;
2355 
2356 		/* This is a good chance to make sure we're not
2357 		 * locking an invalid object.  ocfs2_read_inode_block()
2358 		 * already checked that the inode block is sane.
2359 		 *
2360 		 * We bug on a stale inode here because we checked
2361 		 * above whether it was wiped from disk. The wiping
2362 		 * node provides a guarantee that we receive that
2363 		 * message and can mark the inode before dropping any
2364 		 * locks associated with it. */
2365 		mlog_bug_on_msg(inode->i_generation !=
2366 				le32_to_cpu(fe->i_generation),
2367 				"Invalid dinode %llu disk generation: %u "
2368 				"inode->i_generation: %u\n",
2369 				(unsigned long long)oi->ip_blkno,
2370 				le32_to_cpu(fe->i_generation),
2371 				inode->i_generation);
2372 		mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
2373 				!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
2374 				"Stale dinode %llu dtime: %llu flags: 0x%x\n",
2375 				(unsigned long long)oi->ip_blkno,
2376 				(unsigned long long)le64_to_cpu(fe->i_dtime),
2377 				le32_to_cpu(fe->i_flags));
2378 
2379 		ocfs2_refresh_inode(inode, fe);
2380 		ocfs2_track_lock_refresh(lockres);
2381 	}
2382 
2383 	status = 0;
2384 bail_refresh:
2385 	ocfs2_complete_lock_res_refresh(lockres, status);
2386 bail:
2387 	return status;
2388 }
2389 
2390 static int ocfs2_assign_bh(struct inode *inode,
2391 			   struct buffer_head **ret_bh,
2392 			   struct buffer_head *passed_bh)
2393 {
2394 	int status;
2395 
2396 	if (passed_bh) {
2397 		/* Ok, the update went to disk for us, use the
2398 		 * returned bh. */
2399 		*ret_bh = passed_bh;
2400 		get_bh(*ret_bh);
2401 
2402 		return 0;
2403 	}
2404 
2405 	status = ocfs2_read_inode_block(inode, ret_bh);
2406 	if (status < 0)
2407 		mlog_errno(status);
2408 
2409 	return status;
2410 }
2411 
2412 /*
2413  * returns < 0 error if the callback will never be called, otherwise
2414  * the result of the lock will be communicated via the callback.
2415  */
2416 int ocfs2_inode_lock_full_nested(struct inode *inode,
2417 				 struct buffer_head **ret_bh,
2418 				 int ex,
2419 				 int arg_flags,
2420 				 int subclass)
2421 {
2422 	int status, level, acquired;
2423 	u32 dlm_flags;
2424 	struct ocfs2_lock_res *lockres = NULL;
2425 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2426 	struct buffer_head *local_bh = NULL;
2427 
2428 	mlog(0, "inode %llu, take %s META lock\n",
2429 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2430 	     ex ? "EXMODE" : "PRMODE");
2431 
2432 	status = 0;
2433 	acquired = 0;
2434 	/* We'll allow faking a readonly metadata lock for
2435 	 * rodevices. */
2436 	if (ocfs2_is_hard_readonly(osb)) {
2437 		if (ex)
2438 			status = -EROFS;
2439 		goto getbh;
2440 	}
2441 
2442 	if ((arg_flags & OCFS2_META_LOCK_GETBH) ||
2443 	    ocfs2_mount_local(osb))
2444 		goto update;
2445 
2446 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2447 		ocfs2_wait_for_recovery(osb);
2448 
2449 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2450 	level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2451 	dlm_flags = 0;
2452 	if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
2453 		dlm_flags |= DLM_LKF_NOQUEUE;
2454 
2455 	status = __ocfs2_cluster_lock(osb, lockres, level, dlm_flags,
2456 				      arg_flags, subclass, _RET_IP_);
2457 	if (status < 0) {
2458 		if (status != -EAGAIN)
2459 			mlog_errno(status);
2460 		goto bail;
2461 	}
2462 
2463 	/* Notify the error cleanup path to drop the cluster lock. */
2464 	acquired = 1;
2465 
2466 	/* We wait twice because a node may have died while we were in
2467 	 * the lower dlm layers. The second time though, we've
2468 	 * committed to owning this lock so we don't allow signals to
2469 	 * abort the operation. */
2470 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2471 		ocfs2_wait_for_recovery(osb);
2472 
2473 update:
2474 	/*
2475 	 * We only see this flag if we're being called from
2476 	 * ocfs2_read_locked_inode(). It means we're locking an inode
2477 	 * which hasn't been populated yet, so clear the refresh flag
2478 	 * and let the caller handle it.
2479 	 */
2480 	if (inode->i_state & I_NEW) {
2481 		status = 0;
2482 		if (lockres)
2483 			ocfs2_complete_lock_res_refresh(lockres, 0);
2484 		goto bail;
2485 	}
2486 
2487 	/* This is fun. The caller may want a bh back, or it may
2488 	 * not. ocfs2_inode_lock_update definitely wants one in, but
2489 	 * may or may not read one, depending on what's in the
2490 	 * LVB. The result of all of this is that we've *only* gone to
2491 	 * disk if we have to, so the complexity is worthwhile. */
2492 	status = ocfs2_inode_lock_update(inode, &local_bh);
2493 	if (status < 0) {
2494 		if (status != -ENOENT)
2495 			mlog_errno(status);
2496 		goto bail;
2497 	}
2498 getbh:
2499 	if (ret_bh) {
2500 		status = ocfs2_assign_bh(inode, ret_bh, local_bh);
2501 		if (status < 0) {
2502 			mlog_errno(status);
2503 			goto bail;
2504 		}
2505 	}
2506 
2507 bail:
2508 	if (status < 0) {
2509 		if (ret_bh && (*ret_bh)) {
2510 			brelse(*ret_bh);
2511 			*ret_bh = NULL;
2512 		}
2513 		if (acquired)
2514 			ocfs2_inode_unlock(inode, ex);
2515 	}
2516 
2517 	brelse(local_bh);
2518 	return status;
2519 }
2520 
2521 /*
2522  * This is working around a lock inversion between tasks acquiring DLM
2523  * locks while holding a page lock and the downconvert thread which
2524  * blocks dlm lock acquiry while acquiring page locks.
2525  *
2526  * ** These _with_page variantes are only intended to be called from aop
2527  * methods that hold page locks and return a very specific *positive* error
2528  * code that aop methods pass up to the VFS -- test for errors with != 0. **
2529  *
2530  * The DLM is called such that it returns -EAGAIN if it would have
2531  * blocked waiting for the downconvert thread.  In that case we unlock
2532  * our page so the downconvert thread can make progress.  Once we've
2533  * done this we have to return AOP_TRUNCATED_PAGE so the aop method
2534  * that called us can bubble that back up into the VFS who will then
2535  * immediately retry the aop call.
2536  */
2537 int ocfs2_inode_lock_with_page(struct inode *inode,
2538 			      struct buffer_head **ret_bh,
2539 			      int ex,
2540 			      struct page *page)
2541 {
2542 	int ret;
2543 
2544 	ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
2545 	if (ret == -EAGAIN) {
2546 		unlock_page(page);
2547 		/*
2548 		 * If we can't get inode lock immediately, we should not return
2549 		 * directly here, since this will lead to a softlockup problem.
2550 		 * The method is to get a blocking lock and immediately unlock
2551 		 * before returning, this can avoid CPU resource waste due to
2552 		 * lots of retries, and benefits fairness in getting lock.
2553 		 */
2554 		if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
2555 			ocfs2_inode_unlock(inode, ex);
2556 		ret = AOP_TRUNCATED_PAGE;
2557 	}
2558 
2559 	return ret;
2560 }
2561 
2562 int ocfs2_inode_lock_atime(struct inode *inode,
2563 			  struct vfsmount *vfsmnt,
2564 			  int *level, int wait)
2565 {
2566 	int ret;
2567 
2568 	if (wait)
2569 		ret = ocfs2_inode_lock(inode, NULL, 0);
2570 	else
2571 		ret = ocfs2_try_inode_lock(inode, NULL, 0);
2572 
2573 	if (ret < 0) {
2574 		if (ret != -EAGAIN)
2575 			mlog_errno(ret);
2576 		return ret;
2577 	}
2578 
2579 	/*
2580 	 * If we should update atime, we will get EX lock,
2581 	 * otherwise we just get PR lock.
2582 	 */
2583 	if (ocfs2_should_update_atime(inode, vfsmnt)) {
2584 		struct buffer_head *bh = NULL;
2585 
2586 		ocfs2_inode_unlock(inode, 0);
2587 		if (wait)
2588 			ret = ocfs2_inode_lock(inode, &bh, 1);
2589 		else
2590 			ret = ocfs2_try_inode_lock(inode, &bh, 1);
2591 
2592 		if (ret < 0) {
2593 			if (ret != -EAGAIN)
2594 				mlog_errno(ret);
2595 			return ret;
2596 		}
2597 		*level = 1;
2598 		if (ocfs2_should_update_atime(inode, vfsmnt))
2599 			ocfs2_update_inode_atime(inode, bh);
2600 		brelse(bh);
2601 	} else
2602 		*level = 0;
2603 
2604 	return ret;
2605 }
2606 
2607 void ocfs2_inode_unlock(struct inode *inode,
2608 		       int ex)
2609 {
2610 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2611 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
2612 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2613 
2614 	mlog(0, "inode %llu drop %s META lock\n",
2615 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2616 	     ex ? "EXMODE" : "PRMODE");
2617 
2618 	if (!ocfs2_is_hard_readonly(osb) &&
2619 	    !ocfs2_mount_local(osb))
2620 		ocfs2_cluster_unlock(osb, lockres, level);
2621 }
2622 
2623 /*
2624  * This _tracker variantes are introduced to deal with the recursive cluster
2625  * locking issue. The idea is to keep track of a lock holder on the stack of
2626  * the current process. If there's a lock holder on the stack, we know the
2627  * task context is already protected by cluster locking. Currently, they're
2628  * used in some VFS entry routines.
2629  *
2630  * return < 0 on error, return == 0 if there's no lock holder on the stack
2631  * before this call, return == 1 if this call would be a recursive locking.
2632  * return == -1 if this lock attempt will cause an upgrade which is forbidden.
2633  *
2634  * When taking lock levels into account,we face some different situations.
2635  *
2636  * 1. no lock is held
2637  *    In this case, just lock the inode as requested and return 0
2638  *
2639  * 2. We are holding a lock
2640  *    For this situation, things diverges into several cases
2641  *
2642  *    wanted     holding	     what to do
2643  *    ex		ex	    see 2.1 below
2644  *    ex		pr	    see 2.2 below
2645  *    pr		ex	    see 2.1 below
2646  *    pr		pr	    see 2.1 below
2647  *
2648  *    2.1 lock level that is been held is compatible
2649  *    with the wanted level, so no lock action will be tacken.
2650  *
2651  *    2.2 Otherwise, an upgrade is needed, but it is forbidden.
2652  *
2653  * Reason why upgrade within a process is forbidden is that
2654  * lock upgrade may cause dead lock. The following illustrates
2655  * how it happens.
2656  *
2657  *         thread on node1                             thread on node2
2658  * ocfs2_inode_lock_tracker(ex=0)
2659  *
2660  *                                <======   ocfs2_inode_lock_tracker(ex=1)
2661  *
2662  * ocfs2_inode_lock_tracker(ex=1)
2663  */
2664 int ocfs2_inode_lock_tracker(struct inode *inode,
2665 			     struct buffer_head **ret_bh,
2666 			     int ex,
2667 			     struct ocfs2_lock_holder *oh)
2668 {
2669 	int status = 0;
2670 	struct ocfs2_lock_res *lockres;
2671 	struct ocfs2_lock_holder *tmp_oh;
2672 	struct pid *pid = task_pid(current);
2673 
2674 
2675 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2676 	tmp_oh = ocfs2_pid_holder(lockres, pid);
2677 
2678 	if (!tmp_oh) {
2679 		/*
2680 		 * This corresponds to the case 1.
2681 		 * We haven't got any lock before.
2682 		 */
2683 		status = ocfs2_inode_lock_full(inode, ret_bh, ex, 0);
2684 		if (status < 0) {
2685 			if (status != -ENOENT)
2686 				mlog_errno(status);
2687 			return status;
2688 		}
2689 
2690 		oh->oh_ex = ex;
2691 		ocfs2_add_holder(lockres, oh);
2692 		return 0;
2693 	}
2694 
2695 	if (unlikely(ex && !tmp_oh->oh_ex)) {
2696 		/*
2697 		 * case 2.2 upgrade may cause dead lock, forbid it.
2698 		 */
2699 		mlog(ML_ERROR, "Recursive locking is not permitted to "
2700 		     "upgrade to EX level from PR level.\n");
2701 		dump_stack();
2702 		return -EINVAL;
2703 	}
2704 
2705 	/*
2706 	 *  case 2.1 OCFS2_META_LOCK_GETBH flag make ocfs2_inode_lock_full.
2707 	 *  ignore the lock level and just update it.
2708 	 */
2709 	if (ret_bh) {
2710 		status = ocfs2_inode_lock_full(inode, ret_bh, ex,
2711 					       OCFS2_META_LOCK_GETBH);
2712 		if (status < 0) {
2713 			if (status != -ENOENT)
2714 				mlog_errno(status);
2715 			return status;
2716 		}
2717 	}
2718 	return tmp_oh ? 1 : 0;
2719 }
2720 
2721 void ocfs2_inode_unlock_tracker(struct inode *inode,
2722 				int ex,
2723 				struct ocfs2_lock_holder *oh,
2724 				int had_lock)
2725 {
2726 	struct ocfs2_lock_res *lockres;
2727 
2728 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2729 	/* had_lock means that the currect process already takes the cluster
2730 	 * lock previously.
2731 	 * If had_lock is 1, we have nothing to do here.
2732 	 * If had_lock is 0, we will release the lock.
2733 	 */
2734 	if (!had_lock) {
2735 		ocfs2_inode_unlock(inode, oh->oh_ex);
2736 		ocfs2_remove_holder(lockres, oh);
2737 	}
2738 }
2739 
2740 int ocfs2_orphan_scan_lock(struct ocfs2_super *osb, u32 *seqno)
2741 {
2742 	struct ocfs2_lock_res *lockres;
2743 	struct ocfs2_orphan_scan_lvb *lvb;
2744 	int status = 0;
2745 
2746 	if (ocfs2_is_hard_readonly(osb))
2747 		return -EROFS;
2748 
2749 	if (ocfs2_mount_local(osb))
2750 		return 0;
2751 
2752 	lockres = &osb->osb_orphan_scan.os_lockres;
2753 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2754 	if (status < 0)
2755 		return status;
2756 
2757 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2758 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2759 	    lvb->lvb_version == OCFS2_ORPHAN_LVB_VERSION)
2760 		*seqno = be32_to_cpu(lvb->lvb_os_seqno);
2761 	else
2762 		*seqno = osb->osb_orphan_scan.os_seqno + 1;
2763 
2764 	return status;
2765 }
2766 
2767 void ocfs2_orphan_scan_unlock(struct ocfs2_super *osb, u32 seqno)
2768 {
2769 	struct ocfs2_lock_res *lockres;
2770 	struct ocfs2_orphan_scan_lvb *lvb;
2771 
2772 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) {
2773 		lockres = &osb->osb_orphan_scan.os_lockres;
2774 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2775 		lvb->lvb_version = OCFS2_ORPHAN_LVB_VERSION;
2776 		lvb->lvb_os_seqno = cpu_to_be32(seqno);
2777 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2778 	}
2779 }
2780 
2781 int ocfs2_super_lock(struct ocfs2_super *osb,
2782 		     int ex)
2783 {
2784 	int status = 0;
2785 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2786 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2787 
2788 	if (ocfs2_is_hard_readonly(osb))
2789 		return -EROFS;
2790 
2791 	if (ocfs2_mount_local(osb))
2792 		goto bail;
2793 
2794 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
2795 	if (status < 0) {
2796 		mlog_errno(status);
2797 		goto bail;
2798 	}
2799 
2800 	/* The super block lock path is really in the best position to
2801 	 * know when resources covered by the lock need to be
2802 	 * refreshed, so we do it here. Of course, making sense of
2803 	 * everything is up to the caller :) */
2804 	status = ocfs2_should_refresh_lock_res(lockres);
2805 	if (status) {
2806 		status = ocfs2_refresh_slot_info(osb);
2807 
2808 		ocfs2_complete_lock_res_refresh(lockres, status);
2809 
2810 		if (status < 0) {
2811 			ocfs2_cluster_unlock(osb, lockres, level);
2812 			mlog_errno(status);
2813 		}
2814 		ocfs2_track_lock_refresh(lockres);
2815 	}
2816 bail:
2817 	return status;
2818 }
2819 
2820 void ocfs2_super_unlock(struct ocfs2_super *osb,
2821 			int ex)
2822 {
2823 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2824 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2825 
2826 	if (!ocfs2_mount_local(osb))
2827 		ocfs2_cluster_unlock(osb, lockres, level);
2828 }
2829 
2830 int ocfs2_rename_lock(struct ocfs2_super *osb)
2831 {
2832 	int status;
2833 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2834 
2835 	if (ocfs2_is_hard_readonly(osb))
2836 		return -EROFS;
2837 
2838 	if (ocfs2_mount_local(osb))
2839 		return 0;
2840 
2841 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2842 	if (status < 0)
2843 		mlog_errno(status);
2844 
2845 	return status;
2846 }
2847 
2848 void ocfs2_rename_unlock(struct ocfs2_super *osb)
2849 {
2850 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2851 
2852 	if (!ocfs2_mount_local(osb))
2853 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2854 }
2855 
2856 int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex)
2857 {
2858 	int status;
2859 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2860 
2861 	if (ocfs2_is_hard_readonly(osb))
2862 		return -EROFS;
2863 
2864 	if (ex)
2865 		down_write(&osb->nfs_sync_rwlock);
2866 	else
2867 		down_read(&osb->nfs_sync_rwlock);
2868 
2869 	if (ocfs2_mount_local(osb))
2870 		return 0;
2871 
2872 	status = ocfs2_cluster_lock(osb, lockres, ex ? LKM_EXMODE : LKM_PRMODE,
2873 				    0, 0);
2874 	if (status < 0)
2875 		mlog(ML_ERROR, "lock on nfs sync lock failed %d\n", status);
2876 
2877 	return status;
2878 }
2879 
2880 void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex)
2881 {
2882 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2883 
2884 	if (!ocfs2_mount_local(osb))
2885 		ocfs2_cluster_unlock(osb, lockres,
2886 				     ex ? LKM_EXMODE : LKM_PRMODE);
2887 	if (ex)
2888 		up_write(&osb->nfs_sync_rwlock);
2889 	else
2890 		up_read(&osb->nfs_sync_rwlock);
2891 }
2892 
2893 int ocfs2_trim_fs_lock(struct ocfs2_super *osb,
2894 		       struct ocfs2_trim_fs_info *info, int trylock)
2895 {
2896 	int status;
2897 	struct ocfs2_trim_fs_lvb *lvb;
2898 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
2899 
2900 	if (info)
2901 		info->tf_valid = 0;
2902 
2903 	if (ocfs2_is_hard_readonly(osb))
2904 		return -EROFS;
2905 
2906 	if (ocfs2_mount_local(osb))
2907 		return 0;
2908 
2909 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX,
2910 				    trylock ? DLM_LKF_NOQUEUE : 0, 0);
2911 	if (status < 0) {
2912 		if (status != -EAGAIN)
2913 			mlog_errno(status);
2914 		return status;
2915 	}
2916 
2917 	if (info) {
2918 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2919 		if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2920 		    lvb->lvb_version == OCFS2_TRIMFS_LVB_VERSION) {
2921 			info->tf_valid = 1;
2922 			info->tf_success = lvb->lvb_success;
2923 			info->tf_nodenum = be32_to_cpu(lvb->lvb_nodenum);
2924 			info->tf_start = be64_to_cpu(lvb->lvb_start);
2925 			info->tf_len = be64_to_cpu(lvb->lvb_len);
2926 			info->tf_minlen = be64_to_cpu(lvb->lvb_minlen);
2927 			info->tf_trimlen = be64_to_cpu(lvb->lvb_trimlen);
2928 		}
2929 	}
2930 
2931 	return status;
2932 }
2933 
2934 void ocfs2_trim_fs_unlock(struct ocfs2_super *osb,
2935 			  struct ocfs2_trim_fs_info *info)
2936 {
2937 	struct ocfs2_trim_fs_lvb *lvb;
2938 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
2939 
2940 	if (ocfs2_mount_local(osb))
2941 		return;
2942 
2943 	if (info) {
2944 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2945 		lvb->lvb_version = OCFS2_TRIMFS_LVB_VERSION;
2946 		lvb->lvb_success = info->tf_success;
2947 		lvb->lvb_nodenum = cpu_to_be32(info->tf_nodenum);
2948 		lvb->lvb_start = cpu_to_be64(info->tf_start);
2949 		lvb->lvb_len = cpu_to_be64(info->tf_len);
2950 		lvb->lvb_minlen = cpu_to_be64(info->tf_minlen);
2951 		lvb->lvb_trimlen = cpu_to_be64(info->tf_trimlen);
2952 	}
2953 
2954 	ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2955 }
2956 
2957 int ocfs2_dentry_lock(struct dentry *dentry, int ex)
2958 {
2959 	int ret;
2960 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2961 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2962 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2963 
2964 	BUG_ON(!dl);
2965 
2966 	if (ocfs2_is_hard_readonly(osb)) {
2967 		if (ex)
2968 			return -EROFS;
2969 		return 0;
2970 	}
2971 
2972 	if (ocfs2_mount_local(osb))
2973 		return 0;
2974 
2975 	ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2976 	if (ret < 0)
2977 		mlog_errno(ret);
2978 
2979 	return ret;
2980 }
2981 
2982 void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2983 {
2984 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2985 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2986 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2987 
2988 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
2989 		ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
2990 }
2991 
2992 /* Reference counting of the dlm debug structure. We want this because
2993  * open references on the debug inodes can live on after a mount, so
2994  * we can't rely on the ocfs2_super to always exist. */
2995 static void ocfs2_dlm_debug_free(struct kref *kref)
2996 {
2997 	struct ocfs2_dlm_debug *dlm_debug;
2998 
2999 	dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
3000 
3001 	kfree(dlm_debug);
3002 }
3003 
3004 void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
3005 {
3006 	if (dlm_debug)
3007 		kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
3008 }
3009 
3010 static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
3011 {
3012 	kref_get(&debug->d_refcnt);
3013 }
3014 
3015 struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
3016 {
3017 	struct ocfs2_dlm_debug *dlm_debug;
3018 
3019 	dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
3020 	if (!dlm_debug) {
3021 		mlog_errno(-ENOMEM);
3022 		goto out;
3023 	}
3024 
3025 	kref_init(&dlm_debug->d_refcnt);
3026 	INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
3027 	dlm_debug->d_filter_secs = 0;
3028 out:
3029 	return dlm_debug;
3030 }
3031 
3032 /* Access to this is arbitrated for us via seq_file->sem. */
3033 struct ocfs2_dlm_seq_priv {
3034 	struct ocfs2_dlm_debug *p_dlm_debug;
3035 	struct ocfs2_lock_res p_iter_res;
3036 	struct ocfs2_lock_res p_tmp_res;
3037 };
3038 
3039 static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
3040 						 struct ocfs2_dlm_seq_priv *priv)
3041 {
3042 	struct ocfs2_lock_res *iter, *ret = NULL;
3043 	struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
3044 
3045 	assert_spin_locked(&ocfs2_dlm_tracking_lock);
3046 
3047 	list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
3048 		/* discover the head of the list */
3049 		if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
3050 			mlog(0, "End of list found, %p\n", ret);
3051 			break;
3052 		}
3053 
3054 		/* We track our "dummy" iteration lockres' by a NULL
3055 		 * l_ops field. */
3056 		if (iter->l_ops != NULL) {
3057 			ret = iter;
3058 			break;
3059 		}
3060 	}
3061 
3062 	return ret;
3063 }
3064 
3065 static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
3066 {
3067 	struct ocfs2_dlm_seq_priv *priv = m->private;
3068 	struct ocfs2_lock_res *iter;
3069 
3070 	spin_lock(&ocfs2_dlm_tracking_lock);
3071 	iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
3072 	if (iter) {
3073 		/* Since lockres' have the lifetime of their container
3074 		 * (which can be inodes, ocfs2_supers, etc) we want to
3075 		 * copy this out to a temporary lockres while still
3076 		 * under the spinlock. Obviously after this we can't
3077 		 * trust any pointers on the copy returned, but that's
3078 		 * ok as the information we want isn't typically held
3079 		 * in them. */
3080 		priv->p_tmp_res = *iter;
3081 		iter = &priv->p_tmp_res;
3082 	}
3083 	spin_unlock(&ocfs2_dlm_tracking_lock);
3084 
3085 	return iter;
3086 }
3087 
3088 static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
3089 {
3090 }
3091 
3092 static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
3093 {
3094 	struct ocfs2_dlm_seq_priv *priv = m->private;
3095 	struct ocfs2_lock_res *iter = v;
3096 	struct ocfs2_lock_res *dummy = &priv->p_iter_res;
3097 
3098 	spin_lock(&ocfs2_dlm_tracking_lock);
3099 	iter = ocfs2_dlm_next_res(iter, priv);
3100 	list_del_init(&dummy->l_debug_list);
3101 	if (iter) {
3102 		list_add(&dummy->l_debug_list, &iter->l_debug_list);
3103 		priv->p_tmp_res = *iter;
3104 		iter = &priv->p_tmp_res;
3105 	}
3106 	spin_unlock(&ocfs2_dlm_tracking_lock);
3107 
3108 	return iter;
3109 }
3110 
3111 /*
3112  * Version is used by debugfs.ocfs2 to determine the format being used
3113  *
3114  * New in version 2
3115  *	- Lock stats printed
3116  * New in version 3
3117  *	- Max time in lock stats is in usecs (instead of nsecs)
3118  * New in version 4
3119  *	- Add last pr/ex unlock times and first lock wait time in usecs
3120  */
3121 #define OCFS2_DLM_DEBUG_STR_VERSION 4
3122 static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
3123 {
3124 	int i;
3125 	char *lvb;
3126 	struct ocfs2_lock_res *lockres = v;
3127 #ifdef CONFIG_OCFS2_FS_STATS
3128 	u64 now, last;
3129 	struct ocfs2_dlm_debug *dlm_debug =
3130 			((struct ocfs2_dlm_seq_priv *)m->private)->p_dlm_debug;
3131 #endif
3132 
3133 	if (!lockres)
3134 		return -EINVAL;
3135 
3136 #ifdef CONFIG_OCFS2_FS_STATS
3137 	if (!lockres->l_lock_wait && dlm_debug->d_filter_secs) {
3138 		now = ktime_to_us(ktime_get_real());
3139 		if (lockres->l_lock_prmode.ls_last >
3140 		    lockres->l_lock_exmode.ls_last)
3141 			last = lockres->l_lock_prmode.ls_last;
3142 		else
3143 			last = lockres->l_lock_exmode.ls_last;
3144 		/*
3145 		 * Use d_filter_secs field to filter lock resources dump,
3146 		 * the default d_filter_secs(0) value filters nothing,
3147 		 * otherwise, only dump the last N seconds active lock
3148 		 * resources.
3149 		 */
3150 		if (div_u64(now - last, 1000000) > dlm_debug->d_filter_secs)
3151 			return 0;
3152 	}
3153 #endif
3154 
3155 	seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
3156 
3157 	if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
3158 		seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
3159 			   lockres->l_name,
3160 			   (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
3161 	else
3162 		seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
3163 
3164 	seq_printf(m, "%d\t"
3165 		   "0x%lx\t"
3166 		   "0x%x\t"
3167 		   "0x%x\t"
3168 		   "%u\t"
3169 		   "%u\t"
3170 		   "%d\t"
3171 		   "%d\t",
3172 		   lockres->l_level,
3173 		   lockres->l_flags,
3174 		   lockres->l_action,
3175 		   lockres->l_unlock_action,
3176 		   lockres->l_ro_holders,
3177 		   lockres->l_ex_holders,
3178 		   lockres->l_requested,
3179 		   lockres->l_blocking);
3180 
3181 	/* Dump the raw LVB */
3182 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
3183 	for(i = 0; i < DLM_LVB_LEN; i++)
3184 		seq_printf(m, "0x%x\t", lvb[i]);
3185 
3186 #ifdef CONFIG_OCFS2_FS_STATS
3187 # define lock_num_prmode(_l)		((_l)->l_lock_prmode.ls_gets)
3188 # define lock_num_exmode(_l)		((_l)->l_lock_exmode.ls_gets)
3189 # define lock_num_prmode_failed(_l)	((_l)->l_lock_prmode.ls_fail)
3190 # define lock_num_exmode_failed(_l)	((_l)->l_lock_exmode.ls_fail)
3191 # define lock_total_prmode(_l)		((_l)->l_lock_prmode.ls_total)
3192 # define lock_total_exmode(_l)		((_l)->l_lock_exmode.ls_total)
3193 # define lock_max_prmode(_l)		((_l)->l_lock_prmode.ls_max)
3194 # define lock_max_exmode(_l)		((_l)->l_lock_exmode.ls_max)
3195 # define lock_refresh(_l)		((_l)->l_lock_refresh)
3196 # define lock_last_prmode(_l)		((_l)->l_lock_prmode.ls_last)
3197 # define lock_last_exmode(_l)		((_l)->l_lock_exmode.ls_last)
3198 # define lock_wait(_l)			((_l)->l_lock_wait)
3199 #else
3200 # define lock_num_prmode(_l)		(0)
3201 # define lock_num_exmode(_l)		(0)
3202 # define lock_num_prmode_failed(_l)	(0)
3203 # define lock_num_exmode_failed(_l)	(0)
3204 # define lock_total_prmode(_l)		(0ULL)
3205 # define lock_total_exmode(_l)		(0ULL)
3206 # define lock_max_prmode(_l)		(0)
3207 # define lock_max_exmode(_l)		(0)
3208 # define lock_refresh(_l)		(0)
3209 # define lock_last_prmode(_l)		(0ULL)
3210 # define lock_last_exmode(_l)		(0ULL)
3211 # define lock_wait(_l)			(0ULL)
3212 #endif
3213 	/* The following seq_print was added in version 2 of this output */
3214 	seq_printf(m, "%u\t"
3215 		   "%u\t"
3216 		   "%u\t"
3217 		   "%u\t"
3218 		   "%llu\t"
3219 		   "%llu\t"
3220 		   "%u\t"
3221 		   "%u\t"
3222 		   "%u\t"
3223 		   "%llu\t"
3224 		   "%llu\t"
3225 		   "%llu\t",
3226 		   lock_num_prmode(lockres),
3227 		   lock_num_exmode(lockres),
3228 		   lock_num_prmode_failed(lockres),
3229 		   lock_num_exmode_failed(lockres),
3230 		   lock_total_prmode(lockres),
3231 		   lock_total_exmode(lockres),
3232 		   lock_max_prmode(lockres),
3233 		   lock_max_exmode(lockres),
3234 		   lock_refresh(lockres),
3235 		   lock_last_prmode(lockres),
3236 		   lock_last_exmode(lockres),
3237 		   lock_wait(lockres));
3238 
3239 	/* End the line */
3240 	seq_printf(m, "\n");
3241 	return 0;
3242 }
3243 
3244 static const struct seq_operations ocfs2_dlm_seq_ops = {
3245 	.start =	ocfs2_dlm_seq_start,
3246 	.stop =		ocfs2_dlm_seq_stop,
3247 	.next =		ocfs2_dlm_seq_next,
3248 	.show =		ocfs2_dlm_seq_show,
3249 };
3250 
3251 static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
3252 {
3253 	struct seq_file *seq = file->private_data;
3254 	struct ocfs2_dlm_seq_priv *priv = seq->private;
3255 	struct ocfs2_lock_res *res = &priv->p_iter_res;
3256 
3257 	ocfs2_remove_lockres_tracking(res);
3258 	ocfs2_put_dlm_debug(priv->p_dlm_debug);
3259 	return seq_release_private(inode, file);
3260 }
3261 
3262 static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
3263 {
3264 	struct ocfs2_dlm_seq_priv *priv;
3265 	struct ocfs2_super *osb;
3266 
3267 	priv = __seq_open_private(file, &ocfs2_dlm_seq_ops, sizeof(*priv));
3268 	if (!priv) {
3269 		mlog_errno(-ENOMEM);
3270 		return -ENOMEM;
3271 	}
3272 
3273 	osb = inode->i_private;
3274 	ocfs2_get_dlm_debug(osb->osb_dlm_debug);
3275 	priv->p_dlm_debug = osb->osb_dlm_debug;
3276 	INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
3277 
3278 	ocfs2_add_lockres_tracking(&priv->p_iter_res,
3279 				   priv->p_dlm_debug);
3280 
3281 	return 0;
3282 }
3283 
3284 static const struct file_operations ocfs2_dlm_debug_fops = {
3285 	.open =		ocfs2_dlm_debug_open,
3286 	.release =	ocfs2_dlm_debug_release,
3287 	.read =		seq_read,
3288 	.llseek =	seq_lseek,
3289 };
3290 
3291 static void ocfs2_dlm_init_debug(struct ocfs2_super *osb)
3292 {
3293 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
3294 
3295 	debugfs_create_file("locking_state", S_IFREG|S_IRUSR,
3296 			    osb->osb_debug_root, osb, &ocfs2_dlm_debug_fops);
3297 
3298 	debugfs_create_u32("locking_filter", 0600, osb->osb_debug_root,
3299 			   &dlm_debug->d_filter_secs);
3300 	ocfs2_get_dlm_debug(dlm_debug);
3301 }
3302 
3303 static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
3304 {
3305 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
3306 
3307 	if (dlm_debug)
3308 		ocfs2_put_dlm_debug(dlm_debug);
3309 }
3310 
3311 int ocfs2_dlm_init(struct ocfs2_super *osb)
3312 {
3313 	int status = 0;
3314 	struct ocfs2_cluster_connection *conn = NULL;
3315 
3316 	if (ocfs2_mount_local(osb)) {
3317 		osb->node_num = 0;
3318 		goto local;
3319 	}
3320 
3321 	ocfs2_dlm_init_debug(osb);
3322 
3323 	/* launch downconvert thread */
3324 	osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc-%s",
3325 			osb->uuid_str);
3326 	if (IS_ERR(osb->dc_task)) {
3327 		status = PTR_ERR(osb->dc_task);
3328 		osb->dc_task = NULL;
3329 		mlog_errno(status);
3330 		goto bail;
3331 	}
3332 
3333 	/* for now, uuid == domain */
3334 	status = ocfs2_cluster_connect(osb->osb_cluster_stack,
3335 				       osb->osb_cluster_name,
3336 				       strlen(osb->osb_cluster_name),
3337 				       osb->uuid_str,
3338 				       strlen(osb->uuid_str),
3339 				       &lproto, ocfs2_do_node_down, osb,
3340 				       &conn);
3341 	if (status) {
3342 		mlog_errno(status);
3343 		goto bail;
3344 	}
3345 
3346 	status = ocfs2_cluster_this_node(conn, &osb->node_num);
3347 	if (status < 0) {
3348 		mlog_errno(status);
3349 		mlog(ML_ERROR,
3350 		     "could not find this host's node number\n");
3351 		ocfs2_cluster_disconnect(conn, 0);
3352 		goto bail;
3353 	}
3354 
3355 local:
3356 	ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
3357 	ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
3358 	ocfs2_nfs_sync_lock_init(osb);
3359 	ocfs2_orphan_scan_lock_res_init(&osb->osb_orphan_scan.os_lockres, osb);
3360 
3361 	osb->cconn = conn;
3362 bail:
3363 	if (status < 0) {
3364 		ocfs2_dlm_shutdown_debug(osb);
3365 		if (osb->dc_task)
3366 			kthread_stop(osb->dc_task);
3367 	}
3368 
3369 	return status;
3370 }
3371 
3372 void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
3373 			int hangup_pending)
3374 {
3375 	ocfs2_drop_osb_locks(osb);
3376 
3377 	/*
3378 	 * Now that we have dropped all locks and ocfs2_dismount_volume()
3379 	 * has disabled recovery, the DLM won't be talking to us.  It's
3380 	 * safe to tear things down before disconnecting the cluster.
3381 	 */
3382 
3383 	if (osb->dc_task) {
3384 		kthread_stop(osb->dc_task);
3385 		osb->dc_task = NULL;
3386 	}
3387 
3388 	ocfs2_lock_res_free(&osb->osb_super_lockres);
3389 	ocfs2_lock_res_free(&osb->osb_rename_lockres);
3390 	ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres);
3391 	ocfs2_lock_res_free(&osb->osb_orphan_scan.os_lockres);
3392 
3393 	ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
3394 	osb->cconn = NULL;
3395 
3396 	ocfs2_dlm_shutdown_debug(osb);
3397 }
3398 
3399 static int ocfs2_drop_lock(struct ocfs2_super *osb,
3400 			   struct ocfs2_lock_res *lockres)
3401 {
3402 	int ret;
3403 	unsigned long flags;
3404 	u32 lkm_flags = 0;
3405 
3406 	/* We didn't get anywhere near actually using this lockres. */
3407 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
3408 		goto out;
3409 
3410 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
3411 		lkm_flags |= DLM_LKF_VALBLK;
3412 
3413 	spin_lock_irqsave(&lockres->l_lock, flags);
3414 
3415 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
3416 			"lockres %s, flags 0x%lx\n",
3417 			lockres->l_name, lockres->l_flags);
3418 
3419 	while (lockres->l_flags & OCFS2_LOCK_BUSY) {
3420 		mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
3421 		     "%u, unlock_action = %u\n",
3422 		     lockres->l_name, lockres->l_flags, lockres->l_action,
3423 		     lockres->l_unlock_action);
3424 
3425 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3426 
3427 		/* XXX: Today we just wait on any busy
3428 		 * locks... Perhaps we need to cancel converts in the
3429 		 * future? */
3430 		ocfs2_wait_on_busy_lock(lockres);
3431 
3432 		spin_lock_irqsave(&lockres->l_lock, flags);
3433 	}
3434 
3435 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3436 		if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
3437 		    lockres->l_level == DLM_LOCK_EX &&
3438 		    !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3439 			lockres->l_ops->set_lvb(lockres);
3440 	}
3441 
3442 	if (lockres->l_flags & OCFS2_LOCK_BUSY)
3443 		mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
3444 		     lockres->l_name);
3445 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
3446 		mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
3447 
3448 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
3449 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3450 		goto out;
3451 	}
3452 
3453 	lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
3454 
3455 	/* make sure we never get here while waiting for an ast to
3456 	 * fire. */
3457 	BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
3458 
3459 	/* is this necessary? */
3460 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3461 	lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
3462 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3463 
3464 	mlog(0, "lock %s\n", lockres->l_name);
3465 
3466 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags);
3467 	if (ret) {
3468 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3469 		mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
3470 		ocfs2_dlm_dump_lksb(&lockres->l_lksb);
3471 		BUG();
3472 	}
3473 	mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n",
3474 	     lockres->l_name);
3475 
3476 	ocfs2_wait_on_busy_lock(lockres);
3477 out:
3478 	return 0;
3479 }
3480 
3481 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3482 				       struct ocfs2_lock_res *lockres);
3483 
3484 /* Mark the lockres as being dropped. It will no longer be
3485  * queued if blocking, but we still may have to wait on it
3486  * being dequeued from the downconvert thread before we can consider
3487  * it safe to drop.
3488  *
3489  * You can *not* attempt to call cluster_lock on this lockres anymore. */
3490 void ocfs2_mark_lockres_freeing(struct ocfs2_super *osb,
3491 				struct ocfs2_lock_res *lockres)
3492 {
3493 	int status;
3494 	struct ocfs2_mask_waiter mw;
3495 	unsigned long flags, flags2;
3496 
3497 	ocfs2_init_mask_waiter(&mw);
3498 
3499 	spin_lock_irqsave(&lockres->l_lock, flags);
3500 	lockres->l_flags |= OCFS2_LOCK_FREEING;
3501 	if (lockres->l_flags & OCFS2_LOCK_QUEUED && current == osb->dc_task) {
3502 		/*
3503 		 * We know the downconvert is queued but not in progress
3504 		 * because we are the downconvert thread and processing
3505 		 * different lock. So we can just remove the lock from the
3506 		 * queue. This is not only an optimization but also a way
3507 		 * to avoid the following deadlock:
3508 		 *   ocfs2_dentry_post_unlock()
3509 		 *     ocfs2_dentry_lock_put()
3510 		 *       ocfs2_drop_dentry_lock()
3511 		 *         iput()
3512 		 *           ocfs2_evict_inode()
3513 		 *             ocfs2_clear_inode()
3514 		 *               ocfs2_mark_lockres_freeing()
3515 		 *                 ... blocks waiting for OCFS2_LOCK_QUEUED
3516 		 *                 since we are the downconvert thread which
3517 		 *                 should clear the flag.
3518 		 */
3519 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3520 		spin_lock_irqsave(&osb->dc_task_lock, flags2);
3521 		list_del_init(&lockres->l_blocked_list);
3522 		osb->blocked_lock_count--;
3523 		spin_unlock_irqrestore(&osb->dc_task_lock, flags2);
3524 		/*
3525 		 * Warn if we recurse into another post_unlock call.  Strictly
3526 		 * speaking it isn't a problem but we need to be careful if
3527 		 * that happens (stack overflow, deadlocks, ...) so warn if
3528 		 * ocfs2 grows a path for which this can happen.
3529 		 */
3530 		WARN_ON_ONCE(lockres->l_ops->post_unlock);
3531 		/* Since the lock is freeing we don't do much in the fn below */
3532 		ocfs2_process_blocked_lock(osb, lockres);
3533 		return;
3534 	}
3535 	while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
3536 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
3537 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3538 
3539 		mlog(0, "Waiting on lockres %s\n", lockres->l_name);
3540 
3541 		status = ocfs2_wait_for_mask(&mw);
3542 		if (status)
3543 			mlog_errno(status);
3544 
3545 		spin_lock_irqsave(&lockres->l_lock, flags);
3546 	}
3547 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3548 }
3549 
3550 void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
3551 			       struct ocfs2_lock_res *lockres)
3552 {
3553 	int ret;
3554 
3555 	ocfs2_mark_lockres_freeing(osb, lockres);
3556 	ret = ocfs2_drop_lock(osb, lockres);
3557 	if (ret)
3558 		mlog_errno(ret);
3559 }
3560 
3561 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
3562 {
3563 	ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
3564 	ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
3565 	ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres);
3566 	ocfs2_simple_drop_lockres(osb, &osb->osb_orphan_scan.os_lockres);
3567 }
3568 
3569 int ocfs2_drop_inode_locks(struct inode *inode)
3570 {
3571 	int status, err;
3572 
3573 	/* No need to call ocfs2_mark_lockres_freeing here -
3574 	 * ocfs2_clear_inode has done it for us. */
3575 
3576 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3577 			      &OCFS2_I(inode)->ip_open_lockres);
3578 	if (err < 0)
3579 		mlog_errno(err);
3580 
3581 	status = err;
3582 
3583 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3584 			      &OCFS2_I(inode)->ip_inode_lockres);
3585 	if (err < 0)
3586 		mlog_errno(err);
3587 	if (err < 0 && !status)
3588 		status = err;
3589 
3590 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3591 			      &OCFS2_I(inode)->ip_rw_lockres);
3592 	if (err < 0)
3593 		mlog_errno(err);
3594 	if (err < 0 && !status)
3595 		status = err;
3596 
3597 	return status;
3598 }
3599 
3600 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
3601 					      int new_level)
3602 {
3603 	assert_spin_locked(&lockres->l_lock);
3604 
3605 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
3606 
3607 	if (lockres->l_level <= new_level) {
3608 		mlog(ML_ERROR, "lockres %s, lvl %d <= %d, blcklst %d, mask %d, "
3609 		     "type %d, flags 0x%lx, hold %d %d, act %d %d, req %d, "
3610 		     "block %d, pgen %d\n", lockres->l_name, lockres->l_level,
3611 		     new_level, list_empty(&lockres->l_blocked_list),
3612 		     list_empty(&lockres->l_mask_waiters), lockres->l_type,
3613 		     lockres->l_flags, lockres->l_ro_holders,
3614 		     lockres->l_ex_holders, lockres->l_action,
3615 		     lockres->l_unlock_action, lockres->l_requested,
3616 		     lockres->l_blocking, lockres->l_pending_gen);
3617 		BUG();
3618 	}
3619 
3620 	mlog(ML_BASTS, "lockres %s, level %d => %d, blocking %d\n",
3621 	     lockres->l_name, lockres->l_level, new_level, lockres->l_blocking);
3622 
3623 	lockres->l_action = OCFS2_AST_DOWNCONVERT;
3624 	lockres->l_requested = new_level;
3625 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3626 	return lockres_set_pending(lockres);
3627 }
3628 
3629 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
3630 				  struct ocfs2_lock_res *lockres,
3631 				  int new_level,
3632 				  int lvb,
3633 				  unsigned int generation)
3634 {
3635 	int ret;
3636 	u32 dlm_flags = DLM_LKF_CONVERT;
3637 
3638 	mlog(ML_BASTS, "lockres %s, level %d => %d\n", lockres->l_name,
3639 	     lockres->l_level, new_level);
3640 
3641 	/*
3642 	 * On DLM_LKF_VALBLK, fsdlm behaves differently with o2cb. It always
3643 	 * expects DLM_LKF_VALBLK being set if the LKB has LVB, so that
3644 	 * we can recover correctly from node failure. Otherwise, we may get
3645 	 * invalid LVB in LKB, but without DLM_SBF_VALNOTVALID being set.
3646 	 */
3647 	if (ocfs2_userspace_stack(osb) &&
3648 	    lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
3649 		lvb = 1;
3650 
3651 	if (lvb)
3652 		dlm_flags |= DLM_LKF_VALBLK;
3653 
3654 	ret = ocfs2_dlm_lock(osb->cconn,
3655 			     new_level,
3656 			     &lockres->l_lksb,
3657 			     dlm_flags,
3658 			     lockres->l_name,
3659 			     OCFS2_LOCK_ID_MAX_LEN - 1);
3660 	lockres_clear_pending(lockres, generation, osb);
3661 	if (ret) {
3662 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
3663 		ocfs2_recover_from_dlm_error(lockres, 1);
3664 		goto bail;
3665 	}
3666 
3667 	ret = 0;
3668 bail:
3669 	return ret;
3670 }
3671 
3672 /* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
3673 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
3674 				        struct ocfs2_lock_res *lockres)
3675 {
3676 	assert_spin_locked(&lockres->l_lock);
3677 
3678 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
3679 		/* If we're already trying to cancel a lock conversion
3680 		 * then just drop the spinlock and allow the caller to
3681 		 * requeue this lock. */
3682 		mlog(ML_BASTS, "lockres %s, skip convert\n", lockres->l_name);
3683 		return 0;
3684 	}
3685 
3686 	/* were we in a convert when we got the bast fire? */
3687 	BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
3688 	       lockres->l_action != OCFS2_AST_DOWNCONVERT);
3689 	/* set things up for the unlockast to know to just
3690 	 * clear out the ast_action and unset busy, etc. */
3691 	lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
3692 
3693 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
3694 			"lock %s, invalid flags: 0x%lx\n",
3695 			lockres->l_name, lockres->l_flags);
3696 
3697 	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3698 
3699 	return 1;
3700 }
3701 
3702 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
3703 				struct ocfs2_lock_res *lockres)
3704 {
3705 	int ret;
3706 
3707 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
3708 			       DLM_LKF_CANCEL);
3709 	if (ret) {
3710 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3711 		ocfs2_recover_from_dlm_error(lockres, 0);
3712 	}
3713 
3714 	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3715 
3716 	return ret;
3717 }
3718 
3719 static int ocfs2_unblock_lock(struct ocfs2_super *osb,
3720 			      struct ocfs2_lock_res *lockres,
3721 			      struct ocfs2_unblock_ctl *ctl)
3722 {
3723 	unsigned long flags;
3724 	int blocking;
3725 	int new_level;
3726 	int level;
3727 	int ret = 0;
3728 	int set_lvb = 0;
3729 	unsigned int gen;
3730 
3731 	spin_lock_irqsave(&lockres->l_lock, flags);
3732 
3733 recheck:
3734 	/*
3735 	 * Is it still blocking? If not, we have no more work to do.
3736 	 */
3737 	if (!(lockres->l_flags & OCFS2_LOCK_BLOCKED)) {
3738 		BUG_ON(lockres->l_blocking != DLM_LOCK_NL);
3739 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3740 		ret = 0;
3741 		goto leave;
3742 	}
3743 
3744 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
3745 		/* XXX
3746 		 * This is a *big* race.  The OCFS2_LOCK_PENDING flag
3747 		 * exists entirely for one reason - another thread has set
3748 		 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
3749 		 *
3750 		 * If we do ocfs2_cancel_convert() before the other thread
3751 		 * calls dlm_lock(), our cancel will do nothing.  We will
3752 		 * get no ast, and we will have no way of knowing the
3753 		 * cancel failed.  Meanwhile, the other thread will call
3754 		 * into dlm_lock() and wait...forever.
3755 		 *
3756 		 * Why forever?  Because another node has asked for the
3757 		 * lock first; that's why we're here in unblock_lock().
3758 		 *
3759 		 * The solution is OCFS2_LOCK_PENDING.  When PENDING is
3760 		 * set, we just requeue the unblock.  Only when the other
3761 		 * thread has called dlm_lock() and cleared PENDING will
3762 		 * we then cancel their request.
3763 		 *
3764 		 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
3765 		 * at the same time they set OCFS2_DLM_BUSY.  They must
3766 		 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
3767 		 */
3768 		if (lockres->l_flags & OCFS2_LOCK_PENDING) {
3769 			mlog(ML_BASTS, "lockres %s, ReQ: Pending\n",
3770 			     lockres->l_name);
3771 			goto leave_requeue;
3772 		}
3773 
3774 		ctl->requeue = 1;
3775 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
3776 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3777 		if (ret) {
3778 			ret = ocfs2_cancel_convert(osb, lockres);
3779 			if (ret < 0)
3780 				mlog_errno(ret);
3781 		}
3782 		goto leave;
3783 	}
3784 
3785 	/*
3786 	 * This prevents livelocks. OCFS2_LOCK_UPCONVERT_FINISHING flag is
3787 	 * set when the ast is received for an upconvert just before the
3788 	 * OCFS2_LOCK_BUSY flag is cleared. Now if the fs received a bast
3789 	 * on the heels of the ast, we want to delay the downconvert just
3790 	 * enough to allow the up requestor to do its task. Because this
3791 	 * lock is in the blocked queue, the lock will be downconverted
3792 	 * as soon as the requestor is done with the lock.
3793 	 */
3794 	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING)
3795 		goto leave_requeue;
3796 
3797 	/*
3798 	 * How can we block and yet be at NL?  We were trying to upconvert
3799 	 * from NL and got canceled.  The code comes back here, and now
3800 	 * we notice and clear BLOCKING.
3801 	 */
3802 	if (lockres->l_level == DLM_LOCK_NL) {
3803 		BUG_ON(lockres->l_ex_holders || lockres->l_ro_holders);
3804 		mlog(ML_BASTS, "lockres %s, Aborting dc\n", lockres->l_name);
3805 		lockres->l_blocking = DLM_LOCK_NL;
3806 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
3807 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3808 		goto leave;
3809 	}
3810 
3811 	/* if we're blocking an exclusive and we have *any* holders,
3812 	 * then requeue. */
3813 	if ((lockres->l_blocking == DLM_LOCK_EX)
3814 	    && (lockres->l_ex_holders || lockres->l_ro_holders)) {
3815 		mlog(ML_BASTS, "lockres %s, ReQ: EX/PR Holders %u,%u\n",
3816 		     lockres->l_name, lockres->l_ex_holders,
3817 		     lockres->l_ro_holders);
3818 		goto leave_requeue;
3819 	}
3820 
3821 	/* If it's a PR we're blocking, then only
3822 	 * requeue if we've got any EX holders */
3823 	if (lockres->l_blocking == DLM_LOCK_PR &&
3824 	    lockres->l_ex_holders) {
3825 		mlog(ML_BASTS, "lockres %s, ReQ: EX Holders %u\n",
3826 		     lockres->l_name, lockres->l_ex_holders);
3827 		goto leave_requeue;
3828 	}
3829 
3830 	/*
3831 	 * Can we get a lock in this state if the holder counts are
3832 	 * zero? The meta data unblock code used to check this.
3833 	 */
3834 	if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
3835 	    && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) {
3836 		mlog(ML_BASTS, "lockres %s, ReQ: Lock Refreshing\n",
3837 		     lockres->l_name);
3838 		goto leave_requeue;
3839 	}
3840 
3841 	new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
3842 
3843 	if (lockres->l_ops->check_downconvert
3844 	    && !lockres->l_ops->check_downconvert(lockres, new_level)) {
3845 		mlog(ML_BASTS, "lockres %s, ReQ: Checkpointing\n",
3846 		     lockres->l_name);
3847 		goto leave_requeue;
3848 	}
3849 
3850 	/* If we get here, then we know that there are no more
3851 	 * incompatible holders (and anyone asking for an incompatible
3852 	 * lock is blocked). We can now downconvert the lock */
3853 	if (!lockres->l_ops->downconvert_worker)
3854 		goto downconvert;
3855 
3856 	/* Some lockres types want to do a bit of work before
3857 	 * downconverting a lock. Allow that here. The worker function
3858 	 * may sleep, so we save off a copy of what we're blocking as
3859 	 * it may change while we're not holding the spin lock. */
3860 	blocking = lockres->l_blocking;
3861 	level = lockres->l_level;
3862 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3863 
3864 	ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
3865 
3866 	if (ctl->unblock_action == UNBLOCK_STOP_POST) {
3867 		mlog(ML_BASTS, "lockres %s, UNBLOCK_STOP_POST\n",
3868 		     lockres->l_name);
3869 		goto leave;
3870 	}
3871 
3872 	spin_lock_irqsave(&lockres->l_lock, flags);
3873 	if ((blocking != lockres->l_blocking) || (level != lockres->l_level)) {
3874 		/* If this changed underneath us, then we can't drop
3875 		 * it just yet. */
3876 		mlog(ML_BASTS, "lockres %s, block=%d:%d, level=%d:%d, "
3877 		     "Recheck\n", lockres->l_name, blocking,
3878 		     lockres->l_blocking, level, lockres->l_level);
3879 		goto recheck;
3880 	}
3881 
3882 downconvert:
3883 	ctl->requeue = 0;
3884 
3885 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3886 		if (lockres->l_level == DLM_LOCK_EX)
3887 			set_lvb = 1;
3888 
3889 		/*
3890 		 * We only set the lvb if the lock has been fully
3891 		 * refreshed - otherwise we risk setting stale
3892 		 * data. Otherwise, there's no need to actually clear
3893 		 * out the lvb here as it's value is still valid.
3894 		 */
3895 		if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3896 			lockres->l_ops->set_lvb(lockres);
3897 	}
3898 
3899 	gen = ocfs2_prepare_downconvert(lockres, new_level);
3900 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3901 	ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
3902 				     gen);
3903 
3904 leave:
3905 	if (ret)
3906 		mlog_errno(ret);
3907 	return ret;
3908 
3909 leave_requeue:
3910 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3911 	ctl->requeue = 1;
3912 
3913 	return 0;
3914 }
3915 
3916 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
3917 				     int blocking)
3918 {
3919 	struct inode *inode;
3920 	struct address_space *mapping;
3921 	struct ocfs2_inode_info *oi;
3922 
3923        	inode = ocfs2_lock_res_inode(lockres);
3924 	mapping = inode->i_mapping;
3925 
3926 	if (S_ISDIR(inode->i_mode)) {
3927 		oi = OCFS2_I(inode);
3928 		oi->ip_dir_lock_gen++;
3929 		mlog(0, "generation: %u\n", oi->ip_dir_lock_gen);
3930 		goto out;
3931 	}
3932 
3933 	if (!S_ISREG(inode->i_mode))
3934 		goto out;
3935 
3936 	/*
3937 	 * We need this before the filemap_fdatawrite() so that it can
3938 	 * transfer the dirty bit from the PTE to the
3939 	 * page. Unfortunately this means that even for EX->PR
3940 	 * downconverts, we'll lose our mappings and have to build
3941 	 * them up again.
3942 	 */
3943 	unmap_mapping_range(mapping, 0, 0, 0);
3944 
3945 	if (filemap_fdatawrite(mapping)) {
3946 		mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
3947 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
3948 	}
3949 	sync_mapping_buffers(mapping);
3950 	if (blocking == DLM_LOCK_EX) {
3951 		truncate_inode_pages(mapping, 0);
3952 	} else {
3953 		/* We only need to wait on the I/O if we're not also
3954 		 * truncating pages because truncate_inode_pages waits
3955 		 * for us above. We don't truncate pages if we're
3956 		 * blocking anything < EXMODE because we want to keep
3957 		 * them around in that case. */
3958 		filemap_fdatawait(mapping);
3959 	}
3960 
3961 	forget_all_cached_acls(inode);
3962 
3963 out:
3964 	return UNBLOCK_CONTINUE;
3965 }
3966 
3967 static int ocfs2_ci_checkpointed(struct ocfs2_caching_info *ci,
3968 				 struct ocfs2_lock_res *lockres,
3969 				 int new_level)
3970 {
3971 	int checkpointed = ocfs2_ci_fully_checkpointed(ci);
3972 
3973 	BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
3974 	BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
3975 
3976 	if (checkpointed)
3977 		return 1;
3978 
3979 	ocfs2_start_checkpoint(OCFS2_SB(ocfs2_metadata_cache_get_super(ci)));
3980 	return 0;
3981 }
3982 
3983 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
3984 					int new_level)
3985 {
3986 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3987 
3988 	return ocfs2_ci_checkpointed(INODE_CACHE(inode), lockres, new_level);
3989 }
3990 
3991 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
3992 {
3993 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3994 
3995 	__ocfs2_stuff_meta_lvb(inode);
3996 }
3997 
3998 /*
3999  * Does the final reference drop on our dentry lock. Right now this
4000  * happens in the downconvert thread, but we could choose to simplify the
4001  * dlmglue API and push these off to the ocfs2_wq in the future.
4002  */
4003 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
4004 				     struct ocfs2_lock_res *lockres)
4005 {
4006 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
4007 	ocfs2_dentry_lock_put(osb, dl);
4008 }
4009 
4010 /*
4011  * d_delete() matching dentries before the lock downconvert.
4012  *
4013  * At this point, any process waiting to destroy the
4014  * dentry_lock due to last ref count is stopped by the
4015  * OCFS2_LOCK_QUEUED flag.
4016  *
4017  * We have two potential problems
4018  *
4019  * 1) If we do the last reference drop on our dentry_lock (via dput)
4020  *    we'll wind up in ocfs2_release_dentry_lock(), waiting on
4021  *    the downconvert to finish. Instead we take an elevated
4022  *    reference and push the drop until after we've completed our
4023  *    unblock processing.
4024  *
4025  * 2) There might be another process with a final reference,
4026  *    waiting on us to finish processing. If this is the case, we
4027  *    detect it and exit out - there's no more dentries anyway.
4028  */
4029 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
4030 				       int blocking)
4031 {
4032 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
4033 	struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
4034 	struct dentry *dentry;
4035 	unsigned long flags;
4036 	int extra_ref = 0;
4037 
4038 	/*
4039 	 * This node is blocking another node from getting a read
4040 	 * lock. This happens when we've renamed within a
4041 	 * directory. We've forced the other nodes to d_delete(), but
4042 	 * we never actually dropped our lock because it's still
4043 	 * valid. The downconvert code will retain a PR for this node,
4044 	 * so there's no further work to do.
4045 	 */
4046 	if (blocking == DLM_LOCK_PR)
4047 		return UNBLOCK_CONTINUE;
4048 
4049 	/*
4050 	 * Mark this inode as potentially orphaned. The code in
4051 	 * ocfs2_delete_inode() will figure out whether it actually
4052 	 * needs to be freed or not.
4053 	 */
4054 	spin_lock(&oi->ip_lock);
4055 	oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
4056 	spin_unlock(&oi->ip_lock);
4057 
4058 	/*
4059 	 * Yuck. We need to make sure however that the check of
4060 	 * OCFS2_LOCK_FREEING and the extra reference are atomic with
4061 	 * respect to a reference decrement or the setting of that
4062 	 * flag.
4063 	 */
4064 	spin_lock_irqsave(&lockres->l_lock, flags);
4065 	spin_lock(&dentry_attach_lock);
4066 	if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
4067 	    && dl->dl_count) {
4068 		dl->dl_count++;
4069 		extra_ref = 1;
4070 	}
4071 	spin_unlock(&dentry_attach_lock);
4072 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4073 
4074 	mlog(0, "extra_ref = %d\n", extra_ref);
4075 
4076 	/*
4077 	 * We have a process waiting on us in ocfs2_dentry_iput(),
4078 	 * which means we can't have any more outstanding
4079 	 * aliases. There's no need to do any more work.
4080 	 */
4081 	if (!extra_ref)
4082 		return UNBLOCK_CONTINUE;
4083 
4084 	spin_lock(&dentry_attach_lock);
4085 	while (1) {
4086 		dentry = ocfs2_find_local_alias(dl->dl_inode,
4087 						dl->dl_parent_blkno, 1);
4088 		if (!dentry)
4089 			break;
4090 		spin_unlock(&dentry_attach_lock);
4091 
4092 		if (S_ISDIR(dl->dl_inode->i_mode))
4093 			shrink_dcache_parent(dentry);
4094 
4095 		mlog(0, "d_delete(%pd);\n", dentry);
4096 
4097 		/*
4098 		 * The following dcache calls may do an
4099 		 * iput(). Normally we don't want that from the
4100 		 * downconverting thread, but in this case it's ok
4101 		 * because the requesting node already has an
4102 		 * exclusive lock on the inode, so it can't be queued
4103 		 * for a downconvert.
4104 		 */
4105 		d_delete(dentry);
4106 		dput(dentry);
4107 
4108 		spin_lock(&dentry_attach_lock);
4109 	}
4110 	spin_unlock(&dentry_attach_lock);
4111 
4112 	/*
4113 	 * If we are the last holder of this dentry lock, there is no
4114 	 * reason to downconvert so skip straight to the unlock.
4115 	 */
4116 	if (dl->dl_count == 1)
4117 		return UNBLOCK_STOP_POST;
4118 
4119 	return UNBLOCK_CONTINUE_POST;
4120 }
4121 
4122 static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
4123 					    int new_level)
4124 {
4125 	struct ocfs2_refcount_tree *tree =
4126 				ocfs2_lock_res_refcount_tree(lockres);
4127 
4128 	return ocfs2_ci_checkpointed(&tree->rf_ci, lockres, new_level);
4129 }
4130 
4131 static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
4132 					 int blocking)
4133 {
4134 	struct ocfs2_refcount_tree *tree =
4135 				ocfs2_lock_res_refcount_tree(lockres);
4136 
4137 	ocfs2_metadata_cache_purge(&tree->rf_ci);
4138 
4139 	return UNBLOCK_CONTINUE;
4140 }
4141 
4142 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
4143 {
4144 	struct ocfs2_qinfo_lvb *lvb;
4145 	struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres);
4146 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
4147 					    oinfo->dqi_gi.dqi_type);
4148 
4149 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
4150 	lvb->lvb_version = OCFS2_QINFO_LVB_VERSION;
4151 	lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace);
4152 	lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace);
4153 	lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms);
4154 	lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks);
4155 	lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk);
4156 	lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry);
4157 }
4158 
4159 void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex)
4160 {
4161 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4162 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
4163 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4164 
4165 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
4166 		ocfs2_cluster_unlock(osb, lockres, level);
4167 }
4168 
4169 static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo)
4170 {
4171 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
4172 					    oinfo->dqi_gi.dqi_type);
4173 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4174 	struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
4175 	struct buffer_head *bh = NULL;
4176 	struct ocfs2_global_disk_dqinfo *gdinfo;
4177 	int status = 0;
4178 
4179 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
4180 	    lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) {
4181 		info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace);
4182 		info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace);
4183 		oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms);
4184 		oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks);
4185 		oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk);
4186 		oinfo->dqi_gi.dqi_free_entry =
4187 					be32_to_cpu(lvb->lvb_free_entry);
4188 	} else {
4189 		status = ocfs2_read_quota_phys_block(oinfo->dqi_gqinode,
4190 						     oinfo->dqi_giblk, &bh);
4191 		if (status) {
4192 			mlog_errno(status);
4193 			goto bail;
4194 		}
4195 		gdinfo = (struct ocfs2_global_disk_dqinfo *)
4196 					(bh->b_data + OCFS2_GLOBAL_INFO_OFF);
4197 		info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace);
4198 		info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace);
4199 		oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms);
4200 		oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks);
4201 		oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk);
4202 		oinfo->dqi_gi.dqi_free_entry =
4203 					le32_to_cpu(gdinfo->dqi_free_entry);
4204 		brelse(bh);
4205 		ocfs2_track_lock_refresh(lockres);
4206 	}
4207 
4208 bail:
4209 	return status;
4210 }
4211 
4212 /* Lock quota info, this function expects at least shared lock on the quota file
4213  * so that we can safely refresh quota info from disk. */
4214 int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex)
4215 {
4216 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4217 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
4218 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4219 	int status = 0;
4220 
4221 	/* On RO devices, locking really isn't needed... */
4222 	if (ocfs2_is_hard_readonly(osb)) {
4223 		if (ex)
4224 			status = -EROFS;
4225 		goto bail;
4226 	}
4227 	if (ocfs2_mount_local(osb))
4228 		goto bail;
4229 
4230 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
4231 	if (status < 0) {
4232 		mlog_errno(status);
4233 		goto bail;
4234 	}
4235 	if (!ocfs2_should_refresh_lock_res(lockres))
4236 		goto bail;
4237 	/* OK, we have the lock but we need to refresh the quota info */
4238 	status = ocfs2_refresh_qinfo(oinfo);
4239 	if (status)
4240 		ocfs2_qinfo_unlock(oinfo, ex);
4241 	ocfs2_complete_lock_res_refresh(lockres, status);
4242 bail:
4243 	return status;
4244 }
4245 
4246 int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex)
4247 {
4248 	int status;
4249 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4250 	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
4251 	struct ocfs2_super *osb = lockres->l_priv;
4252 
4253 
4254 	if (ocfs2_is_hard_readonly(osb))
4255 		return -EROFS;
4256 
4257 	if (ocfs2_mount_local(osb))
4258 		return 0;
4259 
4260 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
4261 	if (status < 0)
4262 		mlog_errno(status);
4263 
4264 	return status;
4265 }
4266 
4267 void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex)
4268 {
4269 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4270 	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
4271 	struct ocfs2_super *osb = lockres->l_priv;
4272 
4273 	if (!ocfs2_mount_local(osb))
4274 		ocfs2_cluster_unlock(osb, lockres, level);
4275 }
4276 
4277 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
4278 				       struct ocfs2_lock_res *lockres)
4279 {
4280 	int status;
4281 	struct ocfs2_unblock_ctl ctl = {0, 0,};
4282 	unsigned long flags;
4283 
4284 	/* Our reference to the lockres in this function can be
4285 	 * considered valid until we remove the OCFS2_LOCK_QUEUED
4286 	 * flag. */
4287 
4288 	BUG_ON(!lockres);
4289 	BUG_ON(!lockres->l_ops);
4290 
4291 	mlog(ML_BASTS, "lockres %s blocked\n", lockres->l_name);
4292 
4293 	/* Detect whether a lock has been marked as going away while
4294 	 * the downconvert thread was processing other things. A lock can
4295 	 * still be marked with OCFS2_LOCK_FREEING after this check,
4296 	 * but short circuiting here will still save us some
4297 	 * performance. */
4298 	spin_lock_irqsave(&lockres->l_lock, flags);
4299 	if (lockres->l_flags & OCFS2_LOCK_FREEING)
4300 		goto unqueue;
4301 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4302 
4303 	status = ocfs2_unblock_lock(osb, lockres, &ctl);
4304 	if (status < 0)
4305 		mlog_errno(status);
4306 
4307 	spin_lock_irqsave(&lockres->l_lock, flags);
4308 unqueue:
4309 	if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
4310 		lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
4311 	} else
4312 		ocfs2_schedule_blocked_lock(osb, lockres);
4313 
4314 	mlog(ML_BASTS, "lockres %s, requeue = %s.\n", lockres->l_name,
4315 	     ctl.requeue ? "yes" : "no");
4316 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4317 
4318 	if (ctl.unblock_action != UNBLOCK_CONTINUE
4319 	    && lockres->l_ops->post_unlock)
4320 		lockres->l_ops->post_unlock(osb, lockres);
4321 }
4322 
4323 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
4324 					struct ocfs2_lock_res *lockres)
4325 {
4326 	unsigned long flags;
4327 
4328 	assert_spin_locked(&lockres->l_lock);
4329 
4330 	if (lockres->l_flags & OCFS2_LOCK_FREEING) {
4331 		/* Do not schedule a lock for downconvert when it's on
4332 		 * the way to destruction - any nodes wanting access
4333 		 * to the resource will get it soon. */
4334 		mlog(ML_BASTS, "lockres %s won't be scheduled: flags 0x%lx\n",
4335 		     lockres->l_name, lockres->l_flags);
4336 		return;
4337 	}
4338 
4339 	lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
4340 
4341 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4342 	if (list_empty(&lockres->l_blocked_list)) {
4343 		list_add_tail(&lockres->l_blocked_list,
4344 			      &osb->blocked_lock_list);
4345 		osb->blocked_lock_count++;
4346 	}
4347 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4348 }
4349 
4350 static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
4351 {
4352 	unsigned long processed;
4353 	unsigned long flags;
4354 	struct ocfs2_lock_res *lockres;
4355 
4356 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4357 	/* grab this early so we know to try again if a state change and
4358 	 * wake happens part-way through our work  */
4359 	osb->dc_work_sequence = osb->dc_wake_sequence;
4360 
4361 	processed = osb->blocked_lock_count;
4362 	/*
4363 	 * blocked lock processing in this loop might call iput which can
4364 	 * remove items off osb->blocked_lock_list. Downconvert up to
4365 	 * 'processed' number of locks, but stop short if we had some
4366 	 * removed in ocfs2_mark_lockres_freeing when downconverting.
4367 	 */
4368 	while (processed && !list_empty(&osb->blocked_lock_list)) {
4369 		lockres = list_entry(osb->blocked_lock_list.next,
4370 				     struct ocfs2_lock_res, l_blocked_list);
4371 		list_del_init(&lockres->l_blocked_list);
4372 		osb->blocked_lock_count--;
4373 		spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4374 
4375 		BUG_ON(!processed);
4376 		processed--;
4377 
4378 		ocfs2_process_blocked_lock(osb, lockres);
4379 
4380 		spin_lock_irqsave(&osb->dc_task_lock, flags);
4381 	}
4382 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4383 }
4384 
4385 static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
4386 {
4387 	int empty = 0;
4388 	unsigned long flags;
4389 
4390 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4391 	if (list_empty(&osb->blocked_lock_list))
4392 		empty = 1;
4393 
4394 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4395 	return empty;
4396 }
4397 
4398 static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
4399 {
4400 	int should_wake = 0;
4401 	unsigned long flags;
4402 
4403 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4404 	if (osb->dc_work_sequence != osb->dc_wake_sequence)
4405 		should_wake = 1;
4406 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4407 
4408 	return should_wake;
4409 }
4410 
4411 static int ocfs2_downconvert_thread(void *arg)
4412 {
4413 	struct ocfs2_super *osb = arg;
4414 
4415 	/* only quit once we've been asked to stop and there is no more
4416 	 * work available */
4417 	while (!(kthread_should_stop() &&
4418 		ocfs2_downconvert_thread_lists_empty(osb))) {
4419 
4420 		wait_event_interruptible(osb->dc_event,
4421 					 ocfs2_downconvert_thread_should_wake(osb) ||
4422 					 kthread_should_stop());
4423 
4424 		mlog(0, "downconvert_thread: awoken\n");
4425 
4426 		ocfs2_downconvert_thread_do_work(osb);
4427 	}
4428 
4429 	osb->dc_task = NULL;
4430 	return 0;
4431 }
4432 
4433 void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
4434 {
4435 	unsigned long flags;
4436 
4437 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4438 	/* make sure the voting thread gets a swipe at whatever changes
4439 	 * the caller may have made to the voting state */
4440 	osb->dc_wake_sequence++;
4441 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4442 	wake_up(&osb->dc_event);
4443 }
4444