1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * dlmglue.c
5  *
6  * Code which implements an OCFS2 specific interface to our DLM.
7  *
8  * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation; either
13  * version 2 of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public
21  * License along with this program; if not, write to the
22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23  * Boston, MA 021110-1307, USA.
24  */
25 
26 #include <linux/types.h>
27 #include <linux/slab.h>
28 #include <linux/highmem.h>
29 #include <linux/mm.h>
30 #include <linux/kthread.h>
31 #include <linux/pagemap.h>
32 #include <linux/debugfs.h>
33 #include <linux/seq_file.h>
34 #include <linux/time.h>
35 #include <linux/quotaops.h>
36 #include <linux/sched/signal.h>
37 
38 #define MLOG_MASK_PREFIX ML_DLM_GLUE
39 #include <cluster/masklog.h>
40 
41 #include "ocfs2.h"
42 #include "ocfs2_lockingver.h"
43 
44 #include "alloc.h"
45 #include "dcache.h"
46 #include "dlmglue.h"
47 #include "extent_map.h"
48 #include "file.h"
49 #include "heartbeat.h"
50 #include "inode.h"
51 #include "journal.h"
52 #include "stackglue.h"
53 #include "slot_map.h"
54 #include "super.h"
55 #include "uptodate.h"
56 #include "quota.h"
57 #include "refcounttree.h"
58 #include "acl.h"
59 
60 #include "buffer_head_io.h"
61 
62 struct ocfs2_mask_waiter {
63 	struct list_head	mw_item;
64 	int			mw_status;
65 	struct completion	mw_complete;
66 	unsigned long		mw_mask;
67 	unsigned long		mw_goal;
68 #ifdef CONFIG_OCFS2_FS_STATS
69 	ktime_t			mw_lock_start;
70 #endif
71 };
72 
73 static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres);
74 static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres);
75 static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres);
76 static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres);
77 
78 /*
79  * Return value from ->downconvert_worker functions.
80  *
81  * These control the precise actions of ocfs2_unblock_lock()
82  * and ocfs2_process_blocked_lock()
83  *
84  */
85 enum ocfs2_unblock_action {
86 	UNBLOCK_CONTINUE	= 0, /* Continue downconvert */
87 	UNBLOCK_CONTINUE_POST	= 1, /* Continue downconvert, fire
88 				      * ->post_unlock callback */
89 	UNBLOCK_STOP_POST	= 2, /* Do not downconvert, fire
90 				      * ->post_unlock() callback. */
91 };
92 
93 struct ocfs2_unblock_ctl {
94 	int requeue;
95 	enum ocfs2_unblock_action unblock_action;
96 };
97 
98 /* Lockdep class keys */
99 #ifdef CONFIG_DEBUG_LOCK_ALLOC
100 static struct lock_class_key lockdep_keys[OCFS2_NUM_LOCK_TYPES];
101 #endif
102 
103 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
104 					int new_level);
105 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);
106 
107 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
108 				     int blocking);
109 
110 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
111 				       int blocking);
112 
113 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
114 				     struct ocfs2_lock_res *lockres);
115 
116 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres);
117 
118 static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
119 					    int new_level);
120 static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
121 					 int blocking);
122 
123 #define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
124 
125 /* 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)126 static void ocfs2_dump_meta_lvb_info(u64 level,
127 				     const char *function,
128 				     unsigned int line,
129 				     struct ocfs2_lock_res *lockres)
130 {
131 	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
132 
133 	mlog(level, "LVB information for %s (called from %s:%u):\n",
134 	     lockres->l_name, function, line);
135 	mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
136 	     lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
137 	     be32_to_cpu(lvb->lvb_igeneration));
138 	mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
139 	     (unsigned long long)be64_to_cpu(lvb->lvb_isize),
140 	     be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
141 	     be16_to_cpu(lvb->lvb_imode));
142 	mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
143 	     "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
144 	     (long long)be64_to_cpu(lvb->lvb_iatime_packed),
145 	     (long long)be64_to_cpu(lvb->lvb_ictime_packed),
146 	     (long long)be64_to_cpu(lvb->lvb_imtime_packed),
147 	     be32_to_cpu(lvb->lvb_iattr));
148 }
149 
150 
151 /*
152  * OCFS2 Lock Resource Operations
153  *
154  * These fine tune the behavior of the generic dlmglue locking infrastructure.
155  *
156  * The most basic of lock types can point ->l_priv to their respective
157  * struct ocfs2_super and allow the default actions to manage things.
158  *
159  * Right now, each lock type also needs to implement an init function,
160  * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
161  * should be called when the lock is no longer needed (i.e., object
162  * destruction time).
163  */
164 struct ocfs2_lock_res_ops {
165 	/*
166 	 * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
167 	 * this callback if ->l_priv is not an ocfs2_super pointer
168 	 */
169 	struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
170 
171 	/*
172 	 * Optionally called in the downconvert thread after a
173 	 * successful downconvert. The lockres will not be referenced
174 	 * after this callback is called, so it is safe to free
175 	 * memory, etc.
176 	 *
177 	 * The exact semantics of when this is called are controlled
178 	 * by ->downconvert_worker()
179 	 */
180 	void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
181 
182 	/*
183 	 * Allow a lock type to add checks to determine whether it is
184 	 * safe to downconvert a lock. Return 0 to re-queue the
185 	 * downconvert at a later time, nonzero to continue.
186 	 *
187 	 * For most locks, the default checks that there are no
188 	 * incompatible holders are sufficient.
189 	 *
190 	 * Called with the lockres spinlock held.
191 	 */
192 	int (*check_downconvert)(struct ocfs2_lock_res *, int);
193 
194 	/*
195 	 * Allows a lock type to populate the lock value block. This
196 	 * is called on downconvert, and when we drop a lock.
197 	 *
198 	 * Locks that want to use this should set LOCK_TYPE_USES_LVB
199 	 * in the flags field.
200 	 *
201 	 * Called with the lockres spinlock held.
202 	 */
203 	void (*set_lvb)(struct ocfs2_lock_res *);
204 
205 	/*
206 	 * Called from the downconvert thread when it is determined
207 	 * that a lock will be downconverted. This is called without
208 	 * any locks held so the function can do work that might
209 	 * schedule (syncing out data, etc).
210 	 *
211 	 * This should return any one of the ocfs2_unblock_action
212 	 * values, depending on what it wants the thread to do.
213 	 */
214 	int (*downconvert_worker)(struct ocfs2_lock_res *, int);
215 
216 	/*
217 	 * LOCK_TYPE_* flags which describe the specific requirements
218 	 * of a lock type. Descriptions of each individual flag follow.
219 	 */
220 	int flags;
221 };
222 
223 /*
224  * Some locks want to "refresh" potentially stale data when a
225  * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
226  * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
227  * individual lockres l_flags member from the ast function. It is
228  * expected that the locking wrapper will clear the
229  * OCFS2_LOCK_NEEDS_REFRESH flag when done.
230  */
231 #define LOCK_TYPE_REQUIRES_REFRESH 0x1
232 
233 /*
234  * Indicate that a lock type makes use of the lock value block. The
235  * ->set_lvb lock type callback must be defined.
236  */
237 #define LOCK_TYPE_USES_LVB		0x2
238 
239 static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
240 	.get_osb	= ocfs2_get_inode_osb,
241 	.flags		= 0,
242 };
243 
244 static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = {
245 	.get_osb	= ocfs2_get_inode_osb,
246 	.check_downconvert = ocfs2_check_meta_downconvert,
247 	.set_lvb	= ocfs2_set_meta_lvb,
248 	.downconvert_worker = ocfs2_data_convert_worker,
249 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
250 };
251 
252 static struct ocfs2_lock_res_ops ocfs2_super_lops = {
253 	.flags		= LOCK_TYPE_REQUIRES_REFRESH,
254 };
255 
256 static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
257 	.flags		= 0,
258 };
259 
260 static struct ocfs2_lock_res_ops ocfs2_nfs_sync_lops = {
261 	.flags		= 0,
262 };
263 
264 static struct ocfs2_lock_res_ops ocfs2_trim_fs_lops = {
265 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
266 };
267 
268 static struct ocfs2_lock_res_ops ocfs2_orphan_scan_lops = {
269 	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
270 };
271 
272 static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
273 	.get_osb	= ocfs2_get_dentry_osb,
274 	.post_unlock	= ocfs2_dentry_post_unlock,
275 	.downconvert_worker = ocfs2_dentry_convert_worker,
276 	.flags		= 0,
277 };
278 
279 static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
280 	.get_osb	= ocfs2_get_inode_osb,
281 	.flags		= 0,
282 };
283 
284 static struct ocfs2_lock_res_ops ocfs2_flock_lops = {
285 	.get_osb	= ocfs2_get_file_osb,
286 	.flags		= 0,
287 };
288 
289 static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = {
290 	.set_lvb	= ocfs2_set_qinfo_lvb,
291 	.get_osb	= ocfs2_get_qinfo_osb,
292 	.flags		= LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB,
293 };
294 
295 static struct ocfs2_lock_res_ops ocfs2_refcount_block_lops = {
296 	.check_downconvert = ocfs2_check_refcount_downconvert,
297 	.downconvert_worker = ocfs2_refcount_convert_worker,
298 	.flags		= 0,
299 };
300 
ocfs2_is_inode_lock(struct ocfs2_lock_res * lockres)301 static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
302 {
303 	return lockres->l_type == OCFS2_LOCK_TYPE_META ||
304 		lockres->l_type == OCFS2_LOCK_TYPE_RW ||
305 		lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
306 }
307 
ocfs2_lksb_to_lock_res(struct ocfs2_dlm_lksb * lksb)308 static inline struct ocfs2_lock_res *ocfs2_lksb_to_lock_res(struct ocfs2_dlm_lksb *lksb)
309 {
310 	return container_of(lksb, struct ocfs2_lock_res, l_lksb);
311 }
312 
ocfs2_lock_res_inode(struct ocfs2_lock_res * lockres)313 static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
314 {
315 	BUG_ON(!ocfs2_is_inode_lock(lockres));
316 
317 	return (struct inode *) lockres->l_priv;
318 }
319 
ocfs2_lock_res_dl(struct ocfs2_lock_res * lockres)320 static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
321 {
322 	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
323 
324 	return (struct ocfs2_dentry_lock *)lockres->l_priv;
325 }
326 
ocfs2_lock_res_qinfo(struct ocfs2_lock_res * lockres)327 static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres)
328 {
329 	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO);
330 
331 	return (struct ocfs2_mem_dqinfo *)lockres->l_priv;
332 }
333 
334 static inline struct ocfs2_refcount_tree *
ocfs2_lock_res_refcount_tree(struct ocfs2_lock_res * res)335 ocfs2_lock_res_refcount_tree(struct ocfs2_lock_res *res)
336 {
337 	return container_of(res, struct ocfs2_refcount_tree, rf_lockres);
338 }
339 
ocfs2_get_lockres_osb(struct ocfs2_lock_res * lockres)340 static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
341 {
342 	if (lockres->l_ops->get_osb)
343 		return lockres->l_ops->get_osb(lockres);
344 
345 	return (struct ocfs2_super *)lockres->l_priv;
346 }
347 
348 static int ocfs2_lock_create(struct ocfs2_super *osb,
349 			     struct ocfs2_lock_res *lockres,
350 			     int level,
351 			     u32 dlm_flags);
352 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
353 						     int wanted);
354 static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
355 				   struct ocfs2_lock_res *lockres,
356 				   int level, unsigned long caller_ip);
ocfs2_cluster_unlock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int level)357 static inline void ocfs2_cluster_unlock(struct ocfs2_super *osb,
358 					struct ocfs2_lock_res *lockres,
359 					int level)
360 {
361 	__ocfs2_cluster_unlock(osb, lockres, level, _RET_IP_);
362 }
363 
364 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
365 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
366 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
367 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
368 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
369 					struct ocfs2_lock_res *lockres);
370 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
371 						int convert);
372 #define ocfs2_log_dlm_error(_func, _err, _lockres) do {					\
373 	if ((_lockres)->l_type != OCFS2_LOCK_TYPE_DENTRY)				\
374 		mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n",	\
375 		     _err, _func, _lockres->l_name);					\
376 	else										\
377 		mlog(ML_ERROR, "DLM error %d while calling %s on resource %.*s%08x\n",	\
378 		     _err, _func, OCFS2_DENTRY_LOCK_INO_START - 1, (_lockres)->l_name,	\
379 		     (unsigned int)ocfs2_get_dentry_lock_ino(_lockres));		\
380 } while (0)
381 static int ocfs2_downconvert_thread(void *arg);
382 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
383 					struct ocfs2_lock_res *lockres);
384 static int ocfs2_inode_lock_update(struct inode *inode,
385 				  struct buffer_head **bh);
386 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
387 static inline int ocfs2_highest_compat_lock_level(int level);
388 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
389 					      int new_level);
390 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
391 				  struct ocfs2_lock_res *lockres,
392 				  int new_level,
393 				  int lvb,
394 				  unsigned int generation);
395 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
396 				        struct ocfs2_lock_res *lockres);
397 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
398 				struct ocfs2_lock_res *lockres);
399 
400 
ocfs2_build_lock_name(enum ocfs2_lock_type type,u64 blkno,u32 generation,char * name)401 static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
402 				  u64 blkno,
403 				  u32 generation,
404 				  char *name)
405 {
406 	int len;
407 
408 	BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
409 
410 	len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
411 		       ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
412 		       (long long)blkno, generation);
413 
414 	BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
415 
416 	mlog(0, "built lock resource with name: %s\n", name);
417 }
418 
419 static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
420 
ocfs2_add_lockres_tracking(struct ocfs2_lock_res * res,struct ocfs2_dlm_debug * dlm_debug)421 static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
422 				       struct ocfs2_dlm_debug *dlm_debug)
423 {
424 	mlog(0, "Add tracking for lockres %s\n", res->l_name);
425 
426 	spin_lock(&ocfs2_dlm_tracking_lock);
427 	list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
428 	spin_unlock(&ocfs2_dlm_tracking_lock);
429 }
430 
ocfs2_remove_lockres_tracking(struct ocfs2_lock_res * res)431 static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
432 {
433 	spin_lock(&ocfs2_dlm_tracking_lock);
434 	if (!list_empty(&res->l_debug_list))
435 		list_del_init(&res->l_debug_list);
436 	spin_unlock(&ocfs2_dlm_tracking_lock);
437 }
438 
439 #ifdef CONFIG_OCFS2_FS_STATS
ocfs2_init_lock_stats(struct ocfs2_lock_res * res)440 static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
441 {
442 	res->l_lock_refresh = 0;
443 	memset(&res->l_lock_prmode, 0, sizeof(struct ocfs2_lock_stats));
444 	memset(&res->l_lock_exmode, 0, sizeof(struct ocfs2_lock_stats));
445 }
446 
ocfs2_update_lock_stats(struct ocfs2_lock_res * res,int level,struct ocfs2_mask_waiter * mw,int ret)447 static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level,
448 				    struct ocfs2_mask_waiter *mw, int ret)
449 {
450 	u32 usec;
451 	ktime_t kt;
452 	struct ocfs2_lock_stats *stats;
453 
454 	if (level == LKM_PRMODE)
455 		stats = &res->l_lock_prmode;
456 	else if (level == LKM_EXMODE)
457 		stats = &res->l_lock_exmode;
458 	else
459 		return;
460 
461 	kt = ktime_sub(ktime_get(), mw->mw_lock_start);
462 	usec = ktime_to_us(kt);
463 
464 	stats->ls_gets++;
465 	stats->ls_total += ktime_to_ns(kt);
466 	/* overflow */
467 	if (unlikely(stats->ls_gets == 0)) {
468 		stats->ls_gets++;
469 		stats->ls_total = ktime_to_ns(kt);
470 	}
471 
472 	if (stats->ls_max < usec)
473 		stats->ls_max = usec;
474 
475 	if (ret)
476 		stats->ls_fail++;
477 }
478 
ocfs2_track_lock_refresh(struct ocfs2_lock_res * lockres)479 static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
480 {
481 	lockres->l_lock_refresh++;
482 }
483 
ocfs2_init_start_time(struct ocfs2_mask_waiter * mw)484 static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
485 {
486 	mw->mw_lock_start = ktime_get();
487 }
488 #else
ocfs2_init_lock_stats(struct ocfs2_lock_res * res)489 static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
490 {
491 }
ocfs2_update_lock_stats(struct ocfs2_lock_res * res,int level,struct ocfs2_mask_waiter * mw,int ret)492 static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res,
493 			   int level, struct ocfs2_mask_waiter *mw, int ret)
494 {
495 }
ocfs2_track_lock_refresh(struct ocfs2_lock_res * lockres)496 static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
497 {
498 }
ocfs2_init_start_time(struct ocfs2_mask_waiter * mw)499 static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
500 {
501 }
502 #endif
503 
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)504 static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
505 				       struct ocfs2_lock_res *res,
506 				       enum ocfs2_lock_type type,
507 				       struct ocfs2_lock_res_ops *ops,
508 				       void *priv)
509 {
510 	res->l_type          = type;
511 	res->l_ops           = ops;
512 	res->l_priv          = priv;
513 
514 	res->l_level         = DLM_LOCK_IV;
515 	res->l_requested     = DLM_LOCK_IV;
516 	res->l_blocking      = DLM_LOCK_IV;
517 	res->l_action        = OCFS2_AST_INVALID;
518 	res->l_unlock_action = OCFS2_UNLOCK_INVALID;
519 
520 	res->l_flags         = OCFS2_LOCK_INITIALIZED;
521 
522 	ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
523 
524 	ocfs2_init_lock_stats(res);
525 #ifdef CONFIG_DEBUG_LOCK_ALLOC
526 	if (type != OCFS2_LOCK_TYPE_OPEN)
527 		lockdep_init_map(&res->l_lockdep_map, ocfs2_lock_type_strings[type],
528 				 &lockdep_keys[type], 0);
529 	else
530 		res->l_lockdep_map.key = NULL;
531 #endif
532 }
533 
ocfs2_lock_res_init_once(struct ocfs2_lock_res * res)534 void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
535 {
536 	/* This also clears out the lock status block */
537 	memset(res, 0, sizeof(struct ocfs2_lock_res));
538 	spin_lock_init(&res->l_lock);
539 	init_waitqueue_head(&res->l_event);
540 	INIT_LIST_HEAD(&res->l_blocked_list);
541 	INIT_LIST_HEAD(&res->l_mask_waiters);
542 	INIT_LIST_HEAD(&res->l_holders);
543 }
544 
ocfs2_inode_lock_res_init(struct ocfs2_lock_res * res,enum ocfs2_lock_type type,unsigned int generation,struct inode * inode)545 void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
546 			       enum ocfs2_lock_type type,
547 			       unsigned int generation,
548 			       struct inode *inode)
549 {
550 	struct ocfs2_lock_res_ops *ops;
551 
552 	switch(type) {
553 		case OCFS2_LOCK_TYPE_RW:
554 			ops = &ocfs2_inode_rw_lops;
555 			break;
556 		case OCFS2_LOCK_TYPE_META:
557 			ops = &ocfs2_inode_inode_lops;
558 			break;
559 		case OCFS2_LOCK_TYPE_OPEN:
560 			ops = &ocfs2_inode_open_lops;
561 			break;
562 		default:
563 			mlog_bug_on_msg(1, "type: %d\n", type);
564 			ops = NULL; /* thanks, gcc */
565 			break;
566 	};
567 
568 	ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
569 			      generation, res->l_name);
570 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
571 }
572 
ocfs2_get_inode_osb(struct ocfs2_lock_res * lockres)573 static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
574 {
575 	struct inode *inode = ocfs2_lock_res_inode(lockres);
576 
577 	return OCFS2_SB(inode->i_sb);
578 }
579 
ocfs2_get_qinfo_osb(struct ocfs2_lock_res * lockres)580 static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres)
581 {
582 	struct ocfs2_mem_dqinfo *info = lockres->l_priv;
583 
584 	return OCFS2_SB(info->dqi_gi.dqi_sb);
585 }
586 
ocfs2_get_file_osb(struct ocfs2_lock_res * lockres)587 static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres)
588 {
589 	struct ocfs2_file_private *fp = lockres->l_priv;
590 
591 	return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb);
592 }
593 
ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res * lockres)594 static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
595 {
596 	__be64 inode_blkno_be;
597 
598 	memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
599 	       sizeof(__be64));
600 
601 	return be64_to_cpu(inode_blkno_be);
602 }
603 
ocfs2_get_dentry_osb(struct ocfs2_lock_res * lockres)604 static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
605 {
606 	struct ocfs2_dentry_lock *dl = lockres->l_priv;
607 
608 	return OCFS2_SB(dl->dl_inode->i_sb);
609 }
610 
ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock * dl,u64 parent,struct inode * inode)611 void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
612 				u64 parent, struct inode *inode)
613 {
614 	int len;
615 	u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
616 	__be64 inode_blkno_be = cpu_to_be64(inode_blkno);
617 	struct ocfs2_lock_res *lockres = &dl->dl_lockres;
618 
619 	ocfs2_lock_res_init_once(lockres);
620 
621 	/*
622 	 * Unfortunately, the standard lock naming scheme won't work
623 	 * here because we have two 16 byte values to use. Instead,
624 	 * we'll stuff the inode number as a binary value. We still
625 	 * want error prints to show something without garbling the
626 	 * display, so drop a null byte in there before the inode
627 	 * number. A future version of OCFS2 will likely use all
628 	 * binary lock names. The stringified names have been a
629 	 * tremendous aid in debugging, but now that the debugfs
630 	 * interface exists, we can mangle things there if need be.
631 	 *
632 	 * NOTE: We also drop the standard "pad" value (the total lock
633 	 * name size stays the same though - the last part is all
634 	 * zeros due to the memset in ocfs2_lock_res_init_once()
635 	 */
636 	len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
637 		       "%c%016llx",
638 		       ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
639 		       (long long)parent);
640 
641 	BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
642 
643 	memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
644 	       sizeof(__be64));
645 
646 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
647 				   OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
648 				   dl);
649 }
650 
ocfs2_super_lock_res_init(struct ocfs2_lock_res * res,struct ocfs2_super * osb)651 static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
652 				      struct ocfs2_super *osb)
653 {
654 	/* Superblock lockres doesn't come from a slab so we call init
655 	 * once on it manually.  */
656 	ocfs2_lock_res_init_once(res);
657 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
658 			      0, res->l_name);
659 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
660 				   &ocfs2_super_lops, osb);
661 }
662 
ocfs2_rename_lock_res_init(struct ocfs2_lock_res * res,struct ocfs2_super * osb)663 static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
664 				       struct ocfs2_super *osb)
665 {
666 	/* Rename lockres doesn't come from a slab so we call init
667 	 * once on it manually.  */
668 	ocfs2_lock_res_init_once(res);
669 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
670 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
671 				   &ocfs2_rename_lops, osb);
672 }
673 
ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res * res,struct ocfs2_super * osb)674 static void ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res *res,
675 					 struct ocfs2_super *osb)
676 {
677 	/* nfs_sync lockres doesn't come from a slab so we call init
678 	 * once on it manually.  */
679 	ocfs2_lock_res_init_once(res);
680 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_NFS_SYNC, 0, 0, res->l_name);
681 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_NFS_SYNC,
682 				   &ocfs2_nfs_sync_lops, osb);
683 }
684 
ocfs2_trim_fs_lock_res_init(struct ocfs2_super * osb)685 void ocfs2_trim_fs_lock_res_init(struct ocfs2_super *osb)
686 {
687 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
688 
689 	ocfs2_lock_res_init_once(lockres);
690 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_TRIM_FS, 0, 0, lockres->l_name);
691 	ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_TRIM_FS,
692 				   &ocfs2_trim_fs_lops, osb);
693 }
694 
ocfs2_trim_fs_lock_res_uninit(struct ocfs2_super * osb)695 void ocfs2_trim_fs_lock_res_uninit(struct ocfs2_super *osb)
696 {
697 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
698 
699 	ocfs2_simple_drop_lockres(osb, lockres);
700 	ocfs2_lock_res_free(lockres);
701 }
702 
ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res * res,struct ocfs2_super * osb)703 static void ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res *res,
704 					    struct ocfs2_super *osb)
705 {
706 	ocfs2_lock_res_init_once(res);
707 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_ORPHAN_SCAN, 0, 0, res->l_name);
708 	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_ORPHAN_SCAN,
709 				   &ocfs2_orphan_scan_lops, osb);
710 }
711 
ocfs2_file_lock_res_init(struct ocfs2_lock_res * lockres,struct ocfs2_file_private * fp)712 void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
713 			      struct ocfs2_file_private *fp)
714 {
715 	struct inode *inode = fp->fp_file->f_mapping->host;
716 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
717 
718 	ocfs2_lock_res_init_once(lockres);
719 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
720 			      inode->i_generation, lockres->l_name);
721 	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
722 				   OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
723 				   fp);
724 	lockres->l_flags |= OCFS2_LOCK_NOCACHE;
725 }
726 
ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res * lockres,struct ocfs2_mem_dqinfo * info)727 void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
728 			       struct ocfs2_mem_dqinfo *info)
729 {
730 	ocfs2_lock_res_init_once(lockres);
731 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type,
732 			      0, lockres->l_name);
733 	ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres,
734 				   OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops,
735 				   info);
736 }
737 
ocfs2_refcount_lock_res_init(struct ocfs2_lock_res * lockres,struct ocfs2_super * osb,u64 ref_blkno,unsigned int generation)738 void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres,
739 				  struct ocfs2_super *osb, u64 ref_blkno,
740 				  unsigned int generation)
741 {
742 	ocfs2_lock_res_init_once(lockres);
743 	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_REFCOUNT, ref_blkno,
744 			      generation, lockres->l_name);
745 	ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_REFCOUNT,
746 				   &ocfs2_refcount_block_lops, osb);
747 }
748 
ocfs2_lock_res_free(struct ocfs2_lock_res * res)749 void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
750 {
751 	if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
752 		return;
753 
754 	ocfs2_remove_lockres_tracking(res);
755 
756 	mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
757 			"Lockres %s is on the blocked list\n",
758 			res->l_name);
759 	mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
760 			"Lockres %s has mask waiters pending\n",
761 			res->l_name);
762 	mlog_bug_on_msg(spin_is_locked(&res->l_lock),
763 			"Lockres %s is locked\n",
764 			res->l_name);
765 	mlog_bug_on_msg(res->l_ro_holders,
766 			"Lockres %s has %u ro holders\n",
767 			res->l_name, res->l_ro_holders);
768 	mlog_bug_on_msg(res->l_ex_holders,
769 			"Lockres %s has %u ex holders\n",
770 			res->l_name, res->l_ex_holders);
771 
772 	/* Need to clear out the lock status block for the dlm */
773 	memset(&res->l_lksb, 0, sizeof(res->l_lksb));
774 
775 	res->l_flags = 0UL;
776 }
777 
778 /*
779  * Keep a list of processes who have interest in a lockres.
780  * Note: this is now only uesed for check recursive cluster locking.
781  */
ocfs2_add_holder(struct ocfs2_lock_res * lockres,struct ocfs2_lock_holder * oh)782 static inline void ocfs2_add_holder(struct ocfs2_lock_res *lockres,
783 				   struct ocfs2_lock_holder *oh)
784 {
785 	INIT_LIST_HEAD(&oh->oh_list);
786 	oh->oh_owner_pid = get_pid(task_pid(current));
787 
788 	spin_lock(&lockres->l_lock);
789 	list_add_tail(&oh->oh_list, &lockres->l_holders);
790 	spin_unlock(&lockres->l_lock);
791 }
792 
793 static struct ocfs2_lock_holder *
ocfs2_pid_holder(struct ocfs2_lock_res * lockres,struct pid * pid)794 ocfs2_pid_holder(struct ocfs2_lock_res *lockres,
795 		struct pid *pid)
796 {
797 	struct ocfs2_lock_holder *oh;
798 
799 	spin_lock(&lockres->l_lock);
800 	list_for_each_entry(oh, &lockres->l_holders, oh_list) {
801 		if (oh->oh_owner_pid == pid) {
802 			spin_unlock(&lockres->l_lock);
803 			return oh;
804 		}
805 	}
806 	spin_unlock(&lockres->l_lock);
807 	return NULL;
808 }
809 
ocfs2_remove_holder(struct ocfs2_lock_res * lockres,struct ocfs2_lock_holder * oh)810 static inline void ocfs2_remove_holder(struct ocfs2_lock_res *lockres,
811 				       struct ocfs2_lock_holder *oh)
812 {
813 	spin_lock(&lockres->l_lock);
814 	list_del(&oh->oh_list);
815 	spin_unlock(&lockres->l_lock);
816 
817 	put_pid(oh->oh_owner_pid);
818 }
819 
820 
ocfs2_inc_holders(struct ocfs2_lock_res * lockres,int level)821 static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
822 				     int level)
823 {
824 	BUG_ON(!lockres);
825 
826 	switch(level) {
827 	case DLM_LOCK_EX:
828 		lockres->l_ex_holders++;
829 		break;
830 	case DLM_LOCK_PR:
831 		lockres->l_ro_holders++;
832 		break;
833 	default:
834 		BUG();
835 	}
836 }
837 
ocfs2_dec_holders(struct ocfs2_lock_res * lockres,int level)838 static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
839 				     int level)
840 {
841 	BUG_ON(!lockres);
842 
843 	switch(level) {
844 	case DLM_LOCK_EX:
845 		BUG_ON(!lockres->l_ex_holders);
846 		lockres->l_ex_holders--;
847 		break;
848 	case DLM_LOCK_PR:
849 		BUG_ON(!lockres->l_ro_holders);
850 		lockres->l_ro_holders--;
851 		break;
852 	default:
853 		BUG();
854 	}
855 }
856 
857 /* WARNING: This function lives in a world where the only three lock
858  * levels are EX, PR, and NL. It *will* have to be adjusted when more
859  * lock types are added. */
ocfs2_highest_compat_lock_level(int level)860 static inline int ocfs2_highest_compat_lock_level(int level)
861 {
862 	int new_level = DLM_LOCK_EX;
863 
864 	if (level == DLM_LOCK_EX)
865 		new_level = DLM_LOCK_NL;
866 	else if (level == DLM_LOCK_PR)
867 		new_level = DLM_LOCK_PR;
868 	return new_level;
869 }
870 
lockres_set_flags(struct ocfs2_lock_res * lockres,unsigned long newflags)871 static void lockres_set_flags(struct ocfs2_lock_res *lockres,
872 			      unsigned long newflags)
873 {
874 	struct ocfs2_mask_waiter *mw, *tmp;
875 
876  	assert_spin_locked(&lockres->l_lock);
877 
878 	lockres->l_flags = newflags;
879 
880 	list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
881 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
882 			continue;
883 
884 		list_del_init(&mw->mw_item);
885 		mw->mw_status = 0;
886 		complete(&mw->mw_complete);
887 	}
888 }
lockres_or_flags(struct ocfs2_lock_res * lockres,unsigned long or)889 static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
890 {
891 	lockres_set_flags(lockres, lockres->l_flags | or);
892 }
lockres_clear_flags(struct ocfs2_lock_res * lockres,unsigned long clear)893 static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
894 				unsigned long clear)
895 {
896 	lockres_set_flags(lockres, lockres->l_flags & ~clear);
897 }
898 
ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res * lockres)899 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
900 {
901 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
902 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
903 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
904 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
905 
906 	lockres->l_level = lockres->l_requested;
907 	if (lockres->l_level <=
908 	    ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
909 		lockres->l_blocking = DLM_LOCK_NL;
910 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
911 	}
912 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
913 }
914 
ocfs2_generic_handle_convert_action(struct ocfs2_lock_res * lockres)915 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
916 {
917 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
918 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
919 
920 	/* Convert from RO to EX doesn't really need anything as our
921 	 * information is already up to data. Convert from NL to
922 	 * *anything* however should mark ourselves as needing an
923 	 * update */
924 	if (lockres->l_level == DLM_LOCK_NL &&
925 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
926 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
927 
928 	lockres->l_level = lockres->l_requested;
929 
930 	/*
931 	 * We set the OCFS2_LOCK_UPCONVERT_FINISHING flag before clearing
932 	 * the OCFS2_LOCK_BUSY flag to prevent the dc thread from
933 	 * downconverting the lock before the upconvert has fully completed.
934 	 * Do not prevent the dc thread from downconverting if NONBLOCK lock
935 	 * had already returned.
936 	 */
937 	if (!(lockres->l_flags & OCFS2_LOCK_NONBLOCK_FINISHED))
938 		lockres_or_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
939 	else
940 		lockres_clear_flags(lockres, OCFS2_LOCK_NONBLOCK_FINISHED);
941 
942 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
943 }
944 
ocfs2_generic_handle_attach_action(struct ocfs2_lock_res * lockres)945 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
946 {
947 	BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
948 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
949 
950 	if (lockres->l_requested > DLM_LOCK_NL &&
951 	    !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
952 	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
953 		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
954 
955 	lockres->l_level = lockres->l_requested;
956 	lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
957 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
958 }
959 
ocfs2_generic_handle_bast(struct ocfs2_lock_res * lockres,int level)960 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
961 				     int level)
962 {
963 	int needs_downconvert = 0;
964 
965 	assert_spin_locked(&lockres->l_lock);
966 
967 	if (level > lockres->l_blocking) {
968 		/* only schedule a downconvert if we haven't already scheduled
969 		 * one that goes low enough to satisfy the level we're
970 		 * blocking.  this also catches the case where we get
971 		 * duplicate BASTs */
972 		if (ocfs2_highest_compat_lock_level(level) <
973 		    ocfs2_highest_compat_lock_level(lockres->l_blocking))
974 			needs_downconvert = 1;
975 
976 		lockres->l_blocking = level;
977 	}
978 
979 	mlog(ML_BASTS, "lockres %s, block %d, level %d, l_block %d, dwn %d\n",
980 	     lockres->l_name, level, lockres->l_level, lockres->l_blocking,
981 	     needs_downconvert);
982 
983 	if (needs_downconvert)
984 		lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
985 	mlog(0, "needs_downconvert = %d\n", needs_downconvert);
986 	return needs_downconvert;
987 }
988 
989 /*
990  * OCFS2_LOCK_PENDING and l_pending_gen.
991  *
992  * Why does OCFS2_LOCK_PENDING exist?  To close a race between setting
993  * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock().  See ocfs2_unblock_lock()
994  * for more details on the race.
995  *
996  * OCFS2_LOCK_PENDING closes the race quite nicely.  However, it introduces
997  * a race on itself.  In o2dlm, we can get the ast before ocfs2_dlm_lock()
998  * returns.  The ast clears OCFS2_LOCK_BUSY, and must therefore clear
999  * OCFS2_LOCK_PENDING at the same time.  When ocfs2_dlm_lock() returns,
1000  * the caller is going to try to clear PENDING again.  If nothing else is
1001  * happening, __lockres_clear_pending() sees PENDING is unset and does
1002  * nothing.
1003  *
1004  * But what if another path (eg downconvert thread) has just started a
1005  * new locking action?  The other path has re-set PENDING.  Our path
1006  * cannot clear PENDING, because that will re-open the original race
1007  * window.
1008  *
1009  * [Example]
1010  *
1011  * ocfs2_meta_lock()
1012  *  ocfs2_cluster_lock()
1013  *   set BUSY
1014  *   set PENDING
1015  *   drop l_lock
1016  *   ocfs2_dlm_lock()
1017  *    ocfs2_locking_ast()		ocfs2_downconvert_thread()
1018  *     clear PENDING			 ocfs2_unblock_lock()
1019  *					  take_l_lock
1020  *					  !BUSY
1021  *					  ocfs2_prepare_downconvert()
1022  *					   set BUSY
1023  *					   set PENDING
1024  *					  drop l_lock
1025  *   take l_lock
1026  *   clear PENDING
1027  *   drop l_lock
1028  *			<window>
1029  *					  ocfs2_dlm_lock()
1030  *
1031  * So as you can see, we now have a window where l_lock is not held,
1032  * PENDING is not set, and ocfs2_dlm_lock() has not been called.
1033  *
1034  * The core problem is that ocfs2_cluster_lock() has cleared the PENDING
1035  * set by ocfs2_prepare_downconvert().  That wasn't nice.
1036  *
1037  * To solve this we introduce l_pending_gen.  A call to
1038  * lockres_clear_pending() will only do so when it is passed a generation
1039  * number that matches the lockres.  lockres_set_pending() will return the
1040  * current generation number.  When ocfs2_cluster_lock() goes to clear
1041  * PENDING, it passes the generation it got from set_pending().  In our
1042  * example above, the generation numbers will *not* match.  Thus,
1043  * ocfs2_cluster_lock() will not clear the PENDING set by
1044  * ocfs2_prepare_downconvert().
1045  */
1046 
1047 /* Unlocked version for ocfs2_locking_ast() */
__lockres_clear_pending(struct ocfs2_lock_res * lockres,unsigned int generation,struct ocfs2_super * osb)1048 static void __lockres_clear_pending(struct ocfs2_lock_res *lockres,
1049 				    unsigned int generation,
1050 				    struct ocfs2_super *osb)
1051 {
1052 	assert_spin_locked(&lockres->l_lock);
1053 
1054 	/*
1055 	 * The ast and locking functions can race us here.  The winner
1056 	 * will clear pending, the loser will not.
1057 	 */
1058 	if (!(lockres->l_flags & OCFS2_LOCK_PENDING) ||
1059 	    (lockres->l_pending_gen != generation))
1060 		return;
1061 
1062 	lockres_clear_flags(lockres, OCFS2_LOCK_PENDING);
1063 	lockres->l_pending_gen++;
1064 
1065 	/*
1066 	 * The downconvert thread may have skipped us because we
1067 	 * were PENDING.  Wake it up.
1068 	 */
1069 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1070 		ocfs2_wake_downconvert_thread(osb);
1071 }
1072 
1073 /* Locked version for callers of ocfs2_dlm_lock() */
lockres_clear_pending(struct ocfs2_lock_res * lockres,unsigned int generation,struct ocfs2_super * osb)1074 static void lockres_clear_pending(struct ocfs2_lock_res *lockres,
1075 				  unsigned int generation,
1076 				  struct ocfs2_super *osb)
1077 {
1078 	unsigned long flags;
1079 
1080 	spin_lock_irqsave(&lockres->l_lock, flags);
1081 	__lockres_clear_pending(lockres, generation, osb);
1082 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1083 }
1084 
lockres_set_pending(struct ocfs2_lock_res * lockres)1085 static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres)
1086 {
1087 	assert_spin_locked(&lockres->l_lock);
1088 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
1089 
1090 	lockres_or_flags(lockres, OCFS2_LOCK_PENDING);
1091 
1092 	return lockres->l_pending_gen;
1093 }
1094 
ocfs2_blocking_ast(struct ocfs2_dlm_lksb * lksb,int level)1095 static void ocfs2_blocking_ast(struct ocfs2_dlm_lksb *lksb, int level)
1096 {
1097 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1098 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1099 	int needs_downconvert;
1100 	unsigned long flags;
1101 
1102 	BUG_ON(level <= DLM_LOCK_NL);
1103 
1104 	mlog(ML_BASTS, "BAST fired for lockres %s, blocking %d, level %d, "
1105 	     "type %s\n", lockres->l_name, level, lockres->l_level,
1106 	     ocfs2_lock_type_string(lockres->l_type));
1107 
1108 	/*
1109 	 * We can skip the bast for locks which don't enable caching -
1110 	 * they'll be dropped at the earliest possible time anyway.
1111 	 */
1112 	if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
1113 		return;
1114 
1115 	spin_lock_irqsave(&lockres->l_lock, flags);
1116 	needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
1117 	if (needs_downconvert)
1118 		ocfs2_schedule_blocked_lock(osb, lockres);
1119 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1120 
1121 	wake_up(&lockres->l_event);
1122 
1123 	ocfs2_wake_downconvert_thread(osb);
1124 }
1125 
ocfs2_locking_ast(struct ocfs2_dlm_lksb * lksb)1126 static void ocfs2_locking_ast(struct ocfs2_dlm_lksb *lksb)
1127 {
1128 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1129 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1130 	unsigned long flags;
1131 	int status;
1132 
1133 	spin_lock_irqsave(&lockres->l_lock, flags);
1134 
1135 	status = ocfs2_dlm_lock_status(&lockres->l_lksb);
1136 
1137 	if (status == -EAGAIN) {
1138 		lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1139 		goto out;
1140 	}
1141 
1142 	if (status) {
1143 		mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n",
1144 		     lockres->l_name, status);
1145 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1146 		return;
1147 	}
1148 
1149 	mlog(ML_BASTS, "AST fired for lockres %s, action %d, unlock %d, "
1150 	     "level %d => %d\n", lockres->l_name, lockres->l_action,
1151 	     lockres->l_unlock_action, lockres->l_level, lockres->l_requested);
1152 
1153 	switch(lockres->l_action) {
1154 	case OCFS2_AST_ATTACH:
1155 		ocfs2_generic_handle_attach_action(lockres);
1156 		lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
1157 		break;
1158 	case OCFS2_AST_CONVERT:
1159 		ocfs2_generic_handle_convert_action(lockres);
1160 		break;
1161 	case OCFS2_AST_DOWNCONVERT:
1162 		ocfs2_generic_handle_downconvert_action(lockres);
1163 		break;
1164 	default:
1165 		mlog(ML_ERROR, "lockres %s: AST fired with invalid action: %u, "
1166 		     "flags 0x%lx, unlock: %u\n",
1167 		     lockres->l_name, lockres->l_action, lockres->l_flags,
1168 		     lockres->l_unlock_action);
1169 		BUG();
1170 	}
1171 out:
1172 	/* set it to something invalid so if we get called again we
1173 	 * can catch it. */
1174 	lockres->l_action = OCFS2_AST_INVALID;
1175 
1176 	/* Did we try to cancel this lock?  Clear that state */
1177 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT)
1178 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1179 
1180 	/*
1181 	 * We may have beaten the locking functions here.  We certainly
1182 	 * know that dlm_lock() has been called :-)
1183 	 * Because we can't have two lock calls in flight at once, we
1184 	 * can use lockres->l_pending_gen.
1185 	 */
1186 	__lockres_clear_pending(lockres, lockres->l_pending_gen,  osb);
1187 
1188 	wake_up(&lockres->l_event);
1189 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1190 }
1191 
ocfs2_unlock_ast(struct ocfs2_dlm_lksb * lksb,int error)1192 static void ocfs2_unlock_ast(struct ocfs2_dlm_lksb *lksb, int error)
1193 {
1194 	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1195 	unsigned long flags;
1196 
1197 	mlog(ML_BASTS, "UNLOCK AST fired for lockres %s, action = %d\n",
1198 	     lockres->l_name, lockres->l_unlock_action);
1199 
1200 	spin_lock_irqsave(&lockres->l_lock, flags);
1201 	if (error) {
1202 		mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
1203 		     "unlock_action %d\n", error, lockres->l_name,
1204 		     lockres->l_unlock_action);
1205 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1206 		return;
1207 	}
1208 
1209 	switch(lockres->l_unlock_action) {
1210 	case OCFS2_UNLOCK_CANCEL_CONVERT:
1211 		mlog(0, "Cancel convert success for %s\n", lockres->l_name);
1212 		lockres->l_action = OCFS2_AST_INVALID;
1213 		/* Downconvert thread may have requeued this lock, we
1214 		 * need to wake it. */
1215 		if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1216 			ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres));
1217 		break;
1218 	case OCFS2_UNLOCK_DROP_LOCK:
1219 		lockres->l_level = DLM_LOCK_IV;
1220 		break;
1221 	default:
1222 		BUG();
1223 	}
1224 
1225 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1226 	lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1227 	wake_up(&lockres->l_event);
1228 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1229 }
1230 
1231 /*
1232  * This is the filesystem locking protocol.  It provides the lock handling
1233  * hooks for the underlying DLM.  It has a maximum version number.
1234  * The version number allows interoperability with systems running at
1235  * the same major number and an equal or smaller minor number.
1236  *
1237  * Whenever the filesystem does new things with locks (adds or removes a
1238  * lock, orders them differently, does different things underneath a lock),
1239  * the version must be changed.  The protocol is negotiated when joining
1240  * the dlm domain.  A node may join the domain if its major version is
1241  * identical to all other nodes and its minor version is greater than
1242  * or equal to all other nodes.  When its minor version is greater than
1243  * the other nodes, it will run at the minor version specified by the
1244  * other nodes.
1245  *
1246  * If a locking change is made that will not be compatible with older
1247  * versions, the major number must be increased and the minor version set
1248  * to zero.  If a change merely adds a behavior that can be disabled when
1249  * speaking to older versions, the minor version must be increased.  If a
1250  * change adds a fully backwards compatible change (eg, LVB changes that
1251  * are just ignored by older versions), the version does not need to be
1252  * updated.
1253  */
1254 static struct ocfs2_locking_protocol lproto = {
1255 	.lp_max_version = {
1256 		.pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
1257 		.pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
1258 	},
1259 	.lp_lock_ast		= ocfs2_locking_ast,
1260 	.lp_blocking_ast	= ocfs2_blocking_ast,
1261 	.lp_unlock_ast		= ocfs2_unlock_ast,
1262 };
1263 
ocfs2_set_locking_protocol(void)1264 void ocfs2_set_locking_protocol(void)
1265 {
1266 	ocfs2_stack_glue_set_max_proto_version(&lproto.lp_max_version);
1267 }
1268 
ocfs2_recover_from_dlm_error(struct ocfs2_lock_res * lockres,int convert)1269 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
1270 						int convert)
1271 {
1272 	unsigned long flags;
1273 
1274 	spin_lock_irqsave(&lockres->l_lock, flags);
1275 	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1276 	lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1277 	if (convert)
1278 		lockres->l_action = OCFS2_AST_INVALID;
1279 	else
1280 		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1281 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1282 
1283 	wake_up(&lockres->l_event);
1284 }
1285 
1286 /* Note: If we detect another process working on the lock (i.e.,
1287  * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
1288  * to do the right thing in that case.
1289  */
ocfs2_lock_create(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int level,u32 dlm_flags)1290 static int ocfs2_lock_create(struct ocfs2_super *osb,
1291 			     struct ocfs2_lock_res *lockres,
1292 			     int level,
1293 			     u32 dlm_flags)
1294 {
1295 	int ret = 0;
1296 	unsigned long flags;
1297 	unsigned int gen;
1298 
1299 	mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
1300 	     dlm_flags);
1301 
1302 	spin_lock_irqsave(&lockres->l_lock, flags);
1303 	if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
1304 	    (lockres->l_flags & OCFS2_LOCK_BUSY)) {
1305 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1306 		goto bail;
1307 	}
1308 
1309 	lockres->l_action = OCFS2_AST_ATTACH;
1310 	lockres->l_requested = level;
1311 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1312 	gen = lockres_set_pending(lockres);
1313 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1314 
1315 	ret = ocfs2_dlm_lock(osb->cconn,
1316 			     level,
1317 			     &lockres->l_lksb,
1318 			     dlm_flags,
1319 			     lockres->l_name,
1320 			     OCFS2_LOCK_ID_MAX_LEN - 1);
1321 	lockres_clear_pending(lockres, gen, osb);
1322 	if (ret) {
1323 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
1324 		ocfs2_recover_from_dlm_error(lockres, 1);
1325 	}
1326 
1327 	mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
1328 
1329 bail:
1330 	return ret;
1331 }
1332 
ocfs2_check_wait_flag(struct ocfs2_lock_res * lockres,int flag)1333 static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
1334 					int flag)
1335 {
1336 	unsigned long flags;
1337 	int ret;
1338 
1339 	spin_lock_irqsave(&lockres->l_lock, flags);
1340 	ret = lockres->l_flags & flag;
1341 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1342 
1343 	return ret;
1344 }
1345 
ocfs2_wait_on_busy_lock(struct ocfs2_lock_res * lockres)1346 static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
1347 
1348 {
1349 	wait_event(lockres->l_event,
1350 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
1351 }
1352 
ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res * lockres)1353 static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
1354 
1355 {
1356 	wait_event(lockres->l_event,
1357 		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
1358 }
1359 
1360 /* predict what lock level we'll be dropping down to on behalf
1361  * of another node, and return true if the currently wanted
1362  * level will be compatible with it. */
ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res * lockres,int wanted)1363 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
1364 						     int wanted)
1365 {
1366 	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
1367 
1368 	return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
1369 }
1370 
ocfs2_init_mask_waiter(struct ocfs2_mask_waiter * mw)1371 static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
1372 {
1373 	INIT_LIST_HEAD(&mw->mw_item);
1374 	init_completion(&mw->mw_complete);
1375 	ocfs2_init_start_time(mw);
1376 }
1377 
ocfs2_wait_for_mask(struct ocfs2_mask_waiter * mw)1378 static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
1379 {
1380 	wait_for_completion(&mw->mw_complete);
1381 	/* Re-arm the completion in case we want to wait on it again */
1382 	reinit_completion(&mw->mw_complete);
1383 	return mw->mw_status;
1384 }
1385 
lockres_add_mask_waiter(struct ocfs2_lock_res * lockres,struct ocfs2_mask_waiter * mw,unsigned long mask,unsigned long goal)1386 static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
1387 				    struct ocfs2_mask_waiter *mw,
1388 				    unsigned long mask,
1389 				    unsigned long goal)
1390 {
1391 	BUG_ON(!list_empty(&mw->mw_item));
1392 
1393 	assert_spin_locked(&lockres->l_lock);
1394 
1395 	list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
1396 	mw->mw_mask = mask;
1397 	mw->mw_goal = goal;
1398 }
1399 
1400 /* returns 0 if the mw that was removed was already satisfied, -EBUSY
1401  * if the mask still hadn't reached its goal */
__lockres_remove_mask_waiter(struct ocfs2_lock_res * lockres,struct ocfs2_mask_waiter * mw)1402 static int __lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1403 				      struct ocfs2_mask_waiter *mw)
1404 {
1405 	int ret = 0;
1406 
1407 	assert_spin_locked(&lockres->l_lock);
1408 	if (!list_empty(&mw->mw_item)) {
1409 		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
1410 			ret = -EBUSY;
1411 
1412 		list_del_init(&mw->mw_item);
1413 		init_completion(&mw->mw_complete);
1414 	}
1415 
1416 	return ret;
1417 }
1418 
lockres_remove_mask_waiter(struct ocfs2_lock_res * lockres,struct ocfs2_mask_waiter * mw)1419 static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1420 				      struct ocfs2_mask_waiter *mw)
1421 {
1422 	unsigned long flags;
1423 	int ret = 0;
1424 
1425 	spin_lock_irqsave(&lockres->l_lock, flags);
1426 	ret = __lockres_remove_mask_waiter(lockres, mw);
1427 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1428 
1429 	return ret;
1430 
1431 }
1432 
ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter * mw,struct ocfs2_lock_res * lockres)1433 static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
1434 					     struct ocfs2_lock_res *lockres)
1435 {
1436 	int ret;
1437 
1438 	ret = wait_for_completion_interruptible(&mw->mw_complete);
1439 	if (ret)
1440 		lockres_remove_mask_waiter(lockres, mw);
1441 	else
1442 		ret = mw->mw_status;
1443 	/* Re-arm the completion in case we want to wait on it again */
1444 	reinit_completion(&mw->mw_complete);
1445 	return ret;
1446 }
1447 
__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)1448 static int __ocfs2_cluster_lock(struct ocfs2_super *osb,
1449 				struct ocfs2_lock_res *lockres,
1450 				int level,
1451 				u32 lkm_flags,
1452 				int arg_flags,
1453 				int l_subclass,
1454 				unsigned long caller_ip)
1455 {
1456 	struct ocfs2_mask_waiter mw;
1457 	int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
1458 	int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1459 	unsigned long flags;
1460 	unsigned int gen;
1461 	int noqueue_attempted = 0;
1462 	int dlm_locked = 0;
1463 	int kick_dc = 0;
1464 
1465 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED)) {
1466 		mlog_errno(-EINVAL);
1467 		return -EINVAL;
1468 	}
1469 
1470 	ocfs2_init_mask_waiter(&mw);
1471 
1472 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
1473 		lkm_flags |= DLM_LKF_VALBLK;
1474 
1475 again:
1476 	wait = 0;
1477 
1478 	spin_lock_irqsave(&lockres->l_lock, flags);
1479 
1480 	if (catch_signals && signal_pending(current)) {
1481 		ret = -ERESTARTSYS;
1482 		goto unlock;
1483 	}
1484 
1485 	mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
1486 			"Cluster lock called on freeing lockres %s! flags "
1487 			"0x%lx\n", lockres->l_name, lockres->l_flags);
1488 
1489 	/* We only compare against the currently granted level
1490 	 * here. If the lock is blocked waiting on a downconvert,
1491 	 * we'll get caught below. */
1492 	if (lockres->l_flags & OCFS2_LOCK_BUSY &&
1493 	    level > lockres->l_level) {
1494 		/* is someone sitting in dlm_lock? If so, wait on
1495 		 * them. */
1496 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1497 		wait = 1;
1498 		goto unlock;
1499 	}
1500 
1501 	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING) {
1502 		/*
1503 		 * We've upconverted. If the lock now has a level we can
1504 		 * work with, we take it. If, however, the lock is not at the
1505 		 * required level, we go thru the full cycle. One way this could
1506 		 * happen is if a process requesting an upconvert to PR is
1507 		 * closely followed by another requesting upconvert to an EX.
1508 		 * If the process requesting EX lands here, we want it to
1509 		 * continue attempting to upconvert and let the process
1510 		 * requesting PR take the lock.
1511 		 * If multiple processes request upconvert to PR, the first one
1512 		 * here will take the lock. The others will have to go thru the
1513 		 * OCFS2_LOCK_BLOCKED check to ensure that there is no pending
1514 		 * downconvert request.
1515 		 */
1516 		if (level <= lockres->l_level)
1517 			goto update_holders;
1518 	}
1519 
1520 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
1521 	    !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
1522 		/* is the lock is currently blocked on behalf of
1523 		 * another node */
1524 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
1525 		wait = 1;
1526 		goto unlock;
1527 	}
1528 
1529 	if (level > lockres->l_level) {
1530 		if (noqueue_attempted > 0) {
1531 			ret = -EAGAIN;
1532 			goto unlock;
1533 		}
1534 		if (lkm_flags & DLM_LKF_NOQUEUE)
1535 			noqueue_attempted = 1;
1536 
1537 		if (lockres->l_action != OCFS2_AST_INVALID)
1538 			mlog(ML_ERROR, "lockres %s has action %u pending\n",
1539 			     lockres->l_name, lockres->l_action);
1540 
1541 		if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1542 			lockres->l_action = OCFS2_AST_ATTACH;
1543 			lkm_flags &= ~DLM_LKF_CONVERT;
1544 		} else {
1545 			lockres->l_action = OCFS2_AST_CONVERT;
1546 			lkm_flags |= DLM_LKF_CONVERT;
1547 		}
1548 
1549 		lockres->l_requested = level;
1550 		lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1551 		gen = lockres_set_pending(lockres);
1552 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1553 
1554 		BUG_ON(level == DLM_LOCK_IV);
1555 		BUG_ON(level == DLM_LOCK_NL);
1556 
1557 		mlog(ML_BASTS, "lockres %s, convert from %d to %d\n",
1558 		     lockres->l_name, lockres->l_level, level);
1559 
1560 		/* call dlm_lock to upgrade lock now */
1561 		ret = ocfs2_dlm_lock(osb->cconn,
1562 				     level,
1563 				     &lockres->l_lksb,
1564 				     lkm_flags,
1565 				     lockres->l_name,
1566 				     OCFS2_LOCK_ID_MAX_LEN - 1);
1567 		lockres_clear_pending(lockres, gen, osb);
1568 		if (ret) {
1569 			if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
1570 			    (ret != -EAGAIN)) {
1571 				ocfs2_log_dlm_error("ocfs2_dlm_lock",
1572 						    ret, lockres);
1573 			}
1574 			ocfs2_recover_from_dlm_error(lockres, 1);
1575 			goto out;
1576 		}
1577 		dlm_locked = 1;
1578 
1579 		mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n",
1580 		     lockres->l_name);
1581 
1582 		/* At this point we've gone inside the dlm and need to
1583 		 * complete our work regardless. */
1584 		catch_signals = 0;
1585 
1586 		/* wait for busy to clear and carry on */
1587 		goto again;
1588 	}
1589 
1590 update_holders:
1591 	/* Ok, if we get here then we're good to go. */
1592 	ocfs2_inc_holders(lockres, level);
1593 
1594 	ret = 0;
1595 unlock:
1596 	lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1597 
1598 	/* ocfs2_unblock_lock reques on seeing OCFS2_LOCK_UPCONVERT_FINISHING */
1599 	kick_dc = (lockres->l_flags & OCFS2_LOCK_BLOCKED);
1600 
1601 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1602 	if (kick_dc)
1603 		ocfs2_wake_downconvert_thread(osb);
1604 out:
1605 	/*
1606 	 * This is helping work around a lock inversion between the page lock
1607 	 * and dlm locks.  One path holds the page lock while calling aops
1608 	 * which block acquiring dlm locks.  The voting thread holds dlm
1609 	 * locks while acquiring page locks while down converting data locks.
1610 	 * This block is helping an aop path notice the inversion and back
1611 	 * off to unlock its page lock before trying the dlm lock again.
1612 	 */
1613 	if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1614 	    mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1615 		wait = 0;
1616 		spin_lock_irqsave(&lockres->l_lock, flags);
1617 		if (__lockres_remove_mask_waiter(lockres, &mw)) {
1618 			if (dlm_locked)
1619 				lockres_or_flags(lockres,
1620 					OCFS2_LOCK_NONBLOCK_FINISHED);
1621 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1622 			ret = -EAGAIN;
1623 		} else {
1624 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1625 			goto again;
1626 		}
1627 	}
1628 	if (wait) {
1629 		ret = ocfs2_wait_for_mask(&mw);
1630 		if (ret == 0)
1631 			goto again;
1632 		mlog_errno(ret);
1633 	}
1634 	ocfs2_update_lock_stats(lockres, level, &mw, ret);
1635 
1636 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1637 	if (!ret && lockres->l_lockdep_map.key != NULL) {
1638 		if (level == DLM_LOCK_PR)
1639 			rwsem_acquire_read(&lockres->l_lockdep_map, l_subclass,
1640 				!!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1641 				caller_ip);
1642 		else
1643 			rwsem_acquire(&lockres->l_lockdep_map, l_subclass,
1644 				!!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1645 				caller_ip);
1646 	}
1647 #endif
1648 	return ret;
1649 }
1650 
ocfs2_cluster_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int level,u32 lkm_flags,int arg_flags)1651 static inline int ocfs2_cluster_lock(struct ocfs2_super *osb,
1652 				     struct ocfs2_lock_res *lockres,
1653 				     int level,
1654 				     u32 lkm_flags,
1655 				     int arg_flags)
1656 {
1657 	return __ocfs2_cluster_lock(osb, lockres, level, lkm_flags, arg_flags,
1658 				    0, _RET_IP_);
1659 }
1660 
1661 
__ocfs2_cluster_unlock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int level,unsigned long caller_ip)1662 static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
1663 				   struct ocfs2_lock_res *lockres,
1664 				   int level,
1665 				   unsigned long caller_ip)
1666 {
1667 	unsigned long flags;
1668 
1669 	spin_lock_irqsave(&lockres->l_lock, flags);
1670 	ocfs2_dec_holders(lockres, level);
1671 	ocfs2_downconvert_on_unlock(osb, lockres);
1672 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1673 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1674 	if (lockres->l_lockdep_map.key != NULL)
1675 		rwsem_release(&lockres->l_lockdep_map, 1, caller_ip);
1676 #endif
1677 }
1678 
ocfs2_create_new_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int ex,int local)1679 static int ocfs2_create_new_lock(struct ocfs2_super *osb,
1680 				 struct ocfs2_lock_res *lockres,
1681 				 int ex,
1682 				 int local)
1683 {
1684 	int level =  ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1685 	unsigned long flags;
1686 	u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
1687 
1688 	spin_lock_irqsave(&lockres->l_lock, flags);
1689 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1690 	lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1691 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1692 
1693 	return ocfs2_lock_create(osb, lockres, level, lkm_flags);
1694 }
1695 
1696 /* Grants us an EX lock on the data and metadata resources, skipping
1697  * the normal cluster directory lookup. Use this ONLY on newly created
1698  * inodes which other nodes can't possibly see, and which haven't been
1699  * hashed in the inode hash yet. This can give us a good performance
1700  * increase as it'll skip the network broadcast normally associated
1701  * with creating a new lock resource. */
ocfs2_create_new_inode_locks(struct inode * inode)1702 int ocfs2_create_new_inode_locks(struct inode *inode)
1703 {
1704 	int ret;
1705 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1706 
1707 	BUG_ON(!ocfs2_inode_is_new(inode));
1708 
1709 	mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
1710 
1711 	/* NOTE: That we don't increment any of the holder counts, nor
1712 	 * do we add anything to a journal handle. Since this is
1713 	 * supposed to be a new inode which the cluster doesn't know
1714 	 * about yet, there is no need to.  As far as the LVB handling
1715 	 * is concerned, this is basically like acquiring an EX lock
1716 	 * on a resource which has an invalid one -- we'll set it
1717 	 * valid when we release the EX. */
1718 
1719 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
1720 	if (ret) {
1721 		mlog_errno(ret);
1722 		goto bail;
1723 	}
1724 
1725 	/*
1726 	 * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
1727 	 * don't use a generation in their lock names.
1728 	 */
1729 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
1730 	if (ret) {
1731 		mlog_errno(ret);
1732 		goto bail;
1733 	}
1734 
1735 	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
1736 	if (ret)
1737 		mlog_errno(ret);
1738 
1739 bail:
1740 	return ret;
1741 }
1742 
ocfs2_rw_lock(struct inode * inode,int write)1743 int ocfs2_rw_lock(struct inode *inode, int write)
1744 {
1745 	int status, level;
1746 	struct ocfs2_lock_res *lockres;
1747 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1748 
1749 	mlog(0, "inode %llu take %s RW lock\n",
1750 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1751 	     write ? "EXMODE" : "PRMODE");
1752 
1753 	if (ocfs2_mount_local(osb))
1754 		return 0;
1755 
1756 	lockres = &OCFS2_I(inode)->ip_rw_lockres;
1757 
1758 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1759 
1760 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
1761 	if (status < 0)
1762 		mlog_errno(status);
1763 
1764 	return status;
1765 }
1766 
ocfs2_try_rw_lock(struct inode * inode,int write)1767 int ocfs2_try_rw_lock(struct inode *inode, int write)
1768 {
1769 	int status, level;
1770 	struct ocfs2_lock_res *lockres;
1771 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1772 
1773 	mlog(0, "inode %llu try to take %s RW lock\n",
1774 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1775 	     write ? "EXMODE" : "PRMODE");
1776 
1777 	if (ocfs2_mount_local(osb))
1778 		return 0;
1779 
1780 	lockres = &OCFS2_I(inode)->ip_rw_lockres;
1781 
1782 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1783 
1784 	status = ocfs2_cluster_lock(osb, lockres, level, DLM_LKF_NOQUEUE, 0);
1785 	return status;
1786 }
1787 
ocfs2_rw_unlock(struct inode * inode,int write)1788 void ocfs2_rw_unlock(struct inode *inode, int write)
1789 {
1790 	int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1791 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
1792 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1793 
1794 	mlog(0, "inode %llu drop %s RW lock\n",
1795 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1796 	     write ? "EXMODE" : "PRMODE");
1797 
1798 	if (!ocfs2_mount_local(osb))
1799 		ocfs2_cluster_unlock(osb, lockres, level);
1800 }
1801 
1802 /*
1803  * ocfs2_open_lock always get PR mode lock.
1804  */
ocfs2_open_lock(struct inode * inode)1805 int ocfs2_open_lock(struct inode *inode)
1806 {
1807 	int status = 0;
1808 	struct ocfs2_lock_res *lockres;
1809 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1810 
1811 	mlog(0, "inode %llu take PRMODE open lock\n",
1812 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1813 
1814 	if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1815 		goto out;
1816 
1817 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1818 
1819 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_PR, 0, 0);
1820 	if (status < 0)
1821 		mlog_errno(status);
1822 
1823 out:
1824 	return status;
1825 }
1826 
ocfs2_try_open_lock(struct inode * inode,int write)1827 int ocfs2_try_open_lock(struct inode *inode, int write)
1828 {
1829 	int status = 0, level;
1830 	struct ocfs2_lock_res *lockres;
1831 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1832 
1833 	mlog(0, "inode %llu try to take %s open lock\n",
1834 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
1835 	     write ? "EXMODE" : "PRMODE");
1836 
1837 	if (ocfs2_is_hard_readonly(osb)) {
1838 		if (write)
1839 			status = -EROFS;
1840 		goto out;
1841 	}
1842 
1843 	if (ocfs2_mount_local(osb))
1844 		goto out;
1845 
1846 	lockres = &OCFS2_I(inode)->ip_open_lockres;
1847 
1848 	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1849 
1850 	/*
1851 	 * The file system may already holding a PRMODE/EXMODE open lock.
1852 	 * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
1853 	 * other nodes and the -EAGAIN will indicate to the caller that
1854 	 * this inode is still in use.
1855 	 */
1856 	status = ocfs2_cluster_lock(osb, lockres, level, DLM_LKF_NOQUEUE, 0);
1857 
1858 out:
1859 	return status;
1860 }
1861 
1862 /*
1863  * ocfs2_open_unlock unlock PR and EX mode open locks.
1864  */
ocfs2_open_unlock(struct inode * inode)1865 void ocfs2_open_unlock(struct inode *inode)
1866 {
1867 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
1868 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1869 
1870 	mlog(0, "inode %llu drop open lock\n",
1871 	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
1872 
1873 	if (ocfs2_mount_local(osb))
1874 		goto out;
1875 
1876 	if(lockres->l_ro_holders)
1877 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_PR);
1878 	if(lockres->l_ex_holders)
1879 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
1880 
1881 out:
1882 	return;
1883 }
1884 
ocfs2_flock_handle_signal(struct ocfs2_lock_res * lockres,int level)1885 static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
1886 				     int level)
1887 {
1888 	int ret;
1889 	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1890 	unsigned long flags;
1891 	struct ocfs2_mask_waiter mw;
1892 
1893 	ocfs2_init_mask_waiter(&mw);
1894 
1895 retry_cancel:
1896 	spin_lock_irqsave(&lockres->l_lock, flags);
1897 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
1898 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
1899 		if (ret) {
1900 			spin_unlock_irqrestore(&lockres->l_lock, flags);
1901 			ret = ocfs2_cancel_convert(osb, lockres);
1902 			if (ret < 0) {
1903 				mlog_errno(ret);
1904 				goto out;
1905 			}
1906 			goto retry_cancel;
1907 		}
1908 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1909 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1910 
1911 		ocfs2_wait_for_mask(&mw);
1912 		goto retry_cancel;
1913 	}
1914 
1915 	ret = -ERESTARTSYS;
1916 	/*
1917 	 * We may still have gotten the lock, in which case there's no
1918 	 * point to restarting the syscall.
1919 	 */
1920 	if (lockres->l_level == level)
1921 		ret = 0;
1922 
1923 	mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
1924 	     lockres->l_flags, lockres->l_level, lockres->l_action);
1925 
1926 	spin_unlock_irqrestore(&lockres->l_lock, flags);
1927 
1928 out:
1929 	return ret;
1930 }
1931 
1932 /*
1933  * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
1934  * flock() calls. The locking approach this requires is sufficiently
1935  * different from all other cluster lock types that we implement a
1936  * separate path to the "low-level" dlm calls. In particular:
1937  *
1938  * - No optimization of lock levels is done - we take at exactly
1939  *   what's been requested.
1940  *
1941  * - No lock caching is employed. We immediately downconvert to
1942  *   no-lock at unlock time. This also means flock locks never go on
1943  *   the blocking list).
1944  *
1945  * - Since userspace can trivially deadlock itself with flock, we make
1946  *   sure to allow cancellation of a misbehaving applications flock()
1947  *   request.
1948  *
1949  * - Access to any flock lockres doesn't require concurrency, so we
1950  *   can simplify the code by requiring the caller to guarantee
1951  *   serialization of dlmglue flock calls.
1952  */
ocfs2_file_lock(struct file * file,int ex,int trylock)1953 int ocfs2_file_lock(struct file *file, int ex, int trylock)
1954 {
1955 	int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1956 	unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0;
1957 	unsigned long flags;
1958 	struct ocfs2_file_private *fp = file->private_data;
1959 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
1960 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1961 	struct ocfs2_mask_waiter mw;
1962 
1963 	ocfs2_init_mask_waiter(&mw);
1964 
1965 	if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
1966 	    (lockres->l_level > DLM_LOCK_NL)) {
1967 		mlog(ML_ERROR,
1968 		     "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
1969 		     "level: %u\n", lockres->l_name, lockres->l_flags,
1970 		     lockres->l_level);
1971 		return -EINVAL;
1972 	}
1973 
1974 	spin_lock_irqsave(&lockres->l_lock, flags);
1975 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1976 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1977 		spin_unlock_irqrestore(&lockres->l_lock, flags);
1978 
1979 		/*
1980 		 * Get the lock at NLMODE to start - that way we
1981 		 * can cancel the upconvert request if need be.
1982 		 */
1983 		ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0);
1984 		if (ret < 0) {
1985 			mlog_errno(ret);
1986 			goto out;
1987 		}
1988 
1989 		ret = ocfs2_wait_for_mask(&mw);
1990 		if (ret) {
1991 			mlog_errno(ret);
1992 			goto out;
1993 		}
1994 		spin_lock_irqsave(&lockres->l_lock, flags);
1995 	}
1996 
1997 	lockres->l_action = OCFS2_AST_CONVERT;
1998 	lkm_flags |= DLM_LKF_CONVERT;
1999 	lockres->l_requested = level;
2000 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2001 
2002 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
2003 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2004 
2005 	ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags,
2006 			     lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1);
2007 	if (ret) {
2008 		if (!trylock || (ret != -EAGAIN)) {
2009 			ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
2010 			ret = -EINVAL;
2011 		}
2012 
2013 		ocfs2_recover_from_dlm_error(lockres, 1);
2014 		lockres_remove_mask_waiter(lockres, &mw);
2015 		goto out;
2016 	}
2017 
2018 	ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
2019 	if (ret == -ERESTARTSYS) {
2020 		/*
2021 		 * Userspace can cause deadlock itself with
2022 		 * flock(). Current behavior locally is to allow the
2023 		 * deadlock, but abort the system call if a signal is
2024 		 * received. We follow this example, otherwise a
2025 		 * poorly written program could sit in kernel until
2026 		 * reboot.
2027 		 *
2028 		 * Handling this is a bit more complicated for Ocfs2
2029 		 * though. We can't exit this function with an
2030 		 * outstanding lock request, so a cancel convert is
2031 		 * required. We intentionally overwrite 'ret' - if the
2032 		 * cancel fails and the lock was granted, it's easier
2033 		 * to just bubble success back up to the user.
2034 		 */
2035 		ret = ocfs2_flock_handle_signal(lockres, level);
2036 	} else if (!ret && (level > lockres->l_level)) {
2037 		/* Trylock failed asynchronously */
2038 		BUG_ON(!trylock);
2039 		ret = -EAGAIN;
2040 	}
2041 
2042 out:
2043 
2044 	mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
2045 	     lockres->l_name, ex, trylock, ret);
2046 	return ret;
2047 }
2048 
ocfs2_file_unlock(struct file * file)2049 void ocfs2_file_unlock(struct file *file)
2050 {
2051 	int ret;
2052 	unsigned int gen;
2053 	unsigned long flags;
2054 	struct ocfs2_file_private *fp = file->private_data;
2055 	struct ocfs2_lock_res *lockres = &fp->fp_flock;
2056 	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
2057 	struct ocfs2_mask_waiter mw;
2058 
2059 	ocfs2_init_mask_waiter(&mw);
2060 
2061 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
2062 		return;
2063 
2064 	if (lockres->l_level == DLM_LOCK_NL)
2065 		return;
2066 
2067 	mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
2068 	     lockres->l_name, lockres->l_flags, lockres->l_level,
2069 	     lockres->l_action);
2070 
2071 	spin_lock_irqsave(&lockres->l_lock, flags);
2072 	/*
2073 	 * Fake a blocking ast for the downconvert code.
2074 	 */
2075 	lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
2076 	lockres->l_blocking = DLM_LOCK_EX;
2077 
2078 	gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL);
2079 	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
2080 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2081 
2082 	ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen);
2083 	if (ret) {
2084 		mlog_errno(ret);
2085 		return;
2086 	}
2087 
2088 	ret = ocfs2_wait_for_mask(&mw);
2089 	if (ret)
2090 		mlog_errno(ret);
2091 }
2092 
ocfs2_downconvert_on_unlock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)2093 static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
2094 					struct ocfs2_lock_res *lockres)
2095 {
2096 	int kick = 0;
2097 
2098 	/* If we know that another node is waiting on our lock, kick
2099 	 * the downconvert thread * pre-emptively when we reach a release
2100 	 * condition. */
2101 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
2102 		switch(lockres->l_blocking) {
2103 		case DLM_LOCK_EX:
2104 			if (!lockres->l_ex_holders && !lockres->l_ro_holders)
2105 				kick = 1;
2106 			break;
2107 		case DLM_LOCK_PR:
2108 			if (!lockres->l_ex_holders)
2109 				kick = 1;
2110 			break;
2111 		default:
2112 			BUG();
2113 		}
2114 	}
2115 
2116 	if (kick)
2117 		ocfs2_wake_downconvert_thread(osb);
2118 }
2119 
2120 #define OCFS2_SEC_BITS   34
2121 #define OCFS2_SEC_SHIFT  (64 - 34)
2122 #define OCFS2_NSEC_MASK  ((1ULL << OCFS2_SEC_SHIFT) - 1)
2123 
2124 /* LVB only has room for 64 bits of time here so we pack it for
2125  * now. */
ocfs2_pack_timespec(struct timespec * spec)2126 static u64 ocfs2_pack_timespec(struct timespec *spec)
2127 {
2128 	u64 res;
2129 	u64 sec = spec->tv_sec;
2130 	u32 nsec = spec->tv_nsec;
2131 
2132 	res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
2133 
2134 	return res;
2135 }
2136 
2137 /* Call this with the lockres locked. I am reasonably sure we don't
2138  * need ip_lock in this function as anyone who would be changing those
2139  * values is supposed to be blocked in ocfs2_inode_lock right now. */
__ocfs2_stuff_meta_lvb(struct inode * inode)2140 static void __ocfs2_stuff_meta_lvb(struct inode *inode)
2141 {
2142 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2143 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2144 	struct ocfs2_meta_lvb *lvb;
2145 	struct timespec ts;
2146 
2147 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2148 
2149 	/*
2150 	 * Invalidate the LVB of a deleted inode - this way other
2151 	 * nodes are forced to go to disk and discover the new inode
2152 	 * status.
2153 	 */
2154 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
2155 		lvb->lvb_version = 0;
2156 		goto out;
2157 	}
2158 
2159 	lvb->lvb_version   = OCFS2_LVB_VERSION;
2160 	lvb->lvb_isize	   = cpu_to_be64(i_size_read(inode));
2161 	lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
2162 	lvb->lvb_iuid      = cpu_to_be32(i_uid_read(inode));
2163 	lvb->lvb_igid      = cpu_to_be32(i_gid_read(inode));
2164 	lvb->lvb_imode     = cpu_to_be16(inode->i_mode);
2165 	lvb->lvb_inlink    = cpu_to_be16(inode->i_nlink);
2166 	ts = timespec64_to_timespec(inode->i_atime);
2167 	lvb->lvb_iatime_packed  =
2168 		cpu_to_be64(ocfs2_pack_timespec(&ts));
2169 	ts = timespec64_to_timespec(inode->i_ctime);
2170 	lvb->lvb_ictime_packed =
2171 		cpu_to_be64(ocfs2_pack_timespec(&ts));
2172 	ts = timespec64_to_timespec(inode->i_mtime);
2173 	lvb->lvb_imtime_packed =
2174 		cpu_to_be64(ocfs2_pack_timespec(&ts));
2175 	lvb->lvb_iattr    = cpu_to_be32(oi->ip_attr);
2176 	lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
2177 	lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
2178 
2179 out:
2180 	mlog_meta_lvb(0, lockres);
2181 }
2182 
ocfs2_unpack_timespec(struct timespec * spec,u64 packed_time)2183 static void ocfs2_unpack_timespec(struct timespec *spec,
2184 				  u64 packed_time)
2185 {
2186 	spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
2187 	spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
2188 }
2189 
ocfs2_refresh_inode_from_lvb(struct inode * inode)2190 static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
2191 {
2192 	struct timespec ts;
2193 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2194 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2195 	struct ocfs2_meta_lvb *lvb;
2196 
2197 	mlog_meta_lvb(0, lockres);
2198 
2199 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2200 
2201 	/* We're safe here without the lockres lock... */
2202 	spin_lock(&oi->ip_lock);
2203 	oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
2204 	i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
2205 
2206 	oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
2207 	oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
2208 	ocfs2_set_inode_flags(inode);
2209 
2210 	/* fast-symlinks are a special case */
2211 	if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
2212 		inode->i_blocks = 0;
2213 	else
2214 		inode->i_blocks = ocfs2_inode_sector_count(inode);
2215 
2216 	i_uid_write(inode, be32_to_cpu(lvb->lvb_iuid));
2217 	i_gid_write(inode, be32_to_cpu(lvb->lvb_igid));
2218 	inode->i_mode    = be16_to_cpu(lvb->lvb_imode);
2219 	set_nlink(inode, be16_to_cpu(lvb->lvb_inlink));
2220 	ocfs2_unpack_timespec(&ts,
2221 			      be64_to_cpu(lvb->lvb_iatime_packed));
2222 	inode->i_atime = timespec_to_timespec64(ts);
2223 	ocfs2_unpack_timespec(&ts,
2224 			      be64_to_cpu(lvb->lvb_imtime_packed));
2225 	inode->i_mtime = timespec_to_timespec64(ts);
2226 	ocfs2_unpack_timespec(&ts,
2227 			      be64_to_cpu(lvb->lvb_ictime_packed));
2228 	inode->i_ctime = timespec_to_timespec64(ts);
2229 	spin_unlock(&oi->ip_lock);
2230 }
2231 
ocfs2_meta_lvb_is_trustable(struct inode * inode,struct ocfs2_lock_res * lockres)2232 static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
2233 					      struct ocfs2_lock_res *lockres)
2234 {
2235 	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2236 
2237 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)
2238 	    && lvb->lvb_version == OCFS2_LVB_VERSION
2239 	    && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
2240 		return 1;
2241 	return 0;
2242 }
2243 
2244 /* Determine whether a lock resource needs to be refreshed, and
2245  * arbitrate who gets to refresh it.
2246  *
2247  *   0 means no refresh needed.
2248  *
2249  *   > 0 means you need to refresh this and you MUST call
2250  *   ocfs2_complete_lock_res_refresh afterwards. */
ocfs2_should_refresh_lock_res(struct ocfs2_lock_res * lockres)2251 static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
2252 {
2253 	unsigned long flags;
2254 	int status = 0;
2255 
2256 refresh_check:
2257 	spin_lock_irqsave(&lockres->l_lock, flags);
2258 	if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
2259 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2260 		goto bail;
2261 	}
2262 
2263 	if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
2264 		spin_unlock_irqrestore(&lockres->l_lock, flags);
2265 
2266 		ocfs2_wait_on_refreshing_lock(lockres);
2267 		goto refresh_check;
2268 	}
2269 
2270 	/* Ok, I'll be the one to refresh this lock. */
2271 	lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
2272 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2273 
2274 	status = 1;
2275 bail:
2276 	mlog(0, "status %d\n", status);
2277 	return status;
2278 }
2279 
2280 /* If status is non zero, I'll mark it as not being in refresh
2281  * anymroe, but i won't clear the needs refresh flag. */
ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res * lockres,int status)2282 static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
2283 						   int status)
2284 {
2285 	unsigned long flags;
2286 
2287 	spin_lock_irqsave(&lockres->l_lock, flags);
2288 	lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
2289 	if (!status)
2290 		lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
2291 	spin_unlock_irqrestore(&lockres->l_lock, flags);
2292 
2293 	wake_up(&lockres->l_event);
2294 }
2295 
2296 /* may or may not return a bh if it went to disk. */
ocfs2_inode_lock_update(struct inode * inode,struct buffer_head ** bh)2297 static int ocfs2_inode_lock_update(struct inode *inode,
2298 				  struct buffer_head **bh)
2299 {
2300 	int status = 0;
2301 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2302 	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2303 	struct ocfs2_dinode *fe;
2304 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2305 
2306 	if (ocfs2_mount_local(osb))
2307 		goto bail;
2308 
2309 	spin_lock(&oi->ip_lock);
2310 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
2311 		mlog(0, "Orphaned inode %llu was deleted while we "
2312 		     "were waiting on a lock. ip_flags = 0x%x\n",
2313 		     (unsigned long long)oi->ip_blkno, oi->ip_flags);
2314 		spin_unlock(&oi->ip_lock);
2315 		status = -ENOENT;
2316 		goto bail;
2317 	}
2318 	spin_unlock(&oi->ip_lock);
2319 
2320 	if (!ocfs2_should_refresh_lock_res(lockres))
2321 		goto bail;
2322 
2323 	/* This will discard any caching information we might have had
2324 	 * for the inode metadata. */
2325 	ocfs2_metadata_cache_purge(INODE_CACHE(inode));
2326 
2327 	ocfs2_extent_map_trunc(inode, 0);
2328 
2329 	if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
2330 		mlog(0, "Trusting LVB on inode %llu\n",
2331 		     (unsigned long long)oi->ip_blkno);
2332 		ocfs2_refresh_inode_from_lvb(inode);
2333 	} else {
2334 		/* Boo, we have to go to disk. */
2335 		/* read bh, cast, ocfs2_refresh_inode */
2336 		status = ocfs2_read_inode_block(inode, bh);
2337 		if (status < 0) {
2338 			mlog_errno(status);
2339 			goto bail_refresh;
2340 		}
2341 		fe = (struct ocfs2_dinode *) (*bh)->b_data;
2342 
2343 		/* This is a good chance to make sure we're not
2344 		 * locking an invalid object.  ocfs2_read_inode_block()
2345 		 * already checked that the inode block is sane.
2346 		 *
2347 		 * We bug on a stale inode here because we checked
2348 		 * above whether it was wiped from disk. The wiping
2349 		 * node provides a guarantee that we receive that
2350 		 * message and can mark the inode before dropping any
2351 		 * locks associated with it. */
2352 		mlog_bug_on_msg(inode->i_generation !=
2353 				le32_to_cpu(fe->i_generation),
2354 				"Invalid dinode %llu disk generation: %u "
2355 				"inode->i_generation: %u\n",
2356 				(unsigned long long)oi->ip_blkno,
2357 				le32_to_cpu(fe->i_generation),
2358 				inode->i_generation);
2359 		mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
2360 				!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
2361 				"Stale dinode %llu dtime: %llu flags: 0x%x\n",
2362 				(unsigned long long)oi->ip_blkno,
2363 				(unsigned long long)le64_to_cpu(fe->i_dtime),
2364 				le32_to_cpu(fe->i_flags));
2365 
2366 		ocfs2_refresh_inode(inode, fe);
2367 		ocfs2_track_lock_refresh(lockres);
2368 	}
2369 
2370 	status = 0;
2371 bail_refresh:
2372 	ocfs2_complete_lock_res_refresh(lockres, status);
2373 bail:
2374 	return status;
2375 }
2376 
ocfs2_assign_bh(struct inode * inode,struct buffer_head ** ret_bh,struct buffer_head * passed_bh)2377 static int ocfs2_assign_bh(struct inode *inode,
2378 			   struct buffer_head **ret_bh,
2379 			   struct buffer_head *passed_bh)
2380 {
2381 	int status;
2382 
2383 	if (passed_bh) {
2384 		/* Ok, the update went to disk for us, use the
2385 		 * returned bh. */
2386 		*ret_bh = passed_bh;
2387 		get_bh(*ret_bh);
2388 
2389 		return 0;
2390 	}
2391 
2392 	status = ocfs2_read_inode_block(inode, ret_bh);
2393 	if (status < 0)
2394 		mlog_errno(status);
2395 
2396 	return status;
2397 }
2398 
2399 /*
2400  * returns < 0 error if the callback will never be called, otherwise
2401  * the result of the lock will be communicated via the callback.
2402  */
ocfs2_inode_lock_full_nested(struct inode * inode,struct buffer_head ** ret_bh,int ex,int arg_flags,int subclass)2403 int ocfs2_inode_lock_full_nested(struct inode *inode,
2404 				 struct buffer_head **ret_bh,
2405 				 int ex,
2406 				 int arg_flags,
2407 				 int subclass)
2408 {
2409 	int status, level, acquired;
2410 	u32 dlm_flags;
2411 	struct ocfs2_lock_res *lockres = NULL;
2412 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2413 	struct buffer_head *local_bh = NULL;
2414 
2415 	mlog(0, "inode %llu, take %s META lock\n",
2416 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2417 	     ex ? "EXMODE" : "PRMODE");
2418 
2419 	status = 0;
2420 	acquired = 0;
2421 	/* We'll allow faking a readonly metadata lock for
2422 	 * rodevices. */
2423 	if (ocfs2_is_hard_readonly(osb)) {
2424 		if (ex)
2425 			status = -EROFS;
2426 		goto getbh;
2427 	}
2428 
2429 	if ((arg_flags & OCFS2_META_LOCK_GETBH) ||
2430 	    ocfs2_mount_local(osb))
2431 		goto update;
2432 
2433 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2434 		ocfs2_wait_for_recovery(osb);
2435 
2436 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2437 	level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2438 	dlm_flags = 0;
2439 	if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
2440 		dlm_flags |= DLM_LKF_NOQUEUE;
2441 
2442 	status = __ocfs2_cluster_lock(osb, lockres, level, dlm_flags,
2443 				      arg_flags, subclass, _RET_IP_);
2444 	if (status < 0) {
2445 		if (status != -EAGAIN)
2446 			mlog_errno(status);
2447 		goto bail;
2448 	}
2449 
2450 	/* Notify the error cleanup path to drop the cluster lock. */
2451 	acquired = 1;
2452 
2453 	/* We wait twice because a node may have died while we were in
2454 	 * the lower dlm layers. The second time though, we've
2455 	 * committed to owning this lock so we don't allow signals to
2456 	 * abort the operation. */
2457 	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2458 		ocfs2_wait_for_recovery(osb);
2459 
2460 update:
2461 	/*
2462 	 * We only see this flag if we're being called from
2463 	 * ocfs2_read_locked_inode(). It means we're locking an inode
2464 	 * which hasn't been populated yet, so clear the refresh flag
2465 	 * and let the caller handle it.
2466 	 */
2467 	if (inode->i_state & I_NEW) {
2468 		status = 0;
2469 		if (lockres)
2470 			ocfs2_complete_lock_res_refresh(lockres, 0);
2471 		goto bail;
2472 	}
2473 
2474 	/* This is fun. The caller may want a bh back, or it may
2475 	 * not. ocfs2_inode_lock_update definitely wants one in, but
2476 	 * may or may not read one, depending on what's in the
2477 	 * LVB. The result of all of this is that we've *only* gone to
2478 	 * disk if we have to, so the complexity is worthwhile. */
2479 	status = ocfs2_inode_lock_update(inode, &local_bh);
2480 	if (status < 0) {
2481 		if (status != -ENOENT)
2482 			mlog_errno(status);
2483 		goto bail;
2484 	}
2485 getbh:
2486 	if (ret_bh) {
2487 		status = ocfs2_assign_bh(inode, ret_bh, local_bh);
2488 		if (status < 0) {
2489 			mlog_errno(status);
2490 			goto bail;
2491 		}
2492 	}
2493 
2494 bail:
2495 	if (status < 0) {
2496 		if (ret_bh && (*ret_bh)) {
2497 			brelse(*ret_bh);
2498 			*ret_bh = NULL;
2499 		}
2500 		if (acquired)
2501 			ocfs2_inode_unlock(inode, ex);
2502 	}
2503 
2504 	if (local_bh)
2505 		brelse(local_bh);
2506 
2507 	return status;
2508 }
2509 
2510 /*
2511  * This is working around a lock inversion between tasks acquiring DLM
2512  * locks while holding a page lock and the downconvert thread which
2513  * blocks dlm lock acquiry while acquiring page locks.
2514  *
2515  * ** These _with_page variantes are only intended to be called from aop
2516  * methods that hold page locks and return a very specific *positive* error
2517  * code that aop methods pass up to the VFS -- test for errors with != 0. **
2518  *
2519  * The DLM is called such that it returns -EAGAIN if it would have
2520  * blocked waiting for the downconvert thread.  In that case we unlock
2521  * our page so the downconvert thread can make progress.  Once we've
2522  * done this we have to return AOP_TRUNCATED_PAGE so the aop method
2523  * that called us can bubble that back up into the VFS who will then
2524  * immediately retry the aop call.
2525  */
ocfs2_inode_lock_with_page(struct inode * inode,struct buffer_head ** ret_bh,int ex,struct page * page)2526 int ocfs2_inode_lock_with_page(struct inode *inode,
2527 			      struct buffer_head **ret_bh,
2528 			      int ex,
2529 			      struct page *page)
2530 {
2531 	int ret;
2532 
2533 	ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
2534 	if (ret == -EAGAIN) {
2535 		unlock_page(page);
2536 		/*
2537 		 * If we can't get inode lock immediately, we should not return
2538 		 * directly here, since this will lead to a softlockup problem.
2539 		 * The method is to get a blocking lock and immediately unlock
2540 		 * before returning, this can avoid CPU resource waste due to
2541 		 * lots of retries, and benefits fairness in getting lock.
2542 		 */
2543 		if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
2544 			ocfs2_inode_unlock(inode, ex);
2545 		ret = AOP_TRUNCATED_PAGE;
2546 	}
2547 
2548 	return ret;
2549 }
2550 
ocfs2_inode_lock_atime(struct inode * inode,struct vfsmount * vfsmnt,int * level,int wait)2551 int ocfs2_inode_lock_atime(struct inode *inode,
2552 			  struct vfsmount *vfsmnt,
2553 			  int *level, int wait)
2554 {
2555 	int ret;
2556 
2557 	if (wait)
2558 		ret = ocfs2_inode_lock(inode, NULL, 0);
2559 	else
2560 		ret = ocfs2_try_inode_lock(inode, NULL, 0);
2561 
2562 	if (ret < 0) {
2563 		if (ret != -EAGAIN)
2564 			mlog_errno(ret);
2565 		return ret;
2566 	}
2567 
2568 	/*
2569 	 * If we should update atime, we will get EX lock,
2570 	 * otherwise we just get PR lock.
2571 	 */
2572 	if (ocfs2_should_update_atime(inode, vfsmnt)) {
2573 		struct buffer_head *bh = NULL;
2574 
2575 		ocfs2_inode_unlock(inode, 0);
2576 		if (wait)
2577 			ret = ocfs2_inode_lock(inode, &bh, 1);
2578 		else
2579 			ret = ocfs2_try_inode_lock(inode, &bh, 1);
2580 
2581 		if (ret < 0) {
2582 			if (ret != -EAGAIN)
2583 				mlog_errno(ret);
2584 			return ret;
2585 		}
2586 		*level = 1;
2587 		if (ocfs2_should_update_atime(inode, vfsmnt))
2588 			ocfs2_update_inode_atime(inode, bh);
2589 		if (bh)
2590 			brelse(bh);
2591 	} else
2592 		*level = 0;
2593 
2594 	return ret;
2595 }
2596 
ocfs2_inode_unlock(struct inode * inode,int ex)2597 void ocfs2_inode_unlock(struct inode *inode,
2598 		       int ex)
2599 {
2600 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2601 	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
2602 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2603 
2604 	mlog(0, "inode %llu drop %s META lock\n",
2605 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2606 	     ex ? "EXMODE" : "PRMODE");
2607 
2608 	if (!ocfs2_is_hard_readonly(osb) &&
2609 	    !ocfs2_mount_local(osb))
2610 		ocfs2_cluster_unlock(osb, lockres, level);
2611 }
2612 
2613 /*
2614  * This _tracker variantes are introduced to deal with the recursive cluster
2615  * locking issue. The idea is to keep track of a lock holder on the stack of
2616  * the current process. If there's a lock holder on the stack, we know the
2617  * task context is already protected by cluster locking. Currently, they're
2618  * used in some VFS entry routines.
2619  *
2620  * return < 0 on error, return == 0 if there's no lock holder on the stack
2621  * before this call, return == 1 if this call would be a recursive locking.
2622  * return == -1 if this lock attempt will cause an upgrade which is forbidden.
2623  *
2624  * When taking lock levels into account,we face some different situations.
2625  *
2626  * 1. no lock is held
2627  *    In this case, just lock the inode as requested and return 0
2628  *
2629  * 2. We are holding a lock
2630  *    For this situation, things diverges into several cases
2631  *
2632  *    wanted     holding	     what to do
2633  *    ex		ex	    see 2.1 below
2634  *    ex		pr	    see 2.2 below
2635  *    pr		ex	    see 2.1 below
2636  *    pr		pr	    see 2.1 below
2637  *
2638  *    2.1 lock level that is been held is compatible
2639  *    with the wanted level, so no lock action will be tacken.
2640  *
2641  *    2.2 Otherwise, an upgrade is needed, but it is forbidden.
2642  *
2643  * Reason why upgrade within a process is forbidden is that
2644  * lock upgrade may cause dead lock. The following illustrates
2645  * how it happens.
2646  *
2647  *         thread on node1                             thread on node2
2648  * ocfs2_inode_lock_tracker(ex=0)
2649  *
2650  *                                <======   ocfs2_inode_lock_tracker(ex=1)
2651  *
2652  * ocfs2_inode_lock_tracker(ex=1)
2653  */
ocfs2_inode_lock_tracker(struct inode * inode,struct buffer_head ** ret_bh,int ex,struct ocfs2_lock_holder * oh)2654 int ocfs2_inode_lock_tracker(struct inode *inode,
2655 			     struct buffer_head **ret_bh,
2656 			     int ex,
2657 			     struct ocfs2_lock_holder *oh)
2658 {
2659 	int status = 0;
2660 	struct ocfs2_lock_res *lockres;
2661 	struct ocfs2_lock_holder *tmp_oh;
2662 	struct pid *pid = task_pid(current);
2663 
2664 
2665 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2666 	tmp_oh = ocfs2_pid_holder(lockres, pid);
2667 
2668 	if (!tmp_oh) {
2669 		/*
2670 		 * This corresponds to the case 1.
2671 		 * We haven't got any lock before.
2672 		 */
2673 		status = ocfs2_inode_lock_full(inode, ret_bh, ex, 0);
2674 		if (status < 0) {
2675 			if (status != -ENOENT)
2676 				mlog_errno(status);
2677 			return status;
2678 		}
2679 
2680 		oh->oh_ex = ex;
2681 		ocfs2_add_holder(lockres, oh);
2682 		return 0;
2683 	}
2684 
2685 	if (unlikely(ex && !tmp_oh->oh_ex)) {
2686 		/*
2687 		 * case 2.2 upgrade may cause dead lock, forbid it.
2688 		 */
2689 		mlog(ML_ERROR, "Recursive locking is not permitted to "
2690 		     "upgrade to EX level from PR level.\n");
2691 		dump_stack();
2692 		return -EINVAL;
2693 	}
2694 
2695 	/*
2696 	 *  case 2.1 OCFS2_META_LOCK_GETBH flag make ocfs2_inode_lock_full.
2697 	 *  ignore the lock level and just update it.
2698 	 */
2699 	if (ret_bh) {
2700 		status = ocfs2_inode_lock_full(inode, ret_bh, ex,
2701 					       OCFS2_META_LOCK_GETBH);
2702 		if (status < 0) {
2703 			if (status != -ENOENT)
2704 				mlog_errno(status);
2705 			return status;
2706 		}
2707 	}
2708 	return tmp_oh ? 1 : 0;
2709 }
2710 
ocfs2_inode_unlock_tracker(struct inode * inode,int ex,struct ocfs2_lock_holder * oh,int had_lock)2711 void ocfs2_inode_unlock_tracker(struct inode *inode,
2712 				int ex,
2713 				struct ocfs2_lock_holder *oh,
2714 				int had_lock)
2715 {
2716 	struct ocfs2_lock_res *lockres;
2717 
2718 	lockres = &OCFS2_I(inode)->ip_inode_lockres;
2719 	/* had_lock means that the currect process already takes the cluster
2720 	 * lock previously.
2721 	 * If had_lock is 1, we have nothing to do here.
2722 	 * If had_lock is 0, we will release the lock.
2723 	 */
2724 	if (!had_lock) {
2725 		ocfs2_inode_unlock(inode, oh->oh_ex);
2726 		ocfs2_remove_holder(lockres, oh);
2727 	}
2728 }
2729 
ocfs2_orphan_scan_lock(struct ocfs2_super * osb,u32 * seqno)2730 int ocfs2_orphan_scan_lock(struct ocfs2_super *osb, u32 *seqno)
2731 {
2732 	struct ocfs2_lock_res *lockres;
2733 	struct ocfs2_orphan_scan_lvb *lvb;
2734 	int status = 0;
2735 
2736 	if (ocfs2_is_hard_readonly(osb))
2737 		return -EROFS;
2738 
2739 	if (ocfs2_mount_local(osb))
2740 		return 0;
2741 
2742 	lockres = &osb->osb_orphan_scan.os_lockres;
2743 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2744 	if (status < 0)
2745 		return status;
2746 
2747 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2748 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2749 	    lvb->lvb_version == OCFS2_ORPHAN_LVB_VERSION)
2750 		*seqno = be32_to_cpu(lvb->lvb_os_seqno);
2751 	else
2752 		*seqno = osb->osb_orphan_scan.os_seqno + 1;
2753 
2754 	return status;
2755 }
2756 
ocfs2_orphan_scan_unlock(struct ocfs2_super * osb,u32 seqno)2757 void ocfs2_orphan_scan_unlock(struct ocfs2_super *osb, u32 seqno)
2758 {
2759 	struct ocfs2_lock_res *lockres;
2760 	struct ocfs2_orphan_scan_lvb *lvb;
2761 
2762 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) {
2763 		lockres = &osb->osb_orphan_scan.os_lockres;
2764 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2765 		lvb->lvb_version = OCFS2_ORPHAN_LVB_VERSION;
2766 		lvb->lvb_os_seqno = cpu_to_be32(seqno);
2767 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2768 	}
2769 }
2770 
ocfs2_super_lock(struct ocfs2_super * osb,int ex)2771 int ocfs2_super_lock(struct ocfs2_super *osb,
2772 		     int ex)
2773 {
2774 	int status = 0;
2775 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2776 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2777 
2778 	if (ocfs2_is_hard_readonly(osb))
2779 		return -EROFS;
2780 
2781 	if (ocfs2_mount_local(osb))
2782 		goto bail;
2783 
2784 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
2785 	if (status < 0) {
2786 		mlog_errno(status);
2787 		goto bail;
2788 	}
2789 
2790 	/* The super block lock path is really in the best position to
2791 	 * know when resources covered by the lock need to be
2792 	 * refreshed, so we do it here. Of course, making sense of
2793 	 * everything is up to the caller :) */
2794 	status = ocfs2_should_refresh_lock_res(lockres);
2795 	if (status) {
2796 		status = ocfs2_refresh_slot_info(osb);
2797 
2798 		ocfs2_complete_lock_res_refresh(lockres, status);
2799 
2800 		if (status < 0) {
2801 			ocfs2_cluster_unlock(osb, lockres, level);
2802 			mlog_errno(status);
2803 		}
2804 		ocfs2_track_lock_refresh(lockres);
2805 	}
2806 bail:
2807 	return status;
2808 }
2809 
ocfs2_super_unlock(struct ocfs2_super * osb,int ex)2810 void ocfs2_super_unlock(struct ocfs2_super *osb,
2811 			int ex)
2812 {
2813 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2814 	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2815 
2816 	if (!ocfs2_mount_local(osb))
2817 		ocfs2_cluster_unlock(osb, lockres, level);
2818 }
2819 
ocfs2_rename_lock(struct ocfs2_super * osb)2820 int ocfs2_rename_lock(struct ocfs2_super *osb)
2821 {
2822 	int status;
2823 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2824 
2825 	if (ocfs2_is_hard_readonly(osb))
2826 		return -EROFS;
2827 
2828 	if (ocfs2_mount_local(osb))
2829 		return 0;
2830 
2831 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2832 	if (status < 0)
2833 		mlog_errno(status);
2834 
2835 	return status;
2836 }
2837 
ocfs2_rename_unlock(struct ocfs2_super * osb)2838 void ocfs2_rename_unlock(struct ocfs2_super *osb)
2839 {
2840 	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2841 
2842 	if (!ocfs2_mount_local(osb))
2843 		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2844 }
2845 
ocfs2_nfs_sync_lock(struct ocfs2_super * osb,int ex)2846 int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex)
2847 {
2848 	int status;
2849 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2850 
2851 	if (ocfs2_is_hard_readonly(osb))
2852 		return -EROFS;
2853 
2854 	if (ocfs2_mount_local(osb))
2855 		return 0;
2856 
2857 	status = ocfs2_cluster_lock(osb, lockres, ex ? LKM_EXMODE : LKM_PRMODE,
2858 				    0, 0);
2859 	if (status < 0)
2860 		mlog(ML_ERROR, "lock on nfs sync lock failed %d\n", status);
2861 
2862 	return status;
2863 }
2864 
ocfs2_nfs_sync_unlock(struct ocfs2_super * osb,int ex)2865 void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex)
2866 {
2867 	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2868 
2869 	if (!ocfs2_mount_local(osb))
2870 		ocfs2_cluster_unlock(osb, lockres,
2871 				     ex ? LKM_EXMODE : LKM_PRMODE);
2872 }
2873 
ocfs2_trim_fs_lock(struct ocfs2_super * osb,struct ocfs2_trim_fs_info * info,int trylock)2874 int ocfs2_trim_fs_lock(struct ocfs2_super *osb,
2875 		       struct ocfs2_trim_fs_info *info, int trylock)
2876 {
2877 	int status;
2878 	struct ocfs2_trim_fs_lvb *lvb;
2879 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
2880 
2881 	if (info)
2882 		info->tf_valid = 0;
2883 
2884 	if (ocfs2_is_hard_readonly(osb))
2885 		return -EROFS;
2886 
2887 	if (ocfs2_mount_local(osb))
2888 		return 0;
2889 
2890 	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX,
2891 				    trylock ? DLM_LKF_NOQUEUE : 0, 0);
2892 	if (status < 0) {
2893 		if (status != -EAGAIN)
2894 			mlog_errno(status);
2895 		return status;
2896 	}
2897 
2898 	if (info) {
2899 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2900 		if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2901 		    lvb->lvb_version == OCFS2_TRIMFS_LVB_VERSION) {
2902 			info->tf_valid = 1;
2903 			info->tf_success = lvb->lvb_success;
2904 			info->tf_nodenum = be32_to_cpu(lvb->lvb_nodenum);
2905 			info->tf_start = be64_to_cpu(lvb->lvb_start);
2906 			info->tf_len = be64_to_cpu(lvb->lvb_len);
2907 			info->tf_minlen = be64_to_cpu(lvb->lvb_minlen);
2908 			info->tf_trimlen = be64_to_cpu(lvb->lvb_trimlen);
2909 		}
2910 	}
2911 
2912 	return status;
2913 }
2914 
ocfs2_trim_fs_unlock(struct ocfs2_super * osb,struct ocfs2_trim_fs_info * info)2915 void ocfs2_trim_fs_unlock(struct ocfs2_super *osb,
2916 			  struct ocfs2_trim_fs_info *info)
2917 {
2918 	struct ocfs2_trim_fs_lvb *lvb;
2919 	struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
2920 
2921 	if (ocfs2_mount_local(osb))
2922 		return;
2923 
2924 	if (info) {
2925 		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2926 		lvb->lvb_version = OCFS2_TRIMFS_LVB_VERSION;
2927 		lvb->lvb_success = info->tf_success;
2928 		lvb->lvb_nodenum = cpu_to_be32(info->tf_nodenum);
2929 		lvb->lvb_start = cpu_to_be64(info->tf_start);
2930 		lvb->lvb_len = cpu_to_be64(info->tf_len);
2931 		lvb->lvb_minlen = cpu_to_be64(info->tf_minlen);
2932 		lvb->lvb_trimlen = cpu_to_be64(info->tf_trimlen);
2933 	}
2934 
2935 	ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2936 }
2937 
ocfs2_dentry_lock(struct dentry * dentry,int ex)2938 int ocfs2_dentry_lock(struct dentry *dentry, int ex)
2939 {
2940 	int ret;
2941 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2942 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2943 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2944 
2945 	BUG_ON(!dl);
2946 
2947 	if (ocfs2_is_hard_readonly(osb)) {
2948 		if (ex)
2949 			return -EROFS;
2950 		return 0;
2951 	}
2952 
2953 	if (ocfs2_mount_local(osb))
2954 		return 0;
2955 
2956 	ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2957 	if (ret < 0)
2958 		mlog_errno(ret);
2959 
2960 	return ret;
2961 }
2962 
ocfs2_dentry_unlock(struct dentry * dentry,int ex)2963 void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2964 {
2965 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2966 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2967 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2968 
2969 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
2970 		ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
2971 }
2972 
2973 /* Reference counting of the dlm debug structure. We want this because
2974  * open references on the debug inodes can live on after a mount, so
2975  * we can't rely on the ocfs2_super to always exist. */
ocfs2_dlm_debug_free(struct kref * kref)2976 static void ocfs2_dlm_debug_free(struct kref *kref)
2977 {
2978 	struct ocfs2_dlm_debug *dlm_debug;
2979 
2980 	dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
2981 
2982 	kfree(dlm_debug);
2983 }
2984 
ocfs2_put_dlm_debug(struct ocfs2_dlm_debug * dlm_debug)2985 void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
2986 {
2987 	if (dlm_debug)
2988 		kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
2989 }
2990 
ocfs2_get_dlm_debug(struct ocfs2_dlm_debug * debug)2991 static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
2992 {
2993 	kref_get(&debug->d_refcnt);
2994 }
2995 
ocfs2_new_dlm_debug(void)2996 struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
2997 {
2998 	struct ocfs2_dlm_debug *dlm_debug;
2999 
3000 	dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
3001 	if (!dlm_debug) {
3002 		mlog_errno(-ENOMEM);
3003 		goto out;
3004 	}
3005 
3006 	kref_init(&dlm_debug->d_refcnt);
3007 	INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
3008 	dlm_debug->d_locking_state = NULL;
3009 out:
3010 	return dlm_debug;
3011 }
3012 
3013 /* Access to this is arbitrated for us via seq_file->sem. */
3014 struct ocfs2_dlm_seq_priv {
3015 	struct ocfs2_dlm_debug *p_dlm_debug;
3016 	struct ocfs2_lock_res p_iter_res;
3017 	struct ocfs2_lock_res p_tmp_res;
3018 };
3019 
ocfs2_dlm_next_res(struct ocfs2_lock_res * start,struct ocfs2_dlm_seq_priv * priv)3020 static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
3021 						 struct ocfs2_dlm_seq_priv *priv)
3022 {
3023 	struct ocfs2_lock_res *iter, *ret = NULL;
3024 	struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
3025 
3026 	assert_spin_locked(&ocfs2_dlm_tracking_lock);
3027 
3028 	list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
3029 		/* discover the head of the list */
3030 		if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
3031 			mlog(0, "End of list found, %p\n", ret);
3032 			break;
3033 		}
3034 
3035 		/* We track our "dummy" iteration lockres' by a NULL
3036 		 * l_ops field. */
3037 		if (iter->l_ops != NULL) {
3038 			ret = iter;
3039 			break;
3040 		}
3041 	}
3042 
3043 	return ret;
3044 }
3045 
ocfs2_dlm_seq_start(struct seq_file * m,loff_t * pos)3046 static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
3047 {
3048 	struct ocfs2_dlm_seq_priv *priv = m->private;
3049 	struct ocfs2_lock_res *iter;
3050 
3051 	spin_lock(&ocfs2_dlm_tracking_lock);
3052 	iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
3053 	if (iter) {
3054 		/* Since lockres' have the lifetime of their container
3055 		 * (which can be inodes, ocfs2_supers, etc) we want to
3056 		 * copy this out to a temporary lockres while still
3057 		 * under the spinlock. Obviously after this we can't
3058 		 * trust any pointers on the copy returned, but that's
3059 		 * ok as the information we want isn't typically held
3060 		 * in them. */
3061 		priv->p_tmp_res = *iter;
3062 		iter = &priv->p_tmp_res;
3063 	}
3064 	spin_unlock(&ocfs2_dlm_tracking_lock);
3065 
3066 	return iter;
3067 }
3068 
ocfs2_dlm_seq_stop(struct seq_file * m,void * v)3069 static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
3070 {
3071 }
3072 
ocfs2_dlm_seq_next(struct seq_file * m,void * v,loff_t * pos)3073 static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
3074 {
3075 	struct ocfs2_dlm_seq_priv *priv = m->private;
3076 	struct ocfs2_lock_res *iter = v;
3077 	struct ocfs2_lock_res *dummy = &priv->p_iter_res;
3078 
3079 	spin_lock(&ocfs2_dlm_tracking_lock);
3080 	iter = ocfs2_dlm_next_res(iter, priv);
3081 	list_del_init(&dummy->l_debug_list);
3082 	if (iter) {
3083 		list_add(&dummy->l_debug_list, &iter->l_debug_list);
3084 		priv->p_tmp_res = *iter;
3085 		iter = &priv->p_tmp_res;
3086 	}
3087 	spin_unlock(&ocfs2_dlm_tracking_lock);
3088 
3089 	return iter;
3090 }
3091 
3092 /*
3093  * Version is used by debugfs.ocfs2 to determine the format being used
3094  *
3095  * New in version 2
3096  *	- Lock stats printed
3097  * New in version 3
3098  *	- Max time in lock stats is in usecs (instead of nsecs)
3099  */
3100 #define OCFS2_DLM_DEBUG_STR_VERSION 3
ocfs2_dlm_seq_show(struct seq_file * m,void * v)3101 static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
3102 {
3103 	int i;
3104 	char *lvb;
3105 	struct ocfs2_lock_res *lockres = v;
3106 
3107 	if (!lockres)
3108 		return -EINVAL;
3109 
3110 	seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
3111 
3112 	if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
3113 		seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
3114 			   lockres->l_name,
3115 			   (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
3116 	else
3117 		seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
3118 
3119 	seq_printf(m, "%d\t"
3120 		   "0x%lx\t"
3121 		   "0x%x\t"
3122 		   "0x%x\t"
3123 		   "%u\t"
3124 		   "%u\t"
3125 		   "%d\t"
3126 		   "%d\t",
3127 		   lockres->l_level,
3128 		   lockres->l_flags,
3129 		   lockres->l_action,
3130 		   lockres->l_unlock_action,
3131 		   lockres->l_ro_holders,
3132 		   lockres->l_ex_holders,
3133 		   lockres->l_requested,
3134 		   lockres->l_blocking);
3135 
3136 	/* Dump the raw LVB */
3137 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
3138 	for(i = 0; i < DLM_LVB_LEN; i++)
3139 		seq_printf(m, "0x%x\t", lvb[i]);
3140 
3141 #ifdef CONFIG_OCFS2_FS_STATS
3142 # define lock_num_prmode(_l)		((_l)->l_lock_prmode.ls_gets)
3143 # define lock_num_exmode(_l)		((_l)->l_lock_exmode.ls_gets)
3144 # define lock_num_prmode_failed(_l)	((_l)->l_lock_prmode.ls_fail)
3145 # define lock_num_exmode_failed(_l)	((_l)->l_lock_exmode.ls_fail)
3146 # define lock_total_prmode(_l)		((_l)->l_lock_prmode.ls_total)
3147 # define lock_total_exmode(_l)		((_l)->l_lock_exmode.ls_total)
3148 # define lock_max_prmode(_l)		((_l)->l_lock_prmode.ls_max)
3149 # define lock_max_exmode(_l)		((_l)->l_lock_exmode.ls_max)
3150 # define lock_refresh(_l)		((_l)->l_lock_refresh)
3151 #else
3152 # define lock_num_prmode(_l)		(0)
3153 # define lock_num_exmode(_l)		(0)
3154 # define lock_num_prmode_failed(_l)	(0)
3155 # define lock_num_exmode_failed(_l)	(0)
3156 # define lock_total_prmode(_l)		(0ULL)
3157 # define lock_total_exmode(_l)		(0ULL)
3158 # define lock_max_prmode(_l)		(0)
3159 # define lock_max_exmode(_l)		(0)
3160 # define lock_refresh(_l)		(0)
3161 #endif
3162 	/* The following seq_print was added in version 2 of this output */
3163 	seq_printf(m, "%u\t"
3164 		   "%u\t"
3165 		   "%u\t"
3166 		   "%u\t"
3167 		   "%llu\t"
3168 		   "%llu\t"
3169 		   "%u\t"
3170 		   "%u\t"
3171 		   "%u\t",
3172 		   lock_num_prmode(lockres),
3173 		   lock_num_exmode(lockres),
3174 		   lock_num_prmode_failed(lockres),
3175 		   lock_num_exmode_failed(lockres),
3176 		   lock_total_prmode(lockres),
3177 		   lock_total_exmode(lockres),
3178 		   lock_max_prmode(lockres),
3179 		   lock_max_exmode(lockres),
3180 		   lock_refresh(lockres));
3181 
3182 	/* End the line */
3183 	seq_printf(m, "\n");
3184 	return 0;
3185 }
3186 
3187 static const struct seq_operations ocfs2_dlm_seq_ops = {
3188 	.start =	ocfs2_dlm_seq_start,
3189 	.stop =		ocfs2_dlm_seq_stop,
3190 	.next =		ocfs2_dlm_seq_next,
3191 	.show =		ocfs2_dlm_seq_show,
3192 };
3193 
ocfs2_dlm_debug_release(struct inode * inode,struct file * file)3194 static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
3195 {
3196 	struct seq_file *seq = file->private_data;
3197 	struct ocfs2_dlm_seq_priv *priv = seq->private;
3198 	struct ocfs2_lock_res *res = &priv->p_iter_res;
3199 
3200 	ocfs2_remove_lockres_tracking(res);
3201 	ocfs2_put_dlm_debug(priv->p_dlm_debug);
3202 	return seq_release_private(inode, file);
3203 }
3204 
ocfs2_dlm_debug_open(struct inode * inode,struct file * file)3205 static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
3206 {
3207 	struct ocfs2_dlm_seq_priv *priv;
3208 	struct ocfs2_super *osb;
3209 
3210 	priv = __seq_open_private(file, &ocfs2_dlm_seq_ops, sizeof(*priv));
3211 	if (!priv) {
3212 		mlog_errno(-ENOMEM);
3213 		return -ENOMEM;
3214 	}
3215 
3216 	osb = inode->i_private;
3217 	ocfs2_get_dlm_debug(osb->osb_dlm_debug);
3218 	priv->p_dlm_debug = osb->osb_dlm_debug;
3219 	INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
3220 
3221 	ocfs2_add_lockres_tracking(&priv->p_iter_res,
3222 				   priv->p_dlm_debug);
3223 
3224 	return 0;
3225 }
3226 
3227 static const struct file_operations ocfs2_dlm_debug_fops = {
3228 	.open =		ocfs2_dlm_debug_open,
3229 	.release =	ocfs2_dlm_debug_release,
3230 	.read =		seq_read,
3231 	.llseek =	seq_lseek,
3232 };
3233 
ocfs2_dlm_init_debug(struct ocfs2_super * osb)3234 static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
3235 {
3236 	int ret = 0;
3237 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
3238 
3239 	dlm_debug->d_locking_state = debugfs_create_file("locking_state",
3240 							 S_IFREG|S_IRUSR,
3241 							 osb->osb_debug_root,
3242 							 osb,
3243 							 &ocfs2_dlm_debug_fops);
3244 	if (!dlm_debug->d_locking_state) {
3245 		ret = -EINVAL;
3246 		mlog(ML_ERROR,
3247 		     "Unable to create locking state debugfs file.\n");
3248 		goto out;
3249 	}
3250 
3251 	ocfs2_get_dlm_debug(dlm_debug);
3252 out:
3253 	return ret;
3254 }
3255 
ocfs2_dlm_shutdown_debug(struct ocfs2_super * osb)3256 static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
3257 {
3258 	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
3259 
3260 	if (dlm_debug) {
3261 		debugfs_remove(dlm_debug->d_locking_state);
3262 		ocfs2_put_dlm_debug(dlm_debug);
3263 	}
3264 }
3265 
ocfs2_dlm_init(struct ocfs2_super * osb)3266 int ocfs2_dlm_init(struct ocfs2_super *osb)
3267 {
3268 	int status = 0;
3269 	struct ocfs2_cluster_connection *conn = NULL;
3270 
3271 	if (ocfs2_mount_local(osb)) {
3272 		osb->node_num = 0;
3273 		goto local;
3274 	}
3275 
3276 	status = ocfs2_dlm_init_debug(osb);
3277 	if (status < 0) {
3278 		mlog_errno(status);
3279 		goto bail;
3280 	}
3281 
3282 	/* launch downconvert thread */
3283 	osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc-%s",
3284 			osb->uuid_str);
3285 	if (IS_ERR(osb->dc_task)) {
3286 		status = PTR_ERR(osb->dc_task);
3287 		osb->dc_task = NULL;
3288 		mlog_errno(status);
3289 		goto bail;
3290 	}
3291 
3292 	/* for now, uuid == domain */
3293 	status = ocfs2_cluster_connect(osb->osb_cluster_stack,
3294 				       osb->osb_cluster_name,
3295 				       strlen(osb->osb_cluster_name),
3296 				       osb->uuid_str,
3297 				       strlen(osb->uuid_str),
3298 				       &lproto, ocfs2_do_node_down, osb,
3299 				       &conn);
3300 	if (status) {
3301 		mlog_errno(status);
3302 		goto bail;
3303 	}
3304 
3305 	status = ocfs2_cluster_this_node(conn, &osb->node_num);
3306 	if (status < 0) {
3307 		mlog_errno(status);
3308 		mlog(ML_ERROR,
3309 		     "could not find this host's node number\n");
3310 		ocfs2_cluster_disconnect(conn, 0);
3311 		goto bail;
3312 	}
3313 
3314 local:
3315 	ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
3316 	ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
3317 	ocfs2_nfs_sync_lock_res_init(&osb->osb_nfs_sync_lockres, osb);
3318 	ocfs2_orphan_scan_lock_res_init(&osb->osb_orphan_scan.os_lockres, osb);
3319 
3320 	osb->cconn = conn;
3321 bail:
3322 	if (status < 0) {
3323 		ocfs2_dlm_shutdown_debug(osb);
3324 		if (osb->dc_task)
3325 			kthread_stop(osb->dc_task);
3326 	}
3327 
3328 	return status;
3329 }
3330 
ocfs2_dlm_shutdown(struct ocfs2_super * osb,int hangup_pending)3331 void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
3332 			int hangup_pending)
3333 {
3334 	ocfs2_drop_osb_locks(osb);
3335 
3336 	/*
3337 	 * Now that we have dropped all locks and ocfs2_dismount_volume()
3338 	 * has disabled recovery, the DLM won't be talking to us.  It's
3339 	 * safe to tear things down before disconnecting the cluster.
3340 	 */
3341 
3342 	if (osb->dc_task) {
3343 		kthread_stop(osb->dc_task);
3344 		osb->dc_task = NULL;
3345 	}
3346 
3347 	ocfs2_lock_res_free(&osb->osb_super_lockres);
3348 	ocfs2_lock_res_free(&osb->osb_rename_lockres);
3349 	ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres);
3350 	ocfs2_lock_res_free(&osb->osb_orphan_scan.os_lockres);
3351 
3352 	ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
3353 	osb->cconn = NULL;
3354 
3355 	ocfs2_dlm_shutdown_debug(osb);
3356 }
3357 
ocfs2_drop_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)3358 static int ocfs2_drop_lock(struct ocfs2_super *osb,
3359 			   struct ocfs2_lock_res *lockres)
3360 {
3361 	int ret;
3362 	unsigned long flags;
3363 	u32 lkm_flags = 0;
3364 
3365 	/* We didn't get anywhere near actually using this lockres. */
3366 	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
3367 		goto out;
3368 
3369 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
3370 		lkm_flags |= DLM_LKF_VALBLK;
3371 
3372 	spin_lock_irqsave(&lockres->l_lock, flags);
3373 
3374 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
3375 			"lockres %s, flags 0x%lx\n",
3376 			lockres->l_name, lockres->l_flags);
3377 
3378 	while (lockres->l_flags & OCFS2_LOCK_BUSY) {
3379 		mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
3380 		     "%u, unlock_action = %u\n",
3381 		     lockres->l_name, lockres->l_flags, lockres->l_action,
3382 		     lockres->l_unlock_action);
3383 
3384 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3385 
3386 		/* XXX: Today we just wait on any busy
3387 		 * locks... Perhaps we need to cancel converts in the
3388 		 * future? */
3389 		ocfs2_wait_on_busy_lock(lockres);
3390 
3391 		spin_lock_irqsave(&lockres->l_lock, flags);
3392 	}
3393 
3394 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3395 		if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
3396 		    lockres->l_level == DLM_LOCK_EX &&
3397 		    !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3398 			lockres->l_ops->set_lvb(lockres);
3399 	}
3400 
3401 	if (lockres->l_flags & OCFS2_LOCK_BUSY)
3402 		mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
3403 		     lockres->l_name);
3404 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
3405 		mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
3406 
3407 	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
3408 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3409 		goto out;
3410 	}
3411 
3412 	lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
3413 
3414 	/* make sure we never get here while waiting for an ast to
3415 	 * fire. */
3416 	BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
3417 
3418 	/* is this necessary? */
3419 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3420 	lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
3421 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3422 
3423 	mlog(0, "lock %s\n", lockres->l_name);
3424 
3425 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags);
3426 	if (ret) {
3427 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3428 		mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
3429 		ocfs2_dlm_dump_lksb(&lockres->l_lksb);
3430 		BUG();
3431 	}
3432 	mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n",
3433 	     lockres->l_name);
3434 
3435 	ocfs2_wait_on_busy_lock(lockres);
3436 out:
3437 	return 0;
3438 }
3439 
3440 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3441 				       struct ocfs2_lock_res *lockres);
3442 
3443 /* Mark the lockres as being dropped. It will no longer be
3444  * queued if blocking, but we still may have to wait on it
3445  * being dequeued from the downconvert thread before we can consider
3446  * it safe to drop.
3447  *
3448  * 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)3449 void ocfs2_mark_lockres_freeing(struct ocfs2_super *osb,
3450 				struct ocfs2_lock_res *lockres)
3451 {
3452 	int status;
3453 	struct ocfs2_mask_waiter mw;
3454 	unsigned long flags, flags2;
3455 
3456 	ocfs2_init_mask_waiter(&mw);
3457 
3458 	spin_lock_irqsave(&lockres->l_lock, flags);
3459 	lockres->l_flags |= OCFS2_LOCK_FREEING;
3460 	if (lockres->l_flags & OCFS2_LOCK_QUEUED && current == osb->dc_task) {
3461 		/*
3462 		 * We know the downconvert is queued but not in progress
3463 		 * because we are the downconvert thread and processing
3464 		 * different lock. So we can just remove the lock from the
3465 		 * queue. This is not only an optimization but also a way
3466 		 * to avoid the following deadlock:
3467 		 *   ocfs2_dentry_post_unlock()
3468 		 *     ocfs2_dentry_lock_put()
3469 		 *       ocfs2_drop_dentry_lock()
3470 		 *         iput()
3471 		 *           ocfs2_evict_inode()
3472 		 *             ocfs2_clear_inode()
3473 		 *               ocfs2_mark_lockres_freeing()
3474 		 *                 ... blocks waiting for OCFS2_LOCK_QUEUED
3475 		 *                 since we are the downconvert thread which
3476 		 *                 should clear the flag.
3477 		 */
3478 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3479 		spin_lock_irqsave(&osb->dc_task_lock, flags2);
3480 		list_del_init(&lockres->l_blocked_list);
3481 		osb->blocked_lock_count--;
3482 		spin_unlock_irqrestore(&osb->dc_task_lock, flags2);
3483 		/*
3484 		 * Warn if we recurse into another post_unlock call.  Strictly
3485 		 * speaking it isn't a problem but we need to be careful if
3486 		 * that happens (stack overflow, deadlocks, ...) so warn if
3487 		 * ocfs2 grows a path for which this can happen.
3488 		 */
3489 		WARN_ON_ONCE(lockres->l_ops->post_unlock);
3490 		/* Since the lock is freeing we don't do much in the fn below */
3491 		ocfs2_process_blocked_lock(osb, lockres);
3492 		return;
3493 	}
3494 	while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
3495 		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
3496 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3497 
3498 		mlog(0, "Waiting on lockres %s\n", lockres->l_name);
3499 
3500 		status = ocfs2_wait_for_mask(&mw);
3501 		if (status)
3502 			mlog_errno(status);
3503 
3504 		spin_lock_irqsave(&lockres->l_lock, flags);
3505 	}
3506 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3507 }
3508 
ocfs2_simple_drop_lockres(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)3509 void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
3510 			       struct ocfs2_lock_res *lockres)
3511 {
3512 	int ret;
3513 
3514 	ocfs2_mark_lockres_freeing(osb, lockres);
3515 	ret = ocfs2_drop_lock(osb, lockres);
3516 	if (ret)
3517 		mlog_errno(ret);
3518 }
3519 
ocfs2_drop_osb_locks(struct ocfs2_super * osb)3520 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
3521 {
3522 	ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
3523 	ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
3524 	ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres);
3525 	ocfs2_simple_drop_lockres(osb, &osb->osb_orphan_scan.os_lockres);
3526 }
3527 
ocfs2_drop_inode_locks(struct inode * inode)3528 int ocfs2_drop_inode_locks(struct inode *inode)
3529 {
3530 	int status, err;
3531 
3532 	/* No need to call ocfs2_mark_lockres_freeing here -
3533 	 * ocfs2_clear_inode has done it for us. */
3534 
3535 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3536 			      &OCFS2_I(inode)->ip_open_lockres);
3537 	if (err < 0)
3538 		mlog_errno(err);
3539 
3540 	status = err;
3541 
3542 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3543 			      &OCFS2_I(inode)->ip_inode_lockres);
3544 	if (err < 0)
3545 		mlog_errno(err);
3546 	if (err < 0 && !status)
3547 		status = err;
3548 
3549 	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3550 			      &OCFS2_I(inode)->ip_rw_lockres);
3551 	if (err < 0)
3552 		mlog_errno(err);
3553 	if (err < 0 && !status)
3554 		status = err;
3555 
3556 	return status;
3557 }
3558 
ocfs2_prepare_downconvert(struct ocfs2_lock_res * lockres,int new_level)3559 static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
3560 					      int new_level)
3561 {
3562 	assert_spin_locked(&lockres->l_lock);
3563 
3564 	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
3565 
3566 	if (lockres->l_level <= new_level) {
3567 		mlog(ML_ERROR, "lockres %s, lvl %d <= %d, blcklst %d, mask %d, "
3568 		     "type %d, flags 0x%lx, hold %d %d, act %d %d, req %d, "
3569 		     "block %d, pgen %d\n", lockres->l_name, lockres->l_level,
3570 		     new_level, list_empty(&lockres->l_blocked_list),
3571 		     list_empty(&lockres->l_mask_waiters), lockres->l_type,
3572 		     lockres->l_flags, lockres->l_ro_holders,
3573 		     lockres->l_ex_holders, lockres->l_action,
3574 		     lockres->l_unlock_action, lockres->l_requested,
3575 		     lockres->l_blocking, lockres->l_pending_gen);
3576 		BUG();
3577 	}
3578 
3579 	mlog(ML_BASTS, "lockres %s, level %d => %d, blocking %d\n",
3580 	     lockres->l_name, lockres->l_level, new_level, lockres->l_blocking);
3581 
3582 	lockres->l_action = OCFS2_AST_DOWNCONVERT;
3583 	lockres->l_requested = new_level;
3584 	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3585 	return lockres_set_pending(lockres);
3586 }
3587 
ocfs2_downconvert_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,int new_level,int lvb,unsigned int generation)3588 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
3589 				  struct ocfs2_lock_res *lockres,
3590 				  int new_level,
3591 				  int lvb,
3592 				  unsigned int generation)
3593 {
3594 	int ret;
3595 	u32 dlm_flags = DLM_LKF_CONVERT;
3596 
3597 	mlog(ML_BASTS, "lockres %s, level %d => %d\n", lockres->l_name,
3598 	     lockres->l_level, new_level);
3599 
3600 	/*
3601 	 * On DLM_LKF_VALBLK, fsdlm behaves differently with o2cb. It always
3602 	 * expects DLM_LKF_VALBLK being set if the LKB has LVB, so that
3603 	 * we can recover correctly from node failure. Otherwise, we may get
3604 	 * invalid LVB in LKB, but without DLM_SBF_VALNOTVALID being set.
3605 	 */
3606 	if (!ocfs2_is_o2cb_active() &&
3607 	    lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
3608 		lvb = 1;
3609 
3610 	if (lvb)
3611 		dlm_flags |= DLM_LKF_VALBLK;
3612 
3613 	ret = ocfs2_dlm_lock(osb->cconn,
3614 			     new_level,
3615 			     &lockres->l_lksb,
3616 			     dlm_flags,
3617 			     lockres->l_name,
3618 			     OCFS2_LOCK_ID_MAX_LEN - 1);
3619 	lockres_clear_pending(lockres, generation, osb);
3620 	if (ret) {
3621 		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
3622 		ocfs2_recover_from_dlm_error(lockres, 1);
3623 		goto bail;
3624 	}
3625 
3626 	ret = 0;
3627 bail:
3628 	return ret;
3629 }
3630 
3631 /* 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)3632 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
3633 				        struct ocfs2_lock_res *lockres)
3634 {
3635 	assert_spin_locked(&lockres->l_lock);
3636 
3637 	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
3638 		/* If we're already trying to cancel a lock conversion
3639 		 * then just drop the spinlock and allow the caller to
3640 		 * requeue this lock. */
3641 		mlog(ML_BASTS, "lockres %s, skip convert\n", lockres->l_name);
3642 		return 0;
3643 	}
3644 
3645 	/* were we in a convert when we got the bast fire? */
3646 	BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
3647 	       lockres->l_action != OCFS2_AST_DOWNCONVERT);
3648 	/* set things up for the unlockast to know to just
3649 	 * clear out the ast_action and unset busy, etc. */
3650 	lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
3651 
3652 	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
3653 			"lock %s, invalid flags: 0x%lx\n",
3654 			lockres->l_name, lockres->l_flags);
3655 
3656 	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3657 
3658 	return 1;
3659 }
3660 
ocfs2_cancel_convert(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)3661 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
3662 				struct ocfs2_lock_res *lockres)
3663 {
3664 	int ret;
3665 
3666 	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
3667 			       DLM_LKF_CANCEL);
3668 	if (ret) {
3669 		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3670 		ocfs2_recover_from_dlm_error(lockres, 0);
3671 	}
3672 
3673 	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3674 
3675 	return ret;
3676 }
3677 
ocfs2_unblock_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres,struct ocfs2_unblock_ctl * ctl)3678 static int ocfs2_unblock_lock(struct ocfs2_super *osb,
3679 			      struct ocfs2_lock_res *lockres,
3680 			      struct ocfs2_unblock_ctl *ctl)
3681 {
3682 	unsigned long flags;
3683 	int blocking;
3684 	int new_level;
3685 	int level;
3686 	int ret = 0;
3687 	int set_lvb = 0;
3688 	unsigned int gen;
3689 
3690 	spin_lock_irqsave(&lockres->l_lock, flags);
3691 
3692 recheck:
3693 	/*
3694 	 * Is it still blocking? If not, we have no more work to do.
3695 	 */
3696 	if (!(lockres->l_flags & OCFS2_LOCK_BLOCKED)) {
3697 		BUG_ON(lockres->l_blocking != DLM_LOCK_NL);
3698 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3699 		ret = 0;
3700 		goto leave;
3701 	}
3702 
3703 	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
3704 		/* XXX
3705 		 * This is a *big* race.  The OCFS2_LOCK_PENDING flag
3706 		 * exists entirely for one reason - another thread has set
3707 		 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
3708 		 *
3709 		 * If we do ocfs2_cancel_convert() before the other thread
3710 		 * calls dlm_lock(), our cancel will do nothing.  We will
3711 		 * get no ast, and we will have no way of knowing the
3712 		 * cancel failed.  Meanwhile, the other thread will call
3713 		 * into dlm_lock() and wait...forever.
3714 		 *
3715 		 * Why forever?  Because another node has asked for the
3716 		 * lock first; that's why we're here in unblock_lock().
3717 		 *
3718 		 * The solution is OCFS2_LOCK_PENDING.  When PENDING is
3719 		 * set, we just requeue the unblock.  Only when the other
3720 		 * thread has called dlm_lock() and cleared PENDING will
3721 		 * we then cancel their request.
3722 		 *
3723 		 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
3724 		 * at the same time they set OCFS2_DLM_BUSY.  They must
3725 		 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
3726 		 */
3727 		if (lockres->l_flags & OCFS2_LOCK_PENDING) {
3728 			mlog(ML_BASTS, "lockres %s, ReQ: Pending\n",
3729 			     lockres->l_name);
3730 			goto leave_requeue;
3731 		}
3732 
3733 		ctl->requeue = 1;
3734 		ret = ocfs2_prepare_cancel_convert(osb, lockres);
3735 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3736 		if (ret) {
3737 			ret = ocfs2_cancel_convert(osb, lockres);
3738 			if (ret < 0)
3739 				mlog_errno(ret);
3740 		}
3741 		goto leave;
3742 	}
3743 
3744 	/*
3745 	 * This prevents livelocks. OCFS2_LOCK_UPCONVERT_FINISHING flag is
3746 	 * set when the ast is received for an upconvert just before the
3747 	 * OCFS2_LOCK_BUSY flag is cleared. Now if the fs received a bast
3748 	 * on the heels of the ast, we want to delay the downconvert just
3749 	 * enough to allow the up requestor to do its task. Because this
3750 	 * lock is in the blocked queue, the lock will be downconverted
3751 	 * as soon as the requestor is done with the lock.
3752 	 */
3753 	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING)
3754 		goto leave_requeue;
3755 
3756 	/*
3757 	 * How can we block and yet be at NL?  We were trying to upconvert
3758 	 * from NL and got canceled.  The code comes back here, and now
3759 	 * we notice and clear BLOCKING.
3760 	 */
3761 	if (lockres->l_level == DLM_LOCK_NL) {
3762 		BUG_ON(lockres->l_ex_holders || lockres->l_ro_holders);
3763 		mlog(ML_BASTS, "lockres %s, Aborting dc\n", lockres->l_name);
3764 		lockres->l_blocking = DLM_LOCK_NL;
3765 		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
3766 		spin_unlock_irqrestore(&lockres->l_lock, flags);
3767 		goto leave;
3768 	}
3769 
3770 	/* if we're blocking an exclusive and we have *any* holders,
3771 	 * then requeue. */
3772 	if ((lockres->l_blocking == DLM_LOCK_EX)
3773 	    && (lockres->l_ex_holders || lockres->l_ro_holders)) {
3774 		mlog(ML_BASTS, "lockres %s, ReQ: EX/PR Holders %u,%u\n",
3775 		     lockres->l_name, lockres->l_ex_holders,
3776 		     lockres->l_ro_holders);
3777 		goto leave_requeue;
3778 	}
3779 
3780 	/* If it's a PR we're blocking, then only
3781 	 * requeue if we've got any EX holders */
3782 	if (lockres->l_blocking == DLM_LOCK_PR &&
3783 	    lockres->l_ex_holders) {
3784 		mlog(ML_BASTS, "lockres %s, ReQ: EX Holders %u\n",
3785 		     lockres->l_name, lockres->l_ex_holders);
3786 		goto leave_requeue;
3787 	}
3788 
3789 	/*
3790 	 * Can we get a lock in this state if the holder counts are
3791 	 * zero? The meta data unblock code used to check this.
3792 	 */
3793 	if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
3794 	    && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) {
3795 		mlog(ML_BASTS, "lockres %s, ReQ: Lock Refreshing\n",
3796 		     lockres->l_name);
3797 		goto leave_requeue;
3798 	}
3799 
3800 	new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
3801 
3802 	if (lockres->l_ops->check_downconvert
3803 	    && !lockres->l_ops->check_downconvert(lockres, new_level)) {
3804 		mlog(ML_BASTS, "lockres %s, ReQ: Checkpointing\n",
3805 		     lockres->l_name);
3806 		goto leave_requeue;
3807 	}
3808 
3809 	/* If we get here, then we know that there are no more
3810 	 * incompatible holders (and anyone asking for an incompatible
3811 	 * lock is blocked). We can now downconvert the lock */
3812 	if (!lockres->l_ops->downconvert_worker)
3813 		goto downconvert;
3814 
3815 	/* Some lockres types want to do a bit of work before
3816 	 * downconverting a lock. Allow that here. The worker function
3817 	 * may sleep, so we save off a copy of what we're blocking as
3818 	 * it may change while we're not holding the spin lock. */
3819 	blocking = lockres->l_blocking;
3820 	level = lockres->l_level;
3821 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3822 
3823 	ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
3824 
3825 	if (ctl->unblock_action == UNBLOCK_STOP_POST) {
3826 		mlog(ML_BASTS, "lockres %s, UNBLOCK_STOP_POST\n",
3827 		     lockres->l_name);
3828 		goto leave;
3829 	}
3830 
3831 	spin_lock_irqsave(&lockres->l_lock, flags);
3832 	if ((blocking != lockres->l_blocking) || (level != lockres->l_level)) {
3833 		/* If this changed underneath us, then we can't drop
3834 		 * it just yet. */
3835 		mlog(ML_BASTS, "lockres %s, block=%d:%d, level=%d:%d, "
3836 		     "Recheck\n", lockres->l_name, blocking,
3837 		     lockres->l_blocking, level, lockres->l_level);
3838 		goto recheck;
3839 	}
3840 
3841 downconvert:
3842 	ctl->requeue = 0;
3843 
3844 	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3845 		if (lockres->l_level == DLM_LOCK_EX)
3846 			set_lvb = 1;
3847 
3848 		/*
3849 		 * We only set the lvb if the lock has been fully
3850 		 * refreshed - otherwise we risk setting stale
3851 		 * data. Otherwise, there's no need to actually clear
3852 		 * out the lvb here as it's value is still valid.
3853 		 */
3854 		if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3855 			lockres->l_ops->set_lvb(lockres);
3856 	}
3857 
3858 	gen = ocfs2_prepare_downconvert(lockres, new_level);
3859 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3860 	ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
3861 				     gen);
3862 
3863 leave:
3864 	if (ret)
3865 		mlog_errno(ret);
3866 	return ret;
3867 
3868 leave_requeue:
3869 	spin_unlock_irqrestore(&lockres->l_lock, flags);
3870 	ctl->requeue = 1;
3871 
3872 	return 0;
3873 }
3874 
ocfs2_data_convert_worker(struct ocfs2_lock_res * lockres,int blocking)3875 static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
3876 				     int blocking)
3877 {
3878 	struct inode *inode;
3879 	struct address_space *mapping;
3880 	struct ocfs2_inode_info *oi;
3881 
3882        	inode = ocfs2_lock_res_inode(lockres);
3883 	mapping = inode->i_mapping;
3884 
3885 	if (S_ISDIR(inode->i_mode)) {
3886 		oi = OCFS2_I(inode);
3887 		oi->ip_dir_lock_gen++;
3888 		mlog(0, "generation: %u\n", oi->ip_dir_lock_gen);
3889 		goto out;
3890 	}
3891 
3892 	if (!S_ISREG(inode->i_mode))
3893 		goto out;
3894 
3895 	/*
3896 	 * We need this before the filemap_fdatawrite() so that it can
3897 	 * transfer the dirty bit from the PTE to the
3898 	 * page. Unfortunately this means that even for EX->PR
3899 	 * downconverts, we'll lose our mappings and have to build
3900 	 * them up again.
3901 	 */
3902 	unmap_mapping_range(mapping, 0, 0, 0);
3903 
3904 	if (filemap_fdatawrite(mapping)) {
3905 		mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
3906 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
3907 	}
3908 	sync_mapping_buffers(mapping);
3909 	if (blocking == DLM_LOCK_EX) {
3910 		truncate_inode_pages(mapping, 0);
3911 	} else {
3912 		/* We only need to wait on the I/O if we're not also
3913 		 * truncating pages because truncate_inode_pages waits
3914 		 * for us above. We don't truncate pages if we're
3915 		 * blocking anything < EXMODE because we want to keep
3916 		 * them around in that case. */
3917 		filemap_fdatawait(mapping);
3918 	}
3919 
3920 	forget_all_cached_acls(inode);
3921 
3922 out:
3923 	return UNBLOCK_CONTINUE;
3924 }
3925 
ocfs2_ci_checkpointed(struct ocfs2_caching_info * ci,struct ocfs2_lock_res * lockres,int new_level)3926 static int ocfs2_ci_checkpointed(struct ocfs2_caching_info *ci,
3927 				 struct ocfs2_lock_res *lockres,
3928 				 int new_level)
3929 {
3930 	int checkpointed = ocfs2_ci_fully_checkpointed(ci);
3931 
3932 	BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
3933 	BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
3934 
3935 	if (checkpointed)
3936 		return 1;
3937 
3938 	ocfs2_start_checkpoint(OCFS2_SB(ocfs2_metadata_cache_get_super(ci)));
3939 	return 0;
3940 }
3941 
ocfs2_check_meta_downconvert(struct ocfs2_lock_res * lockres,int new_level)3942 static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
3943 					int new_level)
3944 {
3945 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3946 
3947 	return ocfs2_ci_checkpointed(INODE_CACHE(inode), lockres, new_level);
3948 }
3949 
ocfs2_set_meta_lvb(struct ocfs2_lock_res * lockres)3950 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
3951 {
3952 	struct inode *inode = ocfs2_lock_res_inode(lockres);
3953 
3954 	__ocfs2_stuff_meta_lvb(inode);
3955 }
3956 
3957 /*
3958  * Does the final reference drop on our dentry lock. Right now this
3959  * happens in the downconvert thread, but we could choose to simplify the
3960  * dlmglue API and push these off to the ocfs2_wq in the future.
3961  */
ocfs2_dentry_post_unlock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)3962 static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
3963 				     struct ocfs2_lock_res *lockres)
3964 {
3965 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3966 	ocfs2_dentry_lock_put(osb, dl);
3967 }
3968 
3969 /*
3970  * d_delete() matching dentries before the lock downconvert.
3971  *
3972  * At this point, any process waiting to destroy the
3973  * dentry_lock due to last ref count is stopped by the
3974  * OCFS2_LOCK_QUEUED flag.
3975  *
3976  * We have two potential problems
3977  *
3978  * 1) If we do the last reference drop on our dentry_lock (via dput)
3979  *    we'll wind up in ocfs2_release_dentry_lock(), waiting on
3980  *    the downconvert to finish. Instead we take an elevated
3981  *    reference and push the drop until after we've completed our
3982  *    unblock processing.
3983  *
3984  * 2) There might be another process with a final reference,
3985  *    waiting on us to finish processing. If this is the case, we
3986  *    detect it and exit out - there's no more dentries anyway.
3987  */
ocfs2_dentry_convert_worker(struct ocfs2_lock_res * lockres,int blocking)3988 static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
3989 				       int blocking)
3990 {
3991 	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3992 	struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
3993 	struct dentry *dentry;
3994 	unsigned long flags;
3995 	int extra_ref = 0;
3996 
3997 	/*
3998 	 * This node is blocking another node from getting a read
3999 	 * lock. This happens when we've renamed within a
4000 	 * directory. We've forced the other nodes to d_delete(), but
4001 	 * we never actually dropped our lock because it's still
4002 	 * valid. The downconvert code will retain a PR for this node,
4003 	 * so there's no further work to do.
4004 	 */
4005 	if (blocking == DLM_LOCK_PR)
4006 		return UNBLOCK_CONTINUE;
4007 
4008 	/*
4009 	 * Mark this inode as potentially orphaned. The code in
4010 	 * ocfs2_delete_inode() will figure out whether it actually
4011 	 * needs to be freed or not.
4012 	 */
4013 	spin_lock(&oi->ip_lock);
4014 	oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
4015 	spin_unlock(&oi->ip_lock);
4016 
4017 	/*
4018 	 * Yuck. We need to make sure however that the check of
4019 	 * OCFS2_LOCK_FREEING and the extra reference are atomic with
4020 	 * respect to a reference decrement or the setting of that
4021 	 * flag.
4022 	 */
4023 	spin_lock_irqsave(&lockres->l_lock, flags);
4024 	spin_lock(&dentry_attach_lock);
4025 	if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
4026 	    && dl->dl_count) {
4027 		dl->dl_count++;
4028 		extra_ref = 1;
4029 	}
4030 	spin_unlock(&dentry_attach_lock);
4031 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4032 
4033 	mlog(0, "extra_ref = %d\n", extra_ref);
4034 
4035 	/*
4036 	 * We have a process waiting on us in ocfs2_dentry_iput(),
4037 	 * which means we can't have any more outstanding
4038 	 * aliases. There's no need to do any more work.
4039 	 */
4040 	if (!extra_ref)
4041 		return UNBLOCK_CONTINUE;
4042 
4043 	spin_lock(&dentry_attach_lock);
4044 	while (1) {
4045 		dentry = ocfs2_find_local_alias(dl->dl_inode,
4046 						dl->dl_parent_blkno, 1);
4047 		if (!dentry)
4048 			break;
4049 		spin_unlock(&dentry_attach_lock);
4050 
4051 		if (S_ISDIR(dl->dl_inode->i_mode))
4052 			shrink_dcache_parent(dentry);
4053 
4054 		mlog(0, "d_delete(%pd);\n", dentry);
4055 
4056 		/*
4057 		 * The following dcache calls may do an
4058 		 * iput(). Normally we don't want that from the
4059 		 * downconverting thread, but in this case it's ok
4060 		 * because the requesting node already has an
4061 		 * exclusive lock on the inode, so it can't be queued
4062 		 * for a downconvert.
4063 		 */
4064 		d_delete(dentry);
4065 		dput(dentry);
4066 
4067 		spin_lock(&dentry_attach_lock);
4068 	}
4069 	spin_unlock(&dentry_attach_lock);
4070 
4071 	/*
4072 	 * If we are the last holder of this dentry lock, there is no
4073 	 * reason to downconvert so skip straight to the unlock.
4074 	 */
4075 	if (dl->dl_count == 1)
4076 		return UNBLOCK_STOP_POST;
4077 
4078 	return UNBLOCK_CONTINUE_POST;
4079 }
4080 
ocfs2_check_refcount_downconvert(struct ocfs2_lock_res * lockres,int new_level)4081 static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
4082 					    int new_level)
4083 {
4084 	struct ocfs2_refcount_tree *tree =
4085 				ocfs2_lock_res_refcount_tree(lockres);
4086 
4087 	return ocfs2_ci_checkpointed(&tree->rf_ci, lockres, new_level);
4088 }
4089 
ocfs2_refcount_convert_worker(struct ocfs2_lock_res * lockres,int blocking)4090 static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
4091 					 int blocking)
4092 {
4093 	struct ocfs2_refcount_tree *tree =
4094 				ocfs2_lock_res_refcount_tree(lockres);
4095 
4096 	ocfs2_metadata_cache_purge(&tree->rf_ci);
4097 
4098 	return UNBLOCK_CONTINUE;
4099 }
4100 
ocfs2_set_qinfo_lvb(struct ocfs2_lock_res * lockres)4101 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
4102 {
4103 	struct ocfs2_qinfo_lvb *lvb;
4104 	struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres);
4105 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
4106 					    oinfo->dqi_gi.dqi_type);
4107 
4108 	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
4109 	lvb->lvb_version = OCFS2_QINFO_LVB_VERSION;
4110 	lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace);
4111 	lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace);
4112 	lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms);
4113 	lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks);
4114 	lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk);
4115 	lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry);
4116 }
4117 
ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo * oinfo,int ex)4118 void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex)
4119 {
4120 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4121 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
4122 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4123 
4124 	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
4125 		ocfs2_cluster_unlock(osb, lockres, level);
4126 }
4127 
ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo * oinfo)4128 static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo)
4129 {
4130 	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
4131 					    oinfo->dqi_gi.dqi_type);
4132 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4133 	struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
4134 	struct buffer_head *bh = NULL;
4135 	struct ocfs2_global_disk_dqinfo *gdinfo;
4136 	int status = 0;
4137 
4138 	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
4139 	    lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) {
4140 		info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace);
4141 		info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace);
4142 		oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms);
4143 		oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks);
4144 		oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk);
4145 		oinfo->dqi_gi.dqi_free_entry =
4146 					be32_to_cpu(lvb->lvb_free_entry);
4147 	} else {
4148 		status = ocfs2_read_quota_phys_block(oinfo->dqi_gqinode,
4149 						     oinfo->dqi_giblk, &bh);
4150 		if (status) {
4151 			mlog_errno(status);
4152 			goto bail;
4153 		}
4154 		gdinfo = (struct ocfs2_global_disk_dqinfo *)
4155 					(bh->b_data + OCFS2_GLOBAL_INFO_OFF);
4156 		info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace);
4157 		info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace);
4158 		oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms);
4159 		oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks);
4160 		oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk);
4161 		oinfo->dqi_gi.dqi_free_entry =
4162 					le32_to_cpu(gdinfo->dqi_free_entry);
4163 		brelse(bh);
4164 		ocfs2_track_lock_refresh(lockres);
4165 	}
4166 
4167 bail:
4168 	return status;
4169 }
4170 
4171 /* Lock quota info, this function expects at least shared lock on the quota file
4172  * so that we can safely refresh quota info from disk. */
ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo * oinfo,int ex)4173 int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex)
4174 {
4175 	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
4176 	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
4177 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4178 	int status = 0;
4179 
4180 	/* On RO devices, locking really isn't needed... */
4181 	if (ocfs2_is_hard_readonly(osb)) {
4182 		if (ex)
4183 			status = -EROFS;
4184 		goto bail;
4185 	}
4186 	if (ocfs2_mount_local(osb))
4187 		goto bail;
4188 
4189 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
4190 	if (status < 0) {
4191 		mlog_errno(status);
4192 		goto bail;
4193 	}
4194 	if (!ocfs2_should_refresh_lock_res(lockres))
4195 		goto bail;
4196 	/* OK, we have the lock but we need to refresh the quota info */
4197 	status = ocfs2_refresh_qinfo(oinfo);
4198 	if (status)
4199 		ocfs2_qinfo_unlock(oinfo, ex);
4200 	ocfs2_complete_lock_res_refresh(lockres, status);
4201 bail:
4202 	return status;
4203 }
4204 
ocfs2_refcount_lock(struct ocfs2_refcount_tree * ref_tree,int ex)4205 int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex)
4206 {
4207 	int status;
4208 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4209 	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
4210 	struct ocfs2_super *osb = lockres->l_priv;
4211 
4212 
4213 	if (ocfs2_is_hard_readonly(osb))
4214 		return -EROFS;
4215 
4216 	if (ocfs2_mount_local(osb))
4217 		return 0;
4218 
4219 	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
4220 	if (status < 0)
4221 		mlog_errno(status);
4222 
4223 	return status;
4224 }
4225 
ocfs2_refcount_unlock(struct ocfs2_refcount_tree * ref_tree,int ex)4226 void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex)
4227 {
4228 	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
4229 	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
4230 	struct ocfs2_super *osb = lockres->l_priv;
4231 
4232 	if (!ocfs2_mount_local(osb))
4233 		ocfs2_cluster_unlock(osb, lockres, level);
4234 }
4235 
ocfs2_process_blocked_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)4236 static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
4237 				       struct ocfs2_lock_res *lockres)
4238 {
4239 	int status;
4240 	struct ocfs2_unblock_ctl ctl = {0, 0,};
4241 	unsigned long flags;
4242 
4243 	/* Our reference to the lockres in this function can be
4244 	 * considered valid until we remove the OCFS2_LOCK_QUEUED
4245 	 * flag. */
4246 
4247 	BUG_ON(!lockres);
4248 	BUG_ON(!lockres->l_ops);
4249 
4250 	mlog(ML_BASTS, "lockres %s blocked\n", lockres->l_name);
4251 
4252 	/* Detect whether a lock has been marked as going away while
4253 	 * the downconvert thread was processing other things. A lock can
4254 	 * still be marked with OCFS2_LOCK_FREEING after this check,
4255 	 * but short circuiting here will still save us some
4256 	 * performance. */
4257 	spin_lock_irqsave(&lockres->l_lock, flags);
4258 	if (lockres->l_flags & OCFS2_LOCK_FREEING)
4259 		goto unqueue;
4260 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4261 
4262 	status = ocfs2_unblock_lock(osb, lockres, &ctl);
4263 	if (status < 0)
4264 		mlog_errno(status);
4265 
4266 	spin_lock_irqsave(&lockres->l_lock, flags);
4267 unqueue:
4268 	if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
4269 		lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
4270 	} else
4271 		ocfs2_schedule_blocked_lock(osb, lockres);
4272 
4273 	mlog(ML_BASTS, "lockres %s, requeue = %s.\n", lockres->l_name,
4274 	     ctl.requeue ? "yes" : "no");
4275 	spin_unlock_irqrestore(&lockres->l_lock, flags);
4276 
4277 	if (ctl.unblock_action != UNBLOCK_CONTINUE
4278 	    && lockres->l_ops->post_unlock)
4279 		lockres->l_ops->post_unlock(osb, lockres);
4280 }
4281 
ocfs2_schedule_blocked_lock(struct ocfs2_super * osb,struct ocfs2_lock_res * lockres)4282 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
4283 					struct ocfs2_lock_res *lockres)
4284 {
4285 	unsigned long flags;
4286 
4287 	assert_spin_locked(&lockres->l_lock);
4288 
4289 	if (lockres->l_flags & OCFS2_LOCK_FREEING) {
4290 		/* Do not schedule a lock for downconvert when it's on
4291 		 * the way to destruction - any nodes wanting access
4292 		 * to the resource will get it soon. */
4293 		mlog(ML_BASTS, "lockres %s won't be scheduled: flags 0x%lx\n",
4294 		     lockres->l_name, lockres->l_flags);
4295 		return;
4296 	}
4297 
4298 	lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
4299 
4300 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4301 	if (list_empty(&lockres->l_blocked_list)) {
4302 		list_add_tail(&lockres->l_blocked_list,
4303 			      &osb->blocked_lock_list);
4304 		osb->blocked_lock_count++;
4305 	}
4306 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4307 }
4308 
ocfs2_downconvert_thread_do_work(struct ocfs2_super * osb)4309 static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
4310 {
4311 	unsigned long processed;
4312 	unsigned long flags;
4313 	struct ocfs2_lock_res *lockres;
4314 
4315 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4316 	/* grab this early so we know to try again if a state change and
4317 	 * wake happens part-way through our work  */
4318 	osb->dc_work_sequence = osb->dc_wake_sequence;
4319 
4320 	processed = osb->blocked_lock_count;
4321 	/*
4322 	 * blocked lock processing in this loop might call iput which can
4323 	 * remove items off osb->blocked_lock_list. Downconvert up to
4324 	 * 'processed' number of locks, but stop short if we had some
4325 	 * removed in ocfs2_mark_lockres_freeing when downconverting.
4326 	 */
4327 	while (processed && !list_empty(&osb->blocked_lock_list)) {
4328 		lockres = list_entry(osb->blocked_lock_list.next,
4329 				     struct ocfs2_lock_res, l_blocked_list);
4330 		list_del_init(&lockres->l_blocked_list);
4331 		osb->blocked_lock_count--;
4332 		spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4333 
4334 		BUG_ON(!processed);
4335 		processed--;
4336 
4337 		ocfs2_process_blocked_lock(osb, lockres);
4338 
4339 		spin_lock_irqsave(&osb->dc_task_lock, flags);
4340 	}
4341 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4342 }
4343 
ocfs2_downconvert_thread_lists_empty(struct ocfs2_super * osb)4344 static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
4345 {
4346 	int empty = 0;
4347 	unsigned long flags;
4348 
4349 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4350 	if (list_empty(&osb->blocked_lock_list))
4351 		empty = 1;
4352 
4353 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4354 	return empty;
4355 }
4356 
ocfs2_downconvert_thread_should_wake(struct ocfs2_super * osb)4357 static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
4358 {
4359 	int should_wake = 0;
4360 	unsigned long flags;
4361 
4362 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4363 	if (osb->dc_work_sequence != osb->dc_wake_sequence)
4364 		should_wake = 1;
4365 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4366 
4367 	return should_wake;
4368 }
4369 
ocfs2_downconvert_thread(void * arg)4370 static int ocfs2_downconvert_thread(void *arg)
4371 {
4372 	int status = 0;
4373 	struct ocfs2_super *osb = arg;
4374 
4375 	/* only quit once we've been asked to stop and there is no more
4376 	 * work available */
4377 	while (!(kthread_should_stop() &&
4378 		ocfs2_downconvert_thread_lists_empty(osb))) {
4379 
4380 		wait_event_interruptible(osb->dc_event,
4381 					 ocfs2_downconvert_thread_should_wake(osb) ||
4382 					 kthread_should_stop());
4383 
4384 		mlog(0, "downconvert_thread: awoken\n");
4385 
4386 		ocfs2_downconvert_thread_do_work(osb);
4387 	}
4388 
4389 	osb->dc_task = NULL;
4390 	return status;
4391 }
4392 
ocfs2_wake_downconvert_thread(struct ocfs2_super * osb)4393 void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
4394 {
4395 	unsigned long flags;
4396 
4397 	spin_lock_irqsave(&osb->dc_task_lock, flags);
4398 	/* make sure the voting thread gets a swipe at whatever changes
4399 	 * the caller may have made to the voting state */
4400 	osb->dc_wake_sequence++;
4401 	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4402 	wake_up(&osb->dc_event);
4403 }
4404