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 #include <linux/pid_namespace.h>
37 #include <linux/fdtable.h>
38 #include <linux/file.h>
39 
40 #include "gfs2.h"
41 #include "incore.h"
42 #include "glock.h"
43 #include "glops.h"
44 #include "inode.h"
45 #include "lops.h"
46 #include "meta_io.h"
47 #include "quota.h"
48 #include "super.h"
49 #include "util.h"
50 #include "bmap.h"
51 #define CREATE_TRACE_POINTS
52 #include "trace_gfs2.h"
53 
54 struct gfs2_glock_iter {
55 	struct gfs2_sbd *sdp;		/* incore superblock           */
56 	struct rhashtable_iter hti;	/* rhashtable iterator         */
57 	struct gfs2_glock *gl;		/* current glock struct        */
58 	loff_t last_pos;		/* last position               */
59 };
60 
61 typedef void (*glock_examiner) (struct gfs2_glock * gl);
62 
63 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
64 static void __gfs2_glock_dq(struct gfs2_holder *gh);
65 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
66 			    unsigned long delay, bool remote);
67 
68 static struct dentry *gfs2_root;
69 static struct workqueue_struct *glock_workqueue;
70 static LIST_HEAD(lru_list);
71 static atomic_t lru_count = ATOMIC_INIT(0);
72 static DEFINE_SPINLOCK(lru_lock);
73 
74 #define GFS2_GL_HASH_SHIFT      15
75 #define GFS2_GL_HASH_SIZE       BIT(GFS2_GL_HASH_SHIFT)
76 
77 static const struct rhashtable_params ht_parms = {
78 	.nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
79 	.key_len = offsetofend(struct lm_lockname, ln_type),
80 	.key_offset = offsetof(struct gfs2_glock, gl_name),
81 	.head_offset = offsetof(struct gfs2_glock, gl_node),
82 };
83 
84 static struct rhashtable gl_hash_table;
85 
86 #define GLOCK_WAIT_TABLE_BITS 12
87 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
88 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
89 
90 struct wait_glock_queue {
91 	struct lm_lockname *name;
92 	wait_queue_entry_t wait;
93 };
94 
glock_wake_function(wait_queue_entry_t * wait,unsigned int mode,int sync,void * key)95 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
96 			       int sync, void *key)
97 {
98 	struct wait_glock_queue *wait_glock =
99 		container_of(wait, struct wait_glock_queue, wait);
100 	struct lm_lockname *wait_name = wait_glock->name;
101 	struct lm_lockname *wake_name = key;
102 
103 	if (wake_name->ln_sbd != wait_name->ln_sbd ||
104 	    wake_name->ln_number != wait_name->ln_number ||
105 	    wake_name->ln_type != wait_name->ln_type)
106 		return 0;
107 	return autoremove_wake_function(wait, mode, sync, key);
108 }
109 
glock_waitqueue(struct lm_lockname * name)110 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
111 {
112 	u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
113 
114 	return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
115 }
116 
117 /**
118  * wake_up_glock  -  Wake up waiters on a glock
119  * @gl: the glock
120  */
wake_up_glock(struct gfs2_glock * gl)121 static void wake_up_glock(struct gfs2_glock *gl)
122 {
123 	wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
124 
125 	if (waitqueue_active(wq))
126 		__wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
127 }
128 
gfs2_glock_dealloc(struct rcu_head * rcu)129 static void gfs2_glock_dealloc(struct rcu_head *rcu)
130 {
131 	struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
132 
133 	kfree(gl->gl_lksb.sb_lvbptr);
134 	if (gl->gl_ops->go_flags & GLOF_ASPACE) {
135 		struct gfs2_glock_aspace *gla =
136 			container_of(gl, struct gfs2_glock_aspace, glock);
137 		kmem_cache_free(gfs2_glock_aspace_cachep, gla);
138 	} else
139 		kmem_cache_free(gfs2_glock_cachep, gl);
140 }
141 
142 /**
143  * glock_blocked_by_withdraw - determine if we can still use a glock
144  * @gl: the glock
145  *
146  * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
147  * when we're withdrawn. For example, to maintain metadata integrity, we should
148  * disallow the use of inode and rgrp glocks when withdrawn. Other glocks like
149  * the iopen or freeze glock may be safely used because none of their
150  * metadata goes through the journal. So in general, we should disallow all
151  * glocks that are journaled, and allow all the others. One exception is:
152  * we need to allow our active journal to be promoted and demoted so others
153  * may recover it and we can reacquire it when they're done.
154  */
glock_blocked_by_withdraw(struct gfs2_glock * gl)155 static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
156 {
157 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
158 
159 	if (likely(!gfs2_withdrawn(sdp)))
160 		return false;
161 	if (gl->gl_ops->go_flags & GLOF_NONDISK)
162 		return false;
163 	if (!sdp->sd_jdesc ||
164 	    gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
165 		return false;
166 	return true;
167 }
168 
gfs2_glock_free(struct gfs2_glock * gl)169 void gfs2_glock_free(struct gfs2_glock *gl)
170 {
171 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
172 
173 	gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
174 	rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
175 	smp_mb();
176 	wake_up_glock(gl);
177 	call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
178 	if (atomic_dec_and_test(&sdp->sd_glock_disposal))
179 		wake_up(&sdp->sd_kill_wait);
180 }
181 
182 /**
183  * gfs2_glock_hold() - increment reference count on glock
184  * @gl: The glock to hold
185  *
186  */
187 
gfs2_glock_hold(struct gfs2_glock * gl)188 struct gfs2_glock *gfs2_glock_hold(struct gfs2_glock *gl)
189 {
190 	GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
191 	lockref_get(&gl->gl_lockref);
192 	return gl;
193 }
194 
195 /**
196  * demote_ok - Check to see if it's ok to unlock a glock
197  * @gl: the glock
198  *
199  * Returns: 1 if it's ok
200  */
201 
demote_ok(const struct gfs2_glock * gl)202 static int demote_ok(const struct gfs2_glock *gl)
203 {
204 	const struct gfs2_glock_operations *glops = gl->gl_ops;
205 
206 	if (gl->gl_state == LM_ST_UNLOCKED)
207 		return 0;
208 	if (!list_empty(&gl->gl_holders))
209 		return 0;
210 	if (glops->go_demote_ok)
211 		return glops->go_demote_ok(gl);
212 	return 1;
213 }
214 
215 
gfs2_glock_add_to_lru(struct gfs2_glock * gl)216 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
217 {
218 	if (!(gl->gl_ops->go_flags & GLOF_LRU))
219 		return;
220 
221 	spin_lock(&lru_lock);
222 
223 	list_move_tail(&gl->gl_lru, &lru_list);
224 
225 	if (!test_bit(GLF_LRU, &gl->gl_flags)) {
226 		set_bit(GLF_LRU, &gl->gl_flags);
227 		atomic_inc(&lru_count);
228 	}
229 
230 	spin_unlock(&lru_lock);
231 }
232 
gfs2_glock_remove_from_lru(struct gfs2_glock * gl)233 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
234 {
235 	if (!(gl->gl_ops->go_flags & GLOF_LRU))
236 		return;
237 
238 	spin_lock(&lru_lock);
239 	if (test_bit(GLF_LRU, &gl->gl_flags)) {
240 		list_del_init(&gl->gl_lru);
241 		atomic_dec(&lru_count);
242 		clear_bit(GLF_LRU, &gl->gl_flags);
243 	}
244 	spin_unlock(&lru_lock);
245 }
246 
247 /*
248  * Enqueue the glock on the work queue.  Passes one glock reference on to the
249  * work queue.
250  */
__gfs2_glock_queue_work(struct gfs2_glock * gl,unsigned long delay)251 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
252 	if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
253 		/*
254 		 * We are holding the lockref spinlock, and the work was still
255 		 * queued above.  The queued work (glock_work_func) takes that
256 		 * spinlock before dropping its glock reference(s), so it
257 		 * cannot have dropped them in the meantime.
258 		 */
259 		GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
260 		gl->gl_lockref.count--;
261 	}
262 }
263 
gfs2_glock_queue_work(struct gfs2_glock * gl,unsigned long delay)264 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
265 	spin_lock(&gl->gl_lockref.lock);
266 	__gfs2_glock_queue_work(gl, delay);
267 	spin_unlock(&gl->gl_lockref.lock);
268 }
269 
__gfs2_glock_put(struct gfs2_glock * gl)270 static void __gfs2_glock_put(struct gfs2_glock *gl)
271 {
272 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
273 	struct address_space *mapping = gfs2_glock2aspace(gl);
274 
275 	lockref_mark_dead(&gl->gl_lockref);
276 	spin_unlock(&gl->gl_lockref.lock);
277 	gfs2_glock_remove_from_lru(gl);
278 	GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
279 	if (mapping) {
280 		truncate_inode_pages_final(mapping);
281 		if (!gfs2_withdrawn(sdp))
282 			GLOCK_BUG_ON(gl, !mapping_empty(mapping));
283 	}
284 	trace_gfs2_glock_put(gl);
285 	sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
286 }
287 
288 /*
289  * Cause the glock to be put in work queue context.
290  */
gfs2_glock_queue_put(struct gfs2_glock * gl)291 void gfs2_glock_queue_put(struct gfs2_glock *gl)
292 {
293 	gfs2_glock_queue_work(gl, 0);
294 }
295 
296 /**
297  * gfs2_glock_put() - Decrement reference count on glock
298  * @gl: The glock to put
299  *
300  */
301 
gfs2_glock_put(struct gfs2_glock * gl)302 void gfs2_glock_put(struct gfs2_glock *gl)
303 {
304 	if (lockref_put_or_lock(&gl->gl_lockref))
305 		return;
306 
307 	__gfs2_glock_put(gl);
308 }
309 
310 /**
311  * may_grant - check if it's ok to grant a new lock
312  * @gl: The glock
313  * @current_gh: One of the current holders of @gl
314  * @gh: The lock request which we wish to grant
315  *
316  * With our current compatibility rules, if a glock has one or more active
317  * holders (HIF_HOLDER flag set), any of those holders can be passed in as
318  * @current_gh; they are all the same as far as compatibility with the new @gh
319  * goes.
320  *
321  * Returns true if it's ok to grant the lock.
322  */
323 
may_grant(struct gfs2_glock * gl,struct gfs2_holder * current_gh,struct gfs2_holder * gh)324 static inline bool may_grant(struct gfs2_glock *gl,
325 			     struct gfs2_holder *current_gh,
326 			     struct gfs2_holder *gh)
327 {
328 	if (current_gh) {
329 		GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, &current_gh->gh_iflags));
330 
331 		switch(current_gh->gh_state) {
332 		case LM_ST_EXCLUSIVE:
333 			/*
334 			 * Here we make a special exception to grant holders
335 			 * who agree to share the EX lock with other holders
336 			 * who also have the bit set. If the original holder
337 			 * has the LM_FLAG_NODE_SCOPE bit set, we grant more
338 			 * holders with the bit set.
339 			 */
340 			return gh->gh_state == LM_ST_EXCLUSIVE &&
341 			       (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) &&
342 			       (gh->gh_flags & LM_FLAG_NODE_SCOPE);
343 
344 		case LM_ST_SHARED:
345 		case LM_ST_DEFERRED:
346 			return gh->gh_state == current_gh->gh_state;
347 
348 		default:
349 			return false;
350 		}
351 	}
352 
353 	if (gl->gl_state == gh->gh_state)
354 		return true;
355 	if (gh->gh_flags & GL_EXACT)
356 		return false;
357 	if (gl->gl_state == LM_ST_EXCLUSIVE) {
358 		return gh->gh_state == LM_ST_SHARED ||
359 		       gh->gh_state == LM_ST_DEFERRED;
360 	}
361 	if (gh->gh_flags & LM_FLAG_ANY)
362 		return gl->gl_state != LM_ST_UNLOCKED;
363 	return false;
364 }
365 
gfs2_holder_wake(struct gfs2_holder * gh)366 static void gfs2_holder_wake(struct gfs2_holder *gh)
367 {
368 	clear_bit(HIF_WAIT, &gh->gh_iflags);
369 	smp_mb__after_atomic();
370 	wake_up_bit(&gh->gh_iflags, HIF_WAIT);
371 	if (gh->gh_flags & GL_ASYNC) {
372 		struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
373 
374 		wake_up(&sdp->sd_async_glock_wait);
375 	}
376 }
377 
378 /**
379  * do_error - Something unexpected has happened during a lock request
380  * @gl: The glock
381  * @ret: The status from the DLM
382  */
383 
do_error(struct gfs2_glock * gl,const int ret)384 static void do_error(struct gfs2_glock *gl, const int ret)
385 {
386 	struct gfs2_holder *gh, *tmp;
387 
388 	list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
389 		if (test_bit(HIF_HOLDER, &gh->gh_iflags))
390 			continue;
391 		if (ret & LM_OUT_ERROR)
392 			gh->gh_error = -EIO;
393 		else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
394 			gh->gh_error = GLR_TRYFAILED;
395 		else
396 			continue;
397 		list_del_init(&gh->gh_list);
398 		trace_gfs2_glock_queue(gh, 0);
399 		gfs2_holder_wake(gh);
400 	}
401 }
402 
403 /**
404  * find_first_holder - find the first "holder" gh
405  * @gl: the glock
406  */
407 
find_first_holder(const struct gfs2_glock * gl)408 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
409 {
410 	struct gfs2_holder *gh;
411 
412 	if (!list_empty(&gl->gl_holders)) {
413 		gh = list_first_entry(&gl->gl_holders, struct gfs2_holder,
414 				      gh_list);
415 		if (test_bit(HIF_HOLDER, &gh->gh_iflags))
416 			return gh;
417 	}
418 	return NULL;
419 }
420 
421 /*
422  * gfs2_instantiate - Call the glops instantiate function
423  * @gh: The glock holder
424  *
425  * Returns: 0 if instantiate was successful, or error.
426  */
gfs2_instantiate(struct gfs2_holder * gh)427 int gfs2_instantiate(struct gfs2_holder *gh)
428 {
429 	struct gfs2_glock *gl = gh->gh_gl;
430 	const struct gfs2_glock_operations *glops = gl->gl_ops;
431 	int ret;
432 
433 again:
434 	if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
435 		goto done;
436 
437 	/*
438 	 * Since we unlock the lockref lock, we set a flag to indicate
439 	 * instantiate is in progress.
440 	 */
441 	if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
442 		wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
443 			    TASK_UNINTERRUPTIBLE);
444 		/*
445 		 * Here we just waited for a different instantiate to finish.
446 		 * But that may not have been successful, as when a process
447 		 * locks an inode glock _before_ it has an actual inode to
448 		 * instantiate into. So we check again. This process might
449 		 * have an inode to instantiate, so might be successful.
450 		 */
451 		goto again;
452 	}
453 
454 	ret = glops->go_instantiate(gl);
455 	if (!ret)
456 		clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
457 	clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
458 	if (ret)
459 		return ret;
460 
461 done:
462 	if (glops->go_held)
463 		return glops->go_held(gh);
464 	return 0;
465 }
466 
467 /**
468  * do_promote - promote as many requests as possible on the current queue
469  * @gl: The glock
470  *
471  * Returns true on success (i.e., progress was made or there are no waiters).
472  */
473 
do_promote(struct gfs2_glock * gl)474 static bool do_promote(struct gfs2_glock *gl)
475 {
476 	struct gfs2_holder *gh, *current_gh;
477 
478 	current_gh = find_first_holder(gl);
479 	list_for_each_entry(gh, &gl->gl_holders, gh_list) {
480 		if (test_bit(HIF_HOLDER, &gh->gh_iflags))
481 			continue;
482 		if (!may_grant(gl, current_gh, gh)) {
483 			/*
484 			 * If we get here, it means we may not grant this
485 			 * holder for some reason. If this holder is at the
486 			 * head of the list, it means we have a blocked holder
487 			 * at the head, so return false.
488 			 */
489 			if (list_is_first(&gh->gh_list, &gl->gl_holders))
490 				return false;
491 			do_error(gl, 0);
492 			break;
493 		}
494 		set_bit(HIF_HOLDER, &gh->gh_iflags);
495 		trace_gfs2_promote(gh);
496 		gfs2_holder_wake(gh);
497 		if (!current_gh)
498 			current_gh = gh;
499 	}
500 	return true;
501 }
502 
503 /**
504  * find_first_waiter - find the first gh that's waiting for the glock
505  * @gl: the glock
506  */
507 
find_first_waiter(const struct gfs2_glock * gl)508 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
509 {
510 	struct gfs2_holder *gh;
511 
512 	list_for_each_entry(gh, &gl->gl_holders, gh_list) {
513 		if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
514 			return gh;
515 	}
516 	return NULL;
517 }
518 
519 /**
520  * state_change - record that the glock is now in a different state
521  * @gl: the glock
522  * @new_state: the new state
523  */
524 
state_change(struct gfs2_glock * gl,unsigned int new_state)525 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
526 {
527 	int held1, held2;
528 
529 	held1 = (gl->gl_state != LM_ST_UNLOCKED);
530 	held2 = (new_state != LM_ST_UNLOCKED);
531 
532 	if (held1 != held2) {
533 		GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
534 		if (held2)
535 			gl->gl_lockref.count++;
536 		else
537 			gl->gl_lockref.count--;
538 	}
539 	if (new_state != gl->gl_target)
540 		/* shorten our minimum hold time */
541 		gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
542 				       GL_GLOCK_MIN_HOLD);
543 	gl->gl_state = new_state;
544 	gl->gl_tchange = jiffies;
545 }
546 
gfs2_set_demote(struct gfs2_glock * gl)547 static void gfs2_set_demote(struct gfs2_glock *gl)
548 {
549 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
550 
551 	set_bit(GLF_DEMOTE, &gl->gl_flags);
552 	smp_mb();
553 	wake_up(&sdp->sd_async_glock_wait);
554 }
555 
gfs2_demote_wake(struct gfs2_glock * gl)556 static void gfs2_demote_wake(struct gfs2_glock *gl)
557 {
558 	gl->gl_demote_state = LM_ST_EXCLUSIVE;
559 	clear_bit(GLF_DEMOTE, &gl->gl_flags);
560 	smp_mb__after_atomic();
561 	wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
562 }
563 
564 /**
565  * finish_xmote - The DLM has replied to one of our lock requests
566  * @gl: The glock
567  * @ret: The status from the DLM
568  *
569  */
570 
finish_xmote(struct gfs2_glock * gl,unsigned int ret)571 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
572 {
573 	const struct gfs2_glock_operations *glops = gl->gl_ops;
574 	struct gfs2_holder *gh;
575 	unsigned state = ret & LM_OUT_ST_MASK;
576 
577 	spin_lock(&gl->gl_lockref.lock);
578 	trace_gfs2_glock_state_change(gl, state);
579 	state_change(gl, state);
580 	gh = find_first_waiter(gl);
581 
582 	/* Demote to UN request arrived during demote to SH or DF */
583 	if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
584 	    state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
585 		gl->gl_target = LM_ST_UNLOCKED;
586 
587 	/* Check for state != intended state */
588 	if (unlikely(state != gl->gl_target)) {
589 		if (gh && (ret & LM_OUT_CANCELED))
590 			gfs2_holder_wake(gh);
591 		if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
592 			/* move to back of queue and try next entry */
593 			if (ret & LM_OUT_CANCELED) {
594 				list_move_tail(&gh->gh_list, &gl->gl_holders);
595 				gh = find_first_waiter(gl);
596 				gl->gl_target = gh->gh_state;
597 				if (do_promote(gl))
598 					goto out;
599 				goto retry;
600 			}
601 			/* Some error or failed "try lock" - report it */
602 			if ((ret & LM_OUT_ERROR) ||
603 			    (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
604 				gl->gl_target = gl->gl_state;
605 				do_error(gl, ret);
606 				goto out;
607 			}
608 		}
609 		switch(state) {
610 		/* Unlocked due to conversion deadlock, try again */
611 		case LM_ST_UNLOCKED:
612 retry:
613 			do_xmote(gl, gh, gl->gl_target);
614 			break;
615 		/* Conversion fails, unlock and try again */
616 		case LM_ST_SHARED:
617 		case LM_ST_DEFERRED:
618 			do_xmote(gl, gh, LM_ST_UNLOCKED);
619 			break;
620 		default: /* Everything else */
621 			fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
622 			       gl->gl_target, state);
623 			GLOCK_BUG_ON(gl, 1);
624 		}
625 		spin_unlock(&gl->gl_lockref.lock);
626 		return;
627 	}
628 
629 	/* Fast path - we got what we asked for */
630 	if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
631 		gfs2_demote_wake(gl);
632 	if (state != LM_ST_UNLOCKED) {
633 		if (glops->go_xmote_bh) {
634 			int rv;
635 
636 			spin_unlock(&gl->gl_lockref.lock);
637 			rv = glops->go_xmote_bh(gl);
638 			spin_lock(&gl->gl_lockref.lock);
639 			if (rv) {
640 				do_error(gl, rv);
641 				goto out;
642 			}
643 		}
644 		do_promote(gl);
645 	}
646 out:
647 	clear_bit(GLF_LOCK, &gl->gl_flags);
648 	spin_unlock(&gl->gl_lockref.lock);
649 }
650 
is_system_glock(struct gfs2_glock * gl)651 static bool is_system_glock(struct gfs2_glock *gl)
652 {
653 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
654 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
655 
656 	if (gl == m_ip->i_gl)
657 		return true;
658 	return false;
659 }
660 
661 /**
662  * do_xmote - Calls the DLM to change the state of a lock
663  * @gl: The lock state
664  * @gh: The holder (only for promotes)
665  * @target: The target lock state
666  *
667  */
668 
do_xmote(struct gfs2_glock * gl,struct gfs2_holder * gh,unsigned int target)669 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh,
670 					 unsigned int target)
671 __releases(&gl->gl_lockref.lock)
672 __acquires(&gl->gl_lockref.lock)
673 {
674 	const struct gfs2_glock_operations *glops = gl->gl_ops;
675 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
676 	unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
677 	int ret;
678 
679 	if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
680 	    gh && !(gh->gh_flags & LM_FLAG_NOEXP))
681 		goto skip_inval;
682 
683 	lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP);
684 	GLOCK_BUG_ON(gl, gl->gl_state == target);
685 	GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
686 	if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
687 	    glops->go_inval) {
688 		/*
689 		 * If another process is already doing the invalidate, let that
690 		 * finish first.  The glock state machine will get back to this
691 		 * holder again later.
692 		 */
693 		if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
694 				     &gl->gl_flags))
695 			return;
696 		do_error(gl, 0); /* Fail queued try locks */
697 	}
698 	gl->gl_req = target;
699 	set_bit(GLF_BLOCKING, &gl->gl_flags);
700 	if ((gl->gl_req == LM_ST_UNLOCKED) ||
701 	    (gl->gl_state == LM_ST_EXCLUSIVE) ||
702 	    (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
703 		clear_bit(GLF_BLOCKING, &gl->gl_flags);
704 	spin_unlock(&gl->gl_lockref.lock);
705 	if (glops->go_sync) {
706 		ret = glops->go_sync(gl);
707 		/* If we had a problem syncing (due to io errors or whatever,
708 		 * we should not invalidate the metadata or tell dlm to
709 		 * release the glock to other nodes.
710 		 */
711 		if (ret) {
712 			if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
713 				fs_err(sdp, "Error %d syncing glock \n", ret);
714 				gfs2_dump_glock(NULL, gl, true);
715 			}
716 			goto skip_inval;
717 		}
718 	}
719 	if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
720 		/*
721 		 * The call to go_sync should have cleared out the ail list.
722 		 * If there are still items, we have a problem. We ought to
723 		 * withdraw, but we can't because the withdraw code also uses
724 		 * glocks. Warn about the error, dump the glock, then fall
725 		 * through and wait for logd to do the withdraw for us.
726 		 */
727 		if ((atomic_read(&gl->gl_ail_count) != 0) &&
728 		    (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
729 			gfs2_glock_assert_warn(gl,
730 					       !atomic_read(&gl->gl_ail_count));
731 			gfs2_dump_glock(NULL, gl, true);
732 		}
733 		glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
734 		clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
735 	}
736 
737 skip_inval:
738 	gfs2_glock_hold(gl);
739 	/*
740 	 * Check for an error encountered since we called go_sync and go_inval.
741 	 * If so, we can't withdraw from the glock code because the withdraw
742 	 * code itself uses glocks (see function signal_our_withdraw) to
743 	 * change the mount to read-only. Most importantly, we must not call
744 	 * dlm to unlock the glock until the journal is in a known good state
745 	 * (after journal replay) otherwise other nodes may use the object
746 	 * (rgrp or dinode) and then later, journal replay will corrupt the
747 	 * file system. The best we can do here is wait for the logd daemon
748 	 * to see sd_log_error and withdraw, and in the meantime, requeue the
749 	 * work for later.
750 	 *
751 	 * We make a special exception for some system glocks, such as the
752 	 * system statfs inode glock, which needs to be granted before the
753 	 * gfs2_quotad daemon can exit, and that exit needs to finish before
754 	 * we can unmount the withdrawn file system.
755 	 *
756 	 * However, if we're just unlocking the lock (say, for unmount, when
757 	 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
758 	 * then it's okay to tell dlm to unlock it.
759 	 */
760 	if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
761 		gfs2_withdraw_delayed(sdp);
762 	if (glock_blocked_by_withdraw(gl) &&
763 	    (target != LM_ST_UNLOCKED ||
764 	     test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
765 		if (!is_system_glock(gl)) {
766 			handle_callback(gl, LM_ST_UNLOCKED, 0, false); /* sets demote */
767 			/*
768 			 * Ordinarily, we would call dlm and its callback would call
769 			 * finish_xmote, which would call state_change() to the new state.
770 			 * Since we withdrew, we won't call dlm, so call state_change
771 			 * manually, but to the UNLOCKED state we desire.
772 			 */
773 			state_change(gl, LM_ST_UNLOCKED);
774 			/*
775 			 * We skip telling dlm to do the locking, so we won't get a
776 			 * reply that would otherwise clear GLF_LOCK. So we clear it here.
777 			 */
778 			clear_bit(GLF_LOCK, &gl->gl_flags);
779 			clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
780 			gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
781 			goto out;
782 		} else {
783 			clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
784 		}
785 	}
786 
787 	if (sdp->sd_lockstruct.ls_ops->lm_lock)	{
788 		/* lock_dlm */
789 		ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
790 		if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
791 		    target == LM_ST_UNLOCKED &&
792 		    test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
793 			finish_xmote(gl, target);
794 			gfs2_glock_queue_work(gl, 0);
795 		} else if (ret) {
796 			fs_err(sdp, "lm_lock ret %d\n", ret);
797 			GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
798 		}
799 	} else { /* lock_nolock */
800 		finish_xmote(gl, target);
801 		gfs2_glock_queue_work(gl, 0);
802 	}
803 out:
804 	spin_lock(&gl->gl_lockref.lock);
805 }
806 
807 /**
808  * run_queue - do all outstanding tasks related to a glock
809  * @gl: The glock in question
810  * @nonblock: True if we must not block in run_queue
811  *
812  */
813 
run_queue(struct gfs2_glock * gl,const int nonblock)814 static void run_queue(struct gfs2_glock *gl, const int nonblock)
815 __releases(&gl->gl_lockref.lock)
816 __acquires(&gl->gl_lockref.lock)
817 {
818 	struct gfs2_holder *gh = NULL;
819 
820 	if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
821 		return;
822 
823 	GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
824 
825 	if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
826 	    gl->gl_demote_state != gl->gl_state) {
827 		if (find_first_holder(gl))
828 			goto out_unlock;
829 		if (nonblock)
830 			goto out_sched;
831 		set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
832 		GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
833 		gl->gl_target = gl->gl_demote_state;
834 	} else {
835 		if (test_bit(GLF_DEMOTE, &gl->gl_flags))
836 			gfs2_demote_wake(gl);
837 		if (do_promote(gl))
838 			goto out_unlock;
839 		gh = find_first_waiter(gl);
840 		gl->gl_target = gh->gh_state;
841 		if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
842 			do_error(gl, 0); /* Fail queued try locks */
843 	}
844 	do_xmote(gl, gh, gl->gl_target);
845 	return;
846 
847 out_sched:
848 	clear_bit(GLF_LOCK, &gl->gl_flags);
849 	smp_mb__after_atomic();
850 	gl->gl_lockref.count++;
851 	__gfs2_glock_queue_work(gl, 0);
852 	return;
853 
854 out_unlock:
855 	clear_bit(GLF_LOCK, &gl->gl_flags);
856 	smp_mb__after_atomic();
857 	return;
858 }
859 
860 /**
861  * glock_set_object - set the gl_object field of a glock
862  * @gl: the glock
863  * @object: the object
864  */
glock_set_object(struct gfs2_glock * gl,void * object)865 void glock_set_object(struct gfs2_glock *gl, void *object)
866 {
867 	void *prev_object;
868 
869 	spin_lock(&gl->gl_lockref.lock);
870 	prev_object = gl->gl_object;
871 	gl->gl_object = object;
872 	spin_unlock(&gl->gl_lockref.lock);
873 	if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == NULL)) {
874 		pr_warn("glock=%u/%llx\n",
875 			gl->gl_name.ln_type,
876 			(unsigned long long)gl->gl_name.ln_number);
877 		gfs2_dump_glock(NULL, gl, true);
878 	}
879 }
880 
881 /**
882  * glock_clear_object - clear the gl_object field of a glock
883  * @gl: the glock
884  * @object: object the glock currently points at
885  */
glock_clear_object(struct gfs2_glock * gl,void * object)886 void glock_clear_object(struct gfs2_glock *gl, void *object)
887 {
888 	void *prev_object;
889 
890 	spin_lock(&gl->gl_lockref.lock);
891 	prev_object = gl->gl_object;
892 	gl->gl_object = NULL;
893 	spin_unlock(&gl->gl_lockref.lock);
894 	if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == object)) {
895 		pr_warn("glock=%u/%llx\n",
896 			gl->gl_name.ln_type,
897 			(unsigned long long)gl->gl_name.ln_number);
898 		gfs2_dump_glock(NULL, gl, true);
899 	}
900 }
901 
gfs2_inode_remember_delete(struct gfs2_glock * gl,u64 generation)902 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
903 {
904 	struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
905 
906 	if (ri->ri_magic == 0)
907 		ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
908 	if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
909 		ri->ri_generation_deleted = cpu_to_be64(generation);
910 }
911 
gfs2_inode_already_deleted(struct gfs2_glock * gl,u64 generation)912 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
913 {
914 	struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
915 
916 	if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
917 		return false;
918 	return generation <= be64_to_cpu(ri->ri_generation_deleted);
919 }
920 
gfs2_glock_poke(struct gfs2_glock * gl)921 static void gfs2_glock_poke(struct gfs2_glock *gl)
922 {
923 	int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
924 	struct gfs2_holder gh;
925 	int error;
926 
927 	__gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
928 	error = gfs2_glock_nq(&gh);
929 	if (!error)
930 		gfs2_glock_dq(&gh);
931 	gfs2_holder_uninit(&gh);
932 }
933 
gfs2_try_evict(struct gfs2_glock * gl)934 static bool gfs2_try_evict(struct gfs2_glock *gl)
935 {
936 	struct gfs2_inode *ip;
937 	bool evicted = false;
938 
939 	/*
940 	 * If there is contention on the iopen glock and we have an inode, try
941 	 * to grab and release the inode so that it can be evicted.  This will
942 	 * allow the remote node to go ahead and delete the inode without us
943 	 * having to do it, which will avoid rgrp glock thrashing.
944 	 *
945 	 * The remote node is likely still holding the corresponding inode
946 	 * glock, so it will run before we get to verify that the delete has
947 	 * happened below.
948 	 */
949 	spin_lock(&gl->gl_lockref.lock);
950 	ip = gl->gl_object;
951 	if (ip && !igrab(&ip->i_inode))
952 		ip = NULL;
953 	spin_unlock(&gl->gl_lockref.lock);
954 	if (ip) {
955 		gl->gl_no_formal_ino = ip->i_no_formal_ino;
956 		set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
957 		d_prune_aliases(&ip->i_inode);
958 		iput(&ip->i_inode);
959 
960 		/* If the inode was evicted, gl->gl_object will now be NULL. */
961 		spin_lock(&gl->gl_lockref.lock);
962 		ip = gl->gl_object;
963 		if (ip) {
964 			clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
965 			if (!igrab(&ip->i_inode))
966 				ip = NULL;
967 		}
968 		spin_unlock(&gl->gl_lockref.lock);
969 		if (ip) {
970 			gfs2_glock_poke(ip->i_gl);
971 			iput(&ip->i_inode);
972 		}
973 		evicted = !ip;
974 	}
975 	return evicted;
976 }
977 
gfs2_queue_try_to_evict(struct gfs2_glock * gl)978 bool gfs2_queue_try_to_evict(struct gfs2_glock *gl)
979 {
980 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
981 
982 	if (test_and_set_bit(GLF_TRY_TO_EVICT, &gl->gl_flags))
983 		return false;
984 	return queue_delayed_work(sdp->sd_delete_wq,
985 				  &gl->gl_delete, 0);
986 }
987 
gfs2_queue_verify_evict(struct gfs2_glock * gl)988 static bool gfs2_queue_verify_evict(struct gfs2_glock *gl)
989 {
990 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
991 
992 	if (test_and_set_bit(GLF_VERIFY_EVICT, &gl->gl_flags))
993 		return false;
994 	return queue_delayed_work(sdp->sd_delete_wq,
995 				  &gl->gl_delete, 5 * HZ);
996 }
997 
delete_work_func(struct work_struct * work)998 static void delete_work_func(struct work_struct *work)
999 {
1000 	struct delayed_work *dwork = to_delayed_work(work);
1001 	struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
1002 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1003 	struct inode *inode;
1004 	u64 no_addr = gl->gl_name.ln_number;
1005 
1006 	if (test_and_clear_bit(GLF_TRY_TO_EVICT, &gl->gl_flags)) {
1007 		/*
1008 		 * If we can evict the inode, give the remote node trying to
1009 		 * delete the inode some time before verifying that the delete
1010 		 * has happened.  Otherwise, if we cause contention on the inode glock
1011 		 * immediately, the remote node will think that we still have
1012 		 * the inode in use, and so it will give up waiting.
1013 		 *
1014 		 * If we can't evict the inode, signal to the remote node that
1015 		 * the inode is still in use.  We'll later try to delete the
1016 		 * inode locally in gfs2_evict_inode.
1017 		 *
1018 		 * FIXME: We only need to verify that the remote node has
1019 		 * deleted the inode because nodes before this remote delete
1020 		 * rework won't cooperate.  At a later time, when we no longer
1021 		 * care about compatibility with such nodes, we can skip this
1022 		 * step entirely.
1023 		 */
1024 		if (gfs2_try_evict(gl)) {
1025 			if (test_bit(SDF_KILL, &sdp->sd_flags))
1026 				goto out;
1027 			if (gfs2_queue_verify_evict(gl))
1028 				return;
1029 		}
1030 		goto out;
1031 	}
1032 
1033 	if (test_and_clear_bit(GLF_VERIFY_EVICT, &gl->gl_flags)) {
1034 		inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1035 					    GFS2_BLKST_UNLINKED);
1036 		if (IS_ERR(inode)) {
1037 			if (PTR_ERR(inode) == -EAGAIN &&
1038 			    !test_bit(SDF_KILL, &sdp->sd_flags) &&
1039 			    gfs2_queue_verify_evict(gl))
1040 				return;
1041 		} else {
1042 			d_prune_aliases(inode);
1043 			iput(inode);
1044 		}
1045 	}
1046 
1047 out:
1048 	gfs2_glock_put(gl);
1049 }
1050 
glock_work_func(struct work_struct * work)1051 static void glock_work_func(struct work_struct *work)
1052 {
1053 	unsigned long delay = 0;
1054 	struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1055 	unsigned int drop_refs = 1;
1056 
1057 	if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1058 		finish_xmote(gl, gl->gl_reply);
1059 		drop_refs++;
1060 	}
1061 	spin_lock(&gl->gl_lockref.lock);
1062 	if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1063 	    gl->gl_state != LM_ST_UNLOCKED &&
1064 	    gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1065 		unsigned long holdtime, now = jiffies;
1066 
1067 		holdtime = gl->gl_tchange + gl->gl_hold_time;
1068 		if (time_before(now, holdtime))
1069 			delay = holdtime - now;
1070 
1071 		if (!delay) {
1072 			clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1073 			gfs2_set_demote(gl);
1074 		}
1075 	}
1076 	run_queue(gl, 0);
1077 	if (delay) {
1078 		/* Keep one glock reference for the work we requeue. */
1079 		drop_refs--;
1080 		if (gl->gl_name.ln_type != LM_TYPE_INODE)
1081 			delay = 0;
1082 		__gfs2_glock_queue_work(gl, delay);
1083 	}
1084 
1085 	/*
1086 	 * Drop the remaining glock references manually here. (Mind that
1087 	 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1088 	 * here as well.)
1089 	 */
1090 	gl->gl_lockref.count -= drop_refs;
1091 	if (!gl->gl_lockref.count) {
1092 		__gfs2_glock_put(gl);
1093 		return;
1094 	}
1095 	spin_unlock(&gl->gl_lockref.lock);
1096 }
1097 
find_insert_glock(struct lm_lockname * name,struct gfs2_glock * new)1098 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1099 					    struct gfs2_glock *new)
1100 {
1101 	struct wait_glock_queue wait;
1102 	wait_queue_head_t *wq = glock_waitqueue(name);
1103 	struct gfs2_glock *gl;
1104 
1105 	wait.name = name;
1106 	init_wait(&wait.wait);
1107 	wait.wait.func = glock_wake_function;
1108 
1109 again:
1110 	prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1111 	rcu_read_lock();
1112 	if (new) {
1113 		gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1114 			&new->gl_node, ht_parms);
1115 		if (IS_ERR(gl))
1116 			goto out;
1117 	} else {
1118 		gl = rhashtable_lookup_fast(&gl_hash_table,
1119 			name, ht_parms);
1120 	}
1121 	if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1122 		rcu_read_unlock();
1123 		schedule();
1124 		goto again;
1125 	}
1126 out:
1127 	rcu_read_unlock();
1128 	finish_wait(wq, &wait.wait);
1129 	return gl;
1130 }
1131 
1132 /**
1133  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1134  * @sdp: The GFS2 superblock
1135  * @number: the lock number
1136  * @glops: The glock_operations to use
1137  * @create: If 0, don't create the glock if it doesn't exist
1138  * @glp: the glock is returned here
1139  *
1140  * This does not lock a glock, just finds/creates structures for one.
1141  *
1142  * Returns: errno
1143  */
1144 
gfs2_glock_get(struct gfs2_sbd * sdp,u64 number,const struct gfs2_glock_operations * glops,int create,struct gfs2_glock ** glp)1145 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1146 		   const struct gfs2_glock_operations *glops, int create,
1147 		   struct gfs2_glock **glp)
1148 {
1149 	struct super_block *s = sdp->sd_vfs;
1150 	struct lm_lockname name = { .ln_number = number,
1151 				    .ln_type = glops->go_type,
1152 				    .ln_sbd = sdp };
1153 	struct gfs2_glock *gl, *tmp;
1154 	struct address_space *mapping;
1155 	int ret = 0;
1156 
1157 	gl = find_insert_glock(&name, NULL);
1158 	if (gl) {
1159 		*glp = gl;
1160 		return 0;
1161 	}
1162 	if (!create)
1163 		return -ENOENT;
1164 
1165 	if (glops->go_flags & GLOF_ASPACE) {
1166 		struct gfs2_glock_aspace *gla =
1167 			kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS);
1168 		if (!gla)
1169 			return -ENOMEM;
1170 		gl = &gla->glock;
1171 	} else {
1172 		gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS);
1173 		if (!gl)
1174 			return -ENOMEM;
1175 	}
1176 	memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1177 	gl->gl_ops = glops;
1178 
1179 	if (glops->go_flags & GLOF_LVB) {
1180 		gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1181 		if (!gl->gl_lksb.sb_lvbptr) {
1182 			gfs2_glock_dealloc(&gl->gl_rcu);
1183 			return -ENOMEM;
1184 		}
1185 	}
1186 
1187 	atomic_inc(&sdp->sd_glock_disposal);
1188 	gl->gl_node.next = NULL;
1189 	gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1190 	gl->gl_name = name;
1191 	lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1192 	gl->gl_lockref.count = 1;
1193 	gl->gl_state = LM_ST_UNLOCKED;
1194 	gl->gl_target = LM_ST_UNLOCKED;
1195 	gl->gl_demote_state = LM_ST_EXCLUSIVE;
1196 	gl->gl_dstamp = 0;
1197 	preempt_disable();
1198 	/* We use the global stats to estimate the initial per-glock stats */
1199 	gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1200 	preempt_enable();
1201 	gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1202 	gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1203 	gl->gl_tchange = jiffies;
1204 	gl->gl_object = NULL;
1205 	gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1206 	INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1207 	if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1208 		INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1209 
1210 	mapping = gfs2_glock2aspace(gl);
1211 	if (mapping) {
1212                 mapping->a_ops = &gfs2_meta_aops;
1213 		mapping->host = s->s_bdev->bd_inode;
1214 		mapping->flags = 0;
1215 		mapping_set_gfp_mask(mapping, GFP_NOFS);
1216 		mapping->private_data = NULL;
1217 		mapping->writeback_index = 0;
1218 	}
1219 
1220 	tmp = find_insert_glock(&name, gl);
1221 	if (!tmp) {
1222 		*glp = gl;
1223 		goto out;
1224 	}
1225 	if (IS_ERR(tmp)) {
1226 		ret = PTR_ERR(tmp);
1227 		goto out_free;
1228 	}
1229 	*glp = tmp;
1230 
1231 out_free:
1232 	gfs2_glock_dealloc(&gl->gl_rcu);
1233 	if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1234 		wake_up(&sdp->sd_kill_wait);
1235 
1236 out:
1237 	return ret;
1238 }
1239 
1240 /**
1241  * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1242  * @gl: the glock
1243  * @state: the state we're requesting
1244  * @flags: the modifier flags
1245  * @gh: the holder structure
1246  *
1247  */
1248 
__gfs2_holder_init(struct gfs2_glock * gl,unsigned int state,u16 flags,struct gfs2_holder * gh,unsigned long ip)1249 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1250 			struct gfs2_holder *gh, unsigned long ip)
1251 {
1252 	INIT_LIST_HEAD(&gh->gh_list);
1253 	gh->gh_gl = gfs2_glock_hold(gl);
1254 	gh->gh_ip = ip;
1255 	gh->gh_owner_pid = get_pid(task_pid(current));
1256 	gh->gh_state = state;
1257 	gh->gh_flags = flags;
1258 	gh->gh_iflags = 0;
1259 }
1260 
1261 /**
1262  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1263  * @state: the state we're requesting
1264  * @flags: the modifier flags
1265  * @gh: the holder structure
1266  *
1267  * Don't mess with the glock.
1268  *
1269  */
1270 
gfs2_holder_reinit(unsigned int state,u16 flags,struct gfs2_holder * gh)1271 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1272 {
1273 	gh->gh_state = state;
1274 	gh->gh_flags = flags;
1275 	gh->gh_iflags = 0;
1276 	gh->gh_ip = _RET_IP_;
1277 	put_pid(gh->gh_owner_pid);
1278 	gh->gh_owner_pid = get_pid(task_pid(current));
1279 }
1280 
1281 /**
1282  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1283  * @gh: the holder structure
1284  *
1285  */
1286 
gfs2_holder_uninit(struct gfs2_holder * gh)1287 void gfs2_holder_uninit(struct gfs2_holder *gh)
1288 {
1289 	put_pid(gh->gh_owner_pid);
1290 	gfs2_glock_put(gh->gh_gl);
1291 	gfs2_holder_mark_uninitialized(gh);
1292 	gh->gh_ip = 0;
1293 }
1294 
gfs2_glock_update_hold_time(struct gfs2_glock * gl,unsigned long start_time)1295 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1296 					unsigned long start_time)
1297 {
1298 	/* Have we waited longer that a second? */
1299 	if (time_after(jiffies, start_time + HZ)) {
1300 		/* Lengthen the minimum hold time. */
1301 		gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1302 				       GL_GLOCK_MAX_HOLD);
1303 	}
1304 }
1305 
1306 /**
1307  * gfs2_glock_holder_ready - holder is ready and its error code can be collected
1308  * @gh: the glock holder
1309  *
1310  * Called when a glock holder no longer needs to be waited for because it is
1311  * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has
1312  * failed (gh_error != 0).
1313  */
1314 
gfs2_glock_holder_ready(struct gfs2_holder * gh)1315 int gfs2_glock_holder_ready(struct gfs2_holder *gh)
1316 {
1317 	if (gh->gh_error || (gh->gh_flags & GL_SKIP))
1318 		return gh->gh_error;
1319 	gh->gh_error = gfs2_instantiate(gh);
1320 	if (gh->gh_error)
1321 		gfs2_glock_dq(gh);
1322 	return gh->gh_error;
1323 }
1324 
1325 /**
1326  * gfs2_glock_wait - wait on a glock acquisition
1327  * @gh: the glock holder
1328  *
1329  * Returns: 0 on success
1330  */
1331 
gfs2_glock_wait(struct gfs2_holder * gh)1332 int gfs2_glock_wait(struct gfs2_holder *gh)
1333 {
1334 	unsigned long start_time = jiffies;
1335 
1336 	might_sleep();
1337 	wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1338 	gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1339 	return gfs2_glock_holder_ready(gh);
1340 }
1341 
glocks_pending(unsigned int num_gh,struct gfs2_holder * ghs)1342 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1343 {
1344 	int i;
1345 
1346 	for (i = 0; i < num_gh; i++)
1347 		if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1348 			return 1;
1349 	return 0;
1350 }
1351 
1352 /**
1353  * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1354  * @num_gh: the number of holders in the array
1355  * @ghs: the glock holder array
1356  *
1357  * Returns: 0 on success, meaning all glocks have been granted and are held.
1358  *          -ESTALE if the request timed out, meaning all glocks were released,
1359  *          and the caller should retry the operation.
1360  */
1361 
gfs2_glock_async_wait(unsigned int num_gh,struct gfs2_holder * ghs)1362 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1363 {
1364 	struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1365 	int i, ret = 0, timeout = 0;
1366 	unsigned long start_time = jiffies;
1367 
1368 	might_sleep();
1369 	/*
1370 	 * Total up the (minimum hold time * 2) of all glocks and use that to
1371 	 * determine the max amount of time we should wait.
1372 	 */
1373 	for (i = 0; i < num_gh; i++)
1374 		timeout += ghs[i].gh_gl->gl_hold_time << 1;
1375 
1376 	if (!wait_event_timeout(sdp->sd_async_glock_wait,
1377 				!glocks_pending(num_gh, ghs), timeout)) {
1378 		ret = -ESTALE; /* request timed out. */
1379 		goto out;
1380 	}
1381 
1382 	for (i = 0; i < num_gh; i++) {
1383 		struct gfs2_holder *gh = &ghs[i];
1384 		int ret2;
1385 
1386 		if (test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1387 			gfs2_glock_update_hold_time(gh->gh_gl,
1388 						    start_time);
1389 		}
1390 		ret2 = gfs2_glock_holder_ready(gh);
1391 		if (!ret)
1392 			ret = ret2;
1393 	}
1394 
1395 out:
1396 	if (ret) {
1397 		for (i = 0; i < num_gh; i++) {
1398 			struct gfs2_holder *gh = &ghs[i];
1399 
1400 			gfs2_glock_dq(gh);
1401 		}
1402 	}
1403 	return ret;
1404 }
1405 
1406 /**
1407  * handle_callback - process a demote request
1408  * @gl: the glock
1409  * @state: the state the caller wants us to change to
1410  * @delay: zero to demote immediately; otherwise pending demote
1411  * @remote: true if this came from a different cluster node
1412  *
1413  * There are only two requests that we are going to see in actual
1414  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1415  */
1416 
handle_callback(struct gfs2_glock * gl,unsigned int state,unsigned long delay,bool remote)1417 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1418 			    unsigned long delay, bool remote)
1419 {
1420 	if (delay)
1421 		set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1422 	else
1423 		gfs2_set_demote(gl);
1424 	if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1425 		gl->gl_demote_state = state;
1426 		gl->gl_demote_time = jiffies;
1427 	} else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1428 			gl->gl_demote_state != state) {
1429 		gl->gl_demote_state = LM_ST_UNLOCKED;
1430 	}
1431 	if (gl->gl_ops->go_callback)
1432 		gl->gl_ops->go_callback(gl, remote);
1433 	trace_gfs2_demote_rq(gl, remote);
1434 }
1435 
gfs2_print_dbg(struct seq_file * seq,const char * fmt,...)1436 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1437 {
1438 	struct va_format vaf;
1439 	va_list args;
1440 
1441 	va_start(args, fmt);
1442 
1443 	if (seq) {
1444 		seq_vprintf(seq, fmt, args);
1445 	} else {
1446 		vaf.fmt = fmt;
1447 		vaf.va = &args;
1448 
1449 		pr_err("%pV", &vaf);
1450 	}
1451 
1452 	va_end(args);
1453 }
1454 
pid_is_meaningful(const struct gfs2_holder * gh)1455 static inline bool pid_is_meaningful(const struct gfs2_holder *gh)
1456 {
1457         if (!(gh->gh_flags & GL_NOPID))
1458                 return true;
1459         if (gh->gh_state == LM_ST_UNLOCKED)
1460                 return true;
1461         return false;
1462 }
1463 
1464 /**
1465  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1466  * @gh: the holder structure to add
1467  *
1468  * Eventually we should move the recursive locking trap to a
1469  * debugging option or something like that. This is the fast
1470  * path and needs to have the minimum number of distractions.
1471  *
1472  */
1473 
add_to_queue(struct gfs2_holder * gh)1474 static inline void add_to_queue(struct gfs2_holder *gh)
1475 __releases(&gl->gl_lockref.lock)
1476 __acquires(&gl->gl_lockref.lock)
1477 {
1478 	struct gfs2_glock *gl = gh->gh_gl;
1479 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1480 	struct list_head *insert_pt = NULL;
1481 	struct gfs2_holder *gh2;
1482 	int try_futile = 0;
1483 
1484 	GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1485 	if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1486 		GLOCK_BUG_ON(gl, true);
1487 
1488 	if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1489 		if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1490 			struct gfs2_holder *current_gh;
1491 
1492 			current_gh = find_first_holder(gl);
1493 			try_futile = !may_grant(gl, current_gh, gh);
1494 		}
1495 		if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1496 			goto fail;
1497 	}
1498 
1499 	list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1500 		if (likely(gh2->gh_owner_pid != gh->gh_owner_pid))
1501 			continue;
1502 		if (gh->gh_gl->gl_ops->go_type == LM_TYPE_FLOCK)
1503 			continue;
1504 		if (!pid_is_meaningful(gh2))
1505 			continue;
1506 		goto trap_recursive;
1507 	}
1508 	list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1509 		if (try_futile &&
1510 		    !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1511 fail:
1512 			gh->gh_error = GLR_TRYFAILED;
1513 			gfs2_holder_wake(gh);
1514 			return;
1515 		}
1516 		if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1517 			continue;
1518 	}
1519 	trace_gfs2_glock_queue(gh, 1);
1520 	gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1521 	gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1522 	if (likely(insert_pt == NULL)) {
1523 		list_add_tail(&gh->gh_list, &gl->gl_holders);
1524 		return;
1525 	}
1526 	list_add_tail(&gh->gh_list, insert_pt);
1527 	gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1528 	spin_unlock(&gl->gl_lockref.lock);
1529 	if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1530 		sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1531 	spin_lock(&gl->gl_lockref.lock);
1532 	return;
1533 
1534 trap_recursive:
1535 	fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1536 	fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1537 	fs_err(sdp, "lock type: %d req lock state : %d\n",
1538 	       gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1539 	fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1540 	fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1541 	fs_err(sdp, "lock type: %d req lock state : %d\n",
1542 	       gh->gh_gl->gl_name.ln_type, gh->gh_state);
1543 	gfs2_dump_glock(NULL, gl, true);
1544 	BUG();
1545 }
1546 
1547 /**
1548  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1549  * @gh: the holder structure
1550  *
1551  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1552  *
1553  * Returns: 0, GLR_TRYFAILED, or errno on failure
1554  */
1555 
gfs2_glock_nq(struct gfs2_holder * gh)1556 int gfs2_glock_nq(struct gfs2_holder *gh)
1557 {
1558 	struct gfs2_glock *gl = gh->gh_gl;
1559 	int error = 0;
1560 
1561 	if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1562 		return -EIO;
1563 
1564 	if (test_bit(GLF_LRU, &gl->gl_flags))
1565 		gfs2_glock_remove_from_lru(gl);
1566 
1567 	gh->gh_error = 0;
1568 	spin_lock(&gl->gl_lockref.lock);
1569 	add_to_queue(gh);
1570 	if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1571 		     test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1572 		set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1573 		gl->gl_lockref.count++;
1574 		__gfs2_glock_queue_work(gl, 0);
1575 	}
1576 	run_queue(gl, 1);
1577 	spin_unlock(&gl->gl_lockref.lock);
1578 
1579 	if (!(gh->gh_flags & GL_ASYNC))
1580 		error = gfs2_glock_wait(gh);
1581 
1582 	return error;
1583 }
1584 
1585 /**
1586  * gfs2_glock_poll - poll to see if an async request has been completed
1587  * @gh: the holder
1588  *
1589  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1590  */
1591 
gfs2_glock_poll(struct gfs2_holder * gh)1592 int gfs2_glock_poll(struct gfs2_holder *gh)
1593 {
1594 	return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1595 }
1596 
needs_demote(struct gfs2_glock * gl)1597 static inline bool needs_demote(struct gfs2_glock *gl)
1598 {
1599 	return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1600 		test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1601 }
1602 
__gfs2_glock_dq(struct gfs2_holder * gh)1603 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1604 {
1605 	struct gfs2_glock *gl = gh->gh_gl;
1606 	unsigned delay = 0;
1607 	int fast_path = 0;
1608 
1609 	/*
1610 	 * This holder should not be cached, so mark it for demote.
1611 	 * Note: this should be done before the check for needs_demote
1612 	 * below.
1613 	 */
1614 	if (gh->gh_flags & GL_NOCACHE)
1615 		handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1616 
1617 	list_del_init(&gh->gh_list);
1618 	clear_bit(HIF_HOLDER, &gh->gh_iflags);
1619 	trace_gfs2_glock_queue(gh, 0);
1620 
1621 	/*
1622 	 * If there hasn't been a demote request we are done.
1623 	 * (Let the remaining holders, if any, keep holding it.)
1624 	 */
1625 	if (!needs_demote(gl)) {
1626 		if (list_empty(&gl->gl_holders))
1627 			fast_path = 1;
1628 	}
1629 
1630 	if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1631 		gfs2_glock_add_to_lru(gl);
1632 
1633 	if (unlikely(!fast_path)) {
1634 		gl->gl_lockref.count++;
1635 		if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1636 		    !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1637 		    gl->gl_name.ln_type == LM_TYPE_INODE)
1638 			delay = gl->gl_hold_time;
1639 		__gfs2_glock_queue_work(gl, delay);
1640 	}
1641 }
1642 
1643 /**
1644  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1645  * @gh: the glock holder
1646  *
1647  */
gfs2_glock_dq(struct gfs2_holder * gh)1648 void gfs2_glock_dq(struct gfs2_holder *gh)
1649 {
1650 	struct gfs2_glock *gl = gh->gh_gl;
1651 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1652 
1653 	spin_lock(&gl->gl_lockref.lock);
1654 	if (!gfs2_holder_queued(gh)) {
1655 		/*
1656 		 * May have already been dequeued because the locking request
1657 		 * was GL_ASYNC and it has failed in the meantime.
1658 		 */
1659 		goto out;
1660 	}
1661 
1662 	if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1663 	    !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1664 		spin_unlock(&gl->gl_lockref.lock);
1665 		gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1666 		wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1667 		spin_lock(&gl->gl_lockref.lock);
1668 	}
1669 
1670 	/*
1671 	 * If we're in the process of file system withdraw, we cannot just
1672 	 * dequeue any glocks until our journal is recovered, lest we introduce
1673 	 * file system corruption. We need two exceptions to this rule: We need
1674 	 * to allow unlocking of nondisk glocks and the glock for our own
1675 	 * journal that needs recovery.
1676 	 */
1677 	if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1678 	    glock_blocked_by_withdraw(gl) &&
1679 	    gh->gh_gl != sdp->sd_jinode_gl) {
1680 		sdp->sd_glock_dqs_held++;
1681 		spin_unlock(&gl->gl_lockref.lock);
1682 		might_sleep();
1683 		wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1684 			    TASK_UNINTERRUPTIBLE);
1685 		spin_lock(&gl->gl_lockref.lock);
1686 	}
1687 
1688 	__gfs2_glock_dq(gh);
1689 out:
1690 	spin_unlock(&gl->gl_lockref.lock);
1691 }
1692 
gfs2_glock_dq_wait(struct gfs2_holder * gh)1693 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1694 {
1695 	struct gfs2_glock *gl = gh->gh_gl;
1696 	gfs2_glock_dq(gh);
1697 	might_sleep();
1698 	wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1699 }
1700 
1701 /**
1702  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1703  * @gh: the holder structure
1704  *
1705  */
1706 
gfs2_glock_dq_uninit(struct gfs2_holder * gh)1707 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1708 {
1709 	gfs2_glock_dq(gh);
1710 	gfs2_holder_uninit(gh);
1711 }
1712 
1713 /**
1714  * gfs2_glock_nq_num - acquire a glock based on lock number
1715  * @sdp: the filesystem
1716  * @number: the lock number
1717  * @glops: the glock operations for the type of glock
1718  * @state: the state to acquire the glock in
1719  * @flags: modifier flags for the acquisition
1720  * @gh: the struct gfs2_holder
1721  *
1722  * Returns: errno
1723  */
1724 
gfs2_glock_nq_num(struct gfs2_sbd * sdp,u64 number,const struct gfs2_glock_operations * glops,unsigned int state,u16 flags,struct gfs2_holder * gh)1725 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1726 		      const struct gfs2_glock_operations *glops,
1727 		      unsigned int state, u16 flags, struct gfs2_holder *gh)
1728 {
1729 	struct gfs2_glock *gl;
1730 	int error;
1731 
1732 	error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1733 	if (!error) {
1734 		error = gfs2_glock_nq_init(gl, state, flags, gh);
1735 		gfs2_glock_put(gl);
1736 	}
1737 
1738 	return error;
1739 }
1740 
1741 /**
1742  * glock_compare - Compare two struct gfs2_glock structures for sorting
1743  * @arg_a: the first structure
1744  * @arg_b: the second structure
1745  *
1746  */
1747 
glock_compare(const void * arg_a,const void * arg_b)1748 static int glock_compare(const void *arg_a, const void *arg_b)
1749 {
1750 	const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1751 	const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1752 	const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1753 	const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1754 
1755 	if (a->ln_number > b->ln_number)
1756 		return 1;
1757 	if (a->ln_number < b->ln_number)
1758 		return -1;
1759 	BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1760 	return 0;
1761 }
1762 
1763 /**
1764  * nq_m_sync - synchronously acquire more than one glock in deadlock free order
1765  * @num_gh: the number of structures
1766  * @ghs: an array of struct gfs2_holder structures
1767  * @p: placeholder for the holder structure to pass back
1768  *
1769  * Returns: 0 on success (all glocks acquired),
1770  *          errno on failure (no glocks acquired)
1771  */
1772 
nq_m_sync(unsigned int num_gh,struct gfs2_holder * ghs,struct gfs2_holder ** p)1773 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1774 		     struct gfs2_holder **p)
1775 {
1776 	unsigned int x;
1777 	int error = 0;
1778 
1779 	for (x = 0; x < num_gh; x++)
1780 		p[x] = &ghs[x];
1781 
1782 	sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1783 
1784 	for (x = 0; x < num_gh; x++) {
1785 		error = gfs2_glock_nq(p[x]);
1786 		if (error) {
1787 			while (x--)
1788 				gfs2_glock_dq(p[x]);
1789 			break;
1790 		}
1791 	}
1792 
1793 	return error;
1794 }
1795 
1796 /**
1797  * gfs2_glock_nq_m - acquire multiple glocks
1798  * @num_gh: the number of structures
1799  * @ghs: an array of struct gfs2_holder structures
1800  *
1801  * Returns: 0 on success (all glocks acquired),
1802  *          errno on failure (no glocks acquired)
1803  */
1804 
gfs2_glock_nq_m(unsigned int num_gh,struct gfs2_holder * ghs)1805 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1806 {
1807 	struct gfs2_holder *tmp[4];
1808 	struct gfs2_holder **pph = tmp;
1809 	int error = 0;
1810 
1811 	switch(num_gh) {
1812 	case 0:
1813 		return 0;
1814 	case 1:
1815 		return gfs2_glock_nq(ghs);
1816 	default:
1817 		if (num_gh <= 4)
1818 			break;
1819 		pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1820 				    GFP_NOFS);
1821 		if (!pph)
1822 			return -ENOMEM;
1823 	}
1824 
1825 	error = nq_m_sync(num_gh, ghs, pph);
1826 
1827 	if (pph != tmp)
1828 		kfree(pph);
1829 
1830 	return error;
1831 }
1832 
1833 /**
1834  * gfs2_glock_dq_m - release multiple glocks
1835  * @num_gh: the number of structures
1836  * @ghs: an array of struct gfs2_holder structures
1837  *
1838  */
1839 
gfs2_glock_dq_m(unsigned int num_gh,struct gfs2_holder * ghs)1840 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1841 {
1842 	while (num_gh--)
1843 		gfs2_glock_dq(&ghs[num_gh]);
1844 }
1845 
gfs2_glock_cb(struct gfs2_glock * gl,unsigned int state)1846 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1847 {
1848 	unsigned long delay = 0;
1849 	unsigned long holdtime;
1850 	unsigned long now = jiffies;
1851 
1852 	gfs2_glock_hold(gl);
1853 	spin_lock(&gl->gl_lockref.lock);
1854 	holdtime = gl->gl_tchange + gl->gl_hold_time;
1855 	if (!list_empty(&gl->gl_holders) &&
1856 	    gl->gl_name.ln_type == LM_TYPE_INODE) {
1857 		if (time_before(now, holdtime))
1858 			delay = holdtime - now;
1859 		if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1860 			delay = gl->gl_hold_time;
1861 	}
1862 	handle_callback(gl, state, delay, true);
1863 	__gfs2_glock_queue_work(gl, delay);
1864 	spin_unlock(&gl->gl_lockref.lock);
1865 }
1866 
1867 /**
1868  * gfs2_should_freeze - Figure out if glock should be frozen
1869  * @gl: The glock in question
1870  *
1871  * Glocks are not frozen if (a) the result of the dlm operation is
1872  * an error, (b) the locking operation was an unlock operation or
1873  * (c) if there is a "noexp" flagged request anywhere in the queue
1874  *
1875  * Returns: 1 if freezing should occur, 0 otherwise
1876  */
1877 
gfs2_should_freeze(const struct gfs2_glock * gl)1878 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1879 {
1880 	const struct gfs2_holder *gh;
1881 
1882 	if (gl->gl_reply & ~LM_OUT_ST_MASK)
1883 		return 0;
1884 	if (gl->gl_target == LM_ST_UNLOCKED)
1885 		return 0;
1886 
1887 	list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1888 		if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1889 			continue;
1890 		if (LM_FLAG_NOEXP & gh->gh_flags)
1891 			return 0;
1892 	}
1893 
1894 	return 1;
1895 }
1896 
1897 /**
1898  * gfs2_glock_complete - Callback used by locking
1899  * @gl: Pointer to the glock
1900  * @ret: The return value from the dlm
1901  *
1902  * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1903  * to use a bitfield shared with other glock state fields.
1904  */
1905 
gfs2_glock_complete(struct gfs2_glock * gl,int ret)1906 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1907 {
1908 	struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1909 
1910 	spin_lock(&gl->gl_lockref.lock);
1911 	gl->gl_reply = ret;
1912 
1913 	if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1914 		if (gfs2_should_freeze(gl)) {
1915 			set_bit(GLF_FROZEN, &gl->gl_flags);
1916 			spin_unlock(&gl->gl_lockref.lock);
1917 			return;
1918 		}
1919 	}
1920 
1921 	gl->gl_lockref.count++;
1922 	set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1923 	__gfs2_glock_queue_work(gl, 0);
1924 	spin_unlock(&gl->gl_lockref.lock);
1925 }
1926 
glock_cmp(void * priv,const struct list_head * a,const struct list_head * b)1927 static int glock_cmp(void *priv, const struct list_head *a,
1928 		     const struct list_head *b)
1929 {
1930 	struct gfs2_glock *gla, *glb;
1931 
1932 	gla = list_entry(a, struct gfs2_glock, gl_lru);
1933 	glb = list_entry(b, struct gfs2_glock, gl_lru);
1934 
1935 	if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1936 		return 1;
1937 	if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1938 		return -1;
1939 
1940 	return 0;
1941 }
1942 
1943 /**
1944  * gfs2_dispose_glock_lru - Demote a list of glocks
1945  * @list: The list to dispose of
1946  *
1947  * Disposing of glocks may involve disk accesses, so that here we sort
1948  * the glocks by number (i.e. disk location of the inodes) so that if
1949  * there are any such accesses, they'll be sent in order (mostly).
1950  *
1951  * Must be called under the lru_lock, but may drop and retake this
1952  * lock. While the lru_lock is dropped, entries may vanish from the
1953  * list, but no new entries will appear on the list (since it is
1954  * private)
1955  */
1956 
gfs2_dispose_glock_lru(struct list_head * list)1957 static void gfs2_dispose_glock_lru(struct list_head *list)
1958 __releases(&lru_lock)
1959 __acquires(&lru_lock)
1960 {
1961 	struct gfs2_glock *gl;
1962 
1963 	list_sort(NULL, list, glock_cmp);
1964 
1965 	while(!list_empty(list)) {
1966 		gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1967 		list_del_init(&gl->gl_lru);
1968 		clear_bit(GLF_LRU, &gl->gl_flags);
1969 		if (!spin_trylock(&gl->gl_lockref.lock)) {
1970 add_back_to_lru:
1971 			list_add(&gl->gl_lru, &lru_list);
1972 			set_bit(GLF_LRU, &gl->gl_flags);
1973 			atomic_inc(&lru_count);
1974 			continue;
1975 		}
1976 		if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1977 			spin_unlock(&gl->gl_lockref.lock);
1978 			goto add_back_to_lru;
1979 		}
1980 		gl->gl_lockref.count++;
1981 		if (demote_ok(gl))
1982 			handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1983 		WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1984 		__gfs2_glock_queue_work(gl, 0);
1985 		spin_unlock(&gl->gl_lockref.lock);
1986 		cond_resched_lock(&lru_lock);
1987 	}
1988 }
1989 
1990 /**
1991  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1992  * @nr: The number of entries to scan
1993  *
1994  * This function selects the entries on the LRU which are able to
1995  * be demoted, and then kicks off the process by calling
1996  * gfs2_dispose_glock_lru() above.
1997  */
1998 
gfs2_scan_glock_lru(int nr)1999 static long gfs2_scan_glock_lru(int nr)
2000 {
2001 	struct gfs2_glock *gl, *next;
2002 	LIST_HEAD(dispose);
2003 	long freed = 0;
2004 
2005 	spin_lock(&lru_lock);
2006 	list_for_each_entry_safe(gl, next, &lru_list, gl_lru) {
2007 		if (nr-- <= 0)
2008 			break;
2009 		/* Test for being demotable */
2010 		if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2011 			if (!spin_trylock(&gl->gl_lockref.lock))
2012 				continue;
2013 			if (gl->gl_lockref.count <= 1 &&
2014 			    (gl->gl_state == LM_ST_UNLOCKED ||
2015 			     demote_ok(gl))) {
2016 				list_move(&gl->gl_lru, &dispose);
2017 				atomic_dec(&lru_count);
2018 				freed++;
2019 			}
2020 			spin_unlock(&gl->gl_lockref.lock);
2021 		}
2022 	}
2023 	if (!list_empty(&dispose))
2024 		gfs2_dispose_glock_lru(&dispose);
2025 	spin_unlock(&lru_lock);
2026 
2027 	return freed;
2028 }
2029 
gfs2_glock_shrink_scan(struct shrinker * shrink,struct shrink_control * sc)2030 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2031 					    struct shrink_control *sc)
2032 {
2033 	if (!(sc->gfp_mask & __GFP_FS))
2034 		return SHRINK_STOP;
2035 	return gfs2_scan_glock_lru(sc->nr_to_scan);
2036 }
2037 
gfs2_glock_shrink_count(struct shrinker * shrink,struct shrink_control * sc)2038 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2039 					     struct shrink_control *sc)
2040 {
2041 	return vfs_pressure_ratio(atomic_read(&lru_count));
2042 }
2043 
2044 static struct shrinker glock_shrinker = {
2045 	.seeks = DEFAULT_SEEKS,
2046 	.count_objects = gfs2_glock_shrink_count,
2047 	.scan_objects = gfs2_glock_shrink_scan,
2048 };
2049 
2050 /**
2051  * glock_hash_walk - Call a function for glock in a hash bucket
2052  * @examiner: the function
2053  * @sdp: the filesystem
2054  *
2055  * Note that the function can be called multiple times on the same
2056  * object.  So the user must ensure that the function can cope with
2057  * that.
2058  */
2059 
glock_hash_walk(glock_examiner examiner,const struct gfs2_sbd * sdp)2060 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2061 {
2062 	struct gfs2_glock *gl;
2063 	struct rhashtable_iter iter;
2064 
2065 	rhashtable_walk_enter(&gl_hash_table, &iter);
2066 
2067 	do {
2068 		rhashtable_walk_start(&iter);
2069 
2070 		while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2071 			if (gl->gl_name.ln_sbd == sdp)
2072 				examiner(gl);
2073 		}
2074 
2075 		rhashtable_walk_stop(&iter);
2076 	} while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2077 
2078 	rhashtable_walk_exit(&iter);
2079 }
2080 
gfs2_cancel_delete_work(struct gfs2_glock * gl)2081 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2082 {
2083 	clear_bit(GLF_TRY_TO_EVICT, &gl->gl_flags);
2084 	clear_bit(GLF_VERIFY_EVICT, &gl->gl_flags);
2085 	if (cancel_delayed_work(&gl->gl_delete))
2086 		gfs2_glock_put(gl);
2087 }
2088 
flush_delete_work(struct gfs2_glock * gl)2089 static void flush_delete_work(struct gfs2_glock *gl)
2090 {
2091 	if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2092 		struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2093 
2094 		if (cancel_delayed_work(&gl->gl_delete)) {
2095 			queue_delayed_work(sdp->sd_delete_wq,
2096 					   &gl->gl_delete, 0);
2097 		}
2098 	}
2099 }
2100 
gfs2_flush_delete_work(struct gfs2_sbd * sdp)2101 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2102 {
2103 	glock_hash_walk(flush_delete_work, sdp);
2104 	flush_workqueue(sdp->sd_delete_wq);
2105 }
2106 
2107 /**
2108  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2109  * @gl: The glock to thaw
2110  *
2111  */
2112 
thaw_glock(struct gfs2_glock * gl)2113 static void thaw_glock(struct gfs2_glock *gl)
2114 {
2115 	if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2116 		return;
2117 	if (!lockref_get_not_dead(&gl->gl_lockref))
2118 		return;
2119 	set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2120 	gfs2_glock_queue_work(gl, 0);
2121 }
2122 
2123 /**
2124  * clear_glock - look at a glock and see if we can free it from glock cache
2125  * @gl: the glock to look at
2126  *
2127  */
2128 
clear_glock(struct gfs2_glock * gl)2129 static void clear_glock(struct gfs2_glock *gl)
2130 {
2131 	gfs2_glock_remove_from_lru(gl);
2132 
2133 	spin_lock(&gl->gl_lockref.lock);
2134 	if (!__lockref_is_dead(&gl->gl_lockref)) {
2135 		gl->gl_lockref.count++;
2136 		if (gl->gl_state != LM_ST_UNLOCKED)
2137 			handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2138 		__gfs2_glock_queue_work(gl, 0);
2139 	}
2140 	spin_unlock(&gl->gl_lockref.lock);
2141 }
2142 
2143 /**
2144  * gfs2_glock_thaw - Thaw any frozen glocks
2145  * @sdp: The super block
2146  *
2147  */
2148 
gfs2_glock_thaw(struct gfs2_sbd * sdp)2149 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2150 {
2151 	glock_hash_walk(thaw_glock, sdp);
2152 }
2153 
dump_glock(struct seq_file * seq,struct gfs2_glock * gl,bool fsid)2154 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2155 {
2156 	spin_lock(&gl->gl_lockref.lock);
2157 	gfs2_dump_glock(seq, gl, fsid);
2158 	spin_unlock(&gl->gl_lockref.lock);
2159 }
2160 
dump_glock_func(struct gfs2_glock * gl)2161 static void dump_glock_func(struct gfs2_glock *gl)
2162 {
2163 	dump_glock(NULL, gl, true);
2164 }
2165 
withdraw_dq(struct gfs2_glock * gl)2166 static void withdraw_dq(struct gfs2_glock *gl)
2167 {
2168 	spin_lock(&gl->gl_lockref.lock);
2169 	if (!__lockref_is_dead(&gl->gl_lockref) &&
2170 	    glock_blocked_by_withdraw(gl))
2171 		do_error(gl, LM_OUT_ERROR); /* remove pending waiters */
2172 	spin_unlock(&gl->gl_lockref.lock);
2173 }
2174 
gfs2_gl_dq_holders(struct gfs2_sbd * sdp)2175 void gfs2_gl_dq_holders(struct gfs2_sbd *sdp)
2176 {
2177 	glock_hash_walk(withdraw_dq, sdp);
2178 }
2179 
2180 /**
2181  * gfs2_gl_hash_clear - Empty out the glock hash table
2182  * @sdp: the filesystem
2183  *
2184  * Called when unmounting the filesystem.
2185  */
2186 
gfs2_gl_hash_clear(struct gfs2_sbd * sdp)2187 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2188 {
2189 	set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2190 	flush_workqueue(glock_workqueue);
2191 	glock_hash_walk(clear_glock, sdp);
2192 	flush_workqueue(glock_workqueue);
2193 	wait_event_timeout(sdp->sd_kill_wait,
2194 			   atomic_read(&sdp->sd_glock_disposal) == 0,
2195 			   HZ * 600);
2196 	glock_hash_walk(dump_glock_func, sdp);
2197 }
2198 
state2str(unsigned state)2199 static const char *state2str(unsigned state)
2200 {
2201 	switch(state) {
2202 	case LM_ST_UNLOCKED:
2203 		return "UN";
2204 	case LM_ST_SHARED:
2205 		return "SH";
2206 	case LM_ST_DEFERRED:
2207 		return "DF";
2208 	case LM_ST_EXCLUSIVE:
2209 		return "EX";
2210 	}
2211 	return "??";
2212 }
2213 
hflags2str(char * buf,u16 flags,unsigned long iflags)2214 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2215 {
2216 	char *p = buf;
2217 	if (flags & LM_FLAG_TRY)
2218 		*p++ = 't';
2219 	if (flags & LM_FLAG_TRY_1CB)
2220 		*p++ = 'T';
2221 	if (flags & LM_FLAG_NOEXP)
2222 		*p++ = 'e';
2223 	if (flags & LM_FLAG_ANY)
2224 		*p++ = 'A';
2225 	if (flags & LM_FLAG_NODE_SCOPE)
2226 		*p++ = 'n';
2227 	if (flags & GL_ASYNC)
2228 		*p++ = 'a';
2229 	if (flags & GL_EXACT)
2230 		*p++ = 'E';
2231 	if (flags & GL_NOCACHE)
2232 		*p++ = 'c';
2233 	if (test_bit(HIF_HOLDER, &iflags))
2234 		*p++ = 'H';
2235 	if (test_bit(HIF_WAIT, &iflags))
2236 		*p++ = 'W';
2237 	if (flags & GL_SKIP)
2238 		*p++ = 's';
2239 	*p = 0;
2240 	return buf;
2241 }
2242 
2243 /**
2244  * dump_holder - print information about a glock holder
2245  * @seq: the seq_file struct
2246  * @gh: the glock holder
2247  * @fs_id_buf: pointer to file system id (if requested)
2248  *
2249  */
2250 
dump_holder(struct seq_file * seq,const struct gfs2_holder * gh,const char * fs_id_buf)2251 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2252 			const char *fs_id_buf)
2253 {
2254 	const char *comm = "(none)";
2255 	pid_t owner_pid = 0;
2256 	char flags_buf[32];
2257 
2258 	rcu_read_lock();
2259 	if (pid_is_meaningful(gh)) {
2260 		struct task_struct *gh_owner;
2261 
2262 		comm = "(ended)";
2263 		owner_pid = pid_nr(gh->gh_owner_pid);
2264 		gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2265 		if (gh_owner)
2266 			comm = gh_owner->comm;
2267 	}
2268 	gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2269 		       fs_id_buf, state2str(gh->gh_state),
2270 		       hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2271 		       gh->gh_error, (long)owner_pid, comm, (void *)gh->gh_ip);
2272 	rcu_read_unlock();
2273 }
2274 
gflags2str(char * buf,const struct gfs2_glock * gl)2275 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2276 {
2277 	const unsigned long *gflags = &gl->gl_flags;
2278 	char *p = buf;
2279 
2280 	if (test_bit(GLF_LOCK, gflags))
2281 		*p++ = 'l';
2282 	if (test_bit(GLF_DEMOTE, gflags))
2283 		*p++ = 'D';
2284 	if (test_bit(GLF_PENDING_DEMOTE, gflags))
2285 		*p++ = 'd';
2286 	if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2287 		*p++ = 'p';
2288 	if (test_bit(GLF_DIRTY, gflags))
2289 		*p++ = 'y';
2290 	if (test_bit(GLF_LFLUSH, gflags))
2291 		*p++ = 'f';
2292 	if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2293 		*p++ = 'i';
2294 	if (test_bit(GLF_REPLY_PENDING, gflags))
2295 		*p++ = 'r';
2296 	if (test_bit(GLF_INITIAL, gflags))
2297 		*p++ = 'I';
2298 	if (test_bit(GLF_FROZEN, gflags))
2299 		*p++ = 'F';
2300 	if (!list_empty(&gl->gl_holders))
2301 		*p++ = 'q';
2302 	if (test_bit(GLF_LRU, gflags))
2303 		*p++ = 'L';
2304 	if (gl->gl_object)
2305 		*p++ = 'o';
2306 	if (test_bit(GLF_BLOCKING, gflags))
2307 		*p++ = 'b';
2308 	if (test_bit(GLF_FREEING, gflags))
2309 		*p++ = 'x';
2310 	if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2311 		*p++ = 'n';
2312 	if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2313 		*p++ = 'N';
2314 	if (test_bit(GLF_TRY_TO_EVICT, gflags))
2315 		*p++ = 'e';
2316 	if (test_bit(GLF_VERIFY_EVICT, gflags))
2317 		*p++ = 'E';
2318 	*p = 0;
2319 	return buf;
2320 }
2321 
2322 /**
2323  * gfs2_dump_glock - print information about a glock
2324  * @seq: The seq_file struct
2325  * @gl: the glock
2326  * @fsid: If true, also dump the file system id
2327  *
2328  * The file format is as follows:
2329  * One line per object, capital letters are used to indicate objects
2330  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2331  * other objects are indented by a single space and follow the glock to
2332  * which they are related. Fields are indicated by lower case letters
2333  * followed by a colon and the field value, except for strings which are in
2334  * [] so that its possible to see if they are composed of spaces for
2335  * example. The field's are n = number (id of the object), f = flags,
2336  * t = type, s = state, r = refcount, e = error, p = pid.
2337  *
2338  */
2339 
gfs2_dump_glock(struct seq_file * seq,struct gfs2_glock * gl,bool fsid)2340 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2341 {
2342 	const struct gfs2_glock_operations *glops = gl->gl_ops;
2343 	unsigned long long dtime;
2344 	const struct gfs2_holder *gh;
2345 	char gflags_buf[32];
2346 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2347 	char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2348 	unsigned long nrpages = 0;
2349 
2350 	if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2351 		struct address_space *mapping = gfs2_glock2aspace(gl);
2352 
2353 		nrpages = mapping->nrpages;
2354 	}
2355 	memset(fs_id_buf, 0, sizeof(fs_id_buf));
2356 	if (fsid && sdp) /* safety precaution */
2357 		sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2358 	dtime = jiffies - gl->gl_demote_time;
2359 	dtime *= 1000000/HZ; /* demote time in uSec */
2360 	if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2361 		dtime = 0;
2362 	gfs2_print_dbg(seq, "%sG:  s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2363 		       "v:%d r:%d m:%ld p:%lu\n",
2364 		       fs_id_buf, state2str(gl->gl_state),
2365 		       gl->gl_name.ln_type,
2366 		       (unsigned long long)gl->gl_name.ln_number,
2367 		       gflags2str(gflags_buf, gl),
2368 		       state2str(gl->gl_target),
2369 		       state2str(gl->gl_demote_state), dtime,
2370 		       atomic_read(&gl->gl_ail_count),
2371 		       atomic_read(&gl->gl_revokes),
2372 		       (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2373 
2374 	list_for_each_entry(gh, &gl->gl_holders, gh_list)
2375 		dump_holder(seq, gh, fs_id_buf);
2376 
2377 	if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2378 		glops->go_dump(seq, gl, fs_id_buf);
2379 }
2380 
gfs2_glstats_seq_show(struct seq_file * seq,void * iter_ptr)2381 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2382 {
2383 	struct gfs2_glock *gl = iter_ptr;
2384 
2385 	seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2386 		   gl->gl_name.ln_type,
2387 		   (unsigned long long)gl->gl_name.ln_number,
2388 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2389 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2390 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2391 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2392 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2393 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2394 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2395 		   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2396 	return 0;
2397 }
2398 
2399 static const char *gfs2_gltype[] = {
2400 	"type",
2401 	"reserved",
2402 	"nondisk",
2403 	"inode",
2404 	"rgrp",
2405 	"meta",
2406 	"iopen",
2407 	"flock",
2408 	"plock",
2409 	"quota",
2410 	"journal",
2411 };
2412 
2413 static const char *gfs2_stype[] = {
2414 	[GFS2_LKS_SRTT]		= "srtt",
2415 	[GFS2_LKS_SRTTVAR]	= "srttvar",
2416 	[GFS2_LKS_SRTTB]	= "srttb",
2417 	[GFS2_LKS_SRTTVARB]	= "srttvarb",
2418 	[GFS2_LKS_SIRT]		= "sirt",
2419 	[GFS2_LKS_SIRTVAR]	= "sirtvar",
2420 	[GFS2_LKS_DCOUNT]	= "dlm",
2421 	[GFS2_LKS_QCOUNT]	= "queue",
2422 };
2423 
2424 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2425 
gfs2_sbstats_seq_show(struct seq_file * seq,void * iter_ptr)2426 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2427 {
2428 	struct gfs2_sbd *sdp = seq->private;
2429 	loff_t pos = *(loff_t *)iter_ptr;
2430 	unsigned index = pos >> 3;
2431 	unsigned subindex = pos & 0x07;
2432 	int i;
2433 
2434 	if (index == 0 && subindex != 0)
2435 		return 0;
2436 
2437 	seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2438 		   (index == 0) ? "cpu": gfs2_stype[subindex]);
2439 
2440 	for_each_possible_cpu(i) {
2441                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2442 
2443 		if (index == 0)
2444 			seq_printf(seq, " %15u", i);
2445 		else
2446 			seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2447 				   lkstats[index - 1].stats[subindex]);
2448 	}
2449 	seq_putc(seq, '\n');
2450 	return 0;
2451 }
2452 
gfs2_glock_init(void)2453 int __init gfs2_glock_init(void)
2454 {
2455 	int i, ret;
2456 
2457 	ret = rhashtable_init(&gl_hash_table, &ht_parms);
2458 	if (ret < 0)
2459 		return ret;
2460 
2461 	glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2462 					  WQ_HIGHPRI | WQ_FREEZABLE, 0);
2463 	if (!glock_workqueue) {
2464 		rhashtable_destroy(&gl_hash_table);
2465 		return -ENOMEM;
2466 	}
2467 
2468 	ret = register_shrinker(&glock_shrinker, "gfs2-glock");
2469 	if (ret) {
2470 		destroy_workqueue(glock_workqueue);
2471 		rhashtable_destroy(&gl_hash_table);
2472 		return ret;
2473 	}
2474 
2475 	for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2476 		init_waitqueue_head(glock_wait_table + i);
2477 
2478 	return 0;
2479 }
2480 
gfs2_glock_exit(void)2481 void gfs2_glock_exit(void)
2482 {
2483 	unregister_shrinker(&glock_shrinker);
2484 	rhashtable_destroy(&gl_hash_table);
2485 	destroy_workqueue(glock_workqueue);
2486 }
2487 
gfs2_glock_iter_next(struct gfs2_glock_iter * gi,loff_t n)2488 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2489 {
2490 	struct gfs2_glock *gl = gi->gl;
2491 
2492 	if (gl) {
2493 		if (n == 0)
2494 			return;
2495 		if (!lockref_put_not_zero(&gl->gl_lockref))
2496 			gfs2_glock_queue_put(gl);
2497 	}
2498 	for (;;) {
2499 		gl = rhashtable_walk_next(&gi->hti);
2500 		if (IS_ERR_OR_NULL(gl)) {
2501 			if (gl == ERR_PTR(-EAGAIN)) {
2502 				n = 1;
2503 				continue;
2504 			}
2505 			gl = NULL;
2506 			break;
2507 		}
2508 		if (gl->gl_name.ln_sbd != gi->sdp)
2509 			continue;
2510 		if (n <= 1) {
2511 			if (!lockref_get_not_dead(&gl->gl_lockref))
2512 				continue;
2513 			break;
2514 		} else {
2515 			if (__lockref_is_dead(&gl->gl_lockref))
2516 				continue;
2517 			n--;
2518 		}
2519 	}
2520 	gi->gl = gl;
2521 }
2522 
gfs2_glock_seq_start(struct seq_file * seq,loff_t * pos)2523 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2524 	__acquires(RCU)
2525 {
2526 	struct gfs2_glock_iter *gi = seq->private;
2527 	loff_t n;
2528 
2529 	/*
2530 	 * We can either stay where we are, skip to the next hash table
2531 	 * entry, or start from the beginning.
2532 	 */
2533 	if (*pos < gi->last_pos) {
2534 		rhashtable_walk_exit(&gi->hti);
2535 		rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2536 		n = *pos + 1;
2537 	} else {
2538 		n = *pos - gi->last_pos;
2539 	}
2540 
2541 	rhashtable_walk_start(&gi->hti);
2542 
2543 	gfs2_glock_iter_next(gi, n);
2544 	gi->last_pos = *pos;
2545 	return gi->gl;
2546 }
2547 
gfs2_glock_seq_next(struct seq_file * seq,void * iter_ptr,loff_t * pos)2548 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2549 				 loff_t *pos)
2550 {
2551 	struct gfs2_glock_iter *gi = seq->private;
2552 
2553 	(*pos)++;
2554 	gi->last_pos = *pos;
2555 	gfs2_glock_iter_next(gi, 1);
2556 	return gi->gl;
2557 }
2558 
gfs2_glock_seq_stop(struct seq_file * seq,void * iter_ptr)2559 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2560 	__releases(RCU)
2561 {
2562 	struct gfs2_glock_iter *gi = seq->private;
2563 
2564 	rhashtable_walk_stop(&gi->hti);
2565 }
2566 
gfs2_glock_seq_show(struct seq_file * seq,void * iter_ptr)2567 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2568 {
2569 	dump_glock(seq, iter_ptr, false);
2570 	return 0;
2571 }
2572 
gfs2_sbstats_seq_start(struct seq_file * seq,loff_t * pos)2573 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2574 {
2575 	preempt_disable();
2576 	if (*pos >= GFS2_NR_SBSTATS)
2577 		return NULL;
2578 	return pos;
2579 }
2580 
gfs2_sbstats_seq_next(struct seq_file * seq,void * iter_ptr,loff_t * pos)2581 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2582 				   loff_t *pos)
2583 {
2584 	(*pos)++;
2585 	if (*pos >= GFS2_NR_SBSTATS)
2586 		return NULL;
2587 	return pos;
2588 }
2589 
gfs2_sbstats_seq_stop(struct seq_file * seq,void * iter_ptr)2590 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2591 {
2592 	preempt_enable();
2593 }
2594 
2595 static const struct seq_operations gfs2_glock_seq_ops = {
2596 	.start = gfs2_glock_seq_start,
2597 	.next  = gfs2_glock_seq_next,
2598 	.stop  = gfs2_glock_seq_stop,
2599 	.show  = gfs2_glock_seq_show,
2600 };
2601 
2602 static const struct seq_operations gfs2_glstats_seq_ops = {
2603 	.start = gfs2_glock_seq_start,
2604 	.next  = gfs2_glock_seq_next,
2605 	.stop  = gfs2_glock_seq_stop,
2606 	.show  = gfs2_glstats_seq_show,
2607 };
2608 
2609 static const struct seq_operations gfs2_sbstats_sops = {
2610 	.start = gfs2_sbstats_seq_start,
2611 	.next  = gfs2_sbstats_seq_next,
2612 	.stop  = gfs2_sbstats_seq_stop,
2613 	.show  = gfs2_sbstats_seq_show,
2614 };
2615 
2616 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2617 
__gfs2_glocks_open(struct inode * inode,struct file * file,const struct seq_operations * ops)2618 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2619 			      const struct seq_operations *ops)
2620 {
2621 	int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2622 	if (ret == 0) {
2623 		struct seq_file *seq = file->private_data;
2624 		struct gfs2_glock_iter *gi = seq->private;
2625 
2626 		gi->sdp = inode->i_private;
2627 		seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2628 		if (seq->buf)
2629 			seq->size = GFS2_SEQ_GOODSIZE;
2630 		/*
2631 		 * Initially, we are "before" the first hash table entry; the
2632 		 * first call to rhashtable_walk_next gets us the first entry.
2633 		 */
2634 		gi->last_pos = -1;
2635 		gi->gl = NULL;
2636 		rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2637 	}
2638 	return ret;
2639 }
2640 
gfs2_glocks_open(struct inode * inode,struct file * file)2641 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2642 {
2643 	return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2644 }
2645 
gfs2_glocks_release(struct inode * inode,struct file * file)2646 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2647 {
2648 	struct seq_file *seq = file->private_data;
2649 	struct gfs2_glock_iter *gi = seq->private;
2650 
2651 	if (gi->gl)
2652 		gfs2_glock_put(gi->gl);
2653 	rhashtable_walk_exit(&gi->hti);
2654 	return seq_release_private(inode, file);
2655 }
2656 
gfs2_glstats_open(struct inode * inode,struct file * file)2657 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2658 {
2659 	return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2660 }
2661 
2662 static const struct file_operations gfs2_glocks_fops = {
2663 	.owner   = THIS_MODULE,
2664 	.open    = gfs2_glocks_open,
2665 	.read    = seq_read,
2666 	.llseek  = seq_lseek,
2667 	.release = gfs2_glocks_release,
2668 };
2669 
2670 static const struct file_operations gfs2_glstats_fops = {
2671 	.owner   = THIS_MODULE,
2672 	.open    = gfs2_glstats_open,
2673 	.read    = seq_read,
2674 	.llseek  = seq_lseek,
2675 	.release = gfs2_glocks_release,
2676 };
2677 
2678 struct gfs2_glockfd_iter {
2679 	struct super_block *sb;
2680 	unsigned int tgid;
2681 	struct task_struct *task;
2682 	unsigned int fd;
2683 	struct file *file;
2684 };
2685 
gfs2_glockfd_next_task(struct gfs2_glockfd_iter * i)2686 static struct task_struct *gfs2_glockfd_next_task(struct gfs2_glockfd_iter *i)
2687 {
2688 	struct pid_namespace *ns = task_active_pid_ns(current);
2689 	struct pid *pid;
2690 
2691 	if (i->task)
2692 		put_task_struct(i->task);
2693 
2694 	rcu_read_lock();
2695 retry:
2696 	i->task = NULL;
2697 	pid = find_ge_pid(i->tgid, ns);
2698 	if (pid) {
2699 		i->tgid = pid_nr_ns(pid, ns);
2700 		i->task = pid_task(pid, PIDTYPE_TGID);
2701 		if (!i->task) {
2702 			i->tgid++;
2703 			goto retry;
2704 		}
2705 		get_task_struct(i->task);
2706 	}
2707 	rcu_read_unlock();
2708 	return i->task;
2709 }
2710 
gfs2_glockfd_next_file(struct gfs2_glockfd_iter * i)2711 static struct file *gfs2_glockfd_next_file(struct gfs2_glockfd_iter *i)
2712 {
2713 	if (i->file) {
2714 		fput(i->file);
2715 		i->file = NULL;
2716 	}
2717 
2718 	rcu_read_lock();
2719 	for(;; i->fd++) {
2720 		struct inode *inode;
2721 
2722 		i->file = task_lookup_next_fd_rcu(i->task, &i->fd);
2723 		if (!i->file) {
2724 			i->fd = 0;
2725 			break;
2726 		}
2727 		inode = file_inode(i->file);
2728 		if (inode->i_sb != i->sb)
2729 			continue;
2730 		if (get_file_rcu(i->file))
2731 			break;
2732 	}
2733 	rcu_read_unlock();
2734 	return i->file;
2735 }
2736 
gfs2_glockfd_seq_start(struct seq_file * seq,loff_t * pos)2737 static void *gfs2_glockfd_seq_start(struct seq_file *seq, loff_t *pos)
2738 {
2739 	struct gfs2_glockfd_iter *i = seq->private;
2740 
2741 	if (*pos)
2742 		return NULL;
2743 	while (gfs2_glockfd_next_task(i)) {
2744 		if (gfs2_glockfd_next_file(i))
2745 			return i;
2746 		i->tgid++;
2747 	}
2748 	return NULL;
2749 }
2750 
gfs2_glockfd_seq_next(struct seq_file * seq,void * iter_ptr,loff_t * pos)2751 static void *gfs2_glockfd_seq_next(struct seq_file *seq, void *iter_ptr,
2752 				   loff_t *pos)
2753 {
2754 	struct gfs2_glockfd_iter *i = seq->private;
2755 
2756 	(*pos)++;
2757 	i->fd++;
2758 	do {
2759 		if (gfs2_glockfd_next_file(i))
2760 			return i;
2761 		i->tgid++;
2762 	} while (gfs2_glockfd_next_task(i));
2763 	return NULL;
2764 }
2765 
gfs2_glockfd_seq_stop(struct seq_file * seq,void * iter_ptr)2766 static void gfs2_glockfd_seq_stop(struct seq_file *seq, void *iter_ptr)
2767 {
2768 	struct gfs2_glockfd_iter *i = seq->private;
2769 
2770 	if (i->file)
2771 		fput(i->file);
2772 	if (i->task)
2773 		put_task_struct(i->task);
2774 }
2775 
gfs2_glockfd_seq_show_flock(struct seq_file * seq,struct gfs2_glockfd_iter * i)2776 static void gfs2_glockfd_seq_show_flock(struct seq_file *seq,
2777 					struct gfs2_glockfd_iter *i)
2778 {
2779 	struct gfs2_file *fp = i->file->private_data;
2780 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
2781 	struct lm_lockname gl_name = { .ln_type = LM_TYPE_RESERVED };
2782 
2783 	if (!READ_ONCE(fl_gh->gh_gl))
2784 		return;
2785 
2786 	spin_lock(&i->file->f_lock);
2787 	if (gfs2_holder_initialized(fl_gh))
2788 		gl_name = fl_gh->gh_gl->gl_name;
2789 	spin_unlock(&i->file->f_lock);
2790 
2791 	if (gl_name.ln_type != LM_TYPE_RESERVED) {
2792 		seq_printf(seq, "%d %u %u/%llx\n",
2793 			   i->tgid, i->fd, gl_name.ln_type,
2794 			   (unsigned long long)gl_name.ln_number);
2795 	}
2796 }
2797 
gfs2_glockfd_seq_show(struct seq_file * seq,void * iter_ptr)2798 static int gfs2_glockfd_seq_show(struct seq_file *seq, void *iter_ptr)
2799 {
2800 	struct gfs2_glockfd_iter *i = seq->private;
2801 	struct inode *inode = file_inode(i->file);
2802 	struct gfs2_glock *gl;
2803 
2804 	inode_lock_shared(inode);
2805 	gl = GFS2_I(inode)->i_iopen_gh.gh_gl;
2806 	if (gl) {
2807 		seq_printf(seq, "%d %u %u/%llx\n",
2808 			   i->tgid, i->fd, gl->gl_name.ln_type,
2809 			   (unsigned long long)gl->gl_name.ln_number);
2810 	}
2811 	gfs2_glockfd_seq_show_flock(seq, i);
2812 	inode_unlock_shared(inode);
2813 	return 0;
2814 }
2815 
2816 static const struct seq_operations gfs2_glockfd_seq_ops = {
2817 	.start = gfs2_glockfd_seq_start,
2818 	.next  = gfs2_glockfd_seq_next,
2819 	.stop  = gfs2_glockfd_seq_stop,
2820 	.show  = gfs2_glockfd_seq_show,
2821 };
2822 
gfs2_glockfd_open(struct inode * inode,struct file * file)2823 static int gfs2_glockfd_open(struct inode *inode, struct file *file)
2824 {
2825 	struct gfs2_glockfd_iter *i;
2826 	struct gfs2_sbd *sdp = inode->i_private;
2827 
2828 	i = __seq_open_private(file, &gfs2_glockfd_seq_ops,
2829 			       sizeof(struct gfs2_glockfd_iter));
2830 	if (!i)
2831 		return -ENOMEM;
2832 	i->sb = sdp->sd_vfs;
2833 	return 0;
2834 }
2835 
2836 static const struct file_operations gfs2_glockfd_fops = {
2837 	.owner   = THIS_MODULE,
2838 	.open    = gfs2_glockfd_open,
2839 	.read    = seq_read,
2840 	.llseek  = seq_lseek,
2841 	.release = seq_release_private,
2842 };
2843 
2844 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2845 
gfs2_create_debugfs_file(struct gfs2_sbd * sdp)2846 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2847 {
2848 	sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2849 
2850 	debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2851 			    &gfs2_glocks_fops);
2852 
2853 	debugfs_create_file("glockfd", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2854 			    &gfs2_glockfd_fops);
2855 
2856 	debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2857 			    &gfs2_glstats_fops);
2858 
2859 	debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2860 			    &gfs2_sbstats_fops);
2861 }
2862 
gfs2_delete_debugfs_file(struct gfs2_sbd * sdp)2863 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2864 {
2865 	debugfs_remove_recursive(sdp->debugfs_dir);
2866 	sdp->debugfs_dir = NULL;
2867 }
2868 
gfs2_register_debugfs(void)2869 void gfs2_register_debugfs(void)
2870 {
2871 	gfs2_root = debugfs_create_dir("gfs2", NULL);
2872 }
2873 
gfs2_unregister_debugfs(void)2874 void gfs2_unregister_debugfs(void)
2875 {
2876 	debugfs_remove(gfs2_root);
2877 	gfs2_root = NULL;
2878 }
2879