1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * dlmglue.c
5  *
6  * Code which implements an OCFS2 specific interface to our DLM.
7  *
8  * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation; either
13  * version 2 of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public
21  * License along with this program; if not, write to the
22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23  * Boston, MA 021110-1307, USA.
24  */
25 
26 #include <linux/types.h>
27 #include <linux/slab.h>
28 #include <linux/highmem.h>
29 #include <linux/mm.h>
30 #include <linux/kthread.h>
31 #include <linux/pagemap.h>
32 #include <linux/debugfs.h>
33 #include <linux/seq_file.h>
34 #include <linux/time.h>
35 #include <linux/quotaops.h>
36 
37 #define MLOG_MASK_PREFIX ML_DLM_GLUE
38 #include <cluster/masklog.h>
39 
40 #include "ocfs2.h"
41 #include "ocfs2_lockingver.h"
42 
43 #include "alloc.h"
44 #include "dcache.h"
45 #include "dlmglue.h"
46 #include "extent_map.h"
47 #include "file.h"
48 #include "heartbeat.h"
49 #include "inode.h"
50 #include "journal.h"
51 #include "stackglue.h"
52 #include "slot_map.h"
53 #include "super.h"
54 #include "uptodate.h"
55 #include "quota.h"
56 #include "refcounttree.h"
57 
58 #include "buffer_head_io.h"
59 
60 struct ocfs2_mask_waiter {
61 	struct list_head	mw_item;
62 	int			mw_status;
63 	struct completion	mw_complete;
64 	unsigned long		mw_mask;
65 	unsigned long		mw_goal;
66 #ifdef CONFIG_OCFS2_FS_STATS
67 	ktime_t			mw_lock_start;
68 #endif
69 };
70 
71 static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres);
72 static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres);
73 static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres);
74 static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres);
75 
76 /*
77  * Return value from ->downconvert_worker functions.
78  *
79  * These control the precise actions of ocfs2_unblock_lock()
80  * and ocfs2_process_blocked_lock()
81  *
82  */
83 enum ocfs2_unblock_action {
84 	UNBLOCK_CONTINUE	= 0, /* Continue downconvert */
85 	UNBLOCK_CONTINUE_POST	= 1, /* Continue downconvert, fire
86 				      * ->post_unlock callback */
87 	UNBLOCK_STOP_POST	= 2, /* Do not downconvert, fire
88 				      * ->post_unlock() callback. */
89 };
90 
91 struct ocfs2_unblock_ctl {
92 	int requeue;
93 	enum ocfs2_unblock_action unblock_action;
94 };
95 
96 /* Lockdep class keys */
97 struct lock_class_key lockdep_keys[OCFS2_NUM_LOCK_TYPES];
98 
99 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
100 					int new_level);
101 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);
102 
103 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
104 				     int blocking);
105 
106 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
107 				       int blocking);
108 
109 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
110 				     struct ocfs2_lock_res *lockres);
111 
112 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres);
113 
114 static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
115 					    int new_level);
116 static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
117 					 int blocking);
118 
119 #define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
120 
121 /* This aids in debugging situations where a bad LVB might be involved. */
ocfs2_dump_meta_lvb_info(u64 level,const char * function,unsigned int line,struct ocfs2_lock_res * lockres)122 static void ocfs2_dump_meta_lvb_info(u64 level,
123 				     const char *function,
124 				     unsigned int line,
125 				     struct ocfs2_lock_res *lockres)
126 {
127 	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
128 
129 	mlog(level, "LVB information for %s (called from %s:%u):\n",
130 	     lockres->l_name, function, line);
131 	mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
132 	     lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
133 	     be32_to_cpu(lvb->lvb_igeneration));
134 	mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
135 	     (unsigned long long)be64_to_cpu(lvb->lvb_isize),
136 	     be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
137 	     be16_to_cpu(lvb->lvb_imode));
138 	mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
139 	     "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
140 	     (long long)be64_to_cpu(lvb->lvb_iatime_packed),
141 	     (long long)be64_to_cpu(lvb->lvb_ictime_packed),
142 	     (long long)be64_to_cpu(lvb->lvb_imtime_packed),
143 	     be32_to_cpu(lvb->lvb_iattr));
144 }
145 
146 
147 /*
148  * OCFS2 Lock Resource Operations
149  *
150  * These fine tune the behavior of the generic dlmglue locking infrastructure.
151  *
152  * The most basic of lock types can point ->l_priv to their respective
153  * struct ocfs2_super and allow the default actions to manage things.
154  *
155  * Right now, each lock type also needs to implement an init function,
156  * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
157  * should be called when the lock is no longer needed (i.e., object
158  * destruction time).
159  */
160 struct ocfs2_lock_res_ops {
161 	/*
162 	 * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
163 	 * this callback if ->l_priv is not an ocfs2_super pointer
164 	 */
165 	struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
166 
167 	/*
168 	 * Optionally called in the downconvert thread after a
169 	 * successful downconvert. The lockres will not be referenced
170 	 * after this callback is called, so it is safe to free
171 	 * memory, etc.
172 	 *
173 	 * The exact semantics of when this is called are controlled
174 	 * by ->downconvert_worker()
175 	 */
176 	void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
177 
178 	/*
179 	 * Allow a lock type to add checks to determine whether it is
180 	 * safe to downconvert a lock. Return 0 to re-queue the
181 	 * downconvert at a later time, nonzero to continue.
182 	 *
183 	 * For most locks, the default checks that there are no
184 	 * incompatible holders are sufficient.
185 	 *
186 	 * Called with the lockres spinlock held.
187 	 */
188 	int (*check_downconvert)(struct ocfs2_lock_res *, int);
189 
190 	/*
191 	 * Allows a lock type to populate the lock value block. This
192 	 * is called on downconvert, and when we drop a lock.
193 	 *
194 	 * Locks that want to use this should set LOCK_TYPE_USES_LVB
195 	 * in the flags field.
196 	 *
197 	 * Called with the lockres spinlock held.
198 	 */
199 	void (*set_lvb)(struct ocfs2_lock_res *);
200 
201 	/*
202 	 * Called from the downconvert thread when it is determined
203 	 * that a lock will be downconverted. This is called without
204 	 * any locks held so the function can do work that might
205 	 * schedule (syncing out data, etc).
206 	 *
207 	 * This should return any one of the ocfs2_unblock_action
208 	 * values, depending on what it wants the thread to do.
209 	 */
210 	int (*downconvert_worker)(struct ocfs2_lock_res *, int);
211 
212 	/*
213 	 * LOCK_TYPE_* flags which describe the specific requirements
214 	 * of a lock type. Descriptions of each individual flag follow.
215 	 */
216 	int flags;
217 };
218 
219 /*
220  * Some locks want to "refresh" potentially stale data when a
221  * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
222  * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
223  * individual lockres l_flags member from the ast function. It is
224  * expected that the locking wrapper will clear the
225  * OCFS2_LOCK_NEEDS_REFRESH flag when done.
226  */
227 #define LOCK_TYPE_REQUIRES_REFRESH 0x1
228 
229 /*
230  * Indicate that a lock type makes use of the lock value block. The
231  * ->set_lvb lock type callback must be defined.
232  */
233 #define LOCK_TYPE_USES_LVB		0x2
234 
235 static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
236 	.get_osb	= ocfs2_get_inode_osb,
237 	.flags		= 0,
238 };
239 
240 static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = {
241 	.get_osb	= ocfs2_get_inode_osb,
242 	.check_downconvert = ocfs2_check_meta_downconvert,
243 	.set_lvb	= ocfs2_set_meta_lvb,
244 	.downconvert_worker = ocfs2_data_convert_worker,
245 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
246 };
247 
248 static struct ocfs2_lock_res_ops ocfs2_super_lops = {
249 	.flags		= LOCK_TYPE_REQUIRES_REFRESH,
250 };
251 
252 static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
253 	.flags		= 0,
254 };
255 
256 static struct ocfs2_lock_res_ops ocfs2_nfs_sync_lops = {
257 	.flags		= 0,
258 };
259 
260 static struct ocfs2_lock_res_ops ocfs2_orphan_scan_lops = {
261 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
262 };
263 
264 static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
265 	.get_osb	= ocfs2_get_dentry_osb,
266 	.post_unlock	= ocfs2_dentry_post_unlock,
267 	.downconvert_worker = ocfs2_dentry_convert_worker,
268 	.flags		= 0,
269 };
270 
271 static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
272 	.get_osb	= ocfs2_get_inode_osb,
273 	.flags		= 0,
274 };
275 
276 static struct ocfs2_lock_res_ops ocfs2_flock_lops = {
277 	.get_osb	= ocfs2_get_file_osb,
278 	.flags		= 0,
279 };
280 
281 static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = {
282 	.set_lvb	= ocfs2_set_qinfo_lvb,
283 	.get_osb	= ocfs2_get_qinfo_osb,
284 	.flags		= LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB,
285 };
286 
287 static struct ocfs2_lock_res_ops ocfs2_refcount_block_lops = {
288 	.check_downconvert = ocfs2_check_refcount_downconvert,
289 	.downconvert_worker = ocfs2_refcount_convert_worker,
290 	.flags		= 0,
291 };
292 
ocfs2_is_inode_lock(struct ocfs2_lock_res * lockres)293 static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
294 {
295 	return lockres->l_type == OCFS2_LOCK_TYPE_META ||
296 		lockres->l_type == OCFS2_LOCK_TYPE_RW ||
297 		lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
298 }
299 
ocfs2_lksb_to_lock_res(struct ocfs2_dlm_lksb * lksb)300 static inline struct ocfs2_lock_res *ocfs2_lksb_to_lock_res(struct ocfs2_dlm_lksb *lksb)
301 {
302 	return container_of(lksb, struct ocfs2_lock_res, l_lksb);
303 }
304 
ocfs2_lock_res_inode(struct ocfs2_lock_res * lockres)305 static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
306 {
307 	BUG_ON(!ocfs2_is_inode_lock(lockres));
308 
309 	return (struct inode *) lockres->l_priv;
310 }
311 
ocfs2_lock_res_dl(struct ocfs2_lock_res * lockres)312 static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
313 {
314 	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
315 
316 	return (struct ocfs2_dentry_lock *)lockres->l_priv;
317 }
318 
ocfs2_lock_res_qinfo(struct ocfs2_lock_res * lockres)319 static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres)
320 {
321 	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO);
322 
323 	return (struct ocfs2_mem_dqinfo *)lockres->l_priv;
324 }
325 
326 static inline struct ocfs2_refcount_tree *
ocfs2_lock_res_refcount_tree(struct ocfs2_lock_res * res)327 ocfs2_lock_res_refcount_tree(struct ocfs2_lock_res *res)
328 {
329 	return container_of(res, struct ocfs2_refcount_tree, rf_lockres);
330 }
331 
ocfs2_get_lockres_osb(struct ocfs2_lock_res * lockres)332 static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
333 {
334 	if (lockres->l_ops->get_osb)
335 		return lockres->l_ops->get_osb(lockres);
336 
337 	return (struct ocfs2_super *)lockres->l_priv;
338 }
339 
340 static int ocfs2_lock_create(struct ocfs2_super *osb,
341 			     struct ocfs2_lock_res *lockres,
342 			     int level,
343 			     u32 dlm_flags);
344 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
345 						     int wanted);
346 static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
347 				   struct ocfs2_lock_res *lockres,
348 				   int level, unsigned long caller_ip);
ocfs2_cluster_unlock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int level)349 static inline void ocfs2_cluster_unlock(struct ocfs2_super *osb,
350 					struct ocfs2_lock_res *lockres,
351 					int level)
352 {
353 	__ocfs2_cluster_unlock(osb, lockres, level, _RET_IP_);
354 }
355 
356 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
357 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
358 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
359 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
360 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
361 					struct ocfs2_lock_res *lockres);
362 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
363 						int convert);
364 #define ocfs2_log_dlm_error(_func, _err, _lockres) do {					\
365 	if ((_lockres)->l_type != OCFS2_LOCK_TYPE_DENTRY)				\
366 		mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n",	\
367 		     _err, _func, _lockres->l_name);					\
368 	else										\
369 		mlog(ML_ERROR, "DLM error %d while calling %s on resource %.*s%08x\n",	\
370 		     _err, _func, OCFS2_DENTRY_LOCK_INO_START - 1, (_lockres)->l_name,	\
371 		     (unsigned int)ocfs2_get_dentry_lock_ino(_lockres));		\
372 } while (0)
373 static int ocfs2_downconvert_thread(void *arg);
374 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
375 					struct ocfs2_lock_res *lockres);
376 static int ocfs2_inode_lock_update(struct inode *inode,
377 				  struct buffer_head **bh);
378 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
379 static inline int ocfs2_highest_compat_lock_level(int level);
380 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
381 					      int new_level);
382 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
383 				  struct ocfs2_lock_res *lockres,
384 				  int new_level,
385 				  int lvb,
386 				  unsigned int generation);
387 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
388 				        struct ocfs2_lock_res *lockres);
389 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
390 				struct ocfs2_lock_res *lockres);
391 
392 
ocfs2_build_lock_name(enum ocfs2_lock_type type,u64 blkno,u32 generation,char * name)393 static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
394 				  u64 blkno,
395 				  u32 generation,
396 				  char *name)
397 {
398 	int len;
399 
400 	BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
401 
402 	len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
403 		       ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
404 		       (long long)blkno, generation);
405 
406 	BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
407 
408 	mlog(0, "built lock resource with name: %s\n", name);
409 }
410 
411 static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
412 
ocfs2_add_lockres_tracking(struct ocfs2_lock_res * res,struct ocfs2_dlm_debug * dlm_debug)413 static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
414 				       struct ocfs2_dlm_debug *dlm_debug)
415 {
416 	mlog(0, "Add tracking for lockres %s\n", res->l_name);
417 
418 	spin_lock(&ocfs2_dlm_tracking_lock);
419 	list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
420 	spin_unlock(&ocfs2_dlm_tracking_lock);
421 }
422 
ocfs2_remove_lockres_tracking(struct ocfs2_lock_res * res)423 static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
424 {
425 	spin_lock(&ocfs2_dlm_tracking_lock);
426 	if (!list_empty(&res->l_debug_list))
427 		list_del_init(&res->l_debug_list);
428 	spin_unlock(&ocfs2_dlm_tracking_lock);
429 }
430 
431 #ifdef CONFIG_OCFS2_FS_STATS
ocfs2_init_lock_stats(struct ocfs2_lock_res * res)432 static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
433 {
434 	res->l_lock_refresh = 0;
435 	memset(&res->l_lock_prmode, 0, sizeof(struct ocfs2_lock_stats));
436 	memset(&res->l_lock_exmode, 0, sizeof(struct ocfs2_lock_stats));
437 }
438 
ocfs2_update_lock_stats(struct ocfs2_lock_res * res,int level,struct ocfs2_mask_waiter * mw,int ret)439 static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level,
440 				    struct ocfs2_mask_waiter *mw, int ret)
441 {
442 	u32 usec;
443 	ktime_t kt;
444 	struct ocfs2_lock_stats *stats;
445 
446 	if (level == LKM_PRMODE)
447 		stats = &res->l_lock_prmode;
448 	else if (level == LKM_EXMODE)
449 		stats = &res->l_lock_exmode;
450 	else
451 		return;
452 
453 	kt = ktime_sub(ktime_get(), mw->mw_lock_start);
454 	usec = ktime_to_us(kt);
455 
456 	stats->ls_gets++;
457 	stats->ls_total += ktime_to_ns(kt);
458 	/* overflow */
459 	if (unlikely(stats->ls_gets == 0)) {
460 		stats->ls_gets++;
461 		stats->ls_total = ktime_to_ns(kt);
462 	}
463 
464 	if (stats->ls_max < usec)
465 		stats->ls_max = usec;
466 
467 	if (ret)
468 		stats->ls_fail++;
469 }
470 
ocfs2_track_lock_refresh(struct ocfs2_lock_res * lockres)471 static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
472 {
473 	lockres->l_lock_refresh++;
474 }
475 
ocfs2_init_start_time(struct ocfs2_mask_waiter * mw)476 static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
477 {
478 	mw->mw_lock_start = ktime_get();
479 }
480 #else
ocfs2_init_lock_stats(struct ocfs2_lock_res * res)481 static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
482 {
483 }
ocfs2_update_lock_stats(struct ocfs2_lock_res * res,int level,struct ocfs2_mask_waiter * mw,int ret)484 static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res,
485 			   int level, struct ocfs2_mask_waiter *mw, int ret)
486 {
487 }
ocfs2_track_lock_refresh(struct ocfs2_lock_res * lockres)488 static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
489 {
490 }
ocfs2_init_start_time(struct ocfs2_mask_waiter * mw)491 static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
492 {
493 }
494 #endif
495 
ocfs2_lock_res_init_common(struct ocfs2_super * osb,struct ocfs2_lock_res * res,enum ocfs2_lock_type type,struct ocfs2_lock_res_ops * ops,void * priv)496 static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
497 				       struct ocfs2_lock_res *res,
498 				       enum ocfs2_lock_type type,
499 				       struct ocfs2_lock_res_ops *ops,
500 				       void *priv)
501 {
502 	res->l_type          = type;
503 	res->l_ops           = ops;
504 	res->l_priv          = priv;
505 
506 	res->l_level         = DLM_LOCK_IV;
507 	res->l_requested     = DLM_LOCK_IV;
508 	res->l_blocking      = DLM_LOCK_IV;
509 	res->l_action        = OCFS2_AST_INVALID;
510 	res->l_unlock_action = OCFS2_UNLOCK_INVALID;
511 
512 	res->l_flags         = OCFS2_LOCK_INITIALIZED;
513 
514 	ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
515 
516 	ocfs2_init_lock_stats(res);
517 #ifdef CONFIG_DEBUG_LOCK_ALLOC
518 	if (type != OCFS2_LOCK_TYPE_OPEN)
519 		lockdep_init_map(&res->l_lockdep_map, ocfs2_lock_type_strings[type],
520 				 &lockdep_keys[type], 0);
521 	else
522 		res->l_lockdep_map.key = NULL;
523 #endif
524 }
525 
ocfs2_lock_res_init_once(struct ocfs2_lock_res * res)526 void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
527 {
528 	/* This also clears out the lock status block */
529 	memset(res, 0, sizeof(struct ocfs2_lock_res));
530 	spin_lock_init(&res->l_lock);
531 	init_waitqueue_head(&res->l_event);
532 	INIT_LIST_HEAD(&res->l_blocked_list);
533 	INIT_LIST_HEAD(&res->l_mask_waiters);
534 }
535 
ocfs2_inode_lock_res_init(struct ocfs2_lock_res * res,enum ocfs2_lock_type type,unsigned int generation,struct inode * inode)536 void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
537 			       enum ocfs2_lock_type type,
538 			       unsigned int generation,
539 			       struct inode *inode)
540 {
541 	struct ocfs2_lock_res_ops *ops;
542 
543 	switch(type) {
544 		case OCFS2_LOCK_TYPE_RW:
545 			ops = &ocfs2_inode_rw_lops;
546 			break;
547 		case OCFS2_LOCK_TYPE_META:
548 			ops = &ocfs2_inode_inode_lops;
549 			break;
550 		case OCFS2_LOCK_TYPE_OPEN:
551 			ops = &ocfs2_inode_open_lops;
552 			break;
553 		default:
554 			mlog_bug_on_msg(1, "type: %d\n", type);
555 			ops = NULL; /* thanks, gcc */
556 			break;
557 	};
558 
559 	ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
560 			      generation, res->l_name);
561 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
562 }
563 
ocfs2_get_inode_osb(struct ocfs2_lock_res * lockres)564 static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
565 {
566 	struct inode *inode = ocfs2_lock_res_inode(lockres);
567 
568 	return OCFS2_SB(inode->i_sb);
569 }
570 
ocfs2_get_qinfo_osb(struct ocfs2_lock_res * lockres)571 static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres)
572 {
573 	struct ocfs2_mem_dqinfo *info = lockres->l_priv;
574 
575 	return OCFS2_SB(info->dqi_gi.dqi_sb);
576 }
577 
ocfs2_get_file_osb(struct ocfs2_lock_res * lockres)578 static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres)
579 {
580 	struct ocfs2_file_private *fp = lockres->l_priv;
581 
582 	return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb);
583 }
584 
ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res * lockres)585 static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
586 {
587 	__be64 inode_blkno_be;
588 
589 	memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
590 	       sizeof(__be64));
591 
592 	return be64_to_cpu(inode_blkno_be);
593 }
594 
ocfs2_get_dentry_osb(struct ocfs2_lock_res * lockres)595 static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
596 {
597 	struct ocfs2_dentry_lock *dl = lockres->l_priv;
598 
599 	return OCFS2_SB(dl->dl_inode->i_sb);
600 }
601 
ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock * dl,u64 parent,struct inode * inode)602 void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
603 				u64 parent, struct inode *inode)
604 {
605 	int len;
606 	u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
607 	__be64 inode_blkno_be = cpu_to_be64(inode_blkno);
608 	struct ocfs2_lock_res *lockres = &dl->dl_lockres;
609 
610 	ocfs2_lock_res_init_once(lockres);
611 
612 	/*
613 	 * Unfortunately, the standard lock naming scheme won't work
614 	 * here because we have two 16 byte values to use. Instead,
615 	 * we'll stuff the inode number as a binary value. We still
616 	 * want error prints to show something without garbling the
617 	 * display, so drop a null byte in there before the inode
618 	 * number. A future version of OCFS2 will likely use all
619 	 * binary lock names. The stringified names have been a
620 	 * tremendous aid in debugging, but now that the debugfs
621 	 * interface exists, we can mangle things there if need be.
622 	 *
623 	 * NOTE: We also drop the standard "pad" value (the total lock
624 	 * name size stays the same though - the last part is all
625 	 * zeros due to the memset in ocfs2_lock_res_init_once()
626 	 */
627 	len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
628 		       "%c%016llx",
629 		       ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
630 		       (long long)parent);
631 
632 	BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
633 
634 	memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
635 	       sizeof(__be64));
636 
637 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
638 				   OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
639 				   dl);
640 }
641 
ocfs2_super_lock_res_init(struct ocfs2_lock_res * res,struct ocfs2_super * osb)642 static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
643 				      struct ocfs2_super *osb)
644 {
645 	/* Superblock lockres doesn't come from a slab so we call init
646 	 * once on it manually.  */
647 	ocfs2_lock_res_init_once(res);
648 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
649 			      0, res->l_name);
650 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
651 				   &ocfs2_super_lops, osb);
652 }
653 
ocfs2_rename_lock_res_init(struct ocfs2_lock_res * res,struct ocfs2_super * osb)654 static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
655 				       struct ocfs2_super *osb)
656 {
657 	/* Rename lockres doesn't come from a slab so we call init
658 	 * once on it manually.  */
659 	ocfs2_lock_res_init_once(res);
660 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
661 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
662 				   &ocfs2_rename_lops, osb);
663 }
664 
ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res * res,struct ocfs2_super * osb)665 static void ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res *res,
666 					 struct ocfs2_super *osb)
667 {
668 	/* nfs_sync lockres doesn't come from a slab so we call init
669 	 * once on it manually.  */
670 	ocfs2_lock_res_init_once(res);
671 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_NFS_SYNC, 0, 0, res->l_name);
672 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_NFS_SYNC,
673 				   &ocfs2_nfs_sync_lops, osb);
674 }
675 
ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res * res,struct ocfs2_super * osb)676 static void ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res *res,
677 					    struct ocfs2_super *osb)
678 {
679 	ocfs2_lock_res_init_once(res);
680 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_ORPHAN_SCAN, 0, 0, res->l_name);
681 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_ORPHAN_SCAN,
682 				   &ocfs2_orphan_scan_lops, osb);
683 }
684 
ocfs2_file_lock_res_init(struct ocfs2_lock_res * lockres,struct ocfs2_file_private * fp)685 void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
686 			      struct ocfs2_file_private *fp)
687 {
688 	struct inode *inode = fp->fp_file->f_mapping->host;
689 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
690 
691 	ocfs2_lock_res_init_once(lockres);
692 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
693 			      inode->i_generation, lockres->l_name);
694 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
695 				   OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
696 				   fp);
697 	lockres->l_flags |= OCFS2_LOCK_NOCACHE;
698 }
699 
ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res * lockres,struct ocfs2_mem_dqinfo * info)700 void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
701 			       struct ocfs2_mem_dqinfo *info)
702 {
703 	ocfs2_lock_res_init_once(lockres);
704 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type,
705 			      0, lockres->l_name);
706 	ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres,
707 				   OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops,
708 				   info);
709 }
710 
ocfs2_refcount_lock_res_init(struct ocfs2_lock_res * lockres,struct ocfs2_super * osb,u64 ref_blkno,unsigned int generation)711 void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres,
712 				  struct ocfs2_super *osb, u64 ref_blkno,
713 				  unsigned int generation)
714 {
715 	ocfs2_lock_res_init_once(lockres);
716 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_REFCOUNT, ref_blkno,
717 			      generation, lockres->l_name);
718 	ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_REFCOUNT,
719 				   &ocfs2_refcount_block_lops, osb);
720 }
721 
ocfs2_lock_res_free(struct ocfs2_lock_res * res)722 void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
723 {
724 	if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
725 		return;
726 
727 	ocfs2_remove_lockres_tracking(res);
728 
729 	mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
730 			"Lockres %s is on the blocked list\n",
731 			res->l_name);
732 	mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
733 			"Lockres %s has mask waiters pending\n",
734 			res->l_name);
735 	mlog_bug_on_msg(spin_is_locked(&res->l_lock),
736 			"Lockres %s is locked\n",
737 			res->l_name);
738 	mlog_bug_on_msg(res->l_ro_holders,
739 			"Lockres %s has %u ro holders\n",
740 			res->l_name, res->l_ro_holders);
741 	mlog_bug_on_msg(res->l_ex_holders,
742 			"Lockres %s has %u ex holders\n",
743 			res->l_name, res->l_ex_holders);
744 
745 	/* Need to clear out the lock status block for the dlm */
746 	memset(&res->l_lksb, 0, sizeof(res->l_lksb));
747 
748 	res->l_flags = 0UL;
749 }
750 
ocfs2_inc_holders(struct ocfs2_lock_res * lockres,int level)751 static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
752 				     int level)
753 {
754 	BUG_ON(!lockres);
755 
756 	switch(level) {
757 	case DLM_LOCK_EX:
758 		lockres->l_ex_holders++;
759 		break;
760 	case DLM_LOCK_PR:
761 		lockres->l_ro_holders++;
762 		break;
763 	default:
764 		BUG();
765 	}
766 }
767 
ocfs2_dec_holders(struct ocfs2_lock_res * lockres,int level)768 static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
769 				     int level)
770 {
771 	BUG_ON(!lockres);
772 
773 	switch(level) {
774 	case DLM_LOCK_EX:
775 		BUG_ON(!lockres->l_ex_holders);
776 		lockres->l_ex_holders--;
777 		break;
778 	case DLM_LOCK_PR:
779 		BUG_ON(!lockres->l_ro_holders);
780 		lockres->l_ro_holders--;
781 		break;
782 	default:
783 		BUG();
784 	}
785 }
786 
787 /* WARNING: This function lives in a world where the only three lock
788  * levels are EX, PR, and NL. It *will* have to be adjusted when more
789  * lock types are added. */
ocfs2_highest_compat_lock_level(int level)790 static inline int ocfs2_highest_compat_lock_level(int level)
791 {
792 	int new_level = DLM_LOCK_EX;
793 
794 	if (level == DLM_LOCK_EX)
795 		new_level = DLM_LOCK_NL;
796 	else if (level == DLM_LOCK_PR)
797 		new_level = DLM_LOCK_PR;
798 	return new_level;
799 }
800 
lockres_set_flags(struct ocfs2_lock_res * lockres,unsigned long newflags)801 static void lockres_set_flags(struct ocfs2_lock_res *lockres,
802 			      unsigned long newflags)
803 {
804 	struct ocfs2_mask_waiter *mw, *tmp;
805 
806  	assert_spin_locked(&lockres->l_lock);
807 
808 	lockres->l_flags = newflags;
809 
810 	list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
811 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
812 			continue;
813 
814 		list_del_init(&mw->mw_item);
815 		mw->mw_status = 0;
816 		complete(&mw->mw_complete);
817 	}
818 }
lockres_or_flags(struct ocfs2_lock_res * lockres,unsigned long or)819 static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
820 {
821 	lockres_set_flags(lockres, lockres->l_flags | or);
822 }
lockres_clear_flags(struct ocfs2_lock_res * lockres,unsigned long clear)823 static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
824 				unsigned long clear)
825 {
826 	lockres_set_flags(lockres, lockres->l_flags & ~clear);
827 }
828 
ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res * lockres)829 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
830 {
831 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
832 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
833 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
834 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
835 
836 	lockres->l_level = lockres->l_requested;
837 	if (lockres->l_level <=
838 	    ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
839 		lockres->l_blocking = DLM_LOCK_NL;
840 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
841 	}
842 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
843 }
844 
ocfs2_generic_handle_convert_action(struct ocfs2_lock_res * lockres)845 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
846 {
847 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
848 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
849 
850 	/* Convert from RO to EX doesn't really need anything as our
851 	 * information is already up to data. Convert from NL to
852 	 * *anything* however should mark ourselves as needing an
853 	 * update */
854 	if (lockres->l_level == DLM_LOCK_NL &&
855 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
856 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
857 
858 	lockres->l_level = lockres->l_requested;
859 
860 	/*
861 	 * We set the OCFS2_LOCK_UPCONVERT_FINISHING flag before clearing
862 	 * the OCFS2_LOCK_BUSY flag to prevent the dc thread from
863 	 * downconverting the lock before the upconvert has fully completed.
864 	 * Do not prevent the dc thread from downconverting if NONBLOCK lock
865 	 * had already returned.
866 	 */
867 	if (!(lockres->l_flags & OCFS2_LOCK_NONBLOCK_FINISHED))
868 		lockres_or_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
869 	else
870 		lockres_clear_flags(lockres, OCFS2_LOCK_NONBLOCK_FINISHED);
871 
872 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
873 }
874 
ocfs2_generic_handle_attach_action(struct ocfs2_lock_res * lockres)875 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
876 {
877 	BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
878 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
879 
880 	if (lockres->l_requested > DLM_LOCK_NL &&
881 	    !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
882 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
883 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
884 
885 	lockres->l_level = lockres->l_requested;
886 	lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
887 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
888 }
889 
ocfs2_generic_handle_bast(struct ocfs2_lock_res * lockres,int level)890 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
891 				     int level)
892 {
893 	int needs_downconvert = 0;
894 
895 	assert_spin_locked(&lockres->l_lock);
896 
897 	if (level > lockres->l_blocking) {
898 		/* only schedule a downconvert if we haven't already scheduled
899 		 * one that goes low enough to satisfy the level we're
900 		 * blocking.  this also catches the case where we get
901 		 * duplicate BASTs */
902 		if (ocfs2_highest_compat_lock_level(level) <
903 		    ocfs2_highest_compat_lock_level(lockres->l_blocking))
904 			needs_downconvert = 1;
905 
906 		lockres->l_blocking = level;
907 	}
908 
909 	mlog(ML_BASTS, "lockres %s, block %d, level %d, l_block %d, dwn %d\n",
910 	     lockres->l_name, level, lockres->l_level, lockres->l_blocking,
911 	     needs_downconvert);
912 
913 	if (needs_downconvert)
914 		lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
915 	mlog(0, "needs_downconvert = %d\n", needs_downconvert);
916 	return needs_downconvert;
917 }
918 
919 /*
920  * OCFS2_LOCK_PENDING and l_pending_gen.
921  *
922  * Why does OCFS2_LOCK_PENDING exist?  To close a race between setting
923  * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock().  See ocfs2_unblock_lock()
924  * for more details on the race.
925  *
926  * OCFS2_LOCK_PENDING closes the race quite nicely.  However, it introduces
927  * a race on itself.  In o2dlm, we can get the ast before ocfs2_dlm_lock()
928  * returns.  The ast clears OCFS2_LOCK_BUSY, and must therefore clear
929  * OCFS2_LOCK_PENDING at the same time.  When ocfs2_dlm_lock() returns,
930  * the caller is going to try to clear PENDING again.  If nothing else is
931  * happening, __lockres_clear_pending() sees PENDING is unset and does
932  * nothing.
933  *
934  * But what if another path (eg downconvert thread) has just started a
935  * new locking action?  The other path has re-set PENDING.  Our path
936  * cannot clear PENDING, because that will re-open the original race
937  * window.
938  *
939  * [Example]
940  *
941  * ocfs2_meta_lock()
942  *  ocfs2_cluster_lock()
943  *   set BUSY
944  *   set PENDING
945  *   drop l_lock
946  *   ocfs2_dlm_lock()
947  *    ocfs2_locking_ast()		ocfs2_downconvert_thread()
948  *     clear PENDING			 ocfs2_unblock_lock()
949  *					  take_l_lock
950  *					  !BUSY
951  *					  ocfs2_prepare_downconvert()
952  *					   set BUSY
953  *					   set PENDING
954  *					  drop l_lock
955  *   take l_lock
956  *   clear PENDING
957  *   drop l_lock
958  *			<window>
959  *					  ocfs2_dlm_lock()
960  *
961  * So as you can see, we now have a window where l_lock is not held,
962  * PENDING is not set, and ocfs2_dlm_lock() has not been called.
963  *
964  * The core problem is that ocfs2_cluster_lock() has cleared the PENDING
965  * set by ocfs2_prepare_downconvert().  That wasn't nice.
966  *
967  * To solve this we introduce l_pending_gen.  A call to
968  * lockres_clear_pending() will only do so when it is passed a generation
969  * number that matches the lockres.  lockres_set_pending() will return the
970  * current generation number.  When ocfs2_cluster_lock() goes to clear
971  * PENDING, it passes the generation it got from set_pending().  In our
972  * example above, the generation numbers will *not* match.  Thus,
973  * ocfs2_cluster_lock() will not clear the PENDING set by
974  * ocfs2_prepare_downconvert().
975  */
976 
977 /* Unlocked version for ocfs2_locking_ast() */
__lockres_clear_pending(struct ocfs2_lock_res * lockres,unsigned int generation,struct ocfs2_super * osb)978 static void __lockres_clear_pending(struct ocfs2_lock_res *lockres,
979 				    unsigned int generation,
980 				    struct ocfs2_super *osb)
981 {
982 	assert_spin_locked(&lockres->l_lock);
983 
984 	/*
985 	 * The ast and locking functions can race us here.  The winner
986 	 * will clear pending, the loser will not.
987 	 */
988 	if (!(lockres->l_flags & OCFS2_LOCK_PENDING) ||
989 	    (lockres->l_pending_gen != generation))
990 		return;
991 
992 	lockres_clear_flags(lockres, OCFS2_LOCK_PENDING);
993 	lockres->l_pending_gen++;
994 
995 	/*
996 	 * The downconvert thread may have skipped us because we
997 	 * were PENDING.  Wake it up.
998 	 */
999 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1000 		ocfs2_wake_downconvert_thread(osb);
1001 }
1002 
1003 /* Locked version for callers of ocfs2_dlm_lock() */
lockres_clear_pending(struct ocfs2_lock_res * lockres,unsigned int generation,struct ocfs2_super * osb)1004 static void lockres_clear_pending(struct ocfs2_lock_res *lockres,
1005 				  unsigned int generation,
1006 				  struct ocfs2_super *osb)
1007 {
1008 	unsigned long flags;
1009 
1010 	spin_lock_irqsave(&lockres->l_lock, flags);
1011 	__lockres_clear_pending(lockres, generation, osb);
1012 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1013 }
1014 
lockres_set_pending(struct ocfs2_lock_res * lockres)1015 static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres)
1016 {
1017 	assert_spin_locked(&lockres->l_lock);
1018 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
1019 
1020 	lockres_or_flags(lockres, OCFS2_LOCK_PENDING);
1021 
1022 	return lockres->l_pending_gen;
1023 }
1024 
ocfs2_blocking_ast(struct ocfs2_dlm_lksb * lksb,int level)1025 static void ocfs2_blocking_ast(struct ocfs2_dlm_lksb *lksb, int level)
1026 {
1027 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1028 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1029 	int needs_downconvert;
1030 	unsigned long flags;
1031 
1032 	BUG_ON(level <= DLM_LOCK_NL);
1033 
1034 	mlog(ML_BASTS, "BAST fired for lockres %s, blocking %d, level %d, "
1035 	     "type %s\n", lockres->l_name, level, lockres->l_level,
1036 	     ocfs2_lock_type_string(lockres->l_type));
1037 
1038 	/*
1039 	 * We can skip the bast for locks which don't enable caching -
1040 	 * they'll be dropped at the earliest possible time anyway.
1041 	 */
1042 	if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
1043 		return;
1044 
1045 	spin_lock_irqsave(&lockres->l_lock, flags);
1046 	needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
1047 	if (needs_downconvert)
1048 		ocfs2_schedule_blocked_lock(osb, lockres);
1049 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1050 
1051 	wake_up(&lockres->l_event);
1052 
1053 	ocfs2_wake_downconvert_thread(osb);
1054 }
1055 
ocfs2_locking_ast(struct ocfs2_dlm_lksb * lksb)1056 static void ocfs2_locking_ast(struct ocfs2_dlm_lksb *lksb)
1057 {
1058 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1059 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1060 	unsigned long flags;
1061 	int status;
1062 
1063 	spin_lock_irqsave(&lockres->l_lock, flags);
1064 
1065 	status = ocfs2_dlm_lock_status(&lockres->l_lksb);
1066 
1067 	if (status == -EAGAIN) {
1068 		lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1069 		goto out;
1070 	}
1071 
1072 	if (status) {
1073 		mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n",
1074 		     lockres->l_name, status);
1075 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1076 		return;
1077 	}
1078 
1079 	mlog(ML_BASTS, "AST fired for lockres %s, action %d, unlock %d, "
1080 	     "level %d => %d\n", lockres->l_name, lockres->l_action,
1081 	     lockres->l_unlock_action, lockres->l_level, lockres->l_requested);
1082 
1083 	switch(lockres->l_action) {
1084 	case OCFS2_AST_ATTACH:
1085 		ocfs2_generic_handle_attach_action(lockres);
1086 		lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
1087 		break;
1088 	case OCFS2_AST_CONVERT:
1089 		ocfs2_generic_handle_convert_action(lockres);
1090 		break;
1091 	case OCFS2_AST_DOWNCONVERT:
1092 		ocfs2_generic_handle_downconvert_action(lockres);
1093 		break;
1094 	default:
1095 		mlog(ML_ERROR, "lockres %s: AST fired with invalid action: %u, "
1096 		     "flags 0x%lx, unlock: %u\n",
1097 		     lockres->l_name, lockres->l_action, lockres->l_flags,
1098 		     lockres->l_unlock_action);
1099 		BUG();
1100 	}
1101 out:
1102 	/* set it to something invalid so if we get called again we
1103 	 * can catch it. */
1104 	lockres->l_action = OCFS2_AST_INVALID;
1105 
1106 	/* Did we try to cancel this lock?  Clear that state */
1107 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT)
1108 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1109 
1110 	/*
1111 	 * We may have beaten the locking functions here.  We certainly
1112 	 * know that dlm_lock() has been called :-)
1113 	 * Because we can't have two lock calls in flight at once, we
1114 	 * can use lockres->l_pending_gen.
1115 	 */
1116 	__lockres_clear_pending(lockres, lockres->l_pending_gen,  osb);
1117 
1118 	wake_up(&lockres->l_event);
1119 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1120 }
1121 
ocfs2_unlock_ast(struct ocfs2_dlm_lksb * lksb,int error)1122 static void ocfs2_unlock_ast(struct ocfs2_dlm_lksb *lksb, int error)
1123 {
1124 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1125 	unsigned long flags;
1126 
1127 	mlog(ML_BASTS, "UNLOCK AST fired for lockres %s, action = %d\n",
1128 	     lockres->l_name, lockres->l_unlock_action);
1129 
1130 	spin_lock_irqsave(&lockres->l_lock, flags);
1131 	if (error) {
1132 		mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
1133 		     "unlock_action %d\n", error, lockres->l_name,
1134 		     lockres->l_unlock_action);
1135 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1136 		return;
1137 	}
1138 
1139 	switch(lockres->l_unlock_action) {
1140 	case OCFS2_UNLOCK_CANCEL_CONVERT:
1141 		mlog(0, "Cancel convert success for %s\n", lockres->l_name);
1142 		lockres->l_action = OCFS2_AST_INVALID;
1143 		/* Downconvert thread may have requeued this lock, we
1144 		 * need to wake it. */
1145 		if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1146 			ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres));
1147 		break;
1148 	case OCFS2_UNLOCK_DROP_LOCK:
1149 		lockres->l_level = DLM_LOCK_IV;
1150 		break;
1151 	default:
1152 		BUG();
1153 	}
1154 
1155 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1156 	lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1157 	wake_up(&lockres->l_event);
1158 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1159 }
1160 
1161 /*
1162  * This is the filesystem locking protocol.  It provides the lock handling
1163  * hooks for the underlying DLM.  It has a maximum version number.
1164  * The version number allows interoperability with systems running at
1165  * the same major number and an equal or smaller minor number.
1166  *
1167  * Whenever the filesystem does new things with locks (adds or removes a
1168  * lock, orders them differently, does different things underneath a lock),
1169  * the version must be changed.  The protocol is negotiated when joining
1170  * the dlm domain.  A node may join the domain if its major version is
1171  * identical to all other nodes and its minor version is greater than
1172  * or equal to all other nodes.  When its minor version is greater than
1173  * the other nodes, it will run at the minor version specified by the
1174  * other nodes.
1175  *
1176  * If a locking change is made that will not be compatible with older
1177  * versions, the major number must be increased and the minor version set
1178  * to zero.  If a change merely adds a behavior that can be disabled when
1179  * speaking to older versions, the minor version must be increased.  If a
1180  * change adds a fully backwards compatible change (eg, LVB changes that
1181  * are just ignored by older versions), the version does not need to be
1182  * updated.
1183  */
1184 static struct ocfs2_locking_protocol lproto = {
1185 	.lp_max_version = {
1186 		.pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
1187 		.pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
1188 	},
1189 	.lp_lock_ast		= ocfs2_locking_ast,
1190 	.lp_blocking_ast	= ocfs2_blocking_ast,
1191 	.lp_unlock_ast		= ocfs2_unlock_ast,
1192 };
1193 
ocfs2_set_locking_protocol(void)1194 void ocfs2_set_locking_protocol(void)
1195 {
1196 	ocfs2_stack_glue_set_max_proto_version(&lproto.lp_max_version);
1197 }
1198 
ocfs2_recover_from_dlm_error(struct ocfs2_lock_res * lockres,int convert)1199 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
1200 						int convert)
1201 {
1202 	unsigned long flags;
1203 
1204 	spin_lock_irqsave(&lockres->l_lock, flags);
1205 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1206 	lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1207 	if (convert)
1208 		lockres->l_action = OCFS2_AST_INVALID;
1209 	else
1210 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1211 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1212 
1213 	wake_up(&lockres->l_event);
1214 }
1215 
1216 /* Note: If we detect another process working on the lock (i.e.,
1217  * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
1218  * to do the right thing in that case.
1219  */
ocfs2_lock_create(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int level,u32 dlm_flags)1220 static int ocfs2_lock_create(struct ocfs2_super *osb,
1221 			     struct ocfs2_lock_res *lockres,
1222 			     int level,
1223 			     u32 dlm_flags)
1224 {
1225 	int ret = 0;
1226 	unsigned long flags;
1227 	unsigned int gen;
1228 
1229 	mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
1230 	     dlm_flags);
1231 
1232 	spin_lock_irqsave(&lockres->l_lock, flags);
1233 	if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
1234 	    (lockres->l_flags & OCFS2_LOCK_BUSY)) {
1235 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1236 		goto bail;
1237 	}
1238 
1239 	lockres->l_action = OCFS2_AST_ATTACH;
1240 	lockres->l_requested = level;
1241 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1242 	gen = lockres_set_pending(lockres);
1243 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1244 
1245 	ret = ocfs2_dlm_lock(osb->cconn,
1246 			     level,
1247 			     &lockres->l_lksb,
1248 			     dlm_flags,
1249 			     lockres->l_name,
1250 			     OCFS2_LOCK_ID_MAX_LEN - 1);
1251 	lockres_clear_pending(lockres, gen, osb);
1252 	if (ret) {
1253 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
1254 		ocfs2_recover_from_dlm_error(lockres, 1);
1255 	}
1256 
1257 	mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
1258 
1259 bail:
1260 	return ret;
1261 }
1262 
ocfs2_check_wait_flag(struct ocfs2_lock_res * lockres,int flag)1263 static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
1264 					int flag)
1265 {
1266 	unsigned long flags;
1267 	int ret;
1268 
1269 	spin_lock_irqsave(&lockres->l_lock, flags);
1270 	ret = lockres->l_flags & flag;
1271 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1272 
1273 	return ret;
1274 }
1275 
ocfs2_wait_on_busy_lock(struct ocfs2_lock_res * lockres)1276 static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
1277 
1278 {
1279 	wait_event(lockres->l_event,
1280 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
1281 }
1282 
ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res * lockres)1283 static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
1284 
1285 {
1286 	wait_event(lockres->l_event,
1287 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
1288 }
1289 
1290 /* predict what lock level we'll be dropping down to on behalf
1291  * of another node, and return true if the currently wanted
1292  * level will be compatible with it. */
ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res * lockres,int wanted)1293 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
1294 						     int wanted)
1295 {
1296 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
1297 
1298 	return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
1299 }
1300 
ocfs2_init_mask_waiter(struct ocfs2_mask_waiter * mw)1301 static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
1302 {
1303 	INIT_LIST_HEAD(&mw->mw_item);
1304 	init_completion(&mw->mw_complete);
1305 	ocfs2_init_start_time(mw);
1306 }
1307 
ocfs2_wait_for_mask(struct ocfs2_mask_waiter * mw)1308 static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
1309 {
1310 	wait_for_completion(&mw->mw_complete);
1311 	/* Re-arm the completion in case we want to wait on it again */
1312 	reinit_completion(&mw->mw_complete);
1313 	return mw->mw_status;
1314 }
1315 
lockres_add_mask_waiter(struct ocfs2_lock_res * lockres,struct ocfs2_mask_waiter * mw,unsigned long mask,unsigned long goal)1316 static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
1317 				    struct ocfs2_mask_waiter *mw,
1318 				    unsigned long mask,
1319 				    unsigned long goal)
1320 {
1321 	BUG_ON(!list_empty(&mw->mw_item));
1322 
1323 	assert_spin_locked(&lockres->l_lock);
1324 
1325 	list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
1326 	mw->mw_mask = mask;
1327 	mw->mw_goal = goal;
1328 }
1329 
1330 /* returns 0 if the mw that was removed was already satisfied, -EBUSY
1331  * if the mask still hadn't reached its goal */
__lockres_remove_mask_waiter(struct ocfs2_lock_res * lockres,struct ocfs2_mask_waiter * mw)1332 static int __lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1333 				      struct ocfs2_mask_waiter *mw)
1334 {
1335 	int ret = 0;
1336 
1337 	assert_spin_locked(&lockres->l_lock);
1338 	if (!list_empty(&mw->mw_item)) {
1339 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
1340 			ret = -EBUSY;
1341 
1342 		list_del_init(&mw->mw_item);
1343 		init_completion(&mw->mw_complete);
1344 	}
1345 
1346 	return ret;
1347 }
1348 
lockres_remove_mask_waiter(struct ocfs2_lock_res * lockres,struct ocfs2_mask_waiter * mw)1349 static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1350 				      struct ocfs2_mask_waiter *mw)
1351 {
1352 	unsigned long flags;
1353 	int ret = 0;
1354 
1355 	spin_lock_irqsave(&lockres->l_lock, flags);
1356 	ret = __lockres_remove_mask_waiter(lockres, mw);
1357 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1358 
1359 	return ret;
1360 
1361 }
1362 
ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter * mw,struct ocfs2_lock_res * lockres)1363 static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
1364 					     struct ocfs2_lock_res *lockres)
1365 {
1366 	int ret;
1367 
1368 	ret = wait_for_completion_interruptible(&mw->mw_complete);
1369 	if (ret)
1370 		lockres_remove_mask_waiter(lockres, mw);
1371 	else
1372 		ret = mw->mw_status;
1373 	/* Re-arm the completion in case we want to wait on it again */
1374 	reinit_completion(&mw->mw_complete);
1375 	return ret;
1376 }
1377 
__ocfs2_cluster_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int level,u32 lkm_flags,int arg_flags,int l_subclass,unsigned long caller_ip)1378 static int __ocfs2_cluster_lock(struct ocfs2_super *osb,
1379 				struct ocfs2_lock_res *lockres,
1380 				int level,
1381 				u32 lkm_flags,
1382 				int arg_flags,
1383 				int l_subclass,
1384 				unsigned long caller_ip)
1385 {
1386 	struct ocfs2_mask_waiter mw;
1387 	int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
1388 	int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1389 	unsigned long flags;
1390 	unsigned int gen;
1391 	int noqueue_attempted = 0;
1392 	int dlm_locked = 0;
1393 	int kick_dc = 0;
1394 
1395 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED)) {
1396 		mlog_errno(-EINVAL);
1397 		return -EINVAL;
1398 	}
1399 
1400 	ocfs2_init_mask_waiter(&mw);
1401 
1402 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
1403 		lkm_flags |= DLM_LKF_VALBLK;
1404 
1405 again:
1406 	wait = 0;
1407 
1408 	spin_lock_irqsave(&lockres->l_lock, flags);
1409 
1410 	if (catch_signals && signal_pending(current)) {
1411 		ret = -ERESTARTSYS;
1412 		goto unlock;
1413 	}
1414 
1415 	mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
1416 			"Cluster lock called on freeing lockres %s! flags "
1417 			"0x%lx\n", lockres->l_name, lockres->l_flags);
1418 
1419 	/* We only compare against the currently granted level
1420 	 * here. If the lock is blocked waiting on a downconvert,
1421 	 * we'll get caught below. */
1422 	if (lockres->l_flags & OCFS2_LOCK_BUSY &&
1423 	    level > lockres->l_level) {
1424 		/* is someone sitting in dlm_lock? If so, wait on
1425 		 * them. */
1426 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1427 		wait = 1;
1428 		goto unlock;
1429 	}
1430 
1431 	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING) {
1432 		/*
1433 		 * We've upconverted. If the lock now has a level we can
1434 		 * work with, we take it. If, however, the lock is not at the
1435 		 * required level, we go thru the full cycle. One way this could
1436 		 * happen is if a process requesting an upconvert to PR is
1437 		 * closely followed by another requesting upconvert to an EX.
1438 		 * If the process requesting EX lands here, we want it to
1439 		 * continue attempting to upconvert and let the process
1440 		 * requesting PR take the lock.
1441 		 * If multiple processes request upconvert to PR, the first one
1442 		 * here will take the lock. The others will have to go thru the
1443 		 * OCFS2_LOCK_BLOCKED check to ensure that there is no pending
1444 		 * downconvert request.
1445 		 */
1446 		if (level <= lockres->l_level)
1447 			goto update_holders;
1448 	}
1449 
1450 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
1451 	    !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
1452 		/* is the lock is currently blocked on behalf of
1453 		 * another node */
1454 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
1455 		wait = 1;
1456 		goto unlock;
1457 	}
1458 
1459 	if (level > lockres->l_level) {
1460 		if (noqueue_attempted > 0) {
1461 			ret = -EAGAIN;
1462 			goto unlock;
1463 		}
1464 		if (lkm_flags & DLM_LKF_NOQUEUE)
1465 			noqueue_attempted = 1;
1466 
1467 		if (lockres->l_action != OCFS2_AST_INVALID)
1468 			mlog(ML_ERROR, "lockres %s has action %u pending\n",
1469 			     lockres->l_name, lockres->l_action);
1470 
1471 		if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1472 			lockres->l_action = OCFS2_AST_ATTACH;
1473 			lkm_flags &= ~DLM_LKF_CONVERT;
1474 		} else {
1475 			lockres->l_action = OCFS2_AST_CONVERT;
1476 			lkm_flags |= DLM_LKF_CONVERT;
1477 		}
1478 
1479 		lockres->l_requested = level;
1480 		lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1481 		gen = lockres_set_pending(lockres);
1482 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1483 
1484 		BUG_ON(level == DLM_LOCK_IV);
1485 		BUG_ON(level == DLM_LOCK_NL);
1486 
1487 		mlog(ML_BASTS, "lockres %s, convert from %d to %d\n",
1488 		     lockres->l_name, lockres->l_level, level);
1489 
1490 		/* call dlm_lock to upgrade lock now */
1491 		ret = ocfs2_dlm_lock(osb->cconn,
1492 				     level,
1493 				     &lockres->l_lksb,
1494 				     lkm_flags,
1495 				     lockres->l_name,
1496 				     OCFS2_LOCK_ID_MAX_LEN - 1);
1497 		lockres_clear_pending(lockres, gen, osb);
1498 		if (ret) {
1499 			if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
1500 			    (ret != -EAGAIN)) {
1501 				ocfs2_log_dlm_error("ocfs2_dlm_lock",
1502 						    ret, lockres);
1503 			}
1504 			ocfs2_recover_from_dlm_error(lockres, 1);
1505 			goto out;
1506 		}
1507 		dlm_locked = 1;
1508 
1509 		mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n",
1510 		     lockres->l_name);
1511 
1512 		/* At this point we've gone inside the dlm and need to
1513 		 * complete our work regardless. */
1514 		catch_signals = 0;
1515 
1516 		/* wait for busy to clear and carry on */
1517 		goto again;
1518 	}
1519 
1520 update_holders:
1521 	/* Ok, if we get here then we're good to go. */
1522 	ocfs2_inc_holders(lockres, level);
1523 
1524 	ret = 0;
1525 unlock:
1526 	lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1527 
1528 	/* ocfs2_unblock_lock reques on seeing OCFS2_LOCK_UPCONVERT_FINISHING */
1529 	kick_dc = (lockres->l_flags & OCFS2_LOCK_BLOCKED);
1530 
1531 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1532 	if (kick_dc)
1533 		ocfs2_wake_downconvert_thread(osb);
1534 out:
1535 	/*
1536 	 * This is helping work around a lock inversion between the page lock
1537 	 * and dlm locks.  One path holds the page lock while calling aops
1538 	 * which block acquiring dlm locks.  The voting thread holds dlm
1539 	 * locks while acquiring page locks while down converting data locks.
1540 	 * This block is helping an aop path notice the inversion and back
1541 	 * off to unlock its page lock before trying the dlm lock again.
1542 	 */
1543 	if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1544 	    mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1545 		wait = 0;
1546 		spin_lock_irqsave(&lockres->l_lock, flags);
1547 		if (__lockres_remove_mask_waiter(lockres, &mw)) {
1548 			if (dlm_locked)
1549 				lockres_or_flags(lockres,
1550 					OCFS2_LOCK_NONBLOCK_FINISHED);
1551 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1552 			ret = -EAGAIN;
1553 		} else {
1554 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1555 			goto again;
1556 		}
1557 	}
1558 	if (wait) {
1559 		ret = ocfs2_wait_for_mask(&mw);
1560 		if (ret == 0)
1561 			goto again;
1562 		mlog_errno(ret);
1563 	}
1564 	ocfs2_update_lock_stats(lockres, level, &mw, ret);
1565 
1566 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1567 	if (!ret && lockres->l_lockdep_map.key != NULL) {
1568 		if (level == DLM_LOCK_PR)
1569 			rwsem_acquire_read(&lockres->l_lockdep_map, l_subclass,
1570 				!!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1571 				caller_ip);
1572 		else
1573 			rwsem_acquire(&lockres->l_lockdep_map, l_subclass,
1574 				!!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1575 				caller_ip);
1576 	}
1577 #endif
1578 	return ret;
1579 }
1580 
ocfs2_cluster_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int level,u32 lkm_flags,int arg_flags)1581 static inline int ocfs2_cluster_lock(struct ocfs2_super *osb,
1582 				     struct ocfs2_lock_res *lockres,
1583 				     int level,
1584 				     u32 lkm_flags,
1585 				     int arg_flags)
1586 {
1587 	return __ocfs2_cluster_lock(osb, lockres, level, lkm_flags, arg_flags,
1588 				    0, _RET_IP_);
1589 }
1590 
1591 
__ocfs2_cluster_unlock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int level,unsigned long caller_ip)1592 static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
1593 				   struct ocfs2_lock_res *lockres,
1594 				   int level,
1595 				   unsigned long caller_ip)
1596 {
1597 	unsigned long flags;
1598 
1599 	spin_lock_irqsave(&lockres->l_lock, flags);
1600 	ocfs2_dec_holders(lockres, level);
1601 	ocfs2_downconvert_on_unlock(osb, lockres);
1602 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1603 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1604 	if (lockres->l_lockdep_map.key != NULL)
1605 		rwsem_release(&lockres->l_lockdep_map, 1, caller_ip);
1606 #endif
1607 }
1608 
ocfs2_create_new_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int ex,int local)1609 static int ocfs2_create_new_lock(struct ocfs2_super *osb,
1610 				 struct ocfs2_lock_res *lockres,
1611 				 int ex,
1612 				 int local)
1613 {
1614 	int level =  ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1615 	unsigned long flags;
1616 	u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
1617 
1618 	spin_lock_irqsave(&lockres->l_lock, flags);
1619 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1620 	lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1621 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1622 
1623 	return ocfs2_lock_create(osb, lockres, level, lkm_flags);
1624 }
1625 
1626 /* Grants us an EX lock on the data and metadata resources, skipping
1627  * the normal cluster directory lookup. Use this ONLY on newly created
1628  * inodes which other nodes can't possibly see, and which haven't been
1629  * hashed in the inode hash yet. This can give us a good performance
1630  * increase as it'll skip the network broadcast normally associated
1631  * with creating a new lock resource. */
ocfs2_create_new_inode_locks(struct inode * inode)1632 int ocfs2_create_new_inode_locks(struct inode *inode)
1633 {
1634 	int ret;
1635 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1636 
1637 	BUG_ON(!inode);
1638 	BUG_ON(!ocfs2_inode_is_new(inode));
1639 
1640 	mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
1641 
1642 	/* NOTE: That we don't increment any of the holder counts, nor
1643 	 * do we add anything to a journal handle. Since this is
1644 	 * supposed to be a new inode which the cluster doesn't know
1645 	 * about yet, there is no need to.  As far as the LVB handling
1646 	 * is concerned, this is basically like acquiring an EX lock
1647 	 * on a resource which has an invalid one -- we'll set it
1648 	 * valid when we release the EX. */
1649 
1650 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
1651 	if (ret) {
1652 		mlog_errno(ret);
1653 		goto bail;
1654 	}
1655 
1656 	/*
1657 	 * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
1658 	 * don't use a generation in their lock names.
1659 	 */
1660 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
1661 	if (ret) {
1662 		mlog_errno(ret);
1663 		goto bail;
1664 	}
1665 
1666 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
1667 	if (ret) {
1668 		mlog_errno(ret);
1669 		goto bail;
1670 	}
1671 
1672 bail:
1673 	return ret;
1674 }
1675 
ocfs2_rw_lock(struct inode * inode,int write)1676 int ocfs2_rw_lock(struct inode *inode, int write)
1677 {
1678 	int status, level;
1679 	struct ocfs2_lock_res *lockres;
1680 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1681 
1682 	BUG_ON(!inode);
1683 
1684 	mlog(0, "inode %llu take %s RW lock\n",
1685 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1686 	     write ? "EXMODE" : "PRMODE");
1687 
1688 	if (ocfs2_mount_local(osb))
1689 		return 0;
1690 
1691 	lockres = &OCFS2_I(inode)->ip_rw_lockres;
1692 
1693 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1694 
1695 	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
1696 				    0);
1697 	if (status < 0)
1698 		mlog_errno(status);
1699 
1700 	return status;
1701 }
1702 
ocfs2_rw_unlock(struct inode * inode,int write)1703 void ocfs2_rw_unlock(struct inode *inode, int write)
1704 {
1705 	int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1706 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
1707 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1708 
1709 	mlog(0, "inode %llu drop %s RW lock\n",
1710 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1711 	     write ? "EXMODE" : "PRMODE");
1712 
1713 	if (!ocfs2_mount_local(osb))
1714 		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1715 }
1716 
1717 /*
1718  * ocfs2_open_lock always get PR mode lock.
1719  */
ocfs2_open_lock(struct inode * inode)1720 int ocfs2_open_lock(struct inode *inode)
1721 {
1722 	int status = 0;
1723 	struct ocfs2_lock_res *lockres;
1724 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1725 
1726 	BUG_ON(!inode);
1727 
1728 	mlog(0, "inode %llu take PRMODE open lock\n",
1729 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1730 
1731 	if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1732 		goto out;
1733 
1734 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1735 
1736 	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
1737 				    DLM_LOCK_PR, 0, 0);
1738 	if (status < 0)
1739 		mlog_errno(status);
1740 
1741 out:
1742 	return status;
1743 }
1744 
ocfs2_try_open_lock(struct inode * inode,int write)1745 int ocfs2_try_open_lock(struct inode *inode, int write)
1746 {
1747 	int status = 0, level;
1748 	struct ocfs2_lock_res *lockres;
1749 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1750 
1751 	BUG_ON(!inode);
1752 
1753 	mlog(0, "inode %llu try to take %s open lock\n",
1754 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1755 	     write ? "EXMODE" : "PRMODE");
1756 
1757 	if (ocfs2_is_hard_readonly(osb)) {
1758 		if (write)
1759 			status = -EROFS;
1760 		goto out;
1761 	}
1762 
1763 	if (ocfs2_mount_local(osb))
1764 		goto out;
1765 
1766 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1767 
1768 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1769 
1770 	/*
1771 	 * The file system may already holding a PRMODE/EXMODE open lock.
1772 	 * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
1773 	 * other nodes and the -EAGAIN will indicate to the caller that
1774 	 * this inode is still in use.
1775 	 */
1776 	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
1777 				    level, DLM_LKF_NOQUEUE, 0);
1778 
1779 out:
1780 	return status;
1781 }
1782 
1783 /*
1784  * ocfs2_open_unlock unlock PR and EX mode open locks.
1785  */
ocfs2_open_unlock(struct inode * inode)1786 void ocfs2_open_unlock(struct inode *inode)
1787 {
1788 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
1789 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1790 
1791 	mlog(0, "inode %llu drop open lock\n",
1792 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1793 
1794 	if (ocfs2_mount_local(osb))
1795 		goto out;
1796 
1797 	if(lockres->l_ro_holders)
1798 		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
1799 				     DLM_LOCK_PR);
1800 	if(lockres->l_ex_holders)
1801 		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
1802 				     DLM_LOCK_EX);
1803 
1804 out:
1805 	return;
1806 }
1807 
ocfs2_flock_handle_signal(struct ocfs2_lock_res * lockres,int level)1808 static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
1809 				     int level)
1810 {
1811 	int ret;
1812 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1813 	unsigned long flags;
1814 	struct ocfs2_mask_waiter mw;
1815 
1816 	ocfs2_init_mask_waiter(&mw);
1817 
1818 retry_cancel:
1819 	spin_lock_irqsave(&lockres->l_lock, flags);
1820 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
1821 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
1822 		if (ret) {
1823 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1824 			ret = ocfs2_cancel_convert(osb, lockres);
1825 			if (ret < 0) {
1826 				mlog_errno(ret);
1827 				goto out;
1828 			}
1829 			goto retry_cancel;
1830 		}
1831 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1832 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1833 
1834 		ocfs2_wait_for_mask(&mw);
1835 		goto retry_cancel;
1836 	}
1837 
1838 	ret = -ERESTARTSYS;
1839 	/*
1840 	 * We may still have gotten the lock, in which case there's no
1841 	 * point to restarting the syscall.
1842 	 */
1843 	if (lockres->l_level == level)
1844 		ret = 0;
1845 
1846 	mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
1847 	     lockres->l_flags, lockres->l_level, lockres->l_action);
1848 
1849 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1850 
1851 out:
1852 	return ret;
1853 }
1854 
1855 /*
1856  * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
1857  * flock() calls. The locking approach this requires is sufficiently
1858  * different from all other cluster lock types that we implement a
1859  * separate path to the "low-level" dlm calls. In particular:
1860  *
1861  * - No optimization of lock levels is done - we take at exactly
1862  *   what's been requested.
1863  *
1864  * - No lock caching is employed. We immediately downconvert to
1865  *   no-lock at unlock time. This also means flock locks never go on
1866  *   the blocking list).
1867  *
1868  * - Since userspace can trivially deadlock itself with flock, we make
1869  *   sure to allow cancellation of a misbehaving applications flock()
1870  *   request.
1871  *
1872  * - Access to any flock lockres doesn't require concurrency, so we
1873  *   can simplify the code by requiring the caller to guarantee
1874  *   serialization of dlmglue flock calls.
1875  */
ocfs2_file_lock(struct file * file,int ex,int trylock)1876 int ocfs2_file_lock(struct file *file, int ex, int trylock)
1877 {
1878 	int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1879 	unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0;
1880 	unsigned long flags;
1881 	struct ocfs2_file_private *fp = file->private_data;
1882 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
1883 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1884 	struct ocfs2_mask_waiter mw;
1885 
1886 	ocfs2_init_mask_waiter(&mw);
1887 
1888 	if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
1889 	    (lockres->l_level > DLM_LOCK_NL)) {
1890 		mlog(ML_ERROR,
1891 		     "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
1892 		     "level: %u\n", lockres->l_name, lockres->l_flags,
1893 		     lockres->l_level);
1894 		return -EINVAL;
1895 	}
1896 
1897 	spin_lock_irqsave(&lockres->l_lock, flags);
1898 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1899 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1900 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1901 
1902 		/*
1903 		 * Get the lock at NLMODE to start - that way we
1904 		 * can cancel the upconvert request if need be.
1905 		 */
1906 		ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0);
1907 		if (ret < 0) {
1908 			mlog_errno(ret);
1909 			goto out;
1910 		}
1911 
1912 		ret = ocfs2_wait_for_mask(&mw);
1913 		if (ret) {
1914 			mlog_errno(ret);
1915 			goto out;
1916 		}
1917 		spin_lock_irqsave(&lockres->l_lock, flags);
1918 	}
1919 
1920 	lockres->l_action = OCFS2_AST_CONVERT;
1921 	lkm_flags |= DLM_LKF_CONVERT;
1922 	lockres->l_requested = level;
1923 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1924 
1925 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1926 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1927 
1928 	ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags,
1929 			     lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1);
1930 	if (ret) {
1931 		if (!trylock || (ret != -EAGAIN)) {
1932 			ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
1933 			ret = -EINVAL;
1934 		}
1935 
1936 		ocfs2_recover_from_dlm_error(lockres, 1);
1937 		lockres_remove_mask_waiter(lockres, &mw);
1938 		goto out;
1939 	}
1940 
1941 	ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
1942 	if (ret == -ERESTARTSYS) {
1943 		/*
1944 		 * Userspace can cause deadlock itself with
1945 		 * flock(). Current behavior locally is to allow the
1946 		 * deadlock, but abort the system call if a signal is
1947 		 * received. We follow this example, otherwise a
1948 		 * poorly written program could sit in kernel until
1949 		 * reboot.
1950 		 *
1951 		 * Handling this is a bit more complicated for Ocfs2
1952 		 * though. We can't exit this function with an
1953 		 * outstanding lock request, so a cancel convert is
1954 		 * required. We intentionally overwrite 'ret' - if the
1955 		 * cancel fails and the lock was granted, it's easier
1956 		 * to just bubble success back up to the user.
1957 		 */
1958 		ret = ocfs2_flock_handle_signal(lockres, level);
1959 	} else if (!ret && (level > lockres->l_level)) {
1960 		/* Trylock failed asynchronously */
1961 		BUG_ON(!trylock);
1962 		ret = -EAGAIN;
1963 	}
1964 
1965 out:
1966 
1967 	mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
1968 	     lockres->l_name, ex, trylock, ret);
1969 	return ret;
1970 }
1971 
ocfs2_file_unlock(struct file * file)1972 void ocfs2_file_unlock(struct file *file)
1973 {
1974 	int ret;
1975 	unsigned int gen;
1976 	unsigned long flags;
1977 	struct ocfs2_file_private *fp = file->private_data;
1978 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
1979 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1980 	struct ocfs2_mask_waiter mw;
1981 
1982 	ocfs2_init_mask_waiter(&mw);
1983 
1984 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
1985 		return;
1986 
1987 	if (lockres->l_level == DLM_LOCK_NL)
1988 		return;
1989 
1990 	mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
1991 	     lockres->l_name, lockres->l_flags, lockres->l_level,
1992 	     lockres->l_action);
1993 
1994 	spin_lock_irqsave(&lockres->l_lock, flags);
1995 	/*
1996 	 * Fake a blocking ast for the downconvert code.
1997 	 */
1998 	lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
1999 	lockres->l_blocking = DLM_LOCK_EX;
2000 
2001 	gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL);
2002 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
2003 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2004 
2005 	ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen);
2006 	if (ret) {
2007 		mlog_errno(ret);
2008 		return;
2009 	}
2010 
2011 	ret = ocfs2_wait_for_mask(&mw);
2012 	if (ret)
2013 		mlog_errno(ret);
2014 }
2015 
ocfs2_downconvert_on_unlock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)2016 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
2017 					struct ocfs2_lock_res *lockres)
2018 {
2019 	int kick = 0;
2020 
2021 	/* If we know that another node is waiting on our lock, kick
2022 	 * the downconvert thread * pre-emptively when we reach a release
2023 	 * condition. */
2024 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
2025 		switch(lockres->l_blocking) {
2026 		case DLM_LOCK_EX:
2027 			if (!lockres->l_ex_holders && !lockres->l_ro_holders)
2028 				kick = 1;
2029 			break;
2030 		case DLM_LOCK_PR:
2031 			if (!lockres->l_ex_holders)
2032 				kick = 1;
2033 			break;
2034 		default:
2035 			BUG();
2036 		}
2037 	}
2038 
2039 	if (kick)
2040 		ocfs2_wake_downconvert_thread(osb);
2041 }
2042 
2043 #define OCFS2_SEC_BITS   34
2044 #define OCFS2_SEC_SHIFT  (64 - 34)
2045 #define OCFS2_NSEC_MASK  ((1ULL << OCFS2_SEC_SHIFT) - 1)
2046 
2047 /* LVB only has room for 64 bits of time here so we pack it for
2048  * now. */
ocfs2_pack_timespec(struct timespec * spec)2049 static u64 ocfs2_pack_timespec(struct timespec *spec)
2050 {
2051 	u64 res;
2052 	u64 sec = spec->tv_sec;
2053 	u32 nsec = spec->tv_nsec;
2054 
2055 	res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
2056 
2057 	return res;
2058 }
2059 
2060 /* Call this with the lockres locked. I am reasonably sure we don't
2061  * need ip_lock in this function as anyone who would be changing those
2062  * values is supposed to be blocked in ocfs2_inode_lock right now. */
__ocfs2_stuff_meta_lvb(struct inode * inode)2063 static void __ocfs2_stuff_meta_lvb(struct inode *inode)
2064 {
2065 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2066 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2067 	struct ocfs2_meta_lvb *lvb;
2068 
2069 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2070 
2071 	/*
2072 	 * Invalidate the LVB of a deleted inode - this way other
2073 	 * nodes are forced to go to disk and discover the new inode
2074 	 * status.
2075 	 */
2076 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
2077 		lvb->lvb_version = 0;
2078 		goto out;
2079 	}
2080 
2081 	lvb->lvb_version   = OCFS2_LVB_VERSION;
2082 	lvb->lvb_isize	   = cpu_to_be64(i_size_read(inode));
2083 	lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
2084 	lvb->lvb_iuid      = cpu_to_be32(i_uid_read(inode));
2085 	lvb->lvb_igid      = cpu_to_be32(i_gid_read(inode));
2086 	lvb->lvb_imode     = cpu_to_be16(inode->i_mode);
2087 	lvb->lvb_inlink    = cpu_to_be16(inode->i_nlink);
2088 	lvb->lvb_iatime_packed  =
2089 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
2090 	lvb->lvb_ictime_packed =
2091 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
2092 	lvb->lvb_imtime_packed =
2093 		cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
2094 	lvb->lvb_iattr    = cpu_to_be32(oi->ip_attr);
2095 	lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
2096 	lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
2097 
2098 out:
2099 	mlog_meta_lvb(0, lockres);
2100 }
2101 
ocfs2_unpack_timespec(struct timespec * spec,u64 packed_time)2102 static void ocfs2_unpack_timespec(struct timespec *spec,
2103 				  u64 packed_time)
2104 {
2105 	spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
2106 	spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
2107 }
2108 
ocfs2_refresh_inode_from_lvb(struct inode * inode)2109 static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
2110 {
2111 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2112 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2113 	struct ocfs2_meta_lvb *lvb;
2114 
2115 	mlog_meta_lvb(0, lockres);
2116 
2117 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2118 
2119 	/* We're safe here without the lockres lock... */
2120 	spin_lock(&oi->ip_lock);
2121 	oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
2122 	i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
2123 
2124 	oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
2125 	oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
2126 	ocfs2_set_inode_flags(inode);
2127 
2128 	/* fast-symlinks are a special case */
2129 	if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
2130 		inode->i_blocks = 0;
2131 	else
2132 		inode->i_blocks = ocfs2_inode_sector_count(inode);
2133 
2134 	i_uid_write(inode, be32_to_cpu(lvb->lvb_iuid));
2135 	i_gid_write(inode, be32_to_cpu(lvb->lvb_igid));
2136 	inode->i_mode    = be16_to_cpu(lvb->lvb_imode);
2137 	set_nlink(inode, be16_to_cpu(lvb->lvb_inlink));
2138 	ocfs2_unpack_timespec(&inode->i_atime,
2139 			      be64_to_cpu(lvb->lvb_iatime_packed));
2140 	ocfs2_unpack_timespec(&inode->i_mtime,
2141 			      be64_to_cpu(lvb->lvb_imtime_packed));
2142 	ocfs2_unpack_timespec(&inode->i_ctime,
2143 			      be64_to_cpu(lvb->lvb_ictime_packed));
2144 	spin_unlock(&oi->ip_lock);
2145 }
2146 
ocfs2_meta_lvb_is_trustable(struct inode * inode,struct ocfs2_lock_res * lockres)2147 static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
2148 					      struct ocfs2_lock_res *lockres)
2149 {
2150 	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2151 
2152 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)
2153 	    && lvb->lvb_version == OCFS2_LVB_VERSION
2154 	    && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
2155 		return 1;
2156 	return 0;
2157 }
2158 
2159 /* Determine whether a lock resource needs to be refreshed, and
2160  * arbitrate who gets to refresh it.
2161  *
2162  *   0 means no refresh needed.
2163  *
2164  *   > 0 means you need to refresh this and you MUST call
2165  *   ocfs2_complete_lock_res_refresh afterwards. */
ocfs2_should_refresh_lock_res(struct ocfs2_lock_res * lockres)2166 static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
2167 {
2168 	unsigned long flags;
2169 	int status = 0;
2170 
2171 refresh_check:
2172 	spin_lock_irqsave(&lockres->l_lock, flags);
2173 	if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
2174 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2175 		goto bail;
2176 	}
2177 
2178 	if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
2179 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2180 
2181 		ocfs2_wait_on_refreshing_lock(lockres);
2182 		goto refresh_check;
2183 	}
2184 
2185 	/* Ok, I'll be the one to refresh this lock. */
2186 	lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
2187 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2188 
2189 	status = 1;
2190 bail:
2191 	mlog(0, "status %d\n", status);
2192 	return status;
2193 }
2194 
2195 /* If status is non zero, I'll mark it as not being in refresh
2196  * anymroe, but i won't clear the needs refresh flag. */
ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res * lockres,int status)2197 static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
2198 						   int status)
2199 {
2200 	unsigned long flags;
2201 
2202 	spin_lock_irqsave(&lockres->l_lock, flags);
2203 	lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
2204 	if (!status)
2205 		lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
2206 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2207 
2208 	wake_up(&lockres->l_event);
2209 }
2210 
2211 /* may or may not return a bh if it went to disk. */
ocfs2_inode_lock_update(struct inode * inode,struct buffer_head ** bh)2212 static int ocfs2_inode_lock_update(struct inode *inode,
2213 				  struct buffer_head **bh)
2214 {
2215 	int status = 0;
2216 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2217 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2218 	struct ocfs2_dinode *fe;
2219 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2220 
2221 	if (ocfs2_mount_local(osb))
2222 		goto bail;
2223 
2224 	spin_lock(&oi->ip_lock);
2225 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
2226 		mlog(0, "Orphaned inode %llu was deleted while we "
2227 		     "were waiting on a lock. ip_flags = 0x%x\n",
2228 		     (unsigned long long)oi->ip_blkno, oi->ip_flags);
2229 		spin_unlock(&oi->ip_lock);
2230 		status = -ENOENT;
2231 		goto bail;
2232 	}
2233 	spin_unlock(&oi->ip_lock);
2234 
2235 	if (!ocfs2_should_refresh_lock_res(lockres))
2236 		goto bail;
2237 
2238 	/* This will discard any caching information we might have had
2239 	 * for the inode metadata. */
2240 	ocfs2_metadata_cache_purge(INODE_CACHE(inode));
2241 
2242 	ocfs2_extent_map_trunc(inode, 0);
2243 
2244 	if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
2245 		mlog(0, "Trusting LVB on inode %llu\n",
2246 		     (unsigned long long)oi->ip_blkno);
2247 		ocfs2_refresh_inode_from_lvb(inode);
2248 	} else {
2249 		/* Boo, we have to go to disk. */
2250 		/* read bh, cast, ocfs2_refresh_inode */
2251 		status = ocfs2_read_inode_block(inode, bh);
2252 		if (status < 0) {
2253 			mlog_errno(status);
2254 			goto bail_refresh;
2255 		}
2256 		fe = (struct ocfs2_dinode *) (*bh)->b_data;
2257 
2258 		/* This is a good chance to make sure we're not
2259 		 * locking an invalid object.  ocfs2_read_inode_block()
2260 		 * already checked that the inode block is sane.
2261 		 *
2262 		 * We bug on a stale inode here because we checked
2263 		 * above whether it was wiped from disk. The wiping
2264 		 * node provides a guarantee that we receive that
2265 		 * message and can mark the inode before dropping any
2266 		 * locks associated with it. */
2267 		mlog_bug_on_msg(inode->i_generation !=
2268 				le32_to_cpu(fe->i_generation),
2269 				"Invalid dinode %llu disk generation: %u "
2270 				"inode->i_generation: %u\n",
2271 				(unsigned long long)oi->ip_blkno,
2272 				le32_to_cpu(fe->i_generation),
2273 				inode->i_generation);
2274 		mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
2275 				!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
2276 				"Stale dinode %llu dtime: %llu flags: 0x%x\n",
2277 				(unsigned long long)oi->ip_blkno,
2278 				(unsigned long long)le64_to_cpu(fe->i_dtime),
2279 				le32_to_cpu(fe->i_flags));
2280 
2281 		ocfs2_refresh_inode(inode, fe);
2282 		ocfs2_track_lock_refresh(lockres);
2283 	}
2284 
2285 	status = 0;
2286 bail_refresh:
2287 	ocfs2_complete_lock_res_refresh(lockres, status);
2288 bail:
2289 	return status;
2290 }
2291 
ocfs2_assign_bh(struct inode * inode,struct buffer_head ** ret_bh,struct buffer_head * passed_bh)2292 static int ocfs2_assign_bh(struct inode *inode,
2293 			   struct buffer_head **ret_bh,
2294 			   struct buffer_head *passed_bh)
2295 {
2296 	int status;
2297 
2298 	if (passed_bh) {
2299 		/* Ok, the update went to disk for us, use the
2300 		 * returned bh. */
2301 		*ret_bh = passed_bh;
2302 		get_bh(*ret_bh);
2303 
2304 		return 0;
2305 	}
2306 
2307 	status = ocfs2_read_inode_block(inode, ret_bh);
2308 	if (status < 0)
2309 		mlog_errno(status);
2310 
2311 	return status;
2312 }
2313 
2314 /*
2315  * returns < 0 error if the callback will never be called, otherwise
2316  * the result of the lock will be communicated via the callback.
2317  */
ocfs2_inode_lock_full_nested(struct inode * inode,struct buffer_head ** ret_bh,int ex,int arg_flags,int subclass)2318 int ocfs2_inode_lock_full_nested(struct inode *inode,
2319 				 struct buffer_head **ret_bh,
2320 				 int ex,
2321 				 int arg_flags,
2322 				 int subclass)
2323 {
2324 	int status, level, acquired;
2325 	u32 dlm_flags;
2326 	struct ocfs2_lock_res *lockres = NULL;
2327 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2328 	struct buffer_head *local_bh = NULL;
2329 
2330 	BUG_ON(!inode);
2331 
2332 	mlog(0, "inode %llu, take %s META lock\n",
2333 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2334 	     ex ? "EXMODE" : "PRMODE");
2335 
2336 	status = 0;
2337 	acquired = 0;
2338 	/* We'll allow faking a readonly metadata lock for
2339 	 * rodevices. */
2340 	if (ocfs2_is_hard_readonly(osb)) {
2341 		if (ex)
2342 			status = -EROFS;
2343 		goto getbh;
2344 	}
2345 
2346 	if (ocfs2_mount_local(osb))
2347 		goto local;
2348 
2349 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2350 		ocfs2_wait_for_recovery(osb);
2351 
2352 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2353 	level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2354 	dlm_flags = 0;
2355 	if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
2356 		dlm_flags |= DLM_LKF_NOQUEUE;
2357 
2358 	status = __ocfs2_cluster_lock(osb, lockres, level, dlm_flags,
2359 				      arg_flags, subclass, _RET_IP_);
2360 	if (status < 0) {
2361 		if (status != -EAGAIN)
2362 			mlog_errno(status);
2363 		goto bail;
2364 	}
2365 
2366 	/* Notify the error cleanup path to drop the cluster lock. */
2367 	acquired = 1;
2368 
2369 	/* We wait twice because a node may have died while we were in
2370 	 * the lower dlm layers. The second time though, we've
2371 	 * committed to owning this lock so we don't allow signals to
2372 	 * abort the operation. */
2373 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2374 		ocfs2_wait_for_recovery(osb);
2375 
2376 local:
2377 	/*
2378 	 * We only see this flag if we're being called from
2379 	 * ocfs2_read_locked_inode(). It means we're locking an inode
2380 	 * which hasn't been populated yet, so clear the refresh flag
2381 	 * and let the caller handle it.
2382 	 */
2383 	if (inode->i_state & I_NEW) {
2384 		status = 0;
2385 		if (lockres)
2386 			ocfs2_complete_lock_res_refresh(lockres, 0);
2387 		goto bail;
2388 	}
2389 
2390 	/* This is fun. The caller may want a bh back, or it may
2391 	 * not. ocfs2_inode_lock_update definitely wants one in, but
2392 	 * may or may not read one, depending on what's in the
2393 	 * LVB. The result of all of this is that we've *only* gone to
2394 	 * disk if we have to, so the complexity is worthwhile. */
2395 	status = ocfs2_inode_lock_update(inode, &local_bh);
2396 	if (status < 0) {
2397 		if (status != -ENOENT)
2398 			mlog_errno(status);
2399 		goto bail;
2400 	}
2401 getbh:
2402 	if (ret_bh) {
2403 		status = ocfs2_assign_bh(inode, ret_bh, local_bh);
2404 		if (status < 0) {
2405 			mlog_errno(status);
2406 			goto bail;
2407 		}
2408 	}
2409 
2410 bail:
2411 	if (status < 0) {
2412 		if (ret_bh && (*ret_bh)) {
2413 			brelse(*ret_bh);
2414 			*ret_bh = NULL;
2415 		}
2416 		if (acquired)
2417 			ocfs2_inode_unlock(inode, ex);
2418 	}
2419 
2420 	if (local_bh)
2421 		brelse(local_bh);
2422 
2423 	return status;
2424 }
2425 
2426 /*
2427  * This is working around a lock inversion between tasks acquiring DLM
2428  * locks while holding a page lock and the downconvert thread which
2429  * blocks dlm lock acquiry while acquiring page locks.
2430  *
2431  * ** These _with_page variantes are only intended to be called from aop
2432  * methods that hold page locks and return a very specific *positive* error
2433  * code that aop methods pass up to the VFS -- test for errors with != 0. **
2434  *
2435  * The DLM is called such that it returns -EAGAIN if it would have
2436  * blocked waiting for the downconvert thread.  In that case we unlock
2437  * our page so the downconvert thread can make progress.  Once we've
2438  * done this we have to return AOP_TRUNCATED_PAGE so the aop method
2439  * that called us can bubble that back up into the VFS who will then
2440  * immediately retry the aop call.
2441  *
2442  * We do a blocking lock and immediate unlock before returning, though, so that
2443  * the lock has a great chance of being cached on this node by the time the VFS
2444  * calls back to retry the aop.    This has a potential to livelock as nodes
2445  * ping locks back and forth, but that's a risk we're willing to take to avoid
2446  * the lock inversion simply.
2447  */
ocfs2_inode_lock_with_page(struct inode * inode,struct buffer_head ** ret_bh,int ex,struct page * page)2448 int ocfs2_inode_lock_with_page(struct inode *inode,
2449 			      struct buffer_head **ret_bh,
2450 			      int ex,
2451 			      struct page *page)
2452 {
2453 	int ret;
2454 
2455 	ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
2456 	if (ret == -EAGAIN) {
2457 		unlock_page(page);
2458 		if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
2459 			ocfs2_inode_unlock(inode, ex);
2460 		ret = AOP_TRUNCATED_PAGE;
2461 	}
2462 
2463 	return ret;
2464 }
2465 
ocfs2_inode_lock_atime(struct inode * inode,struct vfsmount * vfsmnt,int * level)2466 int ocfs2_inode_lock_atime(struct inode *inode,
2467 			  struct vfsmount *vfsmnt,
2468 			  int *level)
2469 {
2470 	int ret;
2471 
2472 	ret = ocfs2_inode_lock(inode, NULL, 0);
2473 	if (ret < 0) {
2474 		mlog_errno(ret);
2475 		return ret;
2476 	}
2477 
2478 	/*
2479 	 * If we should update atime, we will get EX lock,
2480 	 * otherwise we just get PR lock.
2481 	 */
2482 	if (ocfs2_should_update_atime(inode, vfsmnt)) {
2483 		struct buffer_head *bh = NULL;
2484 
2485 		ocfs2_inode_unlock(inode, 0);
2486 		ret = ocfs2_inode_lock(inode, &bh, 1);
2487 		if (ret < 0) {
2488 			mlog_errno(ret);
2489 			return ret;
2490 		}
2491 		*level = 1;
2492 		if (ocfs2_should_update_atime(inode, vfsmnt))
2493 			ocfs2_update_inode_atime(inode, bh);
2494 		if (bh)
2495 			brelse(bh);
2496 	} else
2497 		*level = 0;
2498 
2499 	return ret;
2500 }
2501 
ocfs2_inode_unlock(struct inode * inode,int ex)2502 void ocfs2_inode_unlock(struct inode *inode,
2503 		       int ex)
2504 {
2505 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2506 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
2507 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2508 
2509 	mlog(0, "inode %llu drop %s META lock\n",
2510 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2511 	     ex ? "EXMODE" : "PRMODE");
2512 
2513 	if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) &&
2514 	    !ocfs2_mount_local(osb))
2515 		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
2516 }
2517 
ocfs2_orphan_scan_lock(struct ocfs2_super * osb,u32 * seqno)2518 int ocfs2_orphan_scan_lock(struct ocfs2_super *osb, u32 *seqno)
2519 {
2520 	struct ocfs2_lock_res *lockres;
2521 	struct ocfs2_orphan_scan_lvb *lvb;
2522 	int status = 0;
2523 
2524 	if (ocfs2_is_hard_readonly(osb))
2525 		return -EROFS;
2526 
2527 	if (ocfs2_mount_local(osb))
2528 		return 0;
2529 
2530 	lockres = &osb->osb_orphan_scan.os_lockres;
2531 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2532 	if (status < 0)
2533 		return status;
2534 
2535 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2536 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2537 	    lvb->lvb_version == OCFS2_ORPHAN_LVB_VERSION)
2538 		*seqno = be32_to_cpu(lvb->lvb_os_seqno);
2539 	else
2540 		*seqno = osb->osb_orphan_scan.os_seqno + 1;
2541 
2542 	return status;
2543 }
2544 
ocfs2_orphan_scan_unlock(struct ocfs2_super * osb,u32 seqno)2545 void ocfs2_orphan_scan_unlock(struct ocfs2_super *osb, u32 seqno)
2546 {
2547 	struct ocfs2_lock_res *lockres;
2548 	struct ocfs2_orphan_scan_lvb *lvb;
2549 
2550 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) {
2551 		lockres = &osb->osb_orphan_scan.os_lockres;
2552 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2553 		lvb->lvb_version = OCFS2_ORPHAN_LVB_VERSION;
2554 		lvb->lvb_os_seqno = cpu_to_be32(seqno);
2555 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2556 	}
2557 }
2558 
ocfs2_super_lock(struct ocfs2_super * osb,int ex)2559 int ocfs2_super_lock(struct ocfs2_super *osb,
2560 		     int ex)
2561 {
2562 	int status = 0;
2563 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2564 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2565 
2566 	if (ocfs2_is_hard_readonly(osb))
2567 		return -EROFS;
2568 
2569 	if (ocfs2_mount_local(osb))
2570 		goto bail;
2571 
2572 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
2573 	if (status < 0) {
2574 		mlog_errno(status);
2575 		goto bail;
2576 	}
2577 
2578 	/* The super block lock path is really in the best position to
2579 	 * know when resources covered by the lock need to be
2580 	 * refreshed, so we do it here. Of course, making sense of
2581 	 * everything is up to the caller :) */
2582 	status = ocfs2_should_refresh_lock_res(lockres);
2583 	if (status) {
2584 		status = ocfs2_refresh_slot_info(osb);
2585 
2586 		ocfs2_complete_lock_res_refresh(lockres, status);
2587 
2588 		if (status < 0) {
2589 			ocfs2_cluster_unlock(osb, lockres, level);
2590 			mlog_errno(status);
2591 		}
2592 		ocfs2_track_lock_refresh(lockres);
2593 	}
2594 bail:
2595 	return status;
2596 }
2597 
ocfs2_super_unlock(struct ocfs2_super * osb,int ex)2598 void ocfs2_super_unlock(struct ocfs2_super *osb,
2599 			int ex)
2600 {
2601 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2602 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2603 
2604 	if (!ocfs2_mount_local(osb))
2605 		ocfs2_cluster_unlock(osb, lockres, level);
2606 }
2607 
ocfs2_rename_lock(struct ocfs2_super * osb)2608 int ocfs2_rename_lock(struct ocfs2_super *osb)
2609 {
2610 	int status;
2611 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2612 
2613 	if (ocfs2_is_hard_readonly(osb))
2614 		return -EROFS;
2615 
2616 	if (ocfs2_mount_local(osb))
2617 		return 0;
2618 
2619 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2620 	if (status < 0)
2621 		mlog_errno(status);
2622 
2623 	return status;
2624 }
2625 
ocfs2_rename_unlock(struct ocfs2_super * osb)2626 void ocfs2_rename_unlock(struct ocfs2_super *osb)
2627 {
2628 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2629 
2630 	if (!ocfs2_mount_local(osb))
2631 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2632 }
2633 
ocfs2_nfs_sync_lock(struct ocfs2_super * osb,int ex)2634 int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex)
2635 {
2636 	int status;
2637 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2638 
2639 	if (ocfs2_is_hard_readonly(osb))
2640 		return -EROFS;
2641 
2642 	if (ocfs2_mount_local(osb))
2643 		return 0;
2644 
2645 	status = ocfs2_cluster_lock(osb, lockres, ex ? LKM_EXMODE : LKM_PRMODE,
2646 				    0, 0);
2647 	if (status < 0)
2648 		mlog(ML_ERROR, "lock on nfs sync lock failed %d\n", status);
2649 
2650 	return status;
2651 }
2652 
ocfs2_nfs_sync_unlock(struct ocfs2_super * osb,int ex)2653 void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex)
2654 {
2655 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2656 
2657 	if (!ocfs2_mount_local(osb))
2658 		ocfs2_cluster_unlock(osb, lockres,
2659 				     ex ? LKM_EXMODE : LKM_PRMODE);
2660 }
2661 
ocfs2_dentry_lock(struct dentry * dentry,int ex)2662 int ocfs2_dentry_lock(struct dentry *dentry, int ex)
2663 {
2664 	int ret;
2665 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2666 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2667 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2668 
2669 	BUG_ON(!dl);
2670 
2671 	if (ocfs2_is_hard_readonly(osb)) {
2672 		if (ex)
2673 			return -EROFS;
2674 		return 0;
2675 	}
2676 
2677 	if (ocfs2_mount_local(osb))
2678 		return 0;
2679 
2680 	ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2681 	if (ret < 0)
2682 		mlog_errno(ret);
2683 
2684 	return ret;
2685 }
2686 
ocfs2_dentry_unlock(struct dentry * dentry,int ex)2687 void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2688 {
2689 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2690 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2691 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2692 
2693 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
2694 		ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
2695 }
2696 
2697 /* Reference counting of the dlm debug structure. We want this because
2698  * open references on the debug inodes can live on after a mount, so
2699  * we can't rely on the ocfs2_super to always exist. */
ocfs2_dlm_debug_free(struct kref * kref)2700 static void ocfs2_dlm_debug_free(struct kref *kref)
2701 {
2702 	struct ocfs2_dlm_debug *dlm_debug;
2703 
2704 	dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
2705 
2706 	kfree(dlm_debug);
2707 }
2708 
ocfs2_put_dlm_debug(struct ocfs2_dlm_debug * dlm_debug)2709 void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
2710 {
2711 	if (dlm_debug)
2712 		kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
2713 }
2714 
ocfs2_get_dlm_debug(struct ocfs2_dlm_debug * debug)2715 static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
2716 {
2717 	kref_get(&debug->d_refcnt);
2718 }
2719 
ocfs2_new_dlm_debug(void)2720 struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
2721 {
2722 	struct ocfs2_dlm_debug *dlm_debug;
2723 
2724 	dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
2725 	if (!dlm_debug) {
2726 		mlog_errno(-ENOMEM);
2727 		goto out;
2728 	}
2729 
2730 	kref_init(&dlm_debug->d_refcnt);
2731 	INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
2732 	dlm_debug->d_locking_state = NULL;
2733 out:
2734 	return dlm_debug;
2735 }
2736 
2737 /* Access to this is arbitrated for us via seq_file->sem. */
2738 struct ocfs2_dlm_seq_priv {
2739 	struct ocfs2_dlm_debug *p_dlm_debug;
2740 	struct ocfs2_lock_res p_iter_res;
2741 	struct ocfs2_lock_res p_tmp_res;
2742 };
2743 
ocfs2_dlm_next_res(struct ocfs2_lock_res * start,struct ocfs2_dlm_seq_priv * priv)2744 static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
2745 						 struct ocfs2_dlm_seq_priv *priv)
2746 {
2747 	struct ocfs2_lock_res *iter, *ret = NULL;
2748 	struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
2749 
2750 	assert_spin_locked(&ocfs2_dlm_tracking_lock);
2751 
2752 	list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
2753 		/* discover the head of the list */
2754 		if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
2755 			mlog(0, "End of list found, %p\n", ret);
2756 			break;
2757 		}
2758 
2759 		/* We track our "dummy" iteration lockres' by a NULL
2760 		 * l_ops field. */
2761 		if (iter->l_ops != NULL) {
2762 			ret = iter;
2763 			break;
2764 		}
2765 	}
2766 
2767 	return ret;
2768 }
2769 
ocfs2_dlm_seq_start(struct seq_file * m,loff_t * pos)2770 static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
2771 {
2772 	struct ocfs2_dlm_seq_priv *priv = m->private;
2773 	struct ocfs2_lock_res *iter;
2774 
2775 	spin_lock(&ocfs2_dlm_tracking_lock);
2776 	iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
2777 	if (iter) {
2778 		/* Since lockres' have the lifetime of their container
2779 		 * (which can be inodes, ocfs2_supers, etc) we want to
2780 		 * copy this out to a temporary lockres while still
2781 		 * under the spinlock. Obviously after this we can't
2782 		 * trust any pointers on the copy returned, but that's
2783 		 * ok as the information we want isn't typically held
2784 		 * in them. */
2785 		priv->p_tmp_res = *iter;
2786 		iter = &priv->p_tmp_res;
2787 	}
2788 	spin_unlock(&ocfs2_dlm_tracking_lock);
2789 
2790 	return iter;
2791 }
2792 
ocfs2_dlm_seq_stop(struct seq_file * m,void * v)2793 static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
2794 {
2795 }
2796 
ocfs2_dlm_seq_next(struct seq_file * m,void * v,loff_t * pos)2797 static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
2798 {
2799 	struct ocfs2_dlm_seq_priv *priv = m->private;
2800 	struct ocfs2_lock_res *iter = v;
2801 	struct ocfs2_lock_res *dummy = &priv->p_iter_res;
2802 
2803 	spin_lock(&ocfs2_dlm_tracking_lock);
2804 	iter = ocfs2_dlm_next_res(iter, priv);
2805 	list_del_init(&dummy->l_debug_list);
2806 	if (iter) {
2807 		list_add(&dummy->l_debug_list, &iter->l_debug_list);
2808 		priv->p_tmp_res = *iter;
2809 		iter = &priv->p_tmp_res;
2810 	}
2811 	spin_unlock(&ocfs2_dlm_tracking_lock);
2812 
2813 	return iter;
2814 }
2815 
2816 /*
2817  * Version is used by debugfs.ocfs2 to determine the format being used
2818  *
2819  * New in version 2
2820  *	- Lock stats printed
2821  * New in version 3
2822  *	- Max time in lock stats is in usecs (instead of nsecs)
2823  */
2824 #define OCFS2_DLM_DEBUG_STR_VERSION 3
ocfs2_dlm_seq_show(struct seq_file * m,void * v)2825 static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
2826 {
2827 	int i;
2828 	char *lvb;
2829 	struct ocfs2_lock_res *lockres = v;
2830 
2831 	if (!lockres)
2832 		return -EINVAL;
2833 
2834 	seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
2835 
2836 	if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
2837 		seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
2838 			   lockres->l_name,
2839 			   (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
2840 	else
2841 		seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
2842 
2843 	seq_printf(m, "%d\t"
2844 		   "0x%lx\t"
2845 		   "0x%x\t"
2846 		   "0x%x\t"
2847 		   "%u\t"
2848 		   "%u\t"
2849 		   "%d\t"
2850 		   "%d\t",
2851 		   lockres->l_level,
2852 		   lockres->l_flags,
2853 		   lockres->l_action,
2854 		   lockres->l_unlock_action,
2855 		   lockres->l_ro_holders,
2856 		   lockres->l_ex_holders,
2857 		   lockres->l_requested,
2858 		   lockres->l_blocking);
2859 
2860 	/* Dump the raw LVB */
2861 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2862 	for(i = 0; i < DLM_LVB_LEN; i++)
2863 		seq_printf(m, "0x%x\t", lvb[i]);
2864 
2865 #ifdef CONFIG_OCFS2_FS_STATS
2866 # define lock_num_prmode(_l)		((_l)->l_lock_prmode.ls_gets)
2867 # define lock_num_exmode(_l)		((_l)->l_lock_exmode.ls_gets)
2868 # define lock_num_prmode_failed(_l)	((_l)->l_lock_prmode.ls_fail)
2869 # define lock_num_exmode_failed(_l)	((_l)->l_lock_exmode.ls_fail)
2870 # define lock_total_prmode(_l)		((_l)->l_lock_prmode.ls_total)
2871 # define lock_total_exmode(_l)		((_l)->l_lock_exmode.ls_total)
2872 # define lock_max_prmode(_l)		((_l)->l_lock_prmode.ls_max)
2873 # define lock_max_exmode(_l)		((_l)->l_lock_exmode.ls_max)
2874 # define lock_refresh(_l)		((_l)->l_lock_refresh)
2875 #else
2876 # define lock_num_prmode(_l)		(0)
2877 # define lock_num_exmode(_l)		(0)
2878 # define lock_num_prmode_failed(_l)	(0)
2879 # define lock_num_exmode_failed(_l)	(0)
2880 # define lock_total_prmode(_l)		(0ULL)
2881 # define lock_total_exmode(_l)		(0ULL)
2882 # define lock_max_prmode(_l)		(0)
2883 # define lock_max_exmode(_l)		(0)
2884 # define lock_refresh(_l)		(0)
2885 #endif
2886 	/* The following seq_print was added in version 2 of this output */
2887 	seq_printf(m, "%u\t"
2888 		   "%u\t"
2889 		   "%u\t"
2890 		   "%u\t"
2891 		   "%llu\t"
2892 		   "%llu\t"
2893 		   "%u\t"
2894 		   "%u\t"
2895 		   "%u\t",
2896 		   lock_num_prmode(lockres),
2897 		   lock_num_exmode(lockres),
2898 		   lock_num_prmode_failed(lockres),
2899 		   lock_num_exmode_failed(lockres),
2900 		   lock_total_prmode(lockres),
2901 		   lock_total_exmode(lockres),
2902 		   lock_max_prmode(lockres),
2903 		   lock_max_exmode(lockres),
2904 		   lock_refresh(lockres));
2905 
2906 	/* End the line */
2907 	seq_printf(m, "\n");
2908 	return 0;
2909 }
2910 
2911 static const struct seq_operations ocfs2_dlm_seq_ops = {
2912 	.start =	ocfs2_dlm_seq_start,
2913 	.stop =		ocfs2_dlm_seq_stop,
2914 	.next =		ocfs2_dlm_seq_next,
2915 	.show =		ocfs2_dlm_seq_show,
2916 };
2917 
ocfs2_dlm_debug_release(struct inode * inode,struct file * file)2918 static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
2919 {
2920 	struct seq_file *seq = file->private_data;
2921 	struct ocfs2_dlm_seq_priv *priv = seq->private;
2922 	struct ocfs2_lock_res *res = &priv->p_iter_res;
2923 
2924 	ocfs2_remove_lockres_tracking(res);
2925 	ocfs2_put_dlm_debug(priv->p_dlm_debug);
2926 	return seq_release_private(inode, file);
2927 }
2928 
ocfs2_dlm_debug_open(struct inode * inode,struct file * file)2929 static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
2930 {
2931 	struct ocfs2_dlm_seq_priv *priv;
2932 	struct ocfs2_super *osb;
2933 
2934 	priv = __seq_open_private(file, &ocfs2_dlm_seq_ops, sizeof(*priv));
2935 	if (!priv) {
2936 		mlog_errno(-ENOMEM);
2937 		return -ENOMEM;
2938 	}
2939 
2940 	osb = inode->i_private;
2941 	ocfs2_get_dlm_debug(osb->osb_dlm_debug);
2942 	priv->p_dlm_debug = osb->osb_dlm_debug;
2943 	INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
2944 
2945 	ocfs2_add_lockres_tracking(&priv->p_iter_res,
2946 				   priv->p_dlm_debug);
2947 
2948 	return 0;
2949 }
2950 
2951 static const struct file_operations ocfs2_dlm_debug_fops = {
2952 	.open =		ocfs2_dlm_debug_open,
2953 	.release =	ocfs2_dlm_debug_release,
2954 	.read =		seq_read,
2955 	.llseek =	seq_lseek,
2956 };
2957 
ocfs2_dlm_init_debug(struct ocfs2_super * osb)2958 static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
2959 {
2960 	int ret = 0;
2961 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2962 
2963 	dlm_debug->d_locking_state = debugfs_create_file("locking_state",
2964 							 S_IFREG|S_IRUSR,
2965 							 osb->osb_debug_root,
2966 							 osb,
2967 							 &ocfs2_dlm_debug_fops);
2968 	if (!dlm_debug->d_locking_state) {
2969 		ret = -EINVAL;
2970 		mlog(ML_ERROR,
2971 		     "Unable to create locking state debugfs file.\n");
2972 		goto out;
2973 	}
2974 
2975 	ocfs2_get_dlm_debug(dlm_debug);
2976 out:
2977 	return ret;
2978 }
2979 
ocfs2_dlm_shutdown_debug(struct ocfs2_super * osb)2980 static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
2981 {
2982 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2983 
2984 	if (dlm_debug) {
2985 		debugfs_remove(dlm_debug->d_locking_state);
2986 		ocfs2_put_dlm_debug(dlm_debug);
2987 	}
2988 }
2989 
ocfs2_dlm_init(struct ocfs2_super * osb)2990 int ocfs2_dlm_init(struct ocfs2_super *osb)
2991 {
2992 	int status = 0;
2993 	struct ocfs2_cluster_connection *conn = NULL;
2994 
2995 	if (ocfs2_mount_local(osb)) {
2996 		osb->node_num = 0;
2997 		goto local;
2998 	}
2999 
3000 	status = ocfs2_dlm_init_debug(osb);
3001 	if (status < 0) {
3002 		mlog_errno(status);
3003 		goto bail;
3004 	}
3005 
3006 	/* launch downconvert thread */
3007 	osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc-%s",
3008 			osb->uuid_str);
3009 	if (IS_ERR(osb->dc_task)) {
3010 		status = PTR_ERR(osb->dc_task);
3011 		osb->dc_task = NULL;
3012 		mlog_errno(status);
3013 		goto bail;
3014 	}
3015 
3016 	/* for now, uuid == domain */
3017 	status = ocfs2_cluster_connect(osb->osb_cluster_stack,
3018 				       osb->osb_cluster_name,
3019 				       strlen(osb->osb_cluster_name),
3020 				       osb->uuid_str,
3021 				       strlen(osb->uuid_str),
3022 				       &lproto, ocfs2_do_node_down, osb,
3023 				       &conn);
3024 	if (status) {
3025 		mlog_errno(status);
3026 		goto bail;
3027 	}
3028 
3029 	status = ocfs2_cluster_this_node(conn, &osb->node_num);
3030 	if (status < 0) {
3031 		mlog_errno(status);
3032 		mlog(ML_ERROR,
3033 		     "could not find this host's node number\n");
3034 		ocfs2_cluster_disconnect(conn, 0);
3035 		goto bail;
3036 	}
3037 
3038 local:
3039 	ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
3040 	ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
3041 	ocfs2_nfs_sync_lock_res_init(&osb->osb_nfs_sync_lockres, osb);
3042 	ocfs2_orphan_scan_lock_res_init(&osb->osb_orphan_scan.os_lockres, osb);
3043 
3044 	osb->cconn = conn;
3045 bail:
3046 	if (status < 0) {
3047 		ocfs2_dlm_shutdown_debug(osb);
3048 		if (osb->dc_task)
3049 			kthread_stop(osb->dc_task);
3050 	}
3051 
3052 	return status;
3053 }
3054 
ocfs2_dlm_shutdown(struct ocfs2_super * osb,int hangup_pending)3055 void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
3056 			int hangup_pending)
3057 {
3058 	ocfs2_drop_osb_locks(osb);
3059 
3060 	/*
3061 	 * Now that we have dropped all locks and ocfs2_dismount_volume()
3062 	 * has disabled recovery, the DLM won't be talking to us.  It's
3063 	 * safe to tear things down before disconnecting the cluster.
3064 	 */
3065 
3066 	if (osb->dc_task) {
3067 		kthread_stop(osb->dc_task);
3068 		osb->dc_task = NULL;
3069 	}
3070 
3071 	ocfs2_lock_res_free(&osb->osb_super_lockres);
3072 	ocfs2_lock_res_free(&osb->osb_rename_lockres);
3073 	ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres);
3074 	ocfs2_lock_res_free(&osb->osb_orphan_scan.os_lockres);
3075 
3076 	ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
3077 	osb->cconn = NULL;
3078 
3079 	ocfs2_dlm_shutdown_debug(osb);
3080 }
3081 
ocfs2_drop_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)3082 static int ocfs2_drop_lock(struct ocfs2_super *osb,
3083 			   struct ocfs2_lock_res *lockres)
3084 {
3085 	int ret;
3086 	unsigned long flags;
3087 	u32 lkm_flags = 0;
3088 
3089 	/* We didn't get anywhere near actually using this lockres. */
3090 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
3091 		goto out;
3092 
3093 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
3094 		lkm_flags |= DLM_LKF_VALBLK;
3095 
3096 	spin_lock_irqsave(&lockres->l_lock, flags);
3097 
3098 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
3099 			"lockres %s, flags 0x%lx\n",
3100 			lockres->l_name, lockres->l_flags);
3101 
3102 	while (lockres->l_flags & OCFS2_LOCK_BUSY) {
3103 		mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
3104 		     "%u, unlock_action = %u\n",
3105 		     lockres->l_name, lockres->l_flags, lockres->l_action,
3106 		     lockres->l_unlock_action);
3107 
3108 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3109 
3110 		/* XXX: Today we just wait on any busy
3111 		 * locks... Perhaps we need to cancel converts in the
3112 		 * future? */
3113 		ocfs2_wait_on_busy_lock(lockres);
3114 
3115 		spin_lock_irqsave(&lockres->l_lock, flags);
3116 	}
3117 
3118 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3119 		if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
3120 		    lockres->l_level == DLM_LOCK_EX &&
3121 		    !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3122 			lockres->l_ops->set_lvb(lockres);
3123 	}
3124 
3125 	if (lockres->l_flags & OCFS2_LOCK_BUSY)
3126 		mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
3127 		     lockres->l_name);
3128 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
3129 		mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
3130 
3131 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
3132 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3133 		goto out;
3134 	}
3135 
3136 	lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
3137 
3138 	/* make sure we never get here while waiting for an ast to
3139 	 * fire. */
3140 	BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
3141 
3142 	/* is this necessary? */
3143 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3144 	lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
3145 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3146 
3147 	mlog(0, "lock %s\n", lockres->l_name);
3148 
3149 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags);
3150 	if (ret) {
3151 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3152 		mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
3153 		ocfs2_dlm_dump_lksb(&lockres->l_lksb);
3154 		BUG();
3155 	}
3156 	mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n",
3157 	     lockres->l_name);
3158 
3159 	ocfs2_wait_on_busy_lock(lockres);
3160 out:
3161 	return 0;
3162 }
3163 
3164 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3165 				       struct ocfs2_lock_res *lockres);
3166 
3167 /* Mark the lockres as being dropped. It will no longer be
3168  * queued if blocking, but we still may have to wait on it
3169  * being dequeued from the downconvert thread before we can consider
3170  * it safe to drop.
3171  *
3172  * You can *not* attempt to call cluster_lock on this lockres anymore. */
ocfs2_mark_lockres_freeing(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)3173 void ocfs2_mark_lockres_freeing(struct ocfs2_super *osb,
3174 				struct ocfs2_lock_res *lockres)
3175 {
3176 	int status;
3177 	struct ocfs2_mask_waiter mw;
3178 	unsigned long flags, flags2;
3179 
3180 	ocfs2_init_mask_waiter(&mw);
3181 
3182 	spin_lock_irqsave(&lockres->l_lock, flags);
3183 	lockres->l_flags |= OCFS2_LOCK_FREEING;
3184 	if (lockres->l_flags & OCFS2_LOCK_QUEUED && current == osb->dc_task) {
3185 		/*
3186 		 * We know the downconvert is queued but not in progress
3187 		 * because we are the downconvert thread and processing
3188 		 * different lock. So we can just remove the lock from the
3189 		 * queue. This is not only an optimization but also a way
3190 		 * to avoid the following deadlock:
3191 		 *   ocfs2_dentry_post_unlock()
3192 		 *     ocfs2_dentry_lock_put()
3193 		 *       ocfs2_drop_dentry_lock()
3194 		 *         iput()
3195 		 *           ocfs2_evict_inode()
3196 		 *             ocfs2_clear_inode()
3197 		 *               ocfs2_mark_lockres_freeing()
3198 		 *                 ... blocks waiting for OCFS2_LOCK_QUEUED
3199 		 *                 since we are the downconvert thread which
3200 		 *                 should clear the flag.
3201 		 */
3202 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3203 		spin_lock_irqsave(&osb->dc_task_lock, flags2);
3204 		list_del_init(&lockres->l_blocked_list);
3205 		osb->blocked_lock_count--;
3206 		spin_unlock_irqrestore(&osb->dc_task_lock, flags2);
3207 		/*
3208 		 * Warn if we recurse into another post_unlock call.  Strictly
3209 		 * speaking it isn't a problem but we need to be careful if
3210 		 * that happens (stack overflow, deadlocks, ...) so warn if
3211 		 * ocfs2 grows a path for which this can happen.
3212 		 */
3213 		WARN_ON_ONCE(lockres->l_ops->post_unlock);
3214 		/* Since the lock is freeing we don't do much in the fn below */
3215 		ocfs2_process_blocked_lock(osb, lockres);
3216 		return;
3217 	}
3218 	while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
3219 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
3220 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3221 
3222 		mlog(0, "Waiting on lockres %s\n", lockres->l_name);
3223 
3224 		status = ocfs2_wait_for_mask(&mw);
3225 		if (status)
3226 			mlog_errno(status);
3227 
3228 		spin_lock_irqsave(&lockres->l_lock, flags);
3229 	}
3230 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3231 }
3232 
ocfs2_simple_drop_lockres(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)3233 void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
3234 			       struct ocfs2_lock_res *lockres)
3235 {
3236 	int ret;
3237 
3238 	ocfs2_mark_lockres_freeing(osb, lockres);
3239 	ret = ocfs2_drop_lock(osb, lockres);
3240 	if (ret)
3241 		mlog_errno(ret);
3242 }
3243 
ocfs2_drop_osb_locks(struct ocfs2_super * osb)3244 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
3245 {
3246 	ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
3247 	ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
3248 	ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres);
3249 	ocfs2_simple_drop_lockres(osb, &osb->osb_orphan_scan.os_lockres);
3250 }
3251 
ocfs2_drop_inode_locks(struct inode * inode)3252 int ocfs2_drop_inode_locks(struct inode *inode)
3253 {
3254 	int status, err;
3255 
3256 	/* No need to call ocfs2_mark_lockres_freeing here -
3257 	 * ocfs2_clear_inode has done it for us. */
3258 
3259 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3260 			      &OCFS2_I(inode)->ip_open_lockres);
3261 	if (err < 0)
3262 		mlog_errno(err);
3263 
3264 	status = err;
3265 
3266 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3267 			      &OCFS2_I(inode)->ip_inode_lockres);
3268 	if (err < 0)
3269 		mlog_errno(err);
3270 	if (err < 0 && !status)
3271 		status = err;
3272 
3273 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3274 			      &OCFS2_I(inode)->ip_rw_lockres);
3275 	if (err < 0)
3276 		mlog_errno(err);
3277 	if (err < 0 && !status)
3278 		status = err;
3279 
3280 	return status;
3281 }
3282 
ocfs2_prepare_downconvert(struct ocfs2_lock_res * lockres,int new_level)3283 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
3284 					      int new_level)
3285 {
3286 	assert_spin_locked(&lockres->l_lock);
3287 
3288 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
3289 
3290 	if (lockres->l_level <= new_level) {
3291 		mlog(ML_ERROR, "lockres %s, lvl %d <= %d, blcklst %d, mask %d, "
3292 		     "type %d, flags 0x%lx, hold %d %d, act %d %d, req %d, "
3293 		     "block %d, pgen %d\n", lockres->l_name, lockres->l_level,
3294 		     new_level, list_empty(&lockres->l_blocked_list),
3295 		     list_empty(&lockres->l_mask_waiters), lockres->l_type,
3296 		     lockres->l_flags, lockres->l_ro_holders,
3297 		     lockres->l_ex_holders, lockres->l_action,
3298 		     lockres->l_unlock_action, lockres->l_requested,
3299 		     lockres->l_blocking, lockres->l_pending_gen);
3300 		BUG();
3301 	}
3302 
3303 	mlog(ML_BASTS, "lockres %s, level %d => %d, blocking %d\n",
3304 	     lockres->l_name, lockres->l_level, new_level, lockres->l_blocking);
3305 
3306 	lockres->l_action = OCFS2_AST_DOWNCONVERT;
3307 	lockres->l_requested = new_level;
3308 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3309 	return lockres_set_pending(lockres);
3310 }
3311 
ocfs2_downconvert_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int new_level,int lvb,unsigned int generation)3312 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
3313 				  struct ocfs2_lock_res *lockres,
3314 				  int new_level,
3315 				  int lvb,
3316 				  unsigned int generation)
3317 {
3318 	int ret;
3319 	u32 dlm_flags = DLM_LKF_CONVERT;
3320 
3321 	mlog(ML_BASTS, "lockres %s, level %d => %d\n", lockres->l_name,
3322 	     lockres->l_level, new_level);
3323 
3324 	if (lvb)
3325 		dlm_flags |= DLM_LKF_VALBLK;
3326 
3327 	ret = ocfs2_dlm_lock(osb->cconn,
3328 			     new_level,
3329 			     &lockres->l_lksb,
3330 			     dlm_flags,
3331 			     lockres->l_name,
3332 			     OCFS2_LOCK_ID_MAX_LEN - 1);
3333 	lockres_clear_pending(lockres, generation, osb);
3334 	if (ret) {
3335 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
3336 		ocfs2_recover_from_dlm_error(lockres, 1);
3337 		goto bail;
3338 	}
3339 
3340 	ret = 0;
3341 bail:
3342 	return ret;
3343 }
3344 
3345 /* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
ocfs2_prepare_cancel_convert(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)3346 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
3347 				        struct ocfs2_lock_res *lockres)
3348 {
3349 	assert_spin_locked(&lockres->l_lock);
3350 
3351 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
3352 		/* If we're already trying to cancel a lock conversion
3353 		 * then just drop the spinlock and allow the caller to
3354 		 * requeue this lock. */
3355 		mlog(ML_BASTS, "lockres %s, skip convert\n", lockres->l_name);
3356 		return 0;
3357 	}
3358 
3359 	/* were we in a convert when we got the bast fire? */
3360 	BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
3361 	       lockres->l_action != OCFS2_AST_DOWNCONVERT);
3362 	/* set things up for the unlockast to know to just
3363 	 * clear out the ast_action and unset busy, etc. */
3364 	lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
3365 
3366 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
3367 			"lock %s, invalid flags: 0x%lx\n",
3368 			lockres->l_name, lockres->l_flags);
3369 
3370 	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3371 
3372 	return 1;
3373 }
3374 
ocfs2_cancel_convert(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)3375 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
3376 				struct ocfs2_lock_res *lockres)
3377 {
3378 	int ret;
3379 
3380 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
3381 			       DLM_LKF_CANCEL);
3382 	if (ret) {
3383 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3384 		ocfs2_recover_from_dlm_error(lockres, 0);
3385 	}
3386 
3387 	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3388 
3389 	return ret;
3390 }
3391 
ocfs2_unblock_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,struct ocfs2_unblock_ctl * ctl)3392 static int ocfs2_unblock_lock(struct ocfs2_super *osb,
3393 			      struct ocfs2_lock_res *lockres,
3394 			      struct ocfs2_unblock_ctl *ctl)
3395 {
3396 	unsigned long flags;
3397 	int blocking;
3398 	int new_level;
3399 	int level;
3400 	int ret = 0;
3401 	int set_lvb = 0;
3402 	unsigned int gen;
3403 
3404 	spin_lock_irqsave(&lockres->l_lock, flags);
3405 
3406 recheck:
3407 	/*
3408 	 * Is it still blocking? If not, we have no more work to do.
3409 	 */
3410 	if (!(lockres->l_flags & OCFS2_LOCK_BLOCKED)) {
3411 		BUG_ON(lockres->l_blocking != DLM_LOCK_NL);
3412 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3413 		ret = 0;
3414 		goto leave;
3415 	}
3416 
3417 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
3418 		/* XXX
3419 		 * This is a *big* race.  The OCFS2_LOCK_PENDING flag
3420 		 * exists entirely for one reason - another thread has set
3421 		 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
3422 		 *
3423 		 * If we do ocfs2_cancel_convert() before the other thread
3424 		 * calls dlm_lock(), our cancel will do nothing.  We will
3425 		 * get no ast, and we will have no way of knowing the
3426 		 * cancel failed.  Meanwhile, the other thread will call
3427 		 * into dlm_lock() and wait...forever.
3428 		 *
3429 		 * Why forever?  Because another node has asked for the
3430 		 * lock first; that's why we're here in unblock_lock().
3431 		 *
3432 		 * The solution is OCFS2_LOCK_PENDING.  When PENDING is
3433 		 * set, we just requeue the unblock.  Only when the other
3434 		 * thread has called dlm_lock() and cleared PENDING will
3435 		 * we then cancel their request.
3436 		 *
3437 		 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
3438 		 * at the same time they set OCFS2_DLM_BUSY.  They must
3439 		 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
3440 		 */
3441 		if (lockres->l_flags & OCFS2_LOCK_PENDING) {
3442 			mlog(ML_BASTS, "lockres %s, ReQ: Pending\n",
3443 			     lockres->l_name);
3444 			goto leave_requeue;
3445 		}
3446 
3447 		ctl->requeue = 1;
3448 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
3449 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3450 		if (ret) {
3451 			ret = ocfs2_cancel_convert(osb, lockres);
3452 			if (ret < 0)
3453 				mlog_errno(ret);
3454 		}
3455 		goto leave;
3456 	}
3457 
3458 	/*
3459 	 * This prevents livelocks. OCFS2_LOCK_UPCONVERT_FINISHING flag is
3460 	 * set when the ast is received for an upconvert just before the
3461 	 * OCFS2_LOCK_BUSY flag is cleared. Now if the fs received a bast
3462 	 * on the heels of the ast, we want to delay the downconvert just
3463 	 * enough to allow the up requestor to do its task. Because this
3464 	 * lock is in the blocked queue, the lock will be downconverted
3465 	 * as soon as the requestor is done with the lock.
3466 	 */
3467 	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING)
3468 		goto leave_requeue;
3469 
3470 	/*
3471 	 * How can we block and yet be at NL?  We were trying to upconvert
3472 	 * from NL and got canceled.  The code comes back here, and now
3473 	 * we notice and clear BLOCKING.
3474 	 */
3475 	if (lockres->l_level == DLM_LOCK_NL) {
3476 		BUG_ON(lockres->l_ex_holders || lockres->l_ro_holders);
3477 		mlog(ML_BASTS, "lockres %s, Aborting dc\n", lockres->l_name);
3478 		lockres->l_blocking = DLM_LOCK_NL;
3479 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
3480 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3481 		goto leave;
3482 	}
3483 
3484 	/* if we're blocking an exclusive and we have *any* holders,
3485 	 * then requeue. */
3486 	if ((lockres->l_blocking == DLM_LOCK_EX)
3487 	    && (lockres->l_ex_holders || lockres->l_ro_holders)) {
3488 		mlog(ML_BASTS, "lockres %s, ReQ: EX/PR Holders %u,%u\n",
3489 		     lockres->l_name, lockres->l_ex_holders,
3490 		     lockres->l_ro_holders);
3491 		goto leave_requeue;
3492 	}
3493 
3494 	/* If it's a PR we're blocking, then only
3495 	 * requeue if we've got any EX holders */
3496 	if (lockres->l_blocking == DLM_LOCK_PR &&
3497 	    lockres->l_ex_holders) {
3498 		mlog(ML_BASTS, "lockres %s, ReQ: EX Holders %u\n",
3499 		     lockres->l_name, lockres->l_ex_holders);
3500 		goto leave_requeue;
3501 	}
3502 
3503 	/*
3504 	 * Can we get a lock in this state if the holder counts are
3505 	 * zero? The meta data unblock code used to check this.
3506 	 */
3507 	if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
3508 	    && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) {
3509 		mlog(ML_BASTS, "lockres %s, ReQ: Lock Refreshing\n",
3510 		     lockres->l_name);
3511 		goto leave_requeue;
3512 	}
3513 
3514 	new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
3515 
3516 	if (lockres->l_ops->check_downconvert
3517 	    && !lockres->l_ops->check_downconvert(lockres, new_level)) {
3518 		mlog(ML_BASTS, "lockres %s, ReQ: Checkpointing\n",
3519 		     lockres->l_name);
3520 		goto leave_requeue;
3521 	}
3522 
3523 	/* If we get here, then we know that there are no more
3524 	 * incompatible holders (and anyone asking for an incompatible
3525 	 * lock is blocked). We can now downconvert the lock */
3526 	if (!lockres->l_ops->downconvert_worker)
3527 		goto downconvert;
3528 
3529 	/* Some lockres types want to do a bit of work before
3530 	 * downconverting a lock. Allow that here. The worker function
3531 	 * may sleep, so we save off a copy of what we're blocking as
3532 	 * it may change while we're not holding the spin lock. */
3533 	blocking = lockres->l_blocking;
3534 	level = lockres->l_level;
3535 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3536 
3537 	ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
3538 
3539 	if (ctl->unblock_action == UNBLOCK_STOP_POST) {
3540 		mlog(ML_BASTS, "lockres %s, UNBLOCK_STOP_POST\n",
3541 		     lockres->l_name);
3542 		goto leave;
3543 	}
3544 
3545 	spin_lock_irqsave(&lockres->l_lock, flags);
3546 	if ((blocking != lockres->l_blocking) || (level != lockres->l_level)) {
3547 		/* If this changed underneath us, then we can't drop
3548 		 * it just yet. */
3549 		mlog(ML_BASTS, "lockres %s, block=%d:%d, level=%d:%d, "
3550 		     "Recheck\n", lockres->l_name, blocking,
3551 		     lockres->l_blocking, level, lockres->l_level);
3552 		goto recheck;
3553 	}
3554 
3555 downconvert:
3556 	ctl->requeue = 0;
3557 
3558 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3559 		if (lockres->l_level == DLM_LOCK_EX)
3560 			set_lvb = 1;
3561 
3562 		/*
3563 		 * We only set the lvb if the lock has been fully
3564 		 * refreshed - otherwise we risk setting stale
3565 		 * data. Otherwise, there's no need to actually clear
3566 		 * out the lvb here as it's value is still valid.
3567 		 */
3568 		if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3569 			lockres->l_ops->set_lvb(lockres);
3570 	}
3571 
3572 	gen = ocfs2_prepare_downconvert(lockres, new_level);
3573 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3574 	ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
3575 				     gen);
3576 
3577 leave:
3578 	if (ret)
3579 		mlog_errno(ret);
3580 	return ret;
3581 
3582 leave_requeue:
3583 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3584 	ctl->requeue = 1;
3585 
3586 	return 0;
3587 }
3588 
ocfs2_data_convert_worker(struct ocfs2_lock_res * lockres,int blocking)3589 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
3590 				     int blocking)
3591 {
3592 	struct inode *inode;
3593 	struct address_space *mapping;
3594 	struct ocfs2_inode_info *oi;
3595 
3596        	inode = ocfs2_lock_res_inode(lockres);
3597 	mapping = inode->i_mapping;
3598 
3599 	if (S_ISDIR(inode->i_mode)) {
3600 		oi = OCFS2_I(inode);
3601 		oi->ip_dir_lock_gen++;
3602 		mlog(0, "generation: %u\n", oi->ip_dir_lock_gen);
3603 		goto out;
3604 	}
3605 
3606 	if (!S_ISREG(inode->i_mode))
3607 		goto out;
3608 
3609 	/*
3610 	 * We need this before the filemap_fdatawrite() so that it can
3611 	 * transfer the dirty bit from the PTE to the
3612 	 * page. Unfortunately this means that even for EX->PR
3613 	 * downconverts, we'll lose our mappings and have to build
3614 	 * them up again.
3615 	 */
3616 	unmap_mapping_range(mapping, 0, 0, 0);
3617 
3618 	if (filemap_fdatawrite(mapping)) {
3619 		mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
3620 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
3621 	}
3622 	sync_mapping_buffers(mapping);
3623 	if (blocking == DLM_LOCK_EX) {
3624 		truncate_inode_pages(mapping, 0);
3625 	} else {
3626 		/* We only need to wait on the I/O if we're not also
3627 		 * truncating pages because truncate_inode_pages waits
3628 		 * for us above. We don't truncate pages if we're
3629 		 * blocking anything < EXMODE because we want to keep
3630 		 * them around in that case. */
3631 		filemap_fdatawait(mapping);
3632 	}
3633 
3634 out:
3635 	return UNBLOCK_CONTINUE;
3636 }
3637 
ocfs2_ci_checkpointed(struct ocfs2_caching_info * ci,struct ocfs2_lock_res * lockres,int new_level)3638 static int ocfs2_ci_checkpointed(struct ocfs2_caching_info *ci,
3639 				 struct ocfs2_lock_res *lockres,
3640 				 int new_level)
3641 {
3642 	int checkpointed = ocfs2_ci_fully_checkpointed(ci);
3643 
3644 	BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
3645 	BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
3646 
3647 	if (checkpointed)
3648 		return 1;
3649 
3650 	ocfs2_start_checkpoint(OCFS2_SB(ocfs2_metadata_cache_get_super(ci)));
3651 	return 0;
3652 }
3653 
ocfs2_check_meta_downconvert(struct ocfs2_lock_res * lockres,int new_level)3654 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
3655 					int new_level)
3656 {
3657 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3658 
3659 	return ocfs2_ci_checkpointed(INODE_CACHE(inode), lockres, new_level);
3660 }
3661 
ocfs2_set_meta_lvb(struct ocfs2_lock_res * lockres)3662 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
3663 {
3664 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3665 
3666 	__ocfs2_stuff_meta_lvb(inode);
3667 }
3668 
3669 /*
3670  * Does the final reference drop on our dentry lock. Right now this
3671  * happens in the downconvert thread, but we could choose to simplify the
3672  * dlmglue API and push these off to the ocfs2_wq in the future.
3673  */
ocfs2_dentry_post_unlock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)3674 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
3675 				     struct ocfs2_lock_res *lockres)
3676 {
3677 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3678 	ocfs2_dentry_lock_put(osb, dl);
3679 }
3680 
3681 /*
3682  * d_delete() matching dentries before the lock downconvert.
3683  *
3684  * At this point, any process waiting to destroy the
3685  * dentry_lock due to last ref count is stopped by the
3686  * OCFS2_LOCK_QUEUED flag.
3687  *
3688  * We have two potential problems
3689  *
3690  * 1) If we do the last reference drop on our dentry_lock (via dput)
3691  *    we'll wind up in ocfs2_release_dentry_lock(), waiting on
3692  *    the downconvert to finish. Instead we take an elevated
3693  *    reference and push the drop until after we've completed our
3694  *    unblock processing.
3695  *
3696  * 2) There might be another process with a final reference,
3697  *    waiting on us to finish processing. If this is the case, we
3698  *    detect it and exit out - there's no more dentries anyway.
3699  */
ocfs2_dentry_convert_worker(struct ocfs2_lock_res * lockres,int blocking)3700 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
3701 				       int blocking)
3702 {
3703 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3704 	struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
3705 	struct dentry *dentry;
3706 	unsigned long flags;
3707 	int extra_ref = 0;
3708 
3709 	/*
3710 	 * This node is blocking another node from getting a read
3711 	 * lock. This happens when we've renamed within a
3712 	 * directory. We've forced the other nodes to d_delete(), but
3713 	 * we never actually dropped our lock because it's still
3714 	 * valid. The downconvert code will retain a PR for this node,
3715 	 * so there's no further work to do.
3716 	 */
3717 	if (blocking == DLM_LOCK_PR)
3718 		return UNBLOCK_CONTINUE;
3719 
3720 	/*
3721 	 * Mark this inode as potentially orphaned. The code in
3722 	 * ocfs2_delete_inode() will figure out whether it actually
3723 	 * needs to be freed or not.
3724 	 */
3725 	spin_lock(&oi->ip_lock);
3726 	oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
3727 	spin_unlock(&oi->ip_lock);
3728 
3729 	/*
3730 	 * Yuck. We need to make sure however that the check of
3731 	 * OCFS2_LOCK_FREEING and the extra reference are atomic with
3732 	 * respect to a reference decrement or the setting of that
3733 	 * flag.
3734 	 */
3735 	spin_lock_irqsave(&lockres->l_lock, flags);
3736 	spin_lock(&dentry_attach_lock);
3737 	if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
3738 	    && dl->dl_count) {
3739 		dl->dl_count++;
3740 		extra_ref = 1;
3741 	}
3742 	spin_unlock(&dentry_attach_lock);
3743 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3744 
3745 	mlog(0, "extra_ref = %d\n", extra_ref);
3746 
3747 	/*
3748 	 * We have a process waiting on us in ocfs2_dentry_iput(),
3749 	 * which means we can't have any more outstanding
3750 	 * aliases. There's no need to do any more work.
3751 	 */
3752 	if (!extra_ref)
3753 		return UNBLOCK_CONTINUE;
3754 
3755 	spin_lock(&dentry_attach_lock);
3756 	while (1) {
3757 		dentry = ocfs2_find_local_alias(dl->dl_inode,
3758 						dl->dl_parent_blkno, 1);
3759 		if (!dentry)
3760 			break;
3761 		spin_unlock(&dentry_attach_lock);
3762 
3763 		if (S_ISDIR(dl->dl_inode->i_mode))
3764 			shrink_dcache_parent(dentry);
3765 
3766 		mlog(0, "d_delete(%pd);\n", dentry);
3767 
3768 		/*
3769 		 * The following dcache calls may do an
3770 		 * iput(). Normally we don't want that from the
3771 		 * downconverting thread, but in this case it's ok
3772 		 * because the requesting node already has an
3773 		 * exclusive lock on the inode, so it can't be queued
3774 		 * for a downconvert.
3775 		 */
3776 		d_delete(dentry);
3777 		dput(dentry);
3778 
3779 		spin_lock(&dentry_attach_lock);
3780 	}
3781 	spin_unlock(&dentry_attach_lock);
3782 
3783 	/*
3784 	 * If we are the last holder of this dentry lock, there is no
3785 	 * reason to downconvert so skip straight to the unlock.
3786 	 */
3787 	if (dl->dl_count == 1)
3788 		return UNBLOCK_STOP_POST;
3789 
3790 	return UNBLOCK_CONTINUE_POST;
3791 }
3792 
ocfs2_check_refcount_downconvert(struct ocfs2_lock_res * lockres,int new_level)3793 static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
3794 					    int new_level)
3795 {
3796 	struct ocfs2_refcount_tree *tree =
3797 				ocfs2_lock_res_refcount_tree(lockres);
3798 
3799 	return ocfs2_ci_checkpointed(&tree->rf_ci, lockres, new_level);
3800 }
3801 
ocfs2_refcount_convert_worker(struct ocfs2_lock_res * lockres,int blocking)3802 static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
3803 					 int blocking)
3804 {
3805 	struct ocfs2_refcount_tree *tree =
3806 				ocfs2_lock_res_refcount_tree(lockres);
3807 
3808 	ocfs2_metadata_cache_purge(&tree->rf_ci);
3809 
3810 	return UNBLOCK_CONTINUE;
3811 }
3812 
ocfs2_set_qinfo_lvb(struct ocfs2_lock_res * lockres)3813 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
3814 {
3815 	struct ocfs2_qinfo_lvb *lvb;
3816 	struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres);
3817 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
3818 					    oinfo->dqi_gi.dqi_type);
3819 
3820 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
3821 	lvb->lvb_version = OCFS2_QINFO_LVB_VERSION;
3822 	lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace);
3823 	lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace);
3824 	lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms);
3825 	lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks);
3826 	lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk);
3827 	lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry);
3828 }
3829 
ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo * oinfo,int ex)3830 void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex)
3831 {
3832 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3833 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
3834 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3835 
3836 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
3837 		ocfs2_cluster_unlock(osb, lockres, level);
3838 }
3839 
ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo * oinfo)3840 static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo)
3841 {
3842 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
3843 					    oinfo->dqi_gi.dqi_type);
3844 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3845 	struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
3846 	struct buffer_head *bh = NULL;
3847 	struct ocfs2_global_disk_dqinfo *gdinfo;
3848 	int status = 0;
3849 
3850 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
3851 	    lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) {
3852 		info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace);
3853 		info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace);
3854 		oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms);
3855 		oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks);
3856 		oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk);
3857 		oinfo->dqi_gi.dqi_free_entry =
3858 					be32_to_cpu(lvb->lvb_free_entry);
3859 	} else {
3860 		status = ocfs2_read_quota_phys_block(oinfo->dqi_gqinode,
3861 						     oinfo->dqi_giblk, &bh);
3862 		if (status) {
3863 			mlog_errno(status);
3864 			goto bail;
3865 		}
3866 		gdinfo = (struct ocfs2_global_disk_dqinfo *)
3867 					(bh->b_data + OCFS2_GLOBAL_INFO_OFF);
3868 		info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace);
3869 		info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace);
3870 		oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms);
3871 		oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks);
3872 		oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk);
3873 		oinfo->dqi_gi.dqi_free_entry =
3874 					le32_to_cpu(gdinfo->dqi_free_entry);
3875 		brelse(bh);
3876 		ocfs2_track_lock_refresh(lockres);
3877 	}
3878 
3879 bail:
3880 	return status;
3881 }
3882 
3883 /* Lock quota info, this function expects at least shared lock on the quota file
3884  * so that we can safely refresh quota info from disk. */
ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo * oinfo,int ex)3885 int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex)
3886 {
3887 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3888 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
3889 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3890 	int status = 0;
3891 
3892 	/* On RO devices, locking really isn't needed... */
3893 	if (ocfs2_is_hard_readonly(osb)) {
3894 		if (ex)
3895 			status = -EROFS;
3896 		goto bail;
3897 	}
3898 	if (ocfs2_mount_local(osb))
3899 		goto bail;
3900 
3901 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
3902 	if (status < 0) {
3903 		mlog_errno(status);
3904 		goto bail;
3905 	}
3906 	if (!ocfs2_should_refresh_lock_res(lockres))
3907 		goto bail;
3908 	/* OK, we have the lock but we need to refresh the quota info */
3909 	status = ocfs2_refresh_qinfo(oinfo);
3910 	if (status)
3911 		ocfs2_qinfo_unlock(oinfo, ex);
3912 	ocfs2_complete_lock_res_refresh(lockres, status);
3913 bail:
3914 	return status;
3915 }
3916 
ocfs2_refcount_lock(struct ocfs2_refcount_tree * ref_tree,int ex)3917 int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex)
3918 {
3919 	int status;
3920 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3921 	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
3922 	struct ocfs2_super *osb = lockres->l_priv;
3923 
3924 
3925 	if (ocfs2_is_hard_readonly(osb))
3926 		return -EROFS;
3927 
3928 	if (ocfs2_mount_local(osb))
3929 		return 0;
3930 
3931 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
3932 	if (status < 0)
3933 		mlog_errno(status);
3934 
3935 	return status;
3936 }
3937 
ocfs2_refcount_unlock(struct ocfs2_refcount_tree * ref_tree,int ex)3938 void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex)
3939 {
3940 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3941 	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
3942 	struct ocfs2_super *osb = lockres->l_priv;
3943 
3944 	if (!ocfs2_mount_local(osb))
3945 		ocfs2_cluster_unlock(osb, lockres, level);
3946 }
3947 
ocfs2_process_blocked_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)3948 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3949 				       struct ocfs2_lock_res *lockres)
3950 {
3951 	int status;
3952 	struct ocfs2_unblock_ctl ctl = {0, 0,};
3953 	unsigned long flags;
3954 
3955 	/* Our reference to the lockres in this function can be
3956 	 * considered valid until we remove the OCFS2_LOCK_QUEUED
3957 	 * flag. */
3958 
3959 	BUG_ON(!lockres);
3960 	BUG_ON(!lockres->l_ops);
3961 
3962 	mlog(ML_BASTS, "lockres %s blocked\n", lockres->l_name);
3963 
3964 	/* Detect whether a lock has been marked as going away while
3965 	 * the downconvert thread was processing other things. A lock can
3966 	 * still be marked with OCFS2_LOCK_FREEING after this check,
3967 	 * but short circuiting here will still save us some
3968 	 * performance. */
3969 	spin_lock_irqsave(&lockres->l_lock, flags);
3970 	if (lockres->l_flags & OCFS2_LOCK_FREEING)
3971 		goto unqueue;
3972 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3973 
3974 	status = ocfs2_unblock_lock(osb, lockres, &ctl);
3975 	if (status < 0)
3976 		mlog_errno(status);
3977 
3978 	spin_lock_irqsave(&lockres->l_lock, flags);
3979 unqueue:
3980 	if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
3981 		lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
3982 	} else
3983 		ocfs2_schedule_blocked_lock(osb, lockres);
3984 
3985 	mlog(ML_BASTS, "lockres %s, requeue = %s.\n", lockres->l_name,
3986 	     ctl.requeue ? "yes" : "no");
3987 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3988 
3989 	if (ctl.unblock_action != UNBLOCK_CONTINUE
3990 	    && lockres->l_ops->post_unlock)
3991 		lockres->l_ops->post_unlock(osb, lockres);
3992 }
3993 
ocfs2_schedule_blocked_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)3994 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
3995 					struct ocfs2_lock_res *lockres)
3996 {
3997 	unsigned long flags;
3998 
3999 	assert_spin_locked(&lockres->l_lock);
4000 
4001 	if (lockres->l_flags & OCFS2_LOCK_FREEING) {
4002 		/* Do not schedule a lock for downconvert when it's on
4003 		 * the way to destruction - any nodes wanting access
4004 		 * to the resource will get it soon. */
4005 		mlog(ML_BASTS, "lockres %s won't be scheduled: flags 0x%lx\n",
4006 		     lockres->l_name, lockres->l_flags);
4007 		return;
4008 	}
4009 
4010 	lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
4011 
4012 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4013 	if (list_empty(&lockres->l_blocked_list)) {
4014 		list_add_tail(&lockres->l_blocked_list,
4015 			      &osb->blocked_lock_list);
4016 		osb->blocked_lock_count++;
4017 	}
4018 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4019 }
4020 
ocfs2_downconvert_thread_do_work(struct ocfs2_super * osb)4021 static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
4022 {
4023 	unsigned long processed;
4024 	unsigned long flags;
4025 	struct ocfs2_lock_res *lockres;
4026 
4027 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4028 	/* grab this early so we know to try again if a state change and
4029 	 * wake happens part-way through our work  */
4030 	osb->dc_work_sequence = osb->dc_wake_sequence;
4031 
4032 	processed = osb->blocked_lock_count;
4033 	/*
4034 	 * blocked lock processing in this loop might call iput which can
4035 	 * remove items off osb->blocked_lock_list. Downconvert up to
4036 	 * 'processed' number of locks, but stop short if we had some
4037 	 * removed in ocfs2_mark_lockres_freeing when downconverting.
4038 	 */
4039 	while (processed && !list_empty(&osb->blocked_lock_list)) {
4040 		lockres = list_entry(osb->blocked_lock_list.next,
4041 				     struct ocfs2_lock_res, l_blocked_list);
4042 		list_del_init(&lockres->l_blocked_list);
4043 		osb->blocked_lock_count--;
4044 		spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4045 
4046 		BUG_ON(!processed);
4047 		processed--;
4048 
4049 		ocfs2_process_blocked_lock(osb, lockres);
4050 
4051 		spin_lock_irqsave(&osb->dc_task_lock, flags);
4052 	}
4053 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4054 }
4055 
ocfs2_downconvert_thread_lists_empty(struct ocfs2_super * osb)4056 static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
4057 {
4058 	int empty = 0;
4059 	unsigned long flags;
4060 
4061 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4062 	if (list_empty(&osb->blocked_lock_list))
4063 		empty = 1;
4064 
4065 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4066 	return empty;
4067 }
4068 
ocfs2_downconvert_thread_should_wake(struct ocfs2_super * osb)4069 static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
4070 {
4071 	int should_wake = 0;
4072 	unsigned long flags;
4073 
4074 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4075 	if (osb->dc_work_sequence != osb->dc_wake_sequence)
4076 		should_wake = 1;
4077 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4078 
4079 	return should_wake;
4080 }
4081 
ocfs2_downconvert_thread(void * arg)4082 static int ocfs2_downconvert_thread(void *arg)
4083 {
4084 	int status = 0;
4085 	struct ocfs2_super *osb = arg;
4086 
4087 	/* only quit once we've been asked to stop and there is no more
4088 	 * work available */
4089 	while (!(kthread_should_stop() &&
4090 		ocfs2_downconvert_thread_lists_empty(osb))) {
4091 
4092 		wait_event_interruptible(osb->dc_event,
4093 					 ocfs2_downconvert_thread_should_wake(osb) ||
4094 					 kthread_should_stop());
4095 
4096 		mlog(0, "downconvert_thread: awoken\n");
4097 
4098 		ocfs2_downconvert_thread_do_work(osb);
4099 	}
4100 
4101 	osb->dc_task = NULL;
4102 	return status;
4103 }
4104 
ocfs2_wake_downconvert_thread(struct ocfs2_super * osb)4105 void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
4106 {
4107 	unsigned long flags;
4108 
4109 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4110 	/* make sure the voting thread gets a swipe at whatever changes
4111 	 * the caller may have made to the voting state */
4112 	osb->dc_wake_sequence++;
4113 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4114 	wake_up(&osb->dc_event);
4115 }
4116