xref: /openbmc/linux/fs/ocfs2/dlmglue.c (revision 27ab1c1c)
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 		if (ex)
2878 			up_write(&osb->nfs_sync_rwlock);
2879 		else
2880 			up_read(&osb->nfs_sync_rwlock);
2881 	}
2882 
2883 	return status;
2884 }
2885 
2886 void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex)
2887 {
2888 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2889 
2890 	if (!ocfs2_mount_local(osb))
2891 		ocfs2_cluster_unlock(osb, lockres,
2892 				     ex ? LKM_EXMODE : LKM_PRMODE);
2893 	if (ex)
2894 		up_write(&osb->nfs_sync_rwlock);
2895 	else
2896 		up_read(&osb->nfs_sync_rwlock);
2897 }
2898 
2899 int ocfs2_trim_fs_lock(struct ocfs2_super *osb,
2900 		       struct ocfs2_trim_fs_info *info, int trylock)
2901 {
2902 	int status;
2903 	struct ocfs2_trim_fs_lvb *lvb;
2904 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
2905 
2906 	if (info)
2907 		info->tf_valid = 0;
2908 
2909 	if (ocfs2_is_hard_readonly(osb))
2910 		return -EROFS;
2911 
2912 	if (ocfs2_mount_local(osb))
2913 		return 0;
2914 
2915 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX,
2916 				    trylock ? DLM_LKF_NOQUEUE : 0, 0);
2917 	if (status < 0) {
2918 		if (status != -EAGAIN)
2919 			mlog_errno(status);
2920 		return status;
2921 	}
2922 
2923 	if (info) {
2924 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2925 		if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2926 		    lvb->lvb_version == OCFS2_TRIMFS_LVB_VERSION) {
2927 			info->tf_valid = 1;
2928 			info->tf_success = lvb->lvb_success;
2929 			info->tf_nodenum = be32_to_cpu(lvb->lvb_nodenum);
2930 			info->tf_start = be64_to_cpu(lvb->lvb_start);
2931 			info->tf_len = be64_to_cpu(lvb->lvb_len);
2932 			info->tf_minlen = be64_to_cpu(lvb->lvb_minlen);
2933 			info->tf_trimlen = be64_to_cpu(lvb->lvb_trimlen);
2934 		}
2935 	}
2936 
2937 	return status;
2938 }
2939 
2940 void ocfs2_trim_fs_unlock(struct ocfs2_super *osb,
2941 			  struct ocfs2_trim_fs_info *info)
2942 {
2943 	struct ocfs2_trim_fs_lvb *lvb;
2944 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
2945 
2946 	if (ocfs2_mount_local(osb))
2947 		return;
2948 
2949 	if (info) {
2950 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2951 		lvb->lvb_version = OCFS2_TRIMFS_LVB_VERSION;
2952 		lvb->lvb_success = info->tf_success;
2953 		lvb->lvb_nodenum = cpu_to_be32(info->tf_nodenum);
2954 		lvb->lvb_start = cpu_to_be64(info->tf_start);
2955 		lvb->lvb_len = cpu_to_be64(info->tf_len);
2956 		lvb->lvb_minlen = cpu_to_be64(info->tf_minlen);
2957 		lvb->lvb_trimlen = cpu_to_be64(info->tf_trimlen);
2958 	}
2959 
2960 	ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2961 }
2962 
2963 int ocfs2_dentry_lock(struct dentry *dentry, int ex)
2964 {
2965 	int ret;
2966 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2967 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2968 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2969 
2970 	BUG_ON(!dl);
2971 
2972 	if (ocfs2_is_hard_readonly(osb)) {
2973 		if (ex)
2974 			return -EROFS;
2975 		return 0;
2976 	}
2977 
2978 	if (ocfs2_mount_local(osb))
2979 		return 0;
2980 
2981 	ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2982 	if (ret < 0)
2983 		mlog_errno(ret);
2984 
2985 	return ret;
2986 }
2987 
2988 void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2989 {
2990 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2991 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2992 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2993 
2994 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
2995 		ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
2996 }
2997 
2998 /* Reference counting of the dlm debug structure. We want this because
2999  * open references on the debug inodes can live on after a mount, so
3000  * we can't rely on the ocfs2_super to always exist. */
3001 static void ocfs2_dlm_debug_free(struct kref *kref)
3002 {
3003 	struct ocfs2_dlm_debug *dlm_debug;
3004 
3005 	dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
3006 
3007 	kfree(dlm_debug);
3008 }
3009 
3010 void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
3011 {
3012 	if (dlm_debug)
3013 		kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
3014 }
3015 
3016 static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
3017 {
3018 	kref_get(&debug->d_refcnt);
3019 }
3020 
3021 struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
3022 {
3023 	struct ocfs2_dlm_debug *dlm_debug;
3024 
3025 	dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
3026 	if (!dlm_debug) {
3027 		mlog_errno(-ENOMEM);
3028 		goto out;
3029 	}
3030 
3031 	kref_init(&dlm_debug->d_refcnt);
3032 	INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
3033 	dlm_debug->d_filter_secs = 0;
3034 out:
3035 	return dlm_debug;
3036 }
3037 
3038 /* Access to this is arbitrated for us via seq_file->sem. */
3039 struct ocfs2_dlm_seq_priv {
3040 	struct ocfs2_dlm_debug *p_dlm_debug;
3041 	struct ocfs2_lock_res p_iter_res;
3042 	struct ocfs2_lock_res p_tmp_res;
3043 };
3044 
3045 static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
3046 						 struct ocfs2_dlm_seq_priv *priv)
3047 {
3048 	struct ocfs2_lock_res *iter, *ret = NULL;
3049 	struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
3050 
3051 	assert_spin_locked(&ocfs2_dlm_tracking_lock);
3052 
3053 	list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
3054 		/* discover the head of the list */
3055 		if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
3056 			mlog(0, "End of list found, %p\n", ret);
3057 			break;
3058 		}
3059 
3060 		/* We track our "dummy" iteration lockres' by a NULL
3061 		 * l_ops field. */
3062 		if (iter->l_ops != NULL) {
3063 			ret = iter;
3064 			break;
3065 		}
3066 	}
3067 
3068 	return ret;
3069 }
3070 
3071 static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
3072 {
3073 	struct ocfs2_dlm_seq_priv *priv = m->private;
3074 	struct ocfs2_lock_res *iter;
3075 
3076 	spin_lock(&ocfs2_dlm_tracking_lock);
3077 	iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
3078 	if (iter) {
3079 		/* Since lockres' have the lifetime of their container
3080 		 * (which can be inodes, ocfs2_supers, etc) we want to
3081 		 * copy this out to a temporary lockres while still
3082 		 * under the spinlock. Obviously after this we can't
3083 		 * trust any pointers on the copy returned, but that's
3084 		 * ok as the information we want isn't typically held
3085 		 * in them. */
3086 		priv->p_tmp_res = *iter;
3087 		iter = &priv->p_tmp_res;
3088 	}
3089 	spin_unlock(&ocfs2_dlm_tracking_lock);
3090 
3091 	return iter;
3092 }
3093 
3094 static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
3095 {
3096 }
3097 
3098 static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
3099 {
3100 	struct ocfs2_dlm_seq_priv *priv = m->private;
3101 	struct ocfs2_lock_res *iter = v;
3102 	struct ocfs2_lock_res *dummy = &priv->p_iter_res;
3103 
3104 	spin_lock(&ocfs2_dlm_tracking_lock);
3105 	iter = ocfs2_dlm_next_res(iter, priv);
3106 	list_del_init(&dummy->l_debug_list);
3107 	if (iter) {
3108 		list_add(&dummy->l_debug_list, &iter->l_debug_list);
3109 		priv->p_tmp_res = *iter;
3110 		iter = &priv->p_tmp_res;
3111 	}
3112 	spin_unlock(&ocfs2_dlm_tracking_lock);
3113 
3114 	return iter;
3115 }
3116 
3117 /*
3118  * Version is used by debugfs.ocfs2 to determine the format being used
3119  *
3120  * New in version 2
3121  *	- Lock stats printed
3122  * New in version 3
3123  *	- Max time in lock stats is in usecs (instead of nsecs)
3124  * New in version 4
3125  *	- Add last pr/ex unlock times and first lock wait time in usecs
3126  */
3127 #define OCFS2_DLM_DEBUG_STR_VERSION 4
3128 static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
3129 {
3130 	int i;
3131 	char *lvb;
3132 	struct ocfs2_lock_res *lockres = v;
3133 #ifdef CONFIG_OCFS2_FS_STATS
3134 	u64 now, last;
3135 	struct ocfs2_dlm_debug *dlm_debug =
3136 			((struct ocfs2_dlm_seq_priv *)m->private)->p_dlm_debug;
3137 #endif
3138 
3139 	if (!lockres)
3140 		return -EINVAL;
3141 
3142 #ifdef CONFIG_OCFS2_FS_STATS
3143 	if (!lockres->l_lock_wait && dlm_debug->d_filter_secs) {
3144 		now = ktime_to_us(ktime_get_real());
3145 		if (lockres->l_lock_prmode.ls_last >
3146 		    lockres->l_lock_exmode.ls_last)
3147 			last = lockres->l_lock_prmode.ls_last;
3148 		else
3149 			last = lockres->l_lock_exmode.ls_last;
3150 		/*
3151 		 * Use d_filter_secs field to filter lock resources dump,
3152 		 * the default d_filter_secs(0) value filters nothing,
3153 		 * otherwise, only dump the last N seconds active lock
3154 		 * resources.
3155 		 */
3156 		if (div_u64(now - last, 1000000) > dlm_debug->d_filter_secs)
3157 			return 0;
3158 	}
3159 #endif
3160 
3161 	seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
3162 
3163 	if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
3164 		seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
3165 			   lockres->l_name,
3166 			   (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
3167 	else
3168 		seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
3169 
3170 	seq_printf(m, "%d\t"
3171 		   "0x%lx\t"
3172 		   "0x%x\t"
3173 		   "0x%x\t"
3174 		   "%u\t"
3175 		   "%u\t"
3176 		   "%d\t"
3177 		   "%d\t",
3178 		   lockres->l_level,
3179 		   lockres->l_flags,
3180 		   lockres->l_action,
3181 		   lockres->l_unlock_action,
3182 		   lockres->l_ro_holders,
3183 		   lockres->l_ex_holders,
3184 		   lockres->l_requested,
3185 		   lockres->l_blocking);
3186 
3187 	/* Dump the raw LVB */
3188 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
3189 	for(i = 0; i < DLM_LVB_LEN; i++)
3190 		seq_printf(m, "0x%x\t", lvb[i]);
3191 
3192 #ifdef CONFIG_OCFS2_FS_STATS
3193 # define lock_num_prmode(_l)		((_l)->l_lock_prmode.ls_gets)
3194 # define lock_num_exmode(_l)		((_l)->l_lock_exmode.ls_gets)
3195 # define lock_num_prmode_failed(_l)	((_l)->l_lock_prmode.ls_fail)
3196 # define lock_num_exmode_failed(_l)	((_l)->l_lock_exmode.ls_fail)
3197 # define lock_total_prmode(_l)		((_l)->l_lock_prmode.ls_total)
3198 # define lock_total_exmode(_l)		((_l)->l_lock_exmode.ls_total)
3199 # define lock_max_prmode(_l)		((_l)->l_lock_prmode.ls_max)
3200 # define lock_max_exmode(_l)		((_l)->l_lock_exmode.ls_max)
3201 # define lock_refresh(_l)		((_l)->l_lock_refresh)
3202 # define lock_last_prmode(_l)		((_l)->l_lock_prmode.ls_last)
3203 # define lock_last_exmode(_l)		((_l)->l_lock_exmode.ls_last)
3204 # define lock_wait(_l)			((_l)->l_lock_wait)
3205 #else
3206 # define lock_num_prmode(_l)		(0)
3207 # define lock_num_exmode(_l)		(0)
3208 # define lock_num_prmode_failed(_l)	(0)
3209 # define lock_num_exmode_failed(_l)	(0)
3210 # define lock_total_prmode(_l)		(0ULL)
3211 # define lock_total_exmode(_l)		(0ULL)
3212 # define lock_max_prmode(_l)		(0)
3213 # define lock_max_exmode(_l)		(0)
3214 # define lock_refresh(_l)		(0)
3215 # define lock_last_prmode(_l)		(0ULL)
3216 # define lock_last_exmode(_l)		(0ULL)
3217 # define lock_wait(_l)			(0ULL)
3218 #endif
3219 	/* The following seq_print was added in version 2 of this output */
3220 	seq_printf(m, "%u\t"
3221 		   "%u\t"
3222 		   "%u\t"
3223 		   "%u\t"
3224 		   "%llu\t"
3225 		   "%llu\t"
3226 		   "%u\t"
3227 		   "%u\t"
3228 		   "%u\t"
3229 		   "%llu\t"
3230 		   "%llu\t"
3231 		   "%llu\t",
3232 		   lock_num_prmode(lockres),
3233 		   lock_num_exmode(lockres),
3234 		   lock_num_prmode_failed(lockres),
3235 		   lock_num_exmode_failed(lockres),
3236 		   lock_total_prmode(lockres),
3237 		   lock_total_exmode(lockres),
3238 		   lock_max_prmode(lockres),
3239 		   lock_max_exmode(lockres),
3240 		   lock_refresh(lockres),
3241 		   lock_last_prmode(lockres),
3242 		   lock_last_exmode(lockres),
3243 		   lock_wait(lockres));
3244 
3245 	/* End the line */
3246 	seq_printf(m, "\n");
3247 	return 0;
3248 }
3249 
3250 static const struct seq_operations ocfs2_dlm_seq_ops = {
3251 	.start =	ocfs2_dlm_seq_start,
3252 	.stop =		ocfs2_dlm_seq_stop,
3253 	.next =		ocfs2_dlm_seq_next,
3254 	.show =		ocfs2_dlm_seq_show,
3255 };
3256 
3257 static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
3258 {
3259 	struct seq_file *seq = file->private_data;
3260 	struct ocfs2_dlm_seq_priv *priv = seq->private;
3261 	struct ocfs2_lock_res *res = &priv->p_iter_res;
3262 
3263 	ocfs2_remove_lockres_tracking(res);
3264 	ocfs2_put_dlm_debug(priv->p_dlm_debug);
3265 	return seq_release_private(inode, file);
3266 }
3267 
3268 static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
3269 {
3270 	struct ocfs2_dlm_seq_priv *priv;
3271 	struct ocfs2_super *osb;
3272 
3273 	priv = __seq_open_private(file, &ocfs2_dlm_seq_ops, sizeof(*priv));
3274 	if (!priv) {
3275 		mlog_errno(-ENOMEM);
3276 		return -ENOMEM;
3277 	}
3278 
3279 	osb = inode->i_private;
3280 	ocfs2_get_dlm_debug(osb->osb_dlm_debug);
3281 	priv->p_dlm_debug = osb->osb_dlm_debug;
3282 	INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
3283 
3284 	ocfs2_add_lockres_tracking(&priv->p_iter_res,
3285 				   priv->p_dlm_debug);
3286 
3287 	return 0;
3288 }
3289 
3290 static const struct file_operations ocfs2_dlm_debug_fops = {
3291 	.open =		ocfs2_dlm_debug_open,
3292 	.release =	ocfs2_dlm_debug_release,
3293 	.read =		seq_read,
3294 	.llseek =	seq_lseek,
3295 };
3296 
3297 static void ocfs2_dlm_init_debug(struct ocfs2_super *osb)
3298 {
3299 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
3300 
3301 	debugfs_create_file("locking_state", S_IFREG|S_IRUSR,
3302 			    osb->osb_debug_root, osb, &ocfs2_dlm_debug_fops);
3303 
3304 	debugfs_create_u32("locking_filter", 0600, osb->osb_debug_root,
3305 			   &dlm_debug->d_filter_secs);
3306 	ocfs2_get_dlm_debug(dlm_debug);
3307 }
3308 
3309 static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
3310 {
3311 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
3312 
3313 	if (dlm_debug)
3314 		ocfs2_put_dlm_debug(dlm_debug);
3315 }
3316 
3317 int ocfs2_dlm_init(struct ocfs2_super *osb)
3318 {
3319 	int status = 0;
3320 	struct ocfs2_cluster_connection *conn = NULL;
3321 
3322 	if (ocfs2_mount_local(osb)) {
3323 		osb->node_num = 0;
3324 		goto local;
3325 	}
3326 
3327 	ocfs2_dlm_init_debug(osb);
3328 
3329 	/* launch downconvert thread */
3330 	osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc-%s",
3331 			osb->uuid_str);
3332 	if (IS_ERR(osb->dc_task)) {
3333 		status = PTR_ERR(osb->dc_task);
3334 		osb->dc_task = NULL;
3335 		mlog_errno(status);
3336 		goto bail;
3337 	}
3338 
3339 	/* for now, uuid == domain */
3340 	status = ocfs2_cluster_connect(osb->osb_cluster_stack,
3341 				       osb->osb_cluster_name,
3342 				       strlen(osb->osb_cluster_name),
3343 				       osb->uuid_str,
3344 				       strlen(osb->uuid_str),
3345 				       &lproto, ocfs2_do_node_down, osb,
3346 				       &conn);
3347 	if (status) {
3348 		mlog_errno(status);
3349 		goto bail;
3350 	}
3351 
3352 	status = ocfs2_cluster_this_node(conn, &osb->node_num);
3353 	if (status < 0) {
3354 		mlog_errno(status);
3355 		mlog(ML_ERROR,
3356 		     "could not find this host's node number\n");
3357 		ocfs2_cluster_disconnect(conn, 0);
3358 		goto bail;
3359 	}
3360 
3361 local:
3362 	ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
3363 	ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
3364 	ocfs2_nfs_sync_lock_init(osb);
3365 	ocfs2_orphan_scan_lock_res_init(&osb->osb_orphan_scan.os_lockres, osb);
3366 
3367 	osb->cconn = conn;
3368 bail:
3369 	if (status < 0) {
3370 		ocfs2_dlm_shutdown_debug(osb);
3371 		if (osb->dc_task)
3372 			kthread_stop(osb->dc_task);
3373 	}
3374 
3375 	return status;
3376 }
3377 
3378 void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
3379 			int hangup_pending)
3380 {
3381 	ocfs2_drop_osb_locks(osb);
3382 
3383 	/*
3384 	 * Now that we have dropped all locks and ocfs2_dismount_volume()
3385 	 * has disabled recovery, the DLM won't be talking to us.  It's
3386 	 * safe to tear things down before disconnecting the cluster.
3387 	 */
3388 
3389 	if (osb->dc_task) {
3390 		kthread_stop(osb->dc_task);
3391 		osb->dc_task = NULL;
3392 	}
3393 
3394 	ocfs2_lock_res_free(&osb->osb_super_lockres);
3395 	ocfs2_lock_res_free(&osb->osb_rename_lockres);
3396 	ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres);
3397 	ocfs2_lock_res_free(&osb->osb_orphan_scan.os_lockres);
3398 
3399 	ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
3400 	osb->cconn = NULL;
3401 
3402 	ocfs2_dlm_shutdown_debug(osb);
3403 }
3404 
3405 static int ocfs2_drop_lock(struct ocfs2_super *osb,
3406 			   struct ocfs2_lock_res *lockres)
3407 {
3408 	int ret;
3409 	unsigned long flags;
3410 	u32 lkm_flags = 0;
3411 
3412 	/* We didn't get anywhere near actually using this lockres. */
3413 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
3414 		goto out;
3415 
3416 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
3417 		lkm_flags |= DLM_LKF_VALBLK;
3418 
3419 	spin_lock_irqsave(&lockres->l_lock, flags);
3420 
3421 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
3422 			"lockres %s, flags 0x%lx\n",
3423 			lockres->l_name, lockres->l_flags);
3424 
3425 	while (lockres->l_flags & OCFS2_LOCK_BUSY) {
3426 		mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
3427 		     "%u, unlock_action = %u\n",
3428 		     lockres->l_name, lockres->l_flags, lockres->l_action,
3429 		     lockres->l_unlock_action);
3430 
3431 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3432 
3433 		/* XXX: Today we just wait on any busy
3434 		 * locks... Perhaps we need to cancel converts in the
3435 		 * future? */
3436 		ocfs2_wait_on_busy_lock(lockres);
3437 
3438 		spin_lock_irqsave(&lockres->l_lock, flags);
3439 	}
3440 
3441 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3442 		if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
3443 		    lockres->l_level == DLM_LOCK_EX &&
3444 		    !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3445 			lockres->l_ops->set_lvb(lockres);
3446 	}
3447 
3448 	if (lockres->l_flags & OCFS2_LOCK_BUSY)
3449 		mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
3450 		     lockres->l_name);
3451 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
3452 		mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
3453 
3454 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
3455 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3456 		goto out;
3457 	}
3458 
3459 	lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
3460 
3461 	/* make sure we never get here while waiting for an ast to
3462 	 * fire. */
3463 	BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
3464 
3465 	/* is this necessary? */
3466 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3467 	lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
3468 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3469 
3470 	mlog(0, "lock %s\n", lockres->l_name);
3471 
3472 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags);
3473 	if (ret) {
3474 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3475 		mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
3476 		ocfs2_dlm_dump_lksb(&lockres->l_lksb);
3477 		BUG();
3478 	}
3479 	mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n",
3480 	     lockres->l_name);
3481 
3482 	ocfs2_wait_on_busy_lock(lockres);
3483 out:
3484 	return 0;
3485 }
3486 
3487 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3488 				       struct ocfs2_lock_res *lockres);
3489 
3490 /* Mark the lockres as being dropped. It will no longer be
3491  * queued if blocking, but we still may have to wait on it
3492  * being dequeued from the downconvert thread before we can consider
3493  * it safe to drop.
3494  *
3495  * You can *not* attempt to call cluster_lock on this lockres anymore. */
3496 void ocfs2_mark_lockres_freeing(struct ocfs2_super *osb,
3497 				struct ocfs2_lock_res *lockres)
3498 {
3499 	int status;
3500 	struct ocfs2_mask_waiter mw;
3501 	unsigned long flags, flags2;
3502 
3503 	ocfs2_init_mask_waiter(&mw);
3504 
3505 	spin_lock_irqsave(&lockres->l_lock, flags);
3506 	lockres->l_flags |= OCFS2_LOCK_FREEING;
3507 	if (lockres->l_flags & OCFS2_LOCK_QUEUED && current == osb->dc_task) {
3508 		/*
3509 		 * We know the downconvert is queued but not in progress
3510 		 * because we are the downconvert thread and processing
3511 		 * different lock. So we can just remove the lock from the
3512 		 * queue. This is not only an optimization but also a way
3513 		 * to avoid the following deadlock:
3514 		 *   ocfs2_dentry_post_unlock()
3515 		 *     ocfs2_dentry_lock_put()
3516 		 *       ocfs2_drop_dentry_lock()
3517 		 *         iput()
3518 		 *           ocfs2_evict_inode()
3519 		 *             ocfs2_clear_inode()
3520 		 *               ocfs2_mark_lockres_freeing()
3521 		 *                 ... blocks waiting for OCFS2_LOCK_QUEUED
3522 		 *                 since we are the downconvert thread which
3523 		 *                 should clear the flag.
3524 		 */
3525 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3526 		spin_lock_irqsave(&osb->dc_task_lock, flags2);
3527 		list_del_init(&lockres->l_blocked_list);
3528 		osb->blocked_lock_count--;
3529 		spin_unlock_irqrestore(&osb->dc_task_lock, flags2);
3530 		/*
3531 		 * Warn if we recurse into another post_unlock call.  Strictly
3532 		 * speaking it isn't a problem but we need to be careful if
3533 		 * that happens (stack overflow, deadlocks, ...) so warn if
3534 		 * ocfs2 grows a path for which this can happen.
3535 		 */
3536 		WARN_ON_ONCE(lockres->l_ops->post_unlock);
3537 		/* Since the lock is freeing we don't do much in the fn below */
3538 		ocfs2_process_blocked_lock(osb, lockres);
3539 		return;
3540 	}
3541 	while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
3542 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
3543 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3544 
3545 		mlog(0, "Waiting on lockres %s\n", lockres->l_name);
3546 
3547 		status = ocfs2_wait_for_mask(&mw);
3548 		if (status)
3549 			mlog_errno(status);
3550 
3551 		spin_lock_irqsave(&lockres->l_lock, flags);
3552 	}
3553 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3554 }
3555 
3556 void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
3557 			       struct ocfs2_lock_res *lockres)
3558 {
3559 	int ret;
3560 
3561 	ocfs2_mark_lockres_freeing(osb, lockres);
3562 	ret = ocfs2_drop_lock(osb, lockres);
3563 	if (ret)
3564 		mlog_errno(ret);
3565 }
3566 
3567 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
3568 {
3569 	ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
3570 	ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
3571 	ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres);
3572 	ocfs2_simple_drop_lockres(osb, &osb->osb_orphan_scan.os_lockres);
3573 }
3574 
3575 int ocfs2_drop_inode_locks(struct inode *inode)
3576 {
3577 	int status, err;
3578 
3579 	/* No need to call ocfs2_mark_lockres_freeing here -
3580 	 * ocfs2_clear_inode has done it for us. */
3581 
3582 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3583 			      &OCFS2_I(inode)->ip_open_lockres);
3584 	if (err < 0)
3585 		mlog_errno(err);
3586 
3587 	status = err;
3588 
3589 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3590 			      &OCFS2_I(inode)->ip_inode_lockres);
3591 	if (err < 0)
3592 		mlog_errno(err);
3593 	if (err < 0 && !status)
3594 		status = err;
3595 
3596 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3597 			      &OCFS2_I(inode)->ip_rw_lockres);
3598 	if (err < 0)
3599 		mlog_errno(err);
3600 	if (err < 0 && !status)
3601 		status = err;
3602 
3603 	return status;
3604 }
3605 
3606 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
3607 					      int new_level)
3608 {
3609 	assert_spin_locked(&lockres->l_lock);
3610 
3611 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
3612 
3613 	if (lockres->l_level <= new_level) {
3614 		mlog(ML_ERROR, "lockres %s, lvl %d <= %d, blcklst %d, mask %d, "
3615 		     "type %d, flags 0x%lx, hold %d %d, act %d %d, req %d, "
3616 		     "block %d, pgen %d\n", lockres->l_name, lockres->l_level,
3617 		     new_level, list_empty(&lockres->l_blocked_list),
3618 		     list_empty(&lockres->l_mask_waiters), lockres->l_type,
3619 		     lockres->l_flags, lockres->l_ro_holders,
3620 		     lockres->l_ex_holders, lockres->l_action,
3621 		     lockres->l_unlock_action, lockres->l_requested,
3622 		     lockres->l_blocking, lockres->l_pending_gen);
3623 		BUG();
3624 	}
3625 
3626 	mlog(ML_BASTS, "lockres %s, level %d => %d, blocking %d\n",
3627 	     lockres->l_name, lockres->l_level, new_level, lockres->l_blocking);
3628 
3629 	lockres->l_action = OCFS2_AST_DOWNCONVERT;
3630 	lockres->l_requested = new_level;
3631 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3632 	return lockres_set_pending(lockres);
3633 }
3634 
3635 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
3636 				  struct ocfs2_lock_res *lockres,
3637 				  int new_level,
3638 				  int lvb,
3639 				  unsigned int generation)
3640 {
3641 	int ret;
3642 	u32 dlm_flags = DLM_LKF_CONVERT;
3643 
3644 	mlog(ML_BASTS, "lockres %s, level %d => %d\n", lockres->l_name,
3645 	     lockres->l_level, new_level);
3646 
3647 	/*
3648 	 * On DLM_LKF_VALBLK, fsdlm behaves differently with o2cb. It always
3649 	 * expects DLM_LKF_VALBLK being set if the LKB has LVB, so that
3650 	 * we can recover correctly from node failure. Otherwise, we may get
3651 	 * invalid LVB in LKB, but without DLM_SBF_VALNOTVALID being set.
3652 	 */
3653 	if (ocfs2_userspace_stack(osb) &&
3654 	    lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
3655 		lvb = 1;
3656 
3657 	if (lvb)
3658 		dlm_flags |= DLM_LKF_VALBLK;
3659 
3660 	ret = ocfs2_dlm_lock(osb->cconn,
3661 			     new_level,
3662 			     &lockres->l_lksb,
3663 			     dlm_flags,
3664 			     lockres->l_name,
3665 			     OCFS2_LOCK_ID_MAX_LEN - 1);
3666 	lockres_clear_pending(lockres, generation, osb);
3667 	if (ret) {
3668 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
3669 		ocfs2_recover_from_dlm_error(lockres, 1);
3670 		goto bail;
3671 	}
3672 
3673 	ret = 0;
3674 bail:
3675 	return ret;
3676 }
3677 
3678 /* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
3679 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
3680 				        struct ocfs2_lock_res *lockres)
3681 {
3682 	assert_spin_locked(&lockres->l_lock);
3683 
3684 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
3685 		/* If we're already trying to cancel a lock conversion
3686 		 * then just drop the spinlock and allow the caller to
3687 		 * requeue this lock. */
3688 		mlog(ML_BASTS, "lockres %s, skip convert\n", lockres->l_name);
3689 		return 0;
3690 	}
3691 
3692 	/* were we in a convert when we got the bast fire? */
3693 	BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
3694 	       lockres->l_action != OCFS2_AST_DOWNCONVERT);
3695 	/* set things up for the unlockast to know to just
3696 	 * clear out the ast_action and unset busy, etc. */
3697 	lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
3698 
3699 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
3700 			"lock %s, invalid flags: 0x%lx\n",
3701 			lockres->l_name, lockres->l_flags);
3702 
3703 	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3704 
3705 	return 1;
3706 }
3707 
3708 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
3709 				struct ocfs2_lock_res *lockres)
3710 {
3711 	int ret;
3712 
3713 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
3714 			       DLM_LKF_CANCEL);
3715 	if (ret) {
3716 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3717 		ocfs2_recover_from_dlm_error(lockres, 0);
3718 	}
3719 
3720 	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3721 
3722 	return ret;
3723 }
3724 
3725 static int ocfs2_unblock_lock(struct ocfs2_super *osb,
3726 			      struct ocfs2_lock_res *lockres,
3727 			      struct ocfs2_unblock_ctl *ctl)
3728 {
3729 	unsigned long flags;
3730 	int blocking;
3731 	int new_level;
3732 	int level;
3733 	int ret = 0;
3734 	int set_lvb = 0;
3735 	unsigned int gen;
3736 
3737 	spin_lock_irqsave(&lockres->l_lock, flags);
3738 
3739 recheck:
3740 	/*
3741 	 * Is it still blocking? If not, we have no more work to do.
3742 	 */
3743 	if (!(lockres->l_flags & OCFS2_LOCK_BLOCKED)) {
3744 		BUG_ON(lockres->l_blocking != DLM_LOCK_NL);
3745 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3746 		ret = 0;
3747 		goto leave;
3748 	}
3749 
3750 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
3751 		/* XXX
3752 		 * This is a *big* race.  The OCFS2_LOCK_PENDING flag
3753 		 * exists entirely for one reason - another thread has set
3754 		 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
3755 		 *
3756 		 * If we do ocfs2_cancel_convert() before the other thread
3757 		 * calls dlm_lock(), our cancel will do nothing.  We will
3758 		 * get no ast, and we will have no way of knowing the
3759 		 * cancel failed.  Meanwhile, the other thread will call
3760 		 * into dlm_lock() and wait...forever.
3761 		 *
3762 		 * Why forever?  Because another node has asked for the
3763 		 * lock first; that's why we're here in unblock_lock().
3764 		 *
3765 		 * The solution is OCFS2_LOCK_PENDING.  When PENDING is
3766 		 * set, we just requeue the unblock.  Only when the other
3767 		 * thread has called dlm_lock() and cleared PENDING will
3768 		 * we then cancel their request.
3769 		 *
3770 		 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
3771 		 * at the same time they set OCFS2_DLM_BUSY.  They must
3772 		 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
3773 		 */
3774 		if (lockres->l_flags & OCFS2_LOCK_PENDING) {
3775 			mlog(ML_BASTS, "lockres %s, ReQ: Pending\n",
3776 			     lockres->l_name);
3777 			goto leave_requeue;
3778 		}
3779 
3780 		ctl->requeue = 1;
3781 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
3782 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3783 		if (ret) {
3784 			ret = ocfs2_cancel_convert(osb, lockres);
3785 			if (ret < 0)
3786 				mlog_errno(ret);
3787 		}
3788 		goto leave;
3789 	}
3790 
3791 	/*
3792 	 * This prevents livelocks. OCFS2_LOCK_UPCONVERT_FINISHING flag is
3793 	 * set when the ast is received for an upconvert just before the
3794 	 * OCFS2_LOCK_BUSY flag is cleared. Now if the fs received a bast
3795 	 * on the heels of the ast, we want to delay the downconvert just
3796 	 * enough to allow the up requestor to do its task. Because this
3797 	 * lock is in the blocked queue, the lock will be downconverted
3798 	 * as soon as the requestor is done with the lock.
3799 	 */
3800 	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING)
3801 		goto leave_requeue;
3802 
3803 	/*
3804 	 * How can we block and yet be at NL?  We were trying to upconvert
3805 	 * from NL and got canceled.  The code comes back here, and now
3806 	 * we notice and clear BLOCKING.
3807 	 */
3808 	if (lockres->l_level == DLM_LOCK_NL) {
3809 		BUG_ON(lockres->l_ex_holders || lockres->l_ro_holders);
3810 		mlog(ML_BASTS, "lockres %s, Aborting dc\n", lockres->l_name);
3811 		lockres->l_blocking = DLM_LOCK_NL;
3812 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
3813 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3814 		goto leave;
3815 	}
3816 
3817 	/* if we're blocking an exclusive and we have *any* holders,
3818 	 * then requeue. */
3819 	if ((lockres->l_blocking == DLM_LOCK_EX)
3820 	    && (lockres->l_ex_holders || lockres->l_ro_holders)) {
3821 		mlog(ML_BASTS, "lockres %s, ReQ: EX/PR Holders %u,%u\n",
3822 		     lockres->l_name, lockres->l_ex_holders,
3823 		     lockres->l_ro_holders);
3824 		goto leave_requeue;
3825 	}
3826 
3827 	/* If it's a PR we're blocking, then only
3828 	 * requeue if we've got any EX holders */
3829 	if (lockres->l_blocking == DLM_LOCK_PR &&
3830 	    lockres->l_ex_holders) {
3831 		mlog(ML_BASTS, "lockres %s, ReQ: EX Holders %u\n",
3832 		     lockres->l_name, lockres->l_ex_holders);
3833 		goto leave_requeue;
3834 	}
3835 
3836 	/*
3837 	 * Can we get a lock in this state if the holder counts are
3838 	 * zero? The meta data unblock code used to check this.
3839 	 */
3840 	if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
3841 	    && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) {
3842 		mlog(ML_BASTS, "lockres %s, ReQ: Lock Refreshing\n",
3843 		     lockres->l_name);
3844 		goto leave_requeue;
3845 	}
3846 
3847 	new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
3848 
3849 	if (lockres->l_ops->check_downconvert
3850 	    && !lockres->l_ops->check_downconvert(lockres, new_level)) {
3851 		mlog(ML_BASTS, "lockres %s, ReQ: Checkpointing\n",
3852 		     lockres->l_name);
3853 		goto leave_requeue;
3854 	}
3855 
3856 	/* If we get here, then we know that there are no more
3857 	 * incompatible holders (and anyone asking for an incompatible
3858 	 * lock is blocked). We can now downconvert the lock */
3859 	if (!lockres->l_ops->downconvert_worker)
3860 		goto downconvert;
3861 
3862 	/* Some lockres types want to do a bit of work before
3863 	 * downconverting a lock. Allow that here. The worker function
3864 	 * may sleep, so we save off a copy of what we're blocking as
3865 	 * it may change while we're not holding the spin lock. */
3866 	blocking = lockres->l_blocking;
3867 	level = lockres->l_level;
3868 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3869 
3870 	ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
3871 
3872 	if (ctl->unblock_action == UNBLOCK_STOP_POST) {
3873 		mlog(ML_BASTS, "lockres %s, UNBLOCK_STOP_POST\n",
3874 		     lockres->l_name);
3875 		goto leave;
3876 	}
3877 
3878 	spin_lock_irqsave(&lockres->l_lock, flags);
3879 	if ((blocking != lockres->l_blocking) || (level != lockres->l_level)) {
3880 		/* If this changed underneath us, then we can't drop
3881 		 * it just yet. */
3882 		mlog(ML_BASTS, "lockres %s, block=%d:%d, level=%d:%d, "
3883 		     "Recheck\n", lockres->l_name, blocking,
3884 		     lockres->l_blocking, level, lockres->l_level);
3885 		goto recheck;
3886 	}
3887 
3888 downconvert:
3889 	ctl->requeue = 0;
3890 
3891 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3892 		if (lockres->l_level == DLM_LOCK_EX)
3893 			set_lvb = 1;
3894 
3895 		/*
3896 		 * We only set the lvb if the lock has been fully
3897 		 * refreshed - otherwise we risk setting stale
3898 		 * data. Otherwise, there's no need to actually clear
3899 		 * out the lvb here as it's value is still valid.
3900 		 */
3901 		if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3902 			lockres->l_ops->set_lvb(lockres);
3903 	}
3904 
3905 	gen = ocfs2_prepare_downconvert(lockres, new_level);
3906 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3907 	ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
3908 				     gen);
3909 
3910 leave:
3911 	if (ret)
3912 		mlog_errno(ret);
3913 	return ret;
3914 
3915 leave_requeue:
3916 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3917 	ctl->requeue = 1;
3918 
3919 	return 0;
3920 }
3921 
3922 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
3923 				     int blocking)
3924 {
3925 	struct inode *inode;
3926 	struct address_space *mapping;
3927 	struct ocfs2_inode_info *oi;
3928 
3929        	inode = ocfs2_lock_res_inode(lockres);
3930 	mapping = inode->i_mapping;
3931 
3932 	if (S_ISDIR(inode->i_mode)) {
3933 		oi = OCFS2_I(inode);
3934 		oi->ip_dir_lock_gen++;
3935 		mlog(0, "generation: %u\n", oi->ip_dir_lock_gen);
3936 		goto out;
3937 	}
3938 
3939 	if (!S_ISREG(inode->i_mode))
3940 		goto out;
3941 
3942 	/*
3943 	 * We need this before the filemap_fdatawrite() so that it can
3944 	 * transfer the dirty bit from the PTE to the
3945 	 * page. Unfortunately this means that even for EX->PR
3946 	 * downconverts, we'll lose our mappings and have to build
3947 	 * them up again.
3948 	 */
3949 	unmap_mapping_range(mapping, 0, 0, 0);
3950 
3951 	if (filemap_fdatawrite(mapping)) {
3952 		mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
3953 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
3954 	}
3955 	sync_mapping_buffers(mapping);
3956 	if (blocking == DLM_LOCK_EX) {
3957 		truncate_inode_pages(mapping, 0);
3958 	} else {
3959 		/* We only need to wait on the I/O if we're not also
3960 		 * truncating pages because truncate_inode_pages waits
3961 		 * for us above. We don't truncate pages if we're
3962 		 * blocking anything < EXMODE because we want to keep
3963 		 * them around in that case. */
3964 		filemap_fdatawait(mapping);
3965 	}
3966 
3967 	forget_all_cached_acls(inode);
3968 
3969 out:
3970 	return UNBLOCK_CONTINUE;
3971 }
3972 
3973 static int ocfs2_ci_checkpointed(struct ocfs2_caching_info *ci,
3974 				 struct ocfs2_lock_res *lockres,
3975 				 int new_level)
3976 {
3977 	int checkpointed = ocfs2_ci_fully_checkpointed(ci);
3978 
3979 	BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
3980 	BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
3981 
3982 	if (checkpointed)
3983 		return 1;
3984 
3985 	ocfs2_start_checkpoint(OCFS2_SB(ocfs2_metadata_cache_get_super(ci)));
3986 	return 0;
3987 }
3988 
3989 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
3990 					int new_level)
3991 {
3992 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3993 
3994 	return ocfs2_ci_checkpointed(INODE_CACHE(inode), lockres, new_level);
3995 }
3996 
3997 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
3998 {
3999 	struct inode *inode = ocfs2_lock_res_inode(lockres);
4000 
4001 	__ocfs2_stuff_meta_lvb(inode);
4002 }
4003 
4004 /*
4005  * Does the final reference drop on our dentry lock. Right now this
4006  * happens in the downconvert thread, but we could choose to simplify the
4007  * dlmglue API and push these off to the ocfs2_wq in the future.
4008  */
4009 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
4010 				     struct ocfs2_lock_res *lockres)
4011 {
4012 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
4013 	ocfs2_dentry_lock_put(osb, dl);
4014 }
4015 
4016 /*
4017  * d_delete() matching dentries before the lock downconvert.
4018  *
4019  * At this point, any process waiting to destroy the
4020  * dentry_lock due to last ref count is stopped by the
4021  * OCFS2_LOCK_QUEUED flag.
4022  *
4023  * We have two potential problems
4024  *
4025  * 1) If we do the last reference drop on our dentry_lock (via dput)
4026  *    we'll wind up in ocfs2_release_dentry_lock(), waiting on
4027  *    the downconvert to finish. Instead we take an elevated
4028  *    reference and push the drop until after we've completed our
4029  *    unblock processing.
4030  *
4031  * 2) There might be another process with a final reference,
4032  *    waiting on us to finish processing. If this is the case, we
4033  *    detect it and exit out - there's no more dentries anyway.
4034  */
4035 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
4036 				       int blocking)
4037 {
4038 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
4039 	struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
4040 	struct dentry *dentry;
4041 	unsigned long flags;
4042 	int extra_ref = 0;
4043 
4044 	/*
4045 	 * This node is blocking another node from getting a read
4046 	 * lock. This happens when we've renamed within a
4047 	 * directory. We've forced the other nodes to d_delete(), but
4048 	 * we never actually dropped our lock because it's still
4049 	 * valid. The downconvert code will retain a PR for this node,
4050 	 * so there's no further work to do.
4051 	 */
4052 	if (blocking == DLM_LOCK_PR)
4053 		return UNBLOCK_CONTINUE;
4054 
4055 	/*
4056 	 * Mark this inode as potentially orphaned. The code in
4057 	 * ocfs2_delete_inode() will figure out whether it actually
4058 	 * needs to be freed or not.
4059 	 */
4060 	spin_lock(&oi->ip_lock);
4061 	oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
4062 	spin_unlock(&oi->ip_lock);
4063 
4064 	/*
4065 	 * Yuck. We need to make sure however that the check of
4066 	 * OCFS2_LOCK_FREEING and the extra reference are atomic with
4067 	 * respect to a reference decrement or the setting of that
4068 	 * flag.
4069 	 */
4070 	spin_lock_irqsave(&lockres->l_lock, flags);
4071 	spin_lock(&dentry_attach_lock);
4072 	if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
4073 	    && dl->dl_count) {
4074 		dl->dl_count++;
4075 		extra_ref = 1;
4076 	}
4077 	spin_unlock(&dentry_attach_lock);
4078 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4079 
4080 	mlog(0, "extra_ref = %d\n", extra_ref);
4081 
4082 	/*
4083 	 * We have a process waiting on us in ocfs2_dentry_iput(),
4084 	 * which means we can't have any more outstanding
4085 	 * aliases. There's no need to do any more work.
4086 	 */
4087 	if (!extra_ref)
4088 		return UNBLOCK_CONTINUE;
4089 
4090 	spin_lock(&dentry_attach_lock);
4091 	while (1) {
4092 		dentry = ocfs2_find_local_alias(dl->dl_inode,
4093 						dl->dl_parent_blkno, 1);
4094 		if (!dentry)
4095 			break;
4096 		spin_unlock(&dentry_attach_lock);
4097 
4098 		if (S_ISDIR(dl->dl_inode->i_mode))
4099 			shrink_dcache_parent(dentry);
4100 
4101 		mlog(0, "d_delete(%pd);\n", dentry);
4102 
4103 		/*
4104 		 * The following dcache calls may do an
4105 		 * iput(). Normally we don't want that from the
4106 		 * downconverting thread, but in this case it's ok
4107 		 * because the requesting node already has an
4108 		 * exclusive lock on the inode, so it can't be queued
4109 		 * for a downconvert.
4110 		 */
4111 		d_delete(dentry);
4112 		dput(dentry);
4113 
4114 		spin_lock(&dentry_attach_lock);
4115 	}
4116 	spin_unlock(&dentry_attach_lock);
4117 
4118 	/*
4119 	 * If we are the last holder of this dentry lock, there is no
4120 	 * reason to downconvert so skip straight to the unlock.
4121 	 */
4122 	if (dl->dl_count == 1)
4123 		return UNBLOCK_STOP_POST;
4124 
4125 	return UNBLOCK_CONTINUE_POST;
4126 }
4127 
4128 static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
4129 					    int new_level)
4130 {
4131 	struct ocfs2_refcount_tree *tree =
4132 				ocfs2_lock_res_refcount_tree(lockres);
4133 
4134 	return ocfs2_ci_checkpointed(&tree->rf_ci, lockres, new_level);
4135 }
4136 
4137 static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
4138 					 int blocking)
4139 {
4140 	struct ocfs2_refcount_tree *tree =
4141 				ocfs2_lock_res_refcount_tree(lockres);
4142 
4143 	ocfs2_metadata_cache_purge(&tree->rf_ci);
4144 
4145 	return UNBLOCK_CONTINUE;
4146 }
4147 
4148 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
4149 {
4150 	struct ocfs2_qinfo_lvb *lvb;
4151 	struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres);
4152 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
4153 					    oinfo->dqi_gi.dqi_type);
4154 
4155 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
4156 	lvb->lvb_version = OCFS2_QINFO_LVB_VERSION;
4157 	lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace);
4158 	lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace);
4159 	lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms);
4160 	lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks);
4161 	lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk);
4162 	lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry);
4163 }
4164 
4165 void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex)
4166 {
4167 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4168 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
4169 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4170 
4171 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
4172 		ocfs2_cluster_unlock(osb, lockres, level);
4173 }
4174 
4175 static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo)
4176 {
4177 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
4178 					    oinfo->dqi_gi.dqi_type);
4179 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4180 	struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
4181 	struct buffer_head *bh = NULL;
4182 	struct ocfs2_global_disk_dqinfo *gdinfo;
4183 	int status = 0;
4184 
4185 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
4186 	    lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) {
4187 		info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace);
4188 		info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace);
4189 		oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms);
4190 		oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks);
4191 		oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk);
4192 		oinfo->dqi_gi.dqi_free_entry =
4193 					be32_to_cpu(lvb->lvb_free_entry);
4194 	} else {
4195 		status = ocfs2_read_quota_phys_block(oinfo->dqi_gqinode,
4196 						     oinfo->dqi_giblk, &bh);
4197 		if (status) {
4198 			mlog_errno(status);
4199 			goto bail;
4200 		}
4201 		gdinfo = (struct ocfs2_global_disk_dqinfo *)
4202 					(bh->b_data + OCFS2_GLOBAL_INFO_OFF);
4203 		info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace);
4204 		info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace);
4205 		oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms);
4206 		oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks);
4207 		oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk);
4208 		oinfo->dqi_gi.dqi_free_entry =
4209 					le32_to_cpu(gdinfo->dqi_free_entry);
4210 		brelse(bh);
4211 		ocfs2_track_lock_refresh(lockres);
4212 	}
4213 
4214 bail:
4215 	return status;
4216 }
4217 
4218 /* Lock quota info, this function expects at least shared lock on the quota file
4219  * so that we can safely refresh quota info from disk. */
4220 int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex)
4221 {
4222 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4223 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
4224 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4225 	int status = 0;
4226 
4227 	/* On RO devices, locking really isn't needed... */
4228 	if (ocfs2_is_hard_readonly(osb)) {
4229 		if (ex)
4230 			status = -EROFS;
4231 		goto bail;
4232 	}
4233 	if (ocfs2_mount_local(osb))
4234 		goto bail;
4235 
4236 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
4237 	if (status < 0) {
4238 		mlog_errno(status);
4239 		goto bail;
4240 	}
4241 	if (!ocfs2_should_refresh_lock_res(lockres))
4242 		goto bail;
4243 	/* OK, we have the lock but we need to refresh the quota info */
4244 	status = ocfs2_refresh_qinfo(oinfo);
4245 	if (status)
4246 		ocfs2_qinfo_unlock(oinfo, ex);
4247 	ocfs2_complete_lock_res_refresh(lockres, status);
4248 bail:
4249 	return status;
4250 }
4251 
4252 int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex)
4253 {
4254 	int status;
4255 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4256 	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
4257 	struct ocfs2_super *osb = lockres->l_priv;
4258 
4259 
4260 	if (ocfs2_is_hard_readonly(osb))
4261 		return -EROFS;
4262 
4263 	if (ocfs2_mount_local(osb))
4264 		return 0;
4265 
4266 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
4267 	if (status < 0)
4268 		mlog_errno(status);
4269 
4270 	return status;
4271 }
4272 
4273 void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex)
4274 {
4275 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4276 	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
4277 	struct ocfs2_super *osb = lockres->l_priv;
4278 
4279 	if (!ocfs2_mount_local(osb))
4280 		ocfs2_cluster_unlock(osb, lockres, level);
4281 }
4282 
4283 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
4284 				       struct ocfs2_lock_res *lockres)
4285 {
4286 	int status;
4287 	struct ocfs2_unblock_ctl ctl = {0, 0,};
4288 	unsigned long flags;
4289 
4290 	/* Our reference to the lockres in this function can be
4291 	 * considered valid until we remove the OCFS2_LOCK_QUEUED
4292 	 * flag. */
4293 
4294 	BUG_ON(!lockres);
4295 	BUG_ON(!lockres->l_ops);
4296 
4297 	mlog(ML_BASTS, "lockres %s blocked\n", lockres->l_name);
4298 
4299 	/* Detect whether a lock has been marked as going away while
4300 	 * the downconvert thread was processing other things. A lock can
4301 	 * still be marked with OCFS2_LOCK_FREEING after this check,
4302 	 * but short circuiting here will still save us some
4303 	 * performance. */
4304 	spin_lock_irqsave(&lockres->l_lock, flags);
4305 	if (lockres->l_flags & OCFS2_LOCK_FREEING)
4306 		goto unqueue;
4307 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4308 
4309 	status = ocfs2_unblock_lock(osb, lockres, &ctl);
4310 	if (status < 0)
4311 		mlog_errno(status);
4312 
4313 	spin_lock_irqsave(&lockres->l_lock, flags);
4314 unqueue:
4315 	if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
4316 		lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
4317 	} else
4318 		ocfs2_schedule_blocked_lock(osb, lockres);
4319 
4320 	mlog(ML_BASTS, "lockres %s, requeue = %s.\n", lockres->l_name,
4321 	     ctl.requeue ? "yes" : "no");
4322 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4323 
4324 	if (ctl.unblock_action != UNBLOCK_CONTINUE
4325 	    && lockres->l_ops->post_unlock)
4326 		lockres->l_ops->post_unlock(osb, lockres);
4327 }
4328 
4329 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
4330 					struct ocfs2_lock_res *lockres)
4331 {
4332 	unsigned long flags;
4333 
4334 	assert_spin_locked(&lockres->l_lock);
4335 
4336 	if (lockres->l_flags & OCFS2_LOCK_FREEING) {
4337 		/* Do not schedule a lock for downconvert when it's on
4338 		 * the way to destruction - any nodes wanting access
4339 		 * to the resource will get it soon. */
4340 		mlog(ML_BASTS, "lockres %s won't be scheduled: flags 0x%lx\n",
4341 		     lockres->l_name, lockres->l_flags);
4342 		return;
4343 	}
4344 
4345 	lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
4346 
4347 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4348 	if (list_empty(&lockres->l_blocked_list)) {
4349 		list_add_tail(&lockres->l_blocked_list,
4350 			      &osb->blocked_lock_list);
4351 		osb->blocked_lock_count++;
4352 	}
4353 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4354 }
4355 
4356 static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
4357 {
4358 	unsigned long processed;
4359 	unsigned long flags;
4360 	struct ocfs2_lock_res *lockres;
4361 
4362 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4363 	/* grab this early so we know to try again if a state change and
4364 	 * wake happens part-way through our work  */
4365 	osb->dc_work_sequence = osb->dc_wake_sequence;
4366 
4367 	processed = osb->blocked_lock_count;
4368 	/*
4369 	 * blocked lock processing in this loop might call iput which can
4370 	 * remove items off osb->blocked_lock_list. Downconvert up to
4371 	 * 'processed' number of locks, but stop short if we had some
4372 	 * removed in ocfs2_mark_lockres_freeing when downconverting.
4373 	 */
4374 	while (processed && !list_empty(&osb->blocked_lock_list)) {
4375 		lockres = list_entry(osb->blocked_lock_list.next,
4376 				     struct ocfs2_lock_res, l_blocked_list);
4377 		list_del_init(&lockres->l_blocked_list);
4378 		osb->blocked_lock_count--;
4379 		spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4380 
4381 		BUG_ON(!processed);
4382 		processed--;
4383 
4384 		ocfs2_process_blocked_lock(osb, lockres);
4385 
4386 		spin_lock_irqsave(&osb->dc_task_lock, flags);
4387 	}
4388 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4389 }
4390 
4391 static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
4392 {
4393 	int empty = 0;
4394 	unsigned long flags;
4395 
4396 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4397 	if (list_empty(&osb->blocked_lock_list))
4398 		empty = 1;
4399 
4400 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4401 	return empty;
4402 }
4403 
4404 static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
4405 {
4406 	int should_wake = 0;
4407 	unsigned long flags;
4408 
4409 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4410 	if (osb->dc_work_sequence != osb->dc_wake_sequence)
4411 		should_wake = 1;
4412 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4413 
4414 	return should_wake;
4415 }
4416 
4417 static int ocfs2_downconvert_thread(void *arg)
4418 {
4419 	struct ocfs2_super *osb = arg;
4420 
4421 	/* only quit once we've been asked to stop and there is no more
4422 	 * work available */
4423 	while (!(kthread_should_stop() &&
4424 		ocfs2_downconvert_thread_lists_empty(osb))) {
4425 
4426 		wait_event_interruptible(osb->dc_event,
4427 					 ocfs2_downconvert_thread_should_wake(osb) ||
4428 					 kthread_should_stop());
4429 
4430 		mlog(0, "downconvert_thread: awoken\n");
4431 
4432 		ocfs2_downconvert_thread_do_work(osb);
4433 	}
4434 
4435 	osb->dc_task = NULL;
4436 	return 0;
4437 }
4438 
4439 void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
4440 {
4441 	unsigned long flags;
4442 
4443 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4444 	/* make sure the voting thread gets a swipe at whatever changes
4445 	 * the caller may have made to the voting state */
4446 	osb->dc_wake_sequence++;
4447 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4448 	wake_up(&osb->dc_event);
4449 }
4450