xref: /openbmc/linux/fs/gfs2/glock.c (revision 1d1997db)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
5  */
6 
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/buffer_head.h>
13 #include <linux/delay.h>
14 #include <linux/sort.h>
15 #include <linux/hash.h>
16 #include <linux/jhash.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/list.h>
20 #include <linux/wait.h>
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/seq_file.h>
24 #include <linux/debugfs.h>
25 #include <linux/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/workqueue.h>
28 #include <linux/jiffies.h>
29 #include <linux/rcupdate.h>
30 #include <linux/rculist_bl.h>
31 #include <linux/bit_spinlock.h>
32 #include <linux/percpu.h>
33 #include <linux/list_sort.h>
34 #include <linux/lockref.h>
35 #include <linux/rhashtable.h>
36 
37 #include "gfs2.h"
38 #include "incore.h"
39 #include "glock.h"
40 #include "glops.h"
41 #include "inode.h"
42 #include "lops.h"
43 #include "meta_io.h"
44 #include "quota.h"
45 #include "super.h"
46 #include "util.h"
47 #include "bmap.h"
48 #define CREATE_TRACE_POINTS
49 #include "trace_gfs2.h"
50 
51 struct gfs2_glock_iter {
52 	struct gfs2_sbd *sdp;		/* incore superblock           */
53 	struct rhashtable_iter hti;	/* rhashtable iterator         */
54 	struct gfs2_glock *gl;		/* current glock struct        */
55 	loff_t last_pos;		/* last position               */
56 };
57 
58 typedef void (*glock_examiner) (struct gfs2_glock * gl);
59 
60 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
61 
62 static struct dentry *gfs2_root;
63 static struct workqueue_struct *glock_workqueue;
64 struct workqueue_struct *gfs2_delete_workqueue;
65 static LIST_HEAD(lru_list);
66 static atomic_t lru_count = ATOMIC_INIT(0);
67 static DEFINE_SPINLOCK(lru_lock);
68 
69 #define GFS2_GL_HASH_SHIFT      15
70 #define GFS2_GL_HASH_SIZE       BIT(GFS2_GL_HASH_SHIFT)
71 
72 static const struct rhashtable_params ht_parms = {
73 	.nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
74 	.key_len = offsetofend(struct lm_lockname, ln_type),
75 	.key_offset = offsetof(struct gfs2_glock, gl_name),
76 	.head_offset = offsetof(struct gfs2_glock, gl_node),
77 };
78 
79 static struct rhashtable gl_hash_table;
80 
81 #define GLOCK_WAIT_TABLE_BITS 12
82 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
83 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
84 
85 struct wait_glock_queue {
86 	struct lm_lockname *name;
87 	wait_queue_entry_t wait;
88 };
89 
90 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
91 			       int sync, void *key)
92 {
93 	struct wait_glock_queue *wait_glock =
94 		container_of(wait, struct wait_glock_queue, wait);
95 	struct lm_lockname *wait_name = wait_glock->name;
96 	struct lm_lockname *wake_name = key;
97 
98 	if (wake_name->ln_sbd != wait_name->ln_sbd ||
99 	    wake_name->ln_number != wait_name->ln_number ||
100 	    wake_name->ln_type != wait_name->ln_type)
101 		return 0;
102 	return autoremove_wake_function(wait, mode, sync, key);
103 }
104 
105 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
106 {
107 	u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
108 
109 	return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
110 }
111 
112 /**
113  * wake_up_glock  -  Wake up waiters on a glock
114  * @gl: the glock
115  */
116 static void wake_up_glock(struct gfs2_glock *gl)
117 {
118 	wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
119 
120 	if (waitqueue_active(wq))
121 		__wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
122 }
123 
124 static void gfs2_glock_dealloc(struct rcu_head *rcu)
125 {
126 	struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
127 
128 	if (gl->gl_ops->go_flags & GLOF_ASPACE) {
129 		kmem_cache_free(gfs2_glock_aspace_cachep, gl);
130 	} else {
131 		kfree(gl->gl_lksb.sb_lvbptr);
132 		kmem_cache_free(gfs2_glock_cachep, gl);
133 	}
134 }
135 
136 void gfs2_glock_free(struct gfs2_glock *gl)
137 {
138 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
139 
140 	BUG_ON(atomic_read(&gl->gl_revokes));
141 	rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
142 	smp_mb();
143 	wake_up_glock(gl);
144 	call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
145 	if (atomic_dec_and_test(&sdp->sd_glock_disposal))
146 		wake_up(&sdp->sd_glock_wait);
147 }
148 
149 /**
150  * gfs2_glock_hold() - increment reference count on glock
151  * @gl: The glock to hold
152  *
153  */
154 
155 void gfs2_glock_hold(struct gfs2_glock *gl)
156 {
157 	GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
158 	lockref_get(&gl->gl_lockref);
159 }
160 
161 /**
162  * demote_ok - Check to see if it's ok to unlock a glock
163  * @gl: the glock
164  *
165  * Returns: 1 if it's ok
166  */
167 
168 static int demote_ok(const struct gfs2_glock *gl)
169 {
170 	const struct gfs2_glock_operations *glops = gl->gl_ops;
171 
172 	if (gl->gl_state == LM_ST_UNLOCKED)
173 		return 0;
174 	if (!list_empty(&gl->gl_holders))
175 		return 0;
176 	if (glops->go_demote_ok)
177 		return glops->go_demote_ok(gl);
178 	return 1;
179 }
180 
181 
182 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
183 {
184 	if (!(gl->gl_ops->go_flags & GLOF_LRU))
185 		return;
186 
187 	spin_lock(&lru_lock);
188 
189 	list_del(&gl->gl_lru);
190 	list_add_tail(&gl->gl_lru, &lru_list);
191 
192 	if (!test_bit(GLF_LRU, &gl->gl_flags)) {
193 		set_bit(GLF_LRU, &gl->gl_flags);
194 		atomic_inc(&lru_count);
195 	}
196 
197 	spin_unlock(&lru_lock);
198 }
199 
200 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
201 {
202 	if (!(gl->gl_ops->go_flags & GLOF_LRU))
203 		return;
204 
205 	spin_lock(&lru_lock);
206 	if (test_bit(GLF_LRU, &gl->gl_flags)) {
207 		list_del_init(&gl->gl_lru);
208 		atomic_dec(&lru_count);
209 		clear_bit(GLF_LRU, &gl->gl_flags);
210 	}
211 	spin_unlock(&lru_lock);
212 }
213 
214 /*
215  * Enqueue the glock on the work queue.  Passes one glock reference on to the
216  * work queue.
217  */
218 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
219 	if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
220 		/*
221 		 * We are holding the lockref spinlock, and the work was still
222 		 * queued above.  The queued work (glock_work_func) takes that
223 		 * spinlock before dropping its glock reference(s), so it
224 		 * cannot have dropped them in the meantime.
225 		 */
226 		GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
227 		gl->gl_lockref.count--;
228 	}
229 }
230 
231 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
232 	spin_lock(&gl->gl_lockref.lock);
233 	__gfs2_glock_queue_work(gl, delay);
234 	spin_unlock(&gl->gl_lockref.lock);
235 }
236 
237 static void __gfs2_glock_put(struct gfs2_glock *gl)
238 {
239 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
240 	struct address_space *mapping = gfs2_glock2aspace(gl);
241 
242 	lockref_mark_dead(&gl->gl_lockref);
243 
244 	gfs2_glock_remove_from_lru(gl);
245 	spin_unlock(&gl->gl_lockref.lock);
246 	GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
247 	GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
248 	trace_gfs2_glock_put(gl);
249 	sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
250 }
251 
252 /*
253  * Cause the glock to be put in work queue context.
254  */
255 void gfs2_glock_queue_put(struct gfs2_glock *gl)
256 {
257 	gfs2_glock_queue_work(gl, 0);
258 }
259 
260 /**
261  * gfs2_glock_put() - Decrement reference count on glock
262  * @gl: The glock to put
263  *
264  */
265 
266 void gfs2_glock_put(struct gfs2_glock *gl)
267 {
268 	if (lockref_put_or_lock(&gl->gl_lockref))
269 		return;
270 
271 	__gfs2_glock_put(gl);
272 }
273 
274 /**
275  * may_grant - check if its ok to grant a new lock
276  * @gl: The glock
277  * @gh: The lock request which we wish to grant
278  *
279  * Returns: true if its ok to grant the lock
280  */
281 
282 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
283 {
284 	const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
285 	if ((gh->gh_state == LM_ST_EXCLUSIVE ||
286 	     gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
287 		return 0;
288 	if (gl->gl_state == gh->gh_state)
289 		return 1;
290 	if (gh->gh_flags & GL_EXACT)
291 		return 0;
292 	if (gl->gl_state == LM_ST_EXCLUSIVE) {
293 		if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
294 			return 1;
295 		if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
296 			return 1;
297 	}
298 	if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
299 		return 1;
300 	return 0;
301 }
302 
303 static void gfs2_holder_wake(struct gfs2_holder *gh)
304 {
305 	clear_bit(HIF_WAIT, &gh->gh_iflags);
306 	smp_mb__after_atomic();
307 	wake_up_bit(&gh->gh_iflags, HIF_WAIT);
308 	if (gh->gh_flags & GL_ASYNC) {
309 		struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
310 
311 		wake_up(&sdp->sd_async_glock_wait);
312 	}
313 }
314 
315 /**
316  * do_error - Something unexpected has happened during a lock request
317  *
318  */
319 
320 static void do_error(struct gfs2_glock *gl, const int ret)
321 {
322 	struct gfs2_holder *gh, *tmp;
323 
324 	list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
325 		if (test_bit(HIF_HOLDER, &gh->gh_iflags))
326 			continue;
327 		if (ret & LM_OUT_ERROR)
328 			gh->gh_error = -EIO;
329 		else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
330 			gh->gh_error = GLR_TRYFAILED;
331 		else
332 			continue;
333 		list_del_init(&gh->gh_list);
334 		trace_gfs2_glock_queue(gh, 0);
335 		gfs2_holder_wake(gh);
336 	}
337 }
338 
339 /**
340  * do_promote - promote as many requests as possible on the current queue
341  * @gl: The glock
342  *
343  * Returns: 1 if there is a blocked holder at the head of the list, or 2
344  *          if a type specific operation is underway.
345  */
346 
347 static int do_promote(struct gfs2_glock *gl)
348 __releases(&gl->gl_lockref.lock)
349 __acquires(&gl->gl_lockref.lock)
350 {
351 	const struct gfs2_glock_operations *glops = gl->gl_ops;
352 	struct gfs2_holder *gh, *tmp;
353 	int ret;
354 
355 restart:
356 	list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
357 		if (test_bit(HIF_HOLDER, &gh->gh_iflags))
358 			continue;
359 		if (may_grant(gl, gh)) {
360 			if (gh->gh_list.prev == &gl->gl_holders &&
361 			    glops->go_lock) {
362 				spin_unlock(&gl->gl_lockref.lock);
363 				/* FIXME: eliminate this eventually */
364 				ret = glops->go_lock(gh);
365 				spin_lock(&gl->gl_lockref.lock);
366 				if (ret) {
367 					if (ret == 1)
368 						return 2;
369 					gh->gh_error = ret;
370 					list_del_init(&gh->gh_list);
371 					trace_gfs2_glock_queue(gh, 0);
372 					gfs2_holder_wake(gh);
373 					goto restart;
374 				}
375 				set_bit(HIF_HOLDER, &gh->gh_iflags);
376 				trace_gfs2_promote(gh, 1);
377 				gfs2_holder_wake(gh);
378 				goto restart;
379 			}
380 			set_bit(HIF_HOLDER, &gh->gh_iflags);
381 			trace_gfs2_promote(gh, 0);
382 			gfs2_holder_wake(gh);
383 			continue;
384 		}
385 		if (gh->gh_list.prev == &gl->gl_holders)
386 			return 1;
387 		do_error(gl, 0);
388 		break;
389 	}
390 	return 0;
391 }
392 
393 /**
394  * find_first_waiter - find the first gh that's waiting for the glock
395  * @gl: the glock
396  */
397 
398 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
399 {
400 	struct gfs2_holder *gh;
401 
402 	list_for_each_entry(gh, &gl->gl_holders, gh_list) {
403 		if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
404 			return gh;
405 	}
406 	return NULL;
407 }
408 
409 /**
410  * state_change - record that the glock is now in a different state
411  * @gl: the glock
412  * @new_state the new state
413  *
414  */
415 
416 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
417 {
418 	int held1, held2;
419 
420 	held1 = (gl->gl_state != LM_ST_UNLOCKED);
421 	held2 = (new_state != LM_ST_UNLOCKED);
422 
423 	if (held1 != held2) {
424 		GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
425 		if (held2)
426 			gl->gl_lockref.count++;
427 		else
428 			gl->gl_lockref.count--;
429 	}
430 	if (held1 && held2 && list_empty(&gl->gl_holders))
431 		clear_bit(GLF_QUEUED, &gl->gl_flags);
432 
433 	if (new_state != gl->gl_target)
434 		/* shorten our minimum hold time */
435 		gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
436 				       GL_GLOCK_MIN_HOLD);
437 	gl->gl_state = new_state;
438 	gl->gl_tchange = jiffies;
439 }
440 
441 static void gfs2_demote_wake(struct gfs2_glock *gl)
442 {
443 	gl->gl_demote_state = LM_ST_EXCLUSIVE;
444 	clear_bit(GLF_DEMOTE, &gl->gl_flags);
445 	smp_mb__after_atomic();
446 	wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
447 }
448 
449 /**
450  * finish_xmote - The DLM has replied to one of our lock requests
451  * @gl: The glock
452  * @ret: The status from the DLM
453  *
454  */
455 
456 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
457 {
458 	const struct gfs2_glock_operations *glops = gl->gl_ops;
459 	struct gfs2_holder *gh;
460 	unsigned state = ret & LM_OUT_ST_MASK;
461 	int rv;
462 
463 	spin_lock(&gl->gl_lockref.lock);
464 	trace_gfs2_glock_state_change(gl, state);
465 	state_change(gl, state);
466 	gh = find_first_waiter(gl);
467 
468 	/* Demote to UN request arrived during demote to SH or DF */
469 	if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
470 	    state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
471 		gl->gl_target = LM_ST_UNLOCKED;
472 
473 	/* Check for state != intended state */
474 	if (unlikely(state != gl->gl_target)) {
475 		if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
476 			/* move to back of queue and try next entry */
477 			if (ret & LM_OUT_CANCELED) {
478 				if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
479 					list_move_tail(&gh->gh_list, &gl->gl_holders);
480 				gh = find_first_waiter(gl);
481 				gl->gl_target = gh->gh_state;
482 				goto retry;
483 			}
484 			/* Some error or failed "try lock" - report it */
485 			if ((ret & LM_OUT_ERROR) ||
486 			    (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
487 				gl->gl_target = gl->gl_state;
488 				do_error(gl, ret);
489 				goto out;
490 			}
491 		}
492 		switch(state) {
493 		/* Unlocked due to conversion deadlock, try again */
494 		case LM_ST_UNLOCKED:
495 retry:
496 			do_xmote(gl, gh, gl->gl_target);
497 			break;
498 		/* Conversion fails, unlock and try again */
499 		case LM_ST_SHARED:
500 		case LM_ST_DEFERRED:
501 			do_xmote(gl, gh, LM_ST_UNLOCKED);
502 			break;
503 		default: /* Everything else */
504 			fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
505 			       gl->gl_target, state);
506 			GLOCK_BUG_ON(gl, 1);
507 		}
508 		spin_unlock(&gl->gl_lockref.lock);
509 		return;
510 	}
511 
512 	/* Fast path - we got what we asked for */
513 	if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
514 		gfs2_demote_wake(gl);
515 	if (state != LM_ST_UNLOCKED) {
516 		if (glops->go_xmote_bh) {
517 			spin_unlock(&gl->gl_lockref.lock);
518 			rv = glops->go_xmote_bh(gl, gh);
519 			spin_lock(&gl->gl_lockref.lock);
520 			if (rv) {
521 				do_error(gl, rv);
522 				goto out;
523 			}
524 		}
525 		rv = do_promote(gl);
526 		if (rv == 2)
527 			goto out_locked;
528 	}
529 out:
530 	clear_bit(GLF_LOCK, &gl->gl_flags);
531 out_locked:
532 	spin_unlock(&gl->gl_lockref.lock);
533 }
534 
535 /**
536  * do_xmote - Calls the DLM to change the state of a lock
537  * @gl: The lock state
538  * @gh: The holder (only for promotes)
539  * @target: The target lock state
540  *
541  */
542 
543 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
544 __releases(&gl->gl_lockref.lock)
545 __acquires(&gl->gl_lockref.lock)
546 {
547 	const struct gfs2_glock_operations *glops = gl->gl_ops;
548 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
549 	unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
550 	int ret;
551 
552 	if (unlikely(gfs2_withdrawn(sdp)) &&
553 	    target != LM_ST_UNLOCKED)
554 		return;
555 	lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
556 		      LM_FLAG_PRIORITY);
557 	GLOCK_BUG_ON(gl, gl->gl_state == target);
558 	GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
559 	if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
560 	    glops->go_inval) {
561 		/*
562 		 * If another process is already doing the invalidate, let that
563 		 * finish first.  The glock state machine will get back to this
564 		 * holder again later.
565 		 */
566 		if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
567 				     &gl->gl_flags))
568 			return;
569 		do_error(gl, 0); /* Fail queued try locks */
570 	}
571 	gl->gl_req = target;
572 	set_bit(GLF_BLOCKING, &gl->gl_flags);
573 	if ((gl->gl_req == LM_ST_UNLOCKED) ||
574 	    (gl->gl_state == LM_ST_EXCLUSIVE) ||
575 	    (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
576 		clear_bit(GLF_BLOCKING, &gl->gl_flags);
577 	spin_unlock(&gl->gl_lockref.lock);
578 	if (glops->go_sync)
579 		glops->go_sync(gl);
580 	if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
581 		glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
582 	clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
583 
584 	gfs2_glock_hold(gl);
585 	if (sdp->sd_lockstruct.ls_ops->lm_lock)	{
586 		/* lock_dlm */
587 		ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
588 		if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
589 		    target == LM_ST_UNLOCKED &&
590 		    test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
591 			finish_xmote(gl, target);
592 			gfs2_glock_queue_work(gl, 0);
593 		}
594 		else if (ret) {
595 			fs_err(sdp, "lm_lock ret %d\n", ret);
596 			GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
597 		}
598 	} else { /* lock_nolock */
599 		finish_xmote(gl, target);
600 		gfs2_glock_queue_work(gl, 0);
601 	}
602 
603 	spin_lock(&gl->gl_lockref.lock);
604 }
605 
606 /**
607  * find_first_holder - find the first "holder" gh
608  * @gl: the glock
609  */
610 
611 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
612 {
613 	struct gfs2_holder *gh;
614 
615 	if (!list_empty(&gl->gl_holders)) {
616 		gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
617 		if (test_bit(HIF_HOLDER, &gh->gh_iflags))
618 			return gh;
619 	}
620 	return NULL;
621 }
622 
623 /**
624  * run_queue - do all outstanding tasks related to a glock
625  * @gl: The glock in question
626  * @nonblock: True if we must not block in run_queue
627  *
628  */
629 
630 static void run_queue(struct gfs2_glock *gl, const int nonblock)
631 __releases(&gl->gl_lockref.lock)
632 __acquires(&gl->gl_lockref.lock)
633 {
634 	struct gfs2_holder *gh = NULL;
635 	int ret;
636 
637 	if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
638 		return;
639 
640 	GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
641 
642 	if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
643 	    gl->gl_demote_state != gl->gl_state) {
644 		if (find_first_holder(gl))
645 			goto out_unlock;
646 		if (nonblock)
647 			goto out_sched;
648 		set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
649 		GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
650 		gl->gl_target = gl->gl_demote_state;
651 	} else {
652 		if (test_bit(GLF_DEMOTE, &gl->gl_flags))
653 			gfs2_demote_wake(gl);
654 		ret = do_promote(gl);
655 		if (ret == 0)
656 			goto out_unlock;
657 		if (ret == 2)
658 			goto out;
659 		gh = find_first_waiter(gl);
660 		gl->gl_target = gh->gh_state;
661 		if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
662 			do_error(gl, 0); /* Fail queued try locks */
663 	}
664 	do_xmote(gl, gh, gl->gl_target);
665 out:
666 	return;
667 
668 out_sched:
669 	clear_bit(GLF_LOCK, &gl->gl_flags);
670 	smp_mb__after_atomic();
671 	gl->gl_lockref.count++;
672 	__gfs2_glock_queue_work(gl, 0);
673 	return;
674 
675 out_unlock:
676 	clear_bit(GLF_LOCK, &gl->gl_flags);
677 	smp_mb__after_atomic();
678 	return;
679 }
680 
681 static void delete_work_func(struct work_struct *work)
682 {
683 	struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
684 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
685 	struct inode *inode;
686 	u64 no_addr = gl->gl_name.ln_number;
687 
688 	/* If someone's using this glock to create a new dinode, the block must
689 	   have been freed by another node, then re-used, in which case our
690 	   iopen callback is too late after the fact. Ignore it. */
691 	if (test_bit(GLF_INODE_CREATING, &gl->gl_flags))
692 		goto out;
693 
694 	inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
695 	if (!IS_ERR_OR_NULL(inode)) {
696 		d_prune_aliases(inode);
697 		iput(inode);
698 	}
699 out:
700 	gfs2_glock_put(gl);
701 }
702 
703 static void glock_work_func(struct work_struct *work)
704 {
705 	unsigned long delay = 0;
706 	struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
707 	unsigned int drop_refs = 1;
708 
709 	if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
710 		finish_xmote(gl, gl->gl_reply);
711 		drop_refs++;
712 	}
713 	spin_lock(&gl->gl_lockref.lock);
714 	if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
715 	    gl->gl_state != LM_ST_UNLOCKED &&
716 	    gl->gl_demote_state != LM_ST_EXCLUSIVE) {
717 		unsigned long holdtime, now = jiffies;
718 
719 		holdtime = gl->gl_tchange + gl->gl_hold_time;
720 		if (time_before(now, holdtime))
721 			delay = holdtime - now;
722 
723 		if (!delay) {
724 			clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
725 			set_bit(GLF_DEMOTE, &gl->gl_flags);
726 		}
727 	}
728 	run_queue(gl, 0);
729 	if (delay) {
730 		/* Keep one glock reference for the work we requeue. */
731 		drop_refs--;
732 		if (gl->gl_name.ln_type != LM_TYPE_INODE)
733 			delay = 0;
734 		__gfs2_glock_queue_work(gl, delay);
735 	}
736 
737 	/*
738 	 * Drop the remaining glock references manually here. (Mind that
739 	 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
740 	 * here as well.)
741 	 */
742 	gl->gl_lockref.count -= drop_refs;
743 	if (!gl->gl_lockref.count) {
744 		__gfs2_glock_put(gl);
745 		return;
746 	}
747 	spin_unlock(&gl->gl_lockref.lock);
748 }
749 
750 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
751 					    struct gfs2_glock *new)
752 {
753 	struct wait_glock_queue wait;
754 	wait_queue_head_t *wq = glock_waitqueue(name);
755 	struct gfs2_glock *gl;
756 
757 	wait.name = name;
758 	init_wait(&wait.wait);
759 	wait.wait.func = glock_wake_function;
760 
761 again:
762 	prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
763 	rcu_read_lock();
764 	if (new) {
765 		gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
766 			&new->gl_node, ht_parms);
767 		if (IS_ERR(gl))
768 			goto out;
769 	} else {
770 		gl = rhashtable_lookup_fast(&gl_hash_table,
771 			name, ht_parms);
772 	}
773 	if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
774 		rcu_read_unlock();
775 		schedule();
776 		goto again;
777 	}
778 out:
779 	rcu_read_unlock();
780 	finish_wait(wq, &wait.wait);
781 	return gl;
782 }
783 
784 /**
785  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
786  * @sdp: The GFS2 superblock
787  * @number: the lock number
788  * @glops: The glock_operations to use
789  * @create: If 0, don't create the glock if it doesn't exist
790  * @glp: the glock is returned here
791  *
792  * This does not lock a glock, just finds/creates structures for one.
793  *
794  * Returns: errno
795  */
796 
797 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
798 		   const struct gfs2_glock_operations *glops, int create,
799 		   struct gfs2_glock **glp)
800 {
801 	struct super_block *s = sdp->sd_vfs;
802 	struct lm_lockname name = { .ln_number = number,
803 				    .ln_type = glops->go_type,
804 				    .ln_sbd = sdp };
805 	struct gfs2_glock *gl, *tmp;
806 	struct address_space *mapping;
807 	struct kmem_cache *cachep;
808 	int ret = 0;
809 
810 	gl = find_insert_glock(&name, NULL);
811 	if (gl) {
812 		*glp = gl;
813 		return 0;
814 	}
815 	if (!create)
816 		return -ENOENT;
817 
818 	if (glops->go_flags & GLOF_ASPACE)
819 		cachep = gfs2_glock_aspace_cachep;
820 	else
821 		cachep = gfs2_glock_cachep;
822 	gl = kmem_cache_alloc(cachep, GFP_NOFS);
823 	if (!gl)
824 		return -ENOMEM;
825 
826 	memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
827 
828 	if (glops->go_flags & GLOF_LVB) {
829 		gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
830 		if (!gl->gl_lksb.sb_lvbptr) {
831 			kmem_cache_free(cachep, gl);
832 			return -ENOMEM;
833 		}
834 	}
835 
836 	atomic_inc(&sdp->sd_glock_disposal);
837 	gl->gl_node.next = NULL;
838 	gl->gl_flags = 0;
839 	gl->gl_name = name;
840 	gl->gl_lockref.count = 1;
841 	gl->gl_state = LM_ST_UNLOCKED;
842 	gl->gl_target = LM_ST_UNLOCKED;
843 	gl->gl_demote_state = LM_ST_EXCLUSIVE;
844 	gl->gl_ops = glops;
845 	gl->gl_dstamp = 0;
846 	preempt_disable();
847 	/* We use the global stats to estimate the initial per-glock stats */
848 	gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
849 	preempt_enable();
850 	gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
851 	gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
852 	gl->gl_tchange = jiffies;
853 	gl->gl_object = NULL;
854 	gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
855 	INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
856 	INIT_WORK(&gl->gl_delete, delete_work_func);
857 
858 	mapping = gfs2_glock2aspace(gl);
859 	if (mapping) {
860                 mapping->a_ops = &gfs2_meta_aops;
861 		mapping->host = s->s_bdev->bd_inode;
862 		mapping->flags = 0;
863 		mapping_set_gfp_mask(mapping, GFP_NOFS);
864 		mapping->private_data = NULL;
865 		mapping->writeback_index = 0;
866 	}
867 
868 	tmp = find_insert_glock(&name, gl);
869 	if (!tmp) {
870 		*glp = gl;
871 		goto out;
872 	}
873 	if (IS_ERR(tmp)) {
874 		ret = PTR_ERR(tmp);
875 		goto out_free;
876 	}
877 	*glp = tmp;
878 
879 out_free:
880 	kfree(gl->gl_lksb.sb_lvbptr);
881 	kmem_cache_free(cachep, gl);
882 	atomic_dec(&sdp->sd_glock_disposal);
883 
884 out:
885 	return ret;
886 }
887 
888 /**
889  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
890  * @gl: the glock
891  * @state: the state we're requesting
892  * @flags: the modifier flags
893  * @gh: the holder structure
894  *
895  */
896 
897 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
898 		      struct gfs2_holder *gh)
899 {
900 	INIT_LIST_HEAD(&gh->gh_list);
901 	gh->gh_gl = gl;
902 	gh->gh_ip = _RET_IP_;
903 	gh->gh_owner_pid = get_pid(task_pid(current));
904 	gh->gh_state = state;
905 	gh->gh_flags = flags;
906 	gh->gh_error = 0;
907 	gh->gh_iflags = 0;
908 	gfs2_glock_hold(gl);
909 }
910 
911 /**
912  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
913  * @state: the state we're requesting
914  * @flags: the modifier flags
915  * @gh: the holder structure
916  *
917  * Don't mess with the glock.
918  *
919  */
920 
921 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
922 {
923 	gh->gh_state = state;
924 	gh->gh_flags = flags;
925 	gh->gh_iflags = 0;
926 	gh->gh_ip = _RET_IP_;
927 	put_pid(gh->gh_owner_pid);
928 	gh->gh_owner_pid = get_pid(task_pid(current));
929 }
930 
931 /**
932  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
933  * @gh: the holder structure
934  *
935  */
936 
937 void gfs2_holder_uninit(struct gfs2_holder *gh)
938 {
939 	put_pid(gh->gh_owner_pid);
940 	gfs2_glock_put(gh->gh_gl);
941 	gfs2_holder_mark_uninitialized(gh);
942 	gh->gh_ip = 0;
943 }
944 
945 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
946 					unsigned long start_time)
947 {
948 	/* Have we waited longer that a second? */
949 	if (time_after(jiffies, start_time + HZ)) {
950 		/* Lengthen the minimum hold time. */
951 		gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
952 				       GL_GLOCK_MAX_HOLD);
953 	}
954 }
955 
956 /**
957  * gfs2_glock_wait - wait on a glock acquisition
958  * @gh: the glock holder
959  *
960  * Returns: 0 on success
961  */
962 
963 int gfs2_glock_wait(struct gfs2_holder *gh)
964 {
965 	unsigned long start_time = jiffies;
966 
967 	might_sleep();
968 	wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
969 	gfs2_glock_update_hold_time(gh->gh_gl, start_time);
970 	return gh->gh_error;
971 }
972 
973 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
974 {
975 	int i;
976 
977 	for (i = 0; i < num_gh; i++)
978 		if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
979 			return 1;
980 	return 0;
981 }
982 
983 /**
984  * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
985  * @num_gh: the number of holders in the array
986  * @ghs: the glock holder array
987  *
988  * Returns: 0 on success, meaning all glocks have been granted and are held.
989  *          -ESTALE if the request timed out, meaning all glocks were released,
990  *          and the caller should retry the operation.
991  */
992 
993 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
994 {
995 	struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
996 	int i, ret = 0, timeout = 0;
997 	unsigned long start_time = jiffies;
998 	bool keep_waiting;
999 
1000 	might_sleep();
1001 	/*
1002 	 * Total up the (minimum hold time * 2) of all glocks and use that to
1003 	 * determine the max amount of time we should wait.
1004 	 */
1005 	for (i = 0; i < num_gh; i++)
1006 		timeout += ghs[i].gh_gl->gl_hold_time << 1;
1007 
1008 wait_for_dlm:
1009 	if (!wait_event_timeout(sdp->sd_async_glock_wait,
1010 				!glocks_pending(num_gh, ghs), timeout))
1011 		ret = -ESTALE; /* request timed out. */
1012 
1013 	/*
1014 	 * If dlm granted all our requests, we need to adjust the glock
1015 	 * minimum hold time values according to how long we waited.
1016 	 *
1017 	 * If our request timed out, we need to repeatedly release any held
1018 	 * glocks we acquired thus far to allow dlm to acquire the remaining
1019 	 * glocks without deadlocking.  We cannot currently cancel outstanding
1020 	 * glock acquisitions.
1021 	 *
1022 	 * The HIF_WAIT bit tells us which requests still need a response from
1023 	 * dlm.
1024 	 *
1025 	 * If dlm sent us any errors, we return the first error we find.
1026 	 */
1027 	keep_waiting = false;
1028 	for (i = 0; i < num_gh; i++) {
1029 		/* Skip holders we have already dequeued below. */
1030 		if (!gfs2_holder_queued(&ghs[i]))
1031 			continue;
1032 		/* Skip holders with a pending DLM response. */
1033 		if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1034 			keep_waiting = true;
1035 			continue;
1036 		}
1037 
1038 		if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1039 			if (ret == -ESTALE)
1040 				gfs2_glock_dq(&ghs[i]);
1041 			else
1042 				gfs2_glock_update_hold_time(ghs[i].gh_gl,
1043 							    start_time);
1044 		}
1045 		if (!ret)
1046 			ret = ghs[i].gh_error;
1047 	}
1048 
1049 	if (keep_waiting)
1050 		goto wait_for_dlm;
1051 
1052 	/*
1053 	 * At this point, we've either acquired all locks or released them all.
1054 	 */
1055 	return ret;
1056 }
1057 
1058 /**
1059  * handle_callback - process a demote request
1060  * @gl: the glock
1061  * @state: the state the caller wants us to change to
1062  *
1063  * There are only two requests that we are going to see in actual
1064  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1065  */
1066 
1067 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1068 			    unsigned long delay, bool remote)
1069 {
1070 	int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
1071 
1072 	set_bit(bit, &gl->gl_flags);
1073 	if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1074 		gl->gl_demote_state = state;
1075 		gl->gl_demote_time = jiffies;
1076 	} else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1077 			gl->gl_demote_state != state) {
1078 		gl->gl_demote_state = LM_ST_UNLOCKED;
1079 	}
1080 	if (gl->gl_ops->go_callback)
1081 		gl->gl_ops->go_callback(gl, remote);
1082 	trace_gfs2_demote_rq(gl, remote);
1083 }
1084 
1085 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1086 {
1087 	struct va_format vaf;
1088 	va_list args;
1089 
1090 	va_start(args, fmt);
1091 
1092 	if (seq) {
1093 		seq_vprintf(seq, fmt, args);
1094 	} else {
1095 		vaf.fmt = fmt;
1096 		vaf.va = &args;
1097 
1098 		pr_err("%pV", &vaf);
1099 	}
1100 
1101 	va_end(args);
1102 }
1103 
1104 /**
1105  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1106  * @gh: the holder structure to add
1107  *
1108  * Eventually we should move the recursive locking trap to a
1109  * debugging option or something like that. This is the fast
1110  * path and needs to have the minimum number of distractions.
1111  *
1112  */
1113 
1114 static inline void add_to_queue(struct gfs2_holder *gh)
1115 __releases(&gl->gl_lockref.lock)
1116 __acquires(&gl->gl_lockref.lock)
1117 {
1118 	struct gfs2_glock *gl = gh->gh_gl;
1119 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1120 	struct list_head *insert_pt = NULL;
1121 	struct gfs2_holder *gh2;
1122 	int try_futile = 0;
1123 
1124 	GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1125 	if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1126 		GLOCK_BUG_ON(gl, true);
1127 
1128 	if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1129 		if (test_bit(GLF_LOCK, &gl->gl_flags))
1130 			try_futile = !may_grant(gl, gh);
1131 		if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1132 			goto fail;
1133 	}
1134 
1135 	list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1136 		if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1137 		    (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1138 			goto trap_recursive;
1139 		if (try_futile &&
1140 		    !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1141 fail:
1142 			gh->gh_error = GLR_TRYFAILED;
1143 			gfs2_holder_wake(gh);
1144 			return;
1145 		}
1146 		if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1147 			continue;
1148 		if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1149 			insert_pt = &gh2->gh_list;
1150 	}
1151 	set_bit(GLF_QUEUED, &gl->gl_flags);
1152 	trace_gfs2_glock_queue(gh, 1);
1153 	gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1154 	gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1155 	if (likely(insert_pt == NULL)) {
1156 		list_add_tail(&gh->gh_list, &gl->gl_holders);
1157 		if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1158 			goto do_cancel;
1159 		return;
1160 	}
1161 	list_add_tail(&gh->gh_list, insert_pt);
1162 do_cancel:
1163 	gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1164 	if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1165 		spin_unlock(&gl->gl_lockref.lock);
1166 		if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1167 			sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1168 		spin_lock(&gl->gl_lockref.lock);
1169 	}
1170 	return;
1171 
1172 trap_recursive:
1173 	fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1174 	fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1175 	fs_err(sdp, "lock type: %d req lock state : %d\n",
1176 	       gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1177 	fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1178 	fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1179 	fs_err(sdp, "lock type: %d req lock state : %d\n",
1180 	       gh->gh_gl->gl_name.ln_type, gh->gh_state);
1181 	gfs2_dump_glock(NULL, gl, true);
1182 	BUG();
1183 }
1184 
1185 /**
1186  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1187  * @gh: the holder structure
1188  *
1189  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1190  *
1191  * Returns: 0, GLR_TRYFAILED, or errno on failure
1192  */
1193 
1194 int gfs2_glock_nq(struct gfs2_holder *gh)
1195 {
1196 	struct gfs2_glock *gl = gh->gh_gl;
1197 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1198 	int error = 0;
1199 
1200 	if (unlikely(gfs2_withdrawn(sdp)))
1201 		return -EIO;
1202 
1203 	if (test_bit(GLF_LRU, &gl->gl_flags))
1204 		gfs2_glock_remove_from_lru(gl);
1205 
1206 	spin_lock(&gl->gl_lockref.lock);
1207 	add_to_queue(gh);
1208 	if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1209 		     test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1210 		set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1211 		gl->gl_lockref.count++;
1212 		__gfs2_glock_queue_work(gl, 0);
1213 	}
1214 	run_queue(gl, 1);
1215 	spin_unlock(&gl->gl_lockref.lock);
1216 
1217 	if (!(gh->gh_flags & GL_ASYNC))
1218 		error = gfs2_glock_wait(gh);
1219 
1220 	return error;
1221 }
1222 
1223 /**
1224  * gfs2_glock_poll - poll to see if an async request has been completed
1225  * @gh: the holder
1226  *
1227  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1228  */
1229 
1230 int gfs2_glock_poll(struct gfs2_holder *gh)
1231 {
1232 	return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1233 }
1234 
1235 /**
1236  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1237  * @gh: the glock holder
1238  *
1239  */
1240 
1241 void gfs2_glock_dq(struct gfs2_holder *gh)
1242 {
1243 	struct gfs2_glock *gl = gh->gh_gl;
1244 	const struct gfs2_glock_operations *glops = gl->gl_ops;
1245 	unsigned delay = 0;
1246 	int fast_path = 0;
1247 
1248 	spin_lock(&gl->gl_lockref.lock);
1249 	if (gh->gh_flags & GL_NOCACHE)
1250 		handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1251 
1252 	list_del_init(&gh->gh_list);
1253 	clear_bit(HIF_HOLDER, &gh->gh_iflags);
1254 	if (find_first_holder(gl) == NULL) {
1255 		if (glops->go_unlock) {
1256 			GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1257 			spin_unlock(&gl->gl_lockref.lock);
1258 			glops->go_unlock(gh);
1259 			spin_lock(&gl->gl_lockref.lock);
1260 			clear_bit(GLF_LOCK, &gl->gl_flags);
1261 		}
1262 		if (list_empty(&gl->gl_holders) &&
1263 		    !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1264 		    !test_bit(GLF_DEMOTE, &gl->gl_flags))
1265 			fast_path = 1;
1266 	}
1267 	if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1268 		gfs2_glock_add_to_lru(gl);
1269 
1270 	trace_gfs2_glock_queue(gh, 0);
1271 	if (unlikely(!fast_path)) {
1272 		gl->gl_lockref.count++;
1273 		if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1274 		    !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1275 		    gl->gl_name.ln_type == LM_TYPE_INODE)
1276 			delay = gl->gl_hold_time;
1277 		__gfs2_glock_queue_work(gl, delay);
1278 	}
1279 	spin_unlock(&gl->gl_lockref.lock);
1280 }
1281 
1282 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1283 {
1284 	struct gfs2_glock *gl = gh->gh_gl;
1285 	gfs2_glock_dq(gh);
1286 	might_sleep();
1287 	wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1288 }
1289 
1290 /**
1291  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1292  * @gh: the holder structure
1293  *
1294  */
1295 
1296 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1297 {
1298 	gfs2_glock_dq(gh);
1299 	gfs2_holder_uninit(gh);
1300 }
1301 
1302 /**
1303  * gfs2_glock_nq_num - acquire a glock based on lock number
1304  * @sdp: the filesystem
1305  * @number: the lock number
1306  * @glops: the glock operations for the type of glock
1307  * @state: the state to acquire the glock in
1308  * @flags: modifier flags for the acquisition
1309  * @gh: the struct gfs2_holder
1310  *
1311  * Returns: errno
1312  */
1313 
1314 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1315 		      const struct gfs2_glock_operations *glops,
1316 		      unsigned int state, u16 flags, struct gfs2_holder *gh)
1317 {
1318 	struct gfs2_glock *gl;
1319 	int error;
1320 
1321 	error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1322 	if (!error) {
1323 		error = gfs2_glock_nq_init(gl, state, flags, gh);
1324 		gfs2_glock_put(gl);
1325 	}
1326 
1327 	return error;
1328 }
1329 
1330 /**
1331  * glock_compare - Compare two struct gfs2_glock structures for sorting
1332  * @arg_a: the first structure
1333  * @arg_b: the second structure
1334  *
1335  */
1336 
1337 static int glock_compare(const void *arg_a, const void *arg_b)
1338 {
1339 	const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1340 	const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1341 	const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1342 	const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1343 
1344 	if (a->ln_number > b->ln_number)
1345 		return 1;
1346 	if (a->ln_number < b->ln_number)
1347 		return -1;
1348 	BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1349 	return 0;
1350 }
1351 
1352 /**
1353  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1354  * @num_gh: the number of structures
1355  * @ghs: an array of struct gfs2_holder structures
1356  *
1357  * Returns: 0 on success (all glocks acquired),
1358  *          errno on failure (no glocks acquired)
1359  */
1360 
1361 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1362 		     struct gfs2_holder **p)
1363 {
1364 	unsigned int x;
1365 	int error = 0;
1366 
1367 	for (x = 0; x < num_gh; x++)
1368 		p[x] = &ghs[x];
1369 
1370 	sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1371 
1372 	for (x = 0; x < num_gh; x++) {
1373 		p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1374 
1375 		error = gfs2_glock_nq(p[x]);
1376 		if (error) {
1377 			while (x--)
1378 				gfs2_glock_dq(p[x]);
1379 			break;
1380 		}
1381 	}
1382 
1383 	return error;
1384 }
1385 
1386 /**
1387  * gfs2_glock_nq_m - acquire multiple glocks
1388  * @num_gh: the number of structures
1389  * @ghs: an array of struct gfs2_holder structures
1390  *
1391  *
1392  * Returns: 0 on success (all glocks acquired),
1393  *          errno on failure (no glocks acquired)
1394  */
1395 
1396 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1397 {
1398 	struct gfs2_holder *tmp[4];
1399 	struct gfs2_holder **pph = tmp;
1400 	int error = 0;
1401 
1402 	switch(num_gh) {
1403 	case 0:
1404 		return 0;
1405 	case 1:
1406 		ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1407 		return gfs2_glock_nq(ghs);
1408 	default:
1409 		if (num_gh <= 4)
1410 			break;
1411 		pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1412 				    GFP_NOFS);
1413 		if (!pph)
1414 			return -ENOMEM;
1415 	}
1416 
1417 	error = nq_m_sync(num_gh, ghs, pph);
1418 
1419 	if (pph != tmp)
1420 		kfree(pph);
1421 
1422 	return error;
1423 }
1424 
1425 /**
1426  * gfs2_glock_dq_m - release multiple glocks
1427  * @num_gh: the number of structures
1428  * @ghs: an array of struct gfs2_holder structures
1429  *
1430  */
1431 
1432 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1433 {
1434 	while (num_gh--)
1435 		gfs2_glock_dq(&ghs[num_gh]);
1436 }
1437 
1438 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1439 {
1440 	unsigned long delay = 0;
1441 	unsigned long holdtime;
1442 	unsigned long now = jiffies;
1443 
1444 	gfs2_glock_hold(gl);
1445 	holdtime = gl->gl_tchange + gl->gl_hold_time;
1446 	if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1447 	    gl->gl_name.ln_type == LM_TYPE_INODE) {
1448 		if (time_before(now, holdtime))
1449 			delay = holdtime - now;
1450 		if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1451 			delay = gl->gl_hold_time;
1452 	}
1453 
1454 	spin_lock(&gl->gl_lockref.lock);
1455 	handle_callback(gl, state, delay, true);
1456 	__gfs2_glock_queue_work(gl, delay);
1457 	spin_unlock(&gl->gl_lockref.lock);
1458 }
1459 
1460 /**
1461  * gfs2_should_freeze - Figure out if glock should be frozen
1462  * @gl: The glock in question
1463  *
1464  * Glocks are not frozen if (a) the result of the dlm operation is
1465  * an error, (b) the locking operation was an unlock operation or
1466  * (c) if there is a "noexp" flagged request anywhere in the queue
1467  *
1468  * Returns: 1 if freezing should occur, 0 otherwise
1469  */
1470 
1471 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1472 {
1473 	const struct gfs2_holder *gh;
1474 
1475 	if (gl->gl_reply & ~LM_OUT_ST_MASK)
1476 		return 0;
1477 	if (gl->gl_target == LM_ST_UNLOCKED)
1478 		return 0;
1479 
1480 	list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1481 		if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1482 			continue;
1483 		if (LM_FLAG_NOEXP & gh->gh_flags)
1484 			return 0;
1485 	}
1486 
1487 	return 1;
1488 }
1489 
1490 /**
1491  * gfs2_glock_complete - Callback used by locking
1492  * @gl: Pointer to the glock
1493  * @ret: The return value from the dlm
1494  *
1495  * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1496  * to use a bitfield shared with other glock state fields.
1497  */
1498 
1499 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1500 {
1501 	struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1502 
1503 	spin_lock(&gl->gl_lockref.lock);
1504 	gl->gl_reply = ret;
1505 
1506 	if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1507 		if (gfs2_should_freeze(gl)) {
1508 			set_bit(GLF_FROZEN, &gl->gl_flags);
1509 			spin_unlock(&gl->gl_lockref.lock);
1510 			return;
1511 		}
1512 	}
1513 
1514 	gl->gl_lockref.count++;
1515 	set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1516 	__gfs2_glock_queue_work(gl, 0);
1517 	spin_unlock(&gl->gl_lockref.lock);
1518 }
1519 
1520 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1521 {
1522 	struct gfs2_glock *gla, *glb;
1523 
1524 	gla = list_entry(a, struct gfs2_glock, gl_lru);
1525 	glb = list_entry(b, struct gfs2_glock, gl_lru);
1526 
1527 	if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1528 		return 1;
1529 	if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1530 		return -1;
1531 
1532 	return 0;
1533 }
1534 
1535 /**
1536  * gfs2_dispose_glock_lru - Demote a list of glocks
1537  * @list: The list to dispose of
1538  *
1539  * Disposing of glocks may involve disk accesses, so that here we sort
1540  * the glocks by number (i.e. disk location of the inodes) so that if
1541  * there are any such accesses, they'll be sent in order (mostly).
1542  *
1543  * Must be called under the lru_lock, but may drop and retake this
1544  * lock. While the lru_lock is dropped, entries may vanish from the
1545  * list, but no new entries will appear on the list (since it is
1546  * private)
1547  */
1548 
1549 static void gfs2_dispose_glock_lru(struct list_head *list)
1550 __releases(&lru_lock)
1551 __acquires(&lru_lock)
1552 {
1553 	struct gfs2_glock *gl;
1554 
1555 	list_sort(NULL, list, glock_cmp);
1556 
1557 	while(!list_empty(list)) {
1558 		gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1559 		list_del_init(&gl->gl_lru);
1560 		if (!spin_trylock(&gl->gl_lockref.lock)) {
1561 add_back_to_lru:
1562 			list_add(&gl->gl_lru, &lru_list);
1563 			set_bit(GLF_LRU, &gl->gl_flags);
1564 			atomic_inc(&lru_count);
1565 			continue;
1566 		}
1567 		if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1568 			spin_unlock(&gl->gl_lockref.lock);
1569 			goto add_back_to_lru;
1570 		}
1571 		gl->gl_lockref.count++;
1572 		if (demote_ok(gl))
1573 			handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1574 		WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1575 		__gfs2_glock_queue_work(gl, 0);
1576 		spin_unlock(&gl->gl_lockref.lock);
1577 		cond_resched_lock(&lru_lock);
1578 	}
1579 }
1580 
1581 /**
1582  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1583  * @nr: The number of entries to scan
1584  *
1585  * This function selects the entries on the LRU which are able to
1586  * be demoted, and then kicks off the process by calling
1587  * gfs2_dispose_glock_lru() above.
1588  */
1589 
1590 static long gfs2_scan_glock_lru(int nr)
1591 {
1592 	struct gfs2_glock *gl;
1593 	LIST_HEAD(skipped);
1594 	LIST_HEAD(dispose);
1595 	long freed = 0;
1596 
1597 	spin_lock(&lru_lock);
1598 	while ((nr-- >= 0) && !list_empty(&lru_list)) {
1599 		gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1600 
1601 		/* Test for being demotable */
1602 		if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1603 			list_move(&gl->gl_lru, &dispose);
1604 			atomic_dec(&lru_count);
1605 			clear_bit(GLF_LRU, &gl->gl_flags);
1606 			freed++;
1607 			continue;
1608 		}
1609 
1610 		list_move(&gl->gl_lru, &skipped);
1611 	}
1612 	list_splice(&skipped, &lru_list);
1613 	if (!list_empty(&dispose))
1614 		gfs2_dispose_glock_lru(&dispose);
1615 	spin_unlock(&lru_lock);
1616 
1617 	return freed;
1618 }
1619 
1620 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1621 					    struct shrink_control *sc)
1622 {
1623 	if (!(sc->gfp_mask & __GFP_FS))
1624 		return SHRINK_STOP;
1625 	return gfs2_scan_glock_lru(sc->nr_to_scan);
1626 }
1627 
1628 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1629 					     struct shrink_control *sc)
1630 {
1631 	return vfs_pressure_ratio(atomic_read(&lru_count));
1632 }
1633 
1634 static struct shrinker glock_shrinker = {
1635 	.seeks = DEFAULT_SEEKS,
1636 	.count_objects = gfs2_glock_shrink_count,
1637 	.scan_objects = gfs2_glock_shrink_scan,
1638 };
1639 
1640 /**
1641  * examine_bucket - Call a function for glock in a hash bucket
1642  * @examiner: the function
1643  * @sdp: the filesystem
1644  * @bucket: the bucket
1645  *
1646  * Note that the function can be called multiple times on the same
1647  * object.  So the user must ensure that the function can cope with
1648  * that.
1649  */
1650 
1651 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1652 {
1653 	struct gfs2_glock *gl;
1654 	struct rhashtable_iter iter;
1655 
1656 	rhashtable_walk_enter(&gl_hash_table, &iter);
1657 
1658 	do {
1659 		rhashtable_walk_start(&iter);
1660 
1661 		while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1662 			if (gl->gl_name.ln_sbd == sdp &&
1663 			    lockref_get_not_dead(&gl->gl_lockref))
1664 				examiner(gl);
1665 
1666 		rhashtable_walk_stop(&iter);
1667 	} while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1668 
1669 	rhashtable_walk_exit(&iter);
1670 }
1671 
1672 /**
1673  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1674  * @gl: The glock to thaw
1675  *
1676  */
1677 
1678 static void thaw_glock(struct gfs2_glock *gl)
1679 {
1680 	if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) {
1681 		gfs2_glock_put(gl);
1682 		return;
1683 	}
1684 	set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1685 	gfs2_glock_queue_work(gl, 0);
1686 }
1687 
1688 /**
1689  * clear_glock - look at a glock and see if we can free it from glock cache
1690  * @gl: the glock to look at
1691  *
1692  */
1693 
1694 static void clear_glock(struct gfs2_glock *gl)
1695 {
1696 	gfs2_glock_remove_from_lru(gl);
1697 
1698 	spin_lock(&gl->gl_lockref.lock);
1699 	if (gl->gl_state != LM_ST_UNLOCKED)
1700 		handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1701 	__gfs2_glock_queue_work(gl, 0);
1702 	spin_unlock(&gl->gl_lockref.lock);
1703 }
1704 
1705 /**
1706  * gfs2_glock_thaw - Thaw any frozen glocks
1707  * @sdp: The super block
1708  *
1709  */
1710 
1711 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1712 {
1713 	glock_hash_walk(thaw_glock, sdp);
1714 }
1715 
1716 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
1717 {
1718 	spin_lock(&gl->gl_lockref.lock);
1719 	gfs2_dump_glock(seq, gl, fsid);
1720 	spin_unlock(&gl->gl_lockref.lock);
1721 }
1722 
1723 static void dump_glock_func(struct gfs2_glock *gl)
1724 {
1725 	dump_glock(NULL, gl, true);
1726 }
1727 
1728 /**
1729  * gfs2_gl_hash_clear - Empty out the glock hash table
1730  * @sdp: the filesystem
1731  * @wait: wait until it's all gone
1732  *
1733  * Called when unmounting the filesystem.
1734  */
1735 
1736 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1737 {
1738 	set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1739 	flush_workqueue(glock_workqueue);
1740 	glock_hash_walk(clear_glock, sdp);
1741 	flush_workqueue(glock_workqueue);
1742 	wait_event_timeout(sdp->sd_glock_wait,
1743 			   atomic_read(&sdp->sd_glock_disposal) == 0,
1744 			   HZ * 600);
1745 	glock_hash_walk(dump_glock_func, sdp);
1746 }
1747 
1748 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1749 {
1750 	struct gfs2_glock *gl = ip->i_gl;
1751 	int ret;
1752 
1753 	ret = gfs2_truncatei_resume(ip);
1754 	gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1755 
1756 	spin_lock(&gl->gl_lockref.lock);
1757 	clear_bit(GLF_LOCK, &gl->gl_flags);
1758 	run_queue(gl, 1);
1759 	spin_unlock(&gl->gl_lockref.lock);
1760 }
1761 
1762 static const char *state2str(unsigned state)
1763 {
1764 	switch(state) {
1765 	case LM_ST_UNLOCKED:
1766 		return "UN";
1767 	case LM_ST_SHARED:
1768 		return "SH";
1769 	case LM_ST_DEFERRED:
1770 		return "DF";
1771 	case LM_ST_EXCLUSIVE:
1772 		return "EX";
1773 	}
1774 	return "??";
1775 }
1776 
1777 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
1778 {
1779 	char *p = buf;
1780 	if (flags & LM_FLAG_TRY)
1781 		*p++ = 't';
1782 	if (flags & LM_FLAG_TRY_1CB)
1783 		*p++ = 'T';
1784 	if (flags & LM_FLAG_NOEXP)
1785 		*p++ = 'e';
1786 	if (flags & LM_FLAG_ANY)
1787 		*p++ = 'A';
1788 	if (flags & LM_FLAG_PRIORITY)
1789 		*p++ = 'p';
1790 	if (flags & GL_ASYNC)
1791 		*p++ = 'a';
1792 	if (flags & GL_EXACT)
1793 		*p++ = 'E';
1794 	if (flags & GL_NOCACHE)
1795 		*p++ = 'c';
1796 	if (test_bit(HIF_HOLDER, &iflags))
1797 		*p++ = 'H';
1798 	if (test_bit(HIF_WAIT, &iflags))
1799 		*p++ = 'W';
1800 	if (test_bit(HIF_FIRST, &iflags))
1801 		*p++ = 'F';
1802 	*p = 0;
1803 	return buf;
1804 }
1805 
1806 /**
1807  * dump_holder - print information about a glock holder
1808  * @seq: the seq_file struct
1809  * @gh: the glock holder
1810  * @fs_id_buf: pointer to file system id (if requested)
1811  *
1812  */
1813 
1814 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
1815 			const char *fs_id_buf)
1816 {
1817 	struct task_struct *gh_owner = NULL;
1818 	char flags_buf[32];
1819 
1820 	rcu_read_lock();
1821 	if (gh->gh_owner_pid)
1822 		gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1823 	gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1824 		       fs_id_buf, state2str(gh->gh_state),
1825 		       hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1826 		       gh->gh_error,
1827 		       gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1828 		       gh_owner ? gh_owner->comm : "(ended)",
1829 		       (void *)gh->gh_ip);
1830 	rcu_read_unlock();
1831 }
1832 
1833 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1834 {
1835 	const unsigned long *gflags = &gl->gl_flags;
1836 	char *p = buf;
1837 
1838 	if (test_bit(GLF_LOCK, gflags))
1839 		*p++ = 'l';
1840 	if (test_bit(GLF_DEMOTE, gflags))
1841 		*p++ = 'D';
1842 	if (test_bit(GLF_PENDING_DEMOTE, gflags))
1843 		*p++ = 'd';
1844 	if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1845 		*p++ = 'p';
1846 	if (test_bit(GLF_DIRTY, gflags))
1847 		*p++ = 'y';
1848 	if (test_bit(GLF_LFLUSH, gflags))
1849 		*p++ = 'f';
1850 	if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1851 		*p++ = 'i';
1852 	if (test_bit(GLF_REPLY_PENDING, gflags))
1853 		*p++ = 'r';
1854 	if (test_bit(GLF_INITIAL, gflags))
1855 		*p++ = 'I';
1856 	if (test_bit(GLF_FROZEN, gflags))
1857 		*p++ = 'F';
1858 	if (test_bit(GLF_QUEUED, gflags))
1859 		*p++ = 'q';
1860 	if (test_bit(GLF_LRU, gflags))
1861 		*p++ = 'L';
1862 	if (gl->gl_object)
1863 		*p++ = 'o';
1864 	if (test_bit(GLF_BLOCKING, gflags))
1865 		*p++ = 'b';
1866 	*p = 0;
1867 	return buf;
1868 }
1869 
1870 /**
1871  * gfs2_dump_glock - print information about a glock
1872  * @seq: The seq_file struct
1873  * @gl: the glock
1874  * @fsid: If true, also dump the file system id
1875  *
1876  * The file format is as follows:
1877  * One line per object, capital letters are used to indicate objects
1878  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1879  * other objects are indented by a single space and follow the glock to
1880  * which they are related. Fields are indicated by lower case letters
1881  * followed by a colon and the field value, except for strings which are in
1882  * [] so that its possible to see if they are composed of spaces for
1883  * example. The field's are n = number (id of the object), f = flags,
1884  * t = type, s = state, r = refcount, e = error, p = pid.
1885  *
1886  */
1887 
1888 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
1889 {
1890 	const struct gfs2_glock_operations *glops = gl->gl_ops;
1891 	unsigned long long dtime;
1892 	const struct gfs2_holder *gh;
1893 	char gflags_buf[32];
1894 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1895 	char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
1896 
1897 	memset(fs_id_buf, 0, sizeof(fs_id_buf));
1898 	if (fsid && sdp) /* safety precaution */
1899 		sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
1900 	dtime = jiffies - gl->gl_demote_time;
1901 	dtime *= 1000000/HZ; /* demote time in uSec */
1902 	if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1903 		dtime = 0;
1904 	gfs2_print_dbg(seq, "%sG:  s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
1905 		       "v:%d r:%d m:%ld\n", fs_id_buf, state2str(gl->gl_state),
1906 		  gl->gl_name.ln_type,
1907 		  (unsigned long long)gl->gl_name.ln_number,
1908 		  gflags2str(gflags_buf, gl),
1909 		  state2str(gl->gl_target),
1910 		  state2str(gl->gl_demote_state), dtime,
1911 		  atomic_read(&gl->gl_ail_count),
1912 		  atomic_read(&gl->gl_revokes),
1913 		  (int)gl->gl_lockref.count, gl->gl_hold_time);
1914 
1915 	list_for_each_entry(gh, &gl->gl_holders, gh_list)
1916 		dump_holder(seq, gh, fs_id_buf);
1917 
1918 	if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1919 		glops->go_dump(seq, gl, fs_id_buf);
1920 }
1921 
1922 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1923 {
1924 	struct gfs2_glock *gl = iter_ptr;
1925 
1926 	seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1927 		   gl->gl_name.ln_type,
1928 		   (unsigned long long)gl->gl_name.ln_number,
1929 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1930 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1931 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1932 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1933 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1934 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1935 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1936 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1937 	return 0;
1938 }
1939 
1940 static const char *gfs2_gltype[] = {
1941 	"type",
1942 	"reserved",
1943 	"nondisk",
1944 	"inode",
1945 	"rgrp",
1946 	"meta",
1947 	"iopen",
1948 	"flock",
1949 	"plock",
1950 	"quota",
1951 	"journal",
1952 };
1953 
1954 static const char *gfs2_stype[] = {
1955 	[GFS2_LKS_SRTT]		= "srtt",
1956 	[GFS2_LKS_SRTTVAR]	= "srttvar",
1957 	[GFS2_LKS_SRTTB]	= "srttb",
1958 	[GFS2_LKS_SRTTVARB]	= "srttvarb",
1959 	[GFS2_LKS_SIRT]		= "sirt",
1960 	[GFS2_LKS_SIRTVAR]	= "sirtvar",
1961 	[GFS2_LKS_DCOUNT]	= "dlm",
1962 	[GFS2_LKS_QCOUNT]	= "queue",
1963 };
1964 
1965 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1966 
1967 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1968 {
1969 	struct gfs2_sbd *sdp = seq->private;
1970 	loff_t pos = *(loff_t *)iter_ptr;
1971 	unsigned index = pos >> 3;
1972 	unsigned subindex = pos & 0x07;
1973 	int i;
1974 
1975 	if (index == 0 && subindex != 0)
1976 		return 0;
1977 
1978 	seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1979 		   (index == 0) ? "cpu": gfs2_stype[subindex]);
1980 
1981 	for_each_possible_cpu(i) {
1982                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1983 
1984 		if (index == 0)
1985 			seq_printf(seq, " %15u", i);
1986 		else
1987 			seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1988 				   lkstats[index - 1].stats[subindex]);
1989 	}
1990 	seq_putc(seq, '\n');
1991 	return 0;
1992 }
1993 
1994 int __init gfs2_glock_init(void)
1995 {
1996 	int i, ret;
1997 
1998 	ret = rhashtable_init(&gl_hash_table, &ht_parms);
1999 	if (ret < 0)
2000 		return ret;
2001 
2002 	glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2003 					  WQ_HIGHPRI | WQ_FREEZABLE, 0);
2004 	if (!glock_workqueue) {
2005 		rhashtable_destroy(&gl_hash_table);
2006 		return -ENOMEM;
2007 	}
2008 	gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2009 						WQ_MEM_RECLAIM | WQ_FREEZABLE,
2010 						0);
2011 	if (!gfs2_delete_workqueue) {
2012 		destroy_workqueue(glock_workqueue);
2013 		rhashtable_destroy(&gl_hash_table);
2014 		return -ENOMEM;
2015 	}
2016 
2017 	ret = register_shrinker(&glock_shrinker);
2018 	if (ret) {
2019 		destroy_workqueue(gfs2_delete_workqueue);
2020 		destroy_workqueue(glock_workqueue);
2021 		rhashtable_destroy(&gl_hash_table);
2022 		return ret;
2023 	}
2024 
2025 	for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2026 		init_waitqueue_head(glock_wait_table + i);
2027 
2028 	return 0;
2029 }
2030 
2031 void gfs2_glock_exit(void)
2032 {
2033 	unregister_shrinker(&glock_shrinker);
2034 	rhashtable_destroy(&gl_hash_table);
2035 	destroy_workqueue(glock_workqueue);
2036 	destroy_workqueue(gfs2_delete_workqueue);
2037 }
2038 
2039 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2040 {
2041 	struct gfs2_glock *gl = gi->gl;
2042 
2043 	if (gl) {
2044 		if (n == 0)
2045 			return;
2046 		if (!lockref_put_not_zero(&gl->gl_lockref))
2047 			gfs2_glock_queue_put(gl);
2048 	}
2049 	for (;;) {
2050 		gl = rhashtable_walk_next(&gi->hti);
2051 		if (IS_ERR_OR_NULL(gl)) {
2052 			if (gl == ERR_PTR(-EAGAIN)) {
2053 				n = 1;
2054 				continue;
2055 			}
2056 			gl = NULL;
2057 			break;
2058 		}
2059 		if (gl->gl_name.ln_sbd != gi->sdp)
2060 			continue;
2061 		if (n <= 1) {
2062 			if (!lockref_get_not_dead(&gl->gl_lockref))
2063 				continue;
2064 			break;
2065 		} else {
2066 			if (__lockref_is_dead(&gl->gl_lockref))
2067 				continue;
2068 			n--;
2069 		}
2070 	}
2071 	gi->gl = gl;
2072 }
2073 
2074 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2075 	__acquires(RCU)
2076 {
2077 	struct gfs2_glock_iter *gi = seq->private;
2078 	loff_t n;
2079 
2080 	/*
2081 	 * We can either stay where we are, skip to the next hash table
2082 	 * entry, or start from the beginning.
2083 	 */
2084 	if (*pos < gi->last_pos) {
2085 		rhashtable_walk_exit(&gi->hti);
2086 		rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2087 		n = *pos + 1;
2088 	} else {
2089 		n = *pos - gi->last_pos;
2090 	}
2091 
2092 	rhashtable_walk_start(&gi->hti);
2093 
2094 	gfs2_glock_iter_next(gi, n);
2095 	gi->last_pos = *pos;
2096 	return gi->gl;
2097 }
2098 
2099 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2100 				 loff_t *pos)
2101 {
2102 	struct gfs2_glock_iter *gi = seq->private;
2103 
2104 	(*pos)++;
2105 	gi->last_pos = *pos;
2106 	gfs2_glock_iter_next(gi, 1);
2107 	return gi->gl;
2108 }
2109 
2110 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2111 	__releases(RCU)
2112 {
2113 	struct gfs2_glock_iter *gi = seq->private;
2114 
2115 	rhashtable_walk_stop(&gi->hti);
2116 }
2117 
2118 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2119 {
2120 	dump_glock(seq, iter_ptr, false);
2121 	return 0;
2122 }
2123 
2124 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2125 {
2126 	preempt_disable();
2127 	if (*pos >= GFS2_NR_SBSTATS)
2128 		return NULL;
2129 	return pos;
2130 }
2131 
2132 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2133 				   loff_t *pos)
2134 {
2135 	(*pos)++;
2136 	if (*pos >= GFS2_NR_SBSTATS)
2137 		return NULL;
2138 	return pos;
2139 }
2140 
2141 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2142 {
2143 	preempt_enable();
2144 }
2145 
2146 static const struct seq_operations gfs2_glock_seq_ops = {
2147 	.start = gfs2_glock_seq_start,
2148 	.next  = gfs2_glock_seq_next,
2149 	.stop  = gfs2_glock_seq_stop,
2150 	.show  = gfs2_glock_seq_show,
2151 };
2152 
2153 static const struct seq_operations gfs2_glstats_seq_ops = {
2154 	.start = gfs2_glock_seq_start,
2155 	.next  = gfs2_glock_seq_next,
2156 	.stop  = gfs2_glock_seq_stop,
2157 	.show  = gfs2_glstats_seq_show,
2158 };
2159 
2160 static const struct seq_operations gfs2_sbstats_seq_ops = {
2161 	.start = gfs2_sbstats_seq_start,
2162 	.next  = gfs2_sbstats_seq_next,
2163 	.stop  = gfs2_sbstats_seq_stop,
2164 	.show  = gfs2_sbstats_seq_show,
2165 };
2166 
2167 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2168 
2169 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2170 			      const struct seq_operations *ops)
2171 {
2172 	int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2173 	if (ret == 0) {
2174 		struct seq_file *seq = file->private_data;
2175 		struct gfs2_glock_iter *gi = seq->private;
2176 
2177 		gi->sdp = inode->i_private;
2178 		seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2179 		if (seq->buf)
2180 			seq->size = GFS2_SEQ_GOODSIZE;
2181 		/*
2182 		 * Initially, we are "before" the first hash table entry; the
2183 		 * first call to rhashtable_walk_next gets us the first entry.
2184 		 */
2185 		gi->last_pos = -1;
2186 		gi->gl = NULL;
2187 		rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2188 	}
2189 	return ret;
2190 }
2191 
2192 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2193 {
2194 	return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2195 }
2196 
2197 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2198 {
2199 	struct seq_file *seq = file->private_data;
2200 	struct gfs2_glock_iter *gi = seq->private;
2201 
2202 	if (gi->gl)
2203 		gfs2_glock_put(gi->gl);
2204 	rhashtable_walk_exit(&gi->hti);
2205 	return seq_release_private(inode, file);
2206 }
2207 
2208 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2209 {
2210 	return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2211 }
2212 
2213 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
2214 {
2215 	int ret = seq_open(file, &gfs2_sbstats_seq_ops);
2216 	if (ret == 0) {
2217 		struct seq_file *seq = file->private_data;
2218 		seq->private = inode->i_private;  /* sdp */
2219 	}
2220 	return ret;
2221 }
2222 
2223 static const struct file_operations gfs2_glocks_fops = {
2224 	.owner   = THIS_MODULE,
2225 	.open    = gfs2_glocks_open,
2226 	.read    = seq_read,
2227 	.llseek  = seq_lseek,
2228 	.release = gfs2_glocks_release,
2229 };
2230 
2231 static const struct file_operations gfs2_glstats_fops = {
2232 	.owner   = THIS_MODULE,
2233 	.open    = gfs2_glstats_open,
2234 	.read    = seq_read,
2235 	.llseek  = seq_lseek,
2236 	.release = gfs2_glocks_release,
2237 };
2238 
2239 static const struct file_operations gfs2_sbstats_fops = {
2240 	.owner   = THIS_MODULE,
2241 	.open	 = gfs2_sbstats_open,
2242 	.read    = seq_read,
2243 	.llseek  = seq_lseek,
2244 	.release = seq_release,
2245 };
2246 
2247 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2248 {
2249 	sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2250 
2251 	debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2252 			    &gfs2_glocks_fops);
2253 
2254 	debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2255 			    &gfs2_glstats_fops);
2256 
2257 	debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2258 			    &gfs2_sbstats_fops);
2259 }
2260 
2261 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2262 {
2263 	debugfs_remove_recursive(sdp->debugfs_dir);
2264 	sdp->debugfs_dir = NULL;
2265 }
2266 
2267 void gfs2_register_debugfs(void)
2268 {
2269 	gfs2_root = debugfs_create_dir("gfs2", NULL);
2270 }
2271 
2272 void gfs2_unregister_debugfs(void)
2273 {
2274 	debugfs_remove(gfs2_root);
2275 	gfs2_root = NULL;
2276 }
2277