xref: /freebsd/sys/kern/subr_witness.c (revision 729362425c09cf6b362366aabc6fb547eee8035a)
1 /*-
2  * Copyright (c) 1998 Berkeley Software Design, Inc. All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  * 3. Berkeley Software Design Inc's name may not be used to endorse or
13  *    promote products derived from this software without specific prior
14  *    written permission.
15  *
16  * THIS SOFTWARE IS PROVIDED BY BERKELEY SOFTWARE DESIGN INC ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL BERKELEY SOFTWARE DESIGN INC BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  *	from BSDI $Id: mutex_witness.c,v 1.1.2.20 2000/04/27 03:10:27 cp Exp $
29  *	and BSDI $Id: synch_machdep.c,v 2.3.2.39 2000/04/27 03:10:25 cp Exp $
30  * $FreeBSD$
31  */
32 
33 /*
34  * Implementation of the `witness' lock verifier.  Originally implemented for
35  * mutexes in BSD/OS.  Extended to handle generic lock objects and lock
36  * classes in FreeBSD.
37  */
38 
39 /*
40  *	Main Entry: witness
41  *	Pronunciation: 'wit-n&s
42  *	Function: noun
43  *	Etymology: Middle English witnesse, from Old English witnes knowledge,
44  *	    testimony, witness, from 2wit
45  *	Date: before 12th century
46  *	1 : attestation of a fact or event : TESTIMONY
47  *	2 : one that gives evidence; specifically : one who testifies in
48  *	    a cause or before a judicial tribunal
49  *	3 : one asked to be present at a transaction so as to be able to
50  *	    testify to its having taken place
51  *	4 : one who has personal knowledge of something
52  *	5 a : something serving as evidence or proof : SIGN
53  *	  b : public affirmation by word or example of usually
54  *	      religious faith or conviction <the heroic witness to divine
55  *	      life -- Pilot>
56  *	6 capitalized : a member of the Jehovah's Witnesses
57  */
58 
59 /*
60  * Special rules concerning Giant and lock orders:
61  *
62  * 1) Giant must be acquired before any other mutexes.  Stated another way,
63  *    no other mutex may be held when Giant is acquired.
64  *
65  * 2) Giant must be released when blocking on a sleepable lock.
66  *
67  * This rule is less obvious, but is a result of Giant providing the same
68  * semantics as spl().  Basically, when a thread sleeps, it must release
69  * Giant.  When a thread blocks on a sleepable lock, it sleeps.  Hence rule
70  * 2).
71  *
72  * 3) Giant may be acquired before or after sleepable locks.
73  *
74  * This rule is also not quite as obvious.  Giant may be acquired after
75  * a sleepable lock because it is a non-sleepable lock and non-sleepable
76  * locks may always be acquired while holding a sleepable lock.  The second
77  * case, Giant before a sleepable lock, follows from rule 2) above.  Suppose
78  * you have two threads T1 and T2 and a sleepable lock X.  Suppose that T1
79  * acquires X and blocks on Giant.  Then suppose that T2 acquires Giant and
80  * blocks on X.  When T2 blocks on X, T2 will release Giant allowing T1 to
81  * execute.  Thus, acquiring Giant both before and after a sleepable lock
82  * will not result in a lock order reversal.
83  */
84 
85 #include "opt_ddb.h"
86 #include "opt_witness.h"
87 
88 #include <sys/param.h>
89 #include <sys/bus.h>
90 #include <sys/kernel.h>
91 #include <sys/ktr.h>
92 #include <sys/lock.h>
93 #include <sys/malloc.h>
94 #include <sys/mutex.h>
95 #include <sys/proc.h>
96 #include <sys/sysctl.h>
97 #include <sys/systm.h>
98 
99 #include <ddb/ddb.h>
100 
101 #include <machine/stdarg.h>
102 
103 /* Define this to check for blessed mutexes */
104 #undef BLESSING
105 
106 #define WITNESS_COUNT 200
107 #define WITNESS_CHILDCOUNT (WITNESS_COUNT * 4)
108 /*
109  * XXX: This is somewhat bogus, as we assume here that at most 1024 threads
110  * will hold LOCK_NCHILDREN * 2 locks.  We handle failure ok, and we should
111  * probably be safe for the most part, but it's still a SWAG.
112  */
113 #define LOCK_CHILDCOUNT (MAXCPU + 1024) * 2
114 
115 #define	WITNESS_NCHILDREN 6
116 
117 struct witness_child_list_entry;
118 
119 struct witness {
120 	const	char *w_name;
121 	struct	lock_class *w_class;
122 	STAILQ_ENTRY(witness) w_list;		/* List of all witnesses. */
123 	STAILQ_ENTRY(witness) w_typelist;	/* Witnesses of a type. */
124 	struct	witness_child_list_entry *w_children;	/* Great evilness... */
125 	const	char *w_file;
126 	int	w_line;
127 	u_int	w_level;
128 	u_int	w_refcount;
129 	u_char	w_Giant_squawked:1;
130 	u_char	w_other_squawked:1;
131 	u_char	w_same_squawked:1;
132 	u_char	w_displayed:1;
133 };
134 
135 struct witness_child_list_entry {
136 	struct	witness_child_list_entry *wcl_next;
137 	struct	witness *wcl_children[WITNESS_NCHILDREN];
138 	u_int	wcl_count;
139 };
140 
141 STAILQ_HEAD(witness_list, witness);
142 
143 #ifdef BLESSING
144 struct witness_blessed {
145 	const	char *b_lock1;
146 	const	char *b_lock2;
147 };
148 #endif
149 
150 struct witness_order_list_entry {
151 	const	char *w_name;
152 	struct	lock_class *w_class;
153 };
154 
155 #ifdef BLESSING
156 static int	blessed(struct witness *, struct witness *);
157 #endif
158 static int	depart(struct witness *w);
159 static struct	witness *enroll(const char *description,
160 				struct lock_class *lock_class);
161 static int	insertchild(struct witness *parent, struct witness *child);
162 static int	isitmychild(struct witness *parent, struct witness *child);
163 static int	isitmydescendant(struct witness *parent, struct witness *child);
164 static int	itismychild(struct witness *parent, struct witness *child);
165 static int	rebalancetree(struct witness_list *list);
166 static void	removechild(struct witness *parent, struct witness *child);
167 static int	reparentchildren(struct witness *newparent,
168 		    struct witness *oldparent);
169 static int	sysctl_debug_witness_watch(SYSCTL_HANDLER_ARGS);
170 static void	witness_displaydescendants(void(*)(const char *fmt, ...),
171 					   struct witness *, int indent);
172 static const char *fixup_filename(const char *file);
173 static void	witness_leveldescendents(struct witness *parent, int level);
174 static void	witness_levelall(void);
175 static struct	witness *witness_get(void);
176 static void	witness_free(struct witness *m);
177 static struct	witness_child_list_entry *witness_child_get(void);
178 static void	witness_child_free(struct witness_child_list_entry *wcl);
179 static struct	lock_list_entry *witness_lock_list_get(void);
180 static void	witness_lock_list_free(struct lock_list_entry *lle);
181 static struct	lock_instance *find_instance(struct lock_list_entry *lock_list,
182 					     struct lock_object *lock);
183 static void	witness_list_lock(struct lock_instance *instance);
184 #ifdef DDB
185 static void	witness_list(struct thread *td);
186 static void	witness_display_list(void(*prnt)(const char *fmt, ...),
187 				     struct witness_list *list);
188 static void	witness_display(void(*)(const char *fmt, ...));
189 #endif
190 
191 MALLOC_DEFINE(M_WITNESS, "witness", "witness structure");
192 
193 /*
194  * If set to 0, witness is disabled.  If set to 1, witness performs full lock
195  * order checking for all locks.  If set to 2 or higher, then witness skips
196  * the full lock order check if the lock being acquired is at a higher level
197  * (i.e. farther down in the tree) than the current lock.  This last mode is
198  * somewhat experimental and not considered fully safe.  At runtime, this
199  * value may be set to 0 to turn off witness.  witness is not allowed be
200  * turned on once it is turned off, however.
201  */
202 static int witness_watch = 1;
203 TUNABLE_INT("debug.witness_watch", &witness_watch);
204 SYSCTL_PROC(_debug, OID_AUTO, witness_watch, CTLFLAG_RW | CTLTYPE_INT, NULL, 0,
205     sysctl_debug_witness_watch, "I", "witness is watching lock operations");
206 
207 #ifdef DDB
208 /*
209  * When DDB is enabled and witness_ddb is set to 1, it will cause the system to
210  * drop into kdebug() when:
211  *	- a lock heirarchy violation occurs
212  *	- locks are held when going to sleep.
213  */
214 #ifdef WITNESS_DDB
215 int	witness_ddb = 1;
216 #else
217 int	witness_ddb = 0;
218 #endif
219 TUNABLE_INT("debug.witness_ddb", &witness_ddb);
220 SYSCTL_INT(_debug, OID_AUTO, witness_ddb, CTLFLAG_RW, &witness_ddb, 0, "");
221 
222 /*
223  * When DDB is enabled and witness_trace is set to 1, it will cause the system
224  * to print a stack trace:
225  *	- a lock heirarchy violation occurs
226  *	- locks are held when going to sleep.
227  */
228 int	witness_trace = 1;
229 TUNABLE_INT("debug.witness_trace", &witness_trace);
230 SYSCTL_INT(_debug, OID_AUTO, witness_trace, CTLFLAG_RW, &witness_trace, 0, "");
231 #endif /* DDB */
232 
233 #ifdef WITNESS_SKIPSPIN
234 int	witness_skipspin = 1;
235 #else
236 int	witness_skipspin = 0;
237 #endif
238 TUNABLE_INT("debug.witness_skipspin", &witness_skipspin);
239 SYSCTL_INT(_debug, OID_AUTO, witness_skipspin, CTLFLAG_RD, &witness_skipspin, 0,
240     "");
241 
242 static struct mtx w_mtx;
243 static struct witness_list w_free = STAILQ_HEAD_INITIALIZER(w_free);
244 static struct witness_list w_all = STAILQ_HEAD_INITIALIZER(w_all);
245 static struct witness_list w_spin = STAILQ_HEAD_INITIALIZER(w_spin);
246 static struct witness_list w_sleep = STAILQ_HEAD_INITIALIZER(w_sleep);
247 static struct witness_child_list_entry *w_child_free = NULL;
248 static struct lock_list_entry *w_lock_list_free = NULL;
249 
250 static struct witness w_data[WITNESS_COUNT];
251 static struct witness_child_list_entry w_childdata[WITNESS_CHILDCOUNT];
252 static struct lock_list_entry w_locklistdata[LOCK_CHILDCOUNT];
253 
254 static struct witness_order_list_entry order_lists[] = {
255 	{ "proctree", &lock_class_sx },
256 	{ "allproc", &lock_class_sx },
257 	{ "Giant", &lock_class_mtx_sleep },
258 	{ "filedesc structure", &lock_class_mtx_sleep },
259 	{ "pipe mutex", &lock_class_mtx_sleep },
260 	{ "sigio lock", &lock_class_mtx_sleep },
261 	{ "process group", &lock_class_mtx_sleep },
262 	{ "process lock", &lock_class_mtx_sleep },
263 	{ "session", &lock_class_mtx_sleep },
264 	{ "uidinfo hash", &lock_class_mtx_sleep },
265 	{ "uidinfo struct", &lock_class_mtx_sleep },
266 	{ NULL, NULL },
267 	/*
268 	 * spin locks
269 	 */
270 #ifdef SMP
271 	{ "ap boot", &lock_class_mtx_spin },
272 #ifdef __i386__
273 	{ "com", &lock_class_mtx_spin },
274 #endif
275 #endif
276 	{ "sio", &lock_class_mtx_spin },
277 #ifdef __i386__
278 	{ "cy", &lock_class_mtx_spin },
279 #endif
280 	{ "sabtty", &lock_class_mtx_spin },
281 	{ "zstty", &lock_class_mtx_spin },
282 	{ "ng_node", &lock_class_mtx_spin },
283 	{ "ng_worklist", &lock_class_mtx_spin },
284 	{ "ithread table lock", &lock_class_mtx_spin },
285 	{ "sched lock", &lock_class_mtx_spin },
286 	{ "callout", &lock_class_mtx_spin },
287 	/*
288 	 * leaf locks
289 	 */
290 	{ "allpmaps", &lock_class_mtx_spin },
291 	{ "vm page queue free mutex", &lock_class_mtx_spin },
292 	{ "icu", &lock_class_mtx_spin },
293 #ifdef SMP
294 	{ "smp rendezvous", &lock_class_mtx_spin },
295 #if defined(__i386__) && defined(APIC_IO)
296 	{ "tlb", &lock_class_mtx_spin },
297 #endif
298 #ifdef __sparc64__
299 	{ "ipi", &lock_class_mtx_spin },
300 #endif
301 #endif
302 	{ "clk", &lock_class_mtx_spin },
303 	{ "mutex profiling lock", &lock_class_mtx_spin },
304 	{ "kse zombie lock", &lock_class_mtx_spin },
305 	{ "ALD Queue", &lock_class_mtx_spin },
306 #ifdef __ia64__
307 	{ "MCA spin lock", &lock_class_mtx_spin },
308 #endif
309 #ifdef __i386__
310 	{ "pcicfg", &lock_class_mtx_spin },
311 #endif
312 	{ NULL, NULL },
313 	{ NULL, NULL }
314 };
315 
316 #ifdef BLESSING
317 /*
318  * Pairs of locks which have been blessed
319  * Don't complain about order problems with blessed locks
320  */
321 static struct witness_blessed blessed_list[] = {
322 };
323 static int blessed_count =
324 	sizeof(blessed_list) / sizeof(struct witness_blessed);
325 #endif
326 
327 /*
328  * List of all locks in the system.
329  */
330 TAILQ_HEAD(, lock_object) all_locks = TAILQ_HEAD_INITIALIZER(all_locks);
331 
332 static struct mtx all_mtx = {
333 	{ &lock_class_mtx_sleep,	/* mtx_object.lo_class */
334 	  "All locks list",		/* mtx_object.lo_name */
335 	  "All locks list",		/* mtx_object.lo_type */
336 	  LO_INITIALIZED,		/* mtx_object.lo_flags */
337 	  { NULL, NULL },		/* mtx_object.lo_list */
338 	  NULL },			/* mtx_object.lo_witness */
339 	MTX_UNOWNED, 0,			/* mtx_lock, mtx_recurse */
340 	TAILQ_HEAD_INITIALIZER(all_mtx.mtx_blocked),
341 	{ NULL, NULL }			/* mtx_contested */
342 };
343 
344 /*
345  * This global is set to 0 once it becomes safe to use the witness code.
346  */
347 static int witness_cold = 1;
348 
349 /*
350  * Global variables for book keeping.
351  */
352 static int lock_cur_cnt;
353 static int lock_max_cnt;
354 
355 /*
356  * The WITNESS-enabled diagnostic code.
357  */
358 static void
359 witness_initialize(void *dummy __unused)
360 {
361 	struct lock_object *lock;
362 	struct witness_order_list_entry *order;
363 	struct witness *w, *w1;
364 	int i;
365 
366 	/*
367 	 * We have to release Giant before initializing its witness
368 	 * structure so that WITNESS doesn't get confused.
369 	 */
370 	mtx_unlock(&Giant);
371 	mtx_assert(&Giant, MA_NOTOWNED);
372 
373 	CTR1(KTR_WITNESS, "%s: initializing witness", __func__);
374 	TAILQ_INSERT_HEAD(&all_locks, &all_mtx.mtx_object, lo_list);
375 	mtx_init(&w_mtx, "witness lock", NULL, MTX_SPIN | MTX_QUIET |
376 	    MTX_NOWITNESS);
377 	for (i = 0; i < WITNESS_COUNT; i++)
378 		witness_free(&w_data[i]);
379 	for (i = 0; i < WITNESS_CHILDCOUNT; i++)
380 		witness_child_free(&w_childdata[i]);
381 	for (i = 0; i < LOCK_CHILDCOUNT; i++)
382 		witness_lock_list_free(&w_locklistdata[i]);
383 
384 	/* First add in all the specified order lists. */
385 	for (order = order_lists; order->w_name != NULL; order++) {
386 		w = enroll(order->w_name, order->w_class);
387 		if (w == NULL)
388 			continue;
389 		w->w_file = "order list";
390 		for (order++; order->w_name != NULL; order++) {
391 			w1 = enroll(order->w_name, order->w_class);
392 			if (w1 == NULL)
393 				continue;
394 			w1->w_file = "order list";
395 			if (!itismychild(w, w1))
396 				panic("Not enough memory for static orders!");
397 			w = w1;
398 		}
399 	}
400 
401 	/* Iterate through all locks and add them to witness. */
402 	mtx_lock(&all_mtx);
403 	TAILQ_FOREACH(lock, &all_locks, lo_list) {
404 		if (lock->lo_flags & LO_WITNESS)
405 			lock->lo_witness = enroll(lock->lo_type,
406 			    lock->lo_class);
407 		else
408 			lock->lo_witness = NULL;
409 	}
410 	mtx_unlock(&all_mtx);
411 
412 	/* Mark the witness code as being ready for use. */
413 	atomic_store_rel_int(&witness_cold, 0);
414 
415 	mtx_lock(&Giant);
416 }
417 SYSINIT(witness_init, SI_SUB_WITNESS, SI_ORDER_FIRST, witness_initialize, NULL)
418 
419 static int
420 sysctl_debug_witness_watch(SYSCTL_HANDLER_ARGS)
421 {
422 	int error, value;
423 
424 	value = witness_watch;
425 	error = sysctl_handle_int(oidp, &value, 0, req);
426 	if (error != 0 || req->newptr == NULL)
427 		return (error);
428 	error = suser(req->td);
429 	if (error != 0)
430 		return (error);
431 	if (value == witness_watch)
432 		return (0);
433 	if (value != 0)
434 		return (EINVAL);
435 	witness_watch = 0;
436 	return (0);
437 }
438 
439 void
440 witness_init(struct lock_object *lock)
441 {
442 	struct lock_class *class;
443 
444 	class = lock->lo_class;
445 	if (lock->lo_flags & LO_INITIALIZED)
446 		panic("%s: lock (%s) %s is already initialized", __func__,
447 		    class->lc_name, lock->lo_name);
448 	if ((lock->lo_flags & LO_RECURSABLE) != 0 &&
449 	    (class->lc_flags & LC_RECURSABLE) == 0)
450 		panic("%s: lock (%s) %s can not be recursable", __func__,
451 		    class->lc_name, lock->lo_name);
452 	if ((lock->lo_flags & LO_SLEEPABLE) != 0 &&
453 	    (class->lc_flags & LC_SLEEPABLE) == 0)
454 		panic("%s: lock (%s) %s can not be sleepable", __func__,
455 		    class->lc_name, lock->lo_name);
456 	if ((lock->lo_flags & LO_UPGRADABLE) != 0 &&
457 	    (class->lc_flags & LC_UPGRADABLE) == 0)
458 		panic("%s: lock (%s) %s can not be upgradable", __func__,
459 		    class->lc_name, lock->lo_name);
460 
461 	mtx_lock(&all_mtx);
462 	TAILQ_INSERT_TAIL(&all_locks, lock, lo_list);
463 	lock->lo_flags |= LO_INITIALIZED;
464 	lock_cur_cnt++;
465 	if (lock_cur_cnt > lock_max_cnt)
466 		lock_max_cnt = lock_cur_cnt;
467 	mtx_unlock(&all_mtx);
468 	if (!witness_cold && witness_watch != 0 && panicstr == NULL &&
469 	    (lock->lo_flags & LO_WITNESS) != 0)
470 		lock->lo_witness = enroll(lock->lo_type, class);
471 	else
472 		lock->lo_witness = NULL;
473 }
474 
475 void
476 witness_destroy(struct lock_object *lock)
477 {
478 	struct witness *w;
479 
480 	if (witness_cold)
481 		panic("lock (%s) %s destroyed while witness_cold",
482 		    lock->lo_class->lc_name, lock->lo_name);
483 	if ((lock->lo_flags & LO_INITIALIZED) == 0)
484 		panic("%s: lock (%s) %s is not initialized", __func__,
485 		    lock->lo_class->lc_name, lock->lo_name);
486 
487 	/* XXX: need to verify that no one holds the lock */
488 	w = lock->lo_witness;
489 	if (w != NULL) {
490 		mtx_lock_spin(&w_mtx);
491 		MPASS(w->w_refcount > 0);
492 		w->w_refcount--;
493 
494 		/*
495 		 * Lock is already released if we have an allocation failure
496 		 * and depart() fails.
497 		 */
498 		if (w->w_refcount != 0 || depart(w))
499 			mtx_unlock_spin(&w_mtx);
500 	}
501 
502 	mtx_lock(&all_mtx);
503 	lock_cur_cnt--;
504 	TAILQ_REMOVE(&all_locks, lock, lo_list);
505 	lock->lo_flags &= ~LO_INITIALIZED;
506 	mtx_unlock(&all_mtx);
507 }
508 
509 #ifdef DDB
510 static void
511 witness_display_list(void(*prnt)(const char *fmt, ...),
512 		     struct witness_list *list)
513 {
514 	struct witness *w;
515 
516 	STAILQ_FOREACH(w, list, w_typelist) {
517 		if (w->w_file == NULL || w->w_level > 0)
518 			continue;
519 		/*
520 		 * This lock has no anscestors, display its descendants.
521 		 */
522 		witness_displaydescendants(prnt, w, 0);
523 	}
524 }
525 
526 static void
527 witness_display(void(*prnt)(const char *fmt, ...))
528 {
529 	struct witness *w;
530 
531 	KASSERT(!witness_cold, ("%s: witness_cold", __func__));
532 	witness_levelall();
533 
534 	/* Clear all the displayed flags. */
535 	STAILQ_FOREACH(w, &w_all, w_list) {
536 		w->w_displayed = 0;
537 	}
538 
539 	/*
540 	 * First, handle sleep locks which have been acquired at least
541 	 * once.
542 	 */
543 	prnt("Sleep locks:\n");
544 	witness_display_list(prnt, &w_sleep);
545 
546 	/*
547 	 * Now do spin locks which have been acquired at least once.
548 	 */
549 	prnt("\nSpin locks:\n");
550 	witness_display_list(prnt, &w_spin);
551 
552 	/*
553 	 * Finally, any locks which have not been acquired yet.
554 	 */
555 	prnt("\nLocks which were never acquired:\n");
556 	STAILQ_FOREACH(w, &w_all, w_list) {
557 		if (w->w_file != NULL || w->w_refcount == 0)
558 			continue;
559 		prnt("%s\n", w->w_name);
560 	}
561 }
562 #endif /* DDB */
563 
564 /* Trim useless garbage from filenames. */
565 static const char *
566 fixup_filename(const char *file)
567 {
568 
569 	if (file == NULL)
570 		return (NULL);
571 	while (strncmp(file, "../", 3) == 0)
572 		file += 3;
573 	return (file);
574 }
575 
576 void
577 witness_lock(struct lock_object *lock, int flags, const char *file, int line)
578 {
579 	struct lock_list_entry **lock_list, *lle;
580 	struct lock_instance *lock1, *lock2;
581 	struct lock_class *class;
582 	struct witness *w, *w1;
583 	struct thread *td;
584 	int i, j;
585 #ifdef DDB
586 	int go_into_ddb = 0;
587 #endif
588 
589 	if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL ||
590 	    panicstr != NULL)
591 		return;
592 	w = lock->lo_witness;
593 	class = lock->lo_class;
594 	td = curthread;
595 	file = fixup_filename(file);
596 
597 	if (class->lc_flags & LC_SLEEPLOCK) {
598 		/*
599 		 * Since spin locks include a critical section, this check
600 		 * impliclty enforces a lock order of all sleep locks before
601 		 * all spin locks.
602 		 */
603 		if (td->td_critnest != 0 && (flags & LOP_TRYLOCK) == 0)
604 			panic("blockable sleep lock (%s) %s @ %s:%d",
605 			    class->lc_name, lock->lo_name, file, line);
606 		lock_list = &td->td_sleeplocks;
607 	} else
608 		lock_list = PCPU_PTR(spinlocks);
609 
610 	/*
611 	 * Is this the first lock acquired?  If so, then no order checking
612 	 * is needed.
613 	 */
614 	if (*lock_list == NULL)
615 		goto out;
616 
617 	/*
618 	 * Check to see if we are recursing on a lock we already own.
619 	 */
620 	lock1 = find_instance(*lock_list, lock);
621 	if (lock1 != NULL) {
622 		if ((lock1->li_flags & LI_EXCLUSIVE) != 0 &&
623 		    (flags & LOP_EXCLUSIVE) == 0) {
624 			printf("shared lock of (%s) %s @ %s:%d\n",
625 			    class->lc_name, lock->lo_name, file, line);
626 			printf("while exclusively locked from %s:%d\n",
627 			    lock1->li_file, lock1->li_line);
628 			panic("share->excl");
629 		}
630 		if ((lock1->li_flags & LI_EXCLUSIVE) == 0 &&
631 		    (flags & LOP_EXCLUSIVE) != 0) {
632 			printf("exclusive lock of (%s) %s @ %s:%d\n",
633 			    class->lc_name, lock->lo_name, file, line);
634 			printf("while share locked from %s:%d\n",
635 			    lock1->li_file, lock1->li_line);
636 			panic("excl->share");
637 		}
638 		lock1->li_flags++;
639 		if ((lock->lo_flags & LO_RECURSABLE) == 0) {
640 			printf(
641 			"recursed on non-recursive lock (%s) %s @ %s:%d\n",
642 			    class->lc_name, lock->lo_name, file, line);
643 			printf("first acquired @ %s:%d\n", lock1->li_file,
644 			    lock1->li_line);
645 			panic("recurse");
646 		}
647 		CTR4(KTR_WITNESS, "%s: pid %d recursed on %s r=%d", __func__,
648 		    td->td_proc->p_pid, lock->lo_name,
649 		    lock1->li_flags & LI_RECURSEMASK);
650 		lock1->li_file = file;
651 		lock1->li_line = line;
652 		return;
653 	}
654 
655 	/*
656 	 * Try locks do not block if they fail to acquire the lock, thus
657 	 * there is no danger of deadlocks or of switching while holding a
658 	 * spin lock if we acquire a lock via a try operation.
659 	 */
660 	if (flags & LOP_TRYLOCK)
661 		goto out;
662 
663 	/*
664 	 * Check for duplicate locks of the same type.  Note that we only
665 	 * have to check for this on the last lock we just acquired.  Any
666 	 * other cases will be caught as lock order violations.
667 	 */
668 	lock1 = &(*lock_list)->ll_children[(*lock_list)->ll_count - 1];
669 	w1 = lock1->li_lock->lo_witness;
670 	if (w1 == w) {
671 		if (w->w_same_squawked || (lock->lo_flags & LO_DUPOK))
672 			goto out;
673 		w->w_same_squawked = 1;
674 		printf("acquiring duplicate lock of same type: \"%s\"\n",
675 			lock->lo_type);
676 		printf(" 1st %s @ %s:%d\n", lock1->li_lock->lo_name,
677 		    lock1->li_file, lock1->li_line);
678 		printf(" 2nd %s @ %s:%d\n", lock->lo_name, file, line);
679 #ifdef DDB
680 		go_into_ddb = 1;
681 #endif
682 		goto out;
683 	}
684 	MPASS(!mtx_owned(&w_mtx));
685 	mtx_lock_spin(&w_mtx);
686 	/*
687 	 * If we have a known higher number just say ok
688 	 */
689 	if (witness_watch > 1 && w->w_level > w1->w_level) {
690 		mtx_unlock_spin(&w_mtx);
691 		goto out;
692 	}
693 	/*
694 	 * If we know that the the lock we are acquiring comes after
695 	 * the lock we most recently acquired in the lock order tree,
696 	 * then there is no need for any further checks.
697 	 */
698 	if (isitmydescendant(w1, w)) {
699 		mtx_unlock_spin(&w_mtx);
700 		goto out;
701 	}
702 	for (j = 0, lle = *lock_list; lle != NULL; lle = lle->ll_next) {
703 		for (i = lle->ll_count - 1; i >= 0; i--, j++) {
704 
705 			MPASS(j < WITNESS_COUNT);
706 			lock1 = &lle->ll_children[i];
707 			w1 = lock1->li_lock->lo_witness;
708 
709 			/*
710 			 * If this lock doesn't undergo witness checking,
711 			 * then skip it.
712 			 */
713 			if (w1 == NULL) {
714 				KASSERT((lock1->li_lock->lo_flags & LO_WITNESS) == 0,
715 				    ("lock missing witness structure"));
716 				continue;
717 			}
718 			/*
719 			 * If we are locking Giant and this is a sleepable
720 			 * lock, then skip it.
721 			 */
722 			if ((lock1->li_lock->lo_flags & LO_SLEEPABLE) != 0 &&
723 			    lock == &Giant.mtx_object)
724 				continue;
725 			/*
726 			 * If we are locking a sleepable lock and this lock
727 			 * is Giant, then skip it.
728 			 */
729 			if ((lock->lo_flags & LO_SLEEPABLE) != 0 &&
730 			    lock1->li_lock == &Giant.mtx_object)
731 				continue;
732 			/*
733 			 * If we are locking a sleepable lock and this lock
734 			 * isn't sleepable, we want to treat it as a lock
735 			 * order violation to enfore a general lock order of
736 			 * sleepable locks before non-sleepable locks.
737 			 */
738 			if (!((lock->lo_flags & LO_SLEEPABLE) != 0 &&
739 			    (lock1->li_lock->lo_flags & LO_SLEEPABLE) == 0))
740 			    /*
741 			     * Check the lock order hierarchy for a reveresal.
742 			     */
743 			    if (!isitmydescendant(w, w1))
744 				continue;
745 			/*
746 			 * We have a lock order violation, check to see if it
747 			 * is allowed or has already been yelled about.
748 			 */
749 			mtx_unlock_spin(&w_mtx);
750 #ifdef BLESSING
751 			if (blessed(w, w1))
752 				goto out;
753 #endif
754 			if (lock1->li_lock == &Giant.mtx_object) {
755 				if (w1->w_Giant_squawked)
756 					goto out;
757 				else
758 					w1->w_Giant_squawked = 1;
759 			} else {
760 				if (w1->w_other_squawked)
761 					goto out;
762 				else
763 					w1->w_other_squawked = 1;
764 			}
765 			/*
766 			 * Ok, yell about it.
767 			 */
768 			printf("lock order reversal\n");
769 			/*
770 			 * Try to locate an earlier lock with
771 			 * witness w in our list.
772 			 */
773 			do {
774 				lock2 = &lle->ll_children[i];
775 				MPASS(lock2->li_lock != NULL);
776 				if (lock2->li_lock->lo_witness == w)
777 					break;
778 				i--;
779 				if (i == 0 && lle->ll_next != NULL) {
780 					lle = lle->ll_next;
781 					i = lle->ll_count - 1;
782 					MPASS(i >= 0 && i < LOCK_NCHILDREN);
783 				}
784 			} while (i >= 0);
785 			if (i < 0) {
786 				printf(" 1st %p %s (%s) @ %s:%d\n",
787 				    lock1->li_lock, lock1->li_lock->lo_name,
788 				    lock1->li_lock->lo_type, lock1->li_file,
789 				    lock1->li_line);
790 				printf(" 2nd %p %s (%s) @ %s:%d\n", lock,
791 				    lock->lo_name, lock->lo_type, file, line);
792 			} else {
793 				printf(" 1st %p %s (%s) @ %s:%d\n",
794 				    lock2->li_lock, lock2->li_lock->lo_name,
795 				    lock2->li_lock->lo_type, lock2->li_file,
796 				    lock2->li_line);
797 				printf(" 2nd %p %s (%s) @ %s:%d\n",
798 				    lock1->li_lock, lock1->li_lock->lo_name,
799 				    lock1->li_lock->lo_type, lock1->li_file,
800 				    lock1->li_line);
801 				printf(" 3rd %p %s (%s) @ %s:%d\n", lock,
802 				    lock->lo_name, lock->lo_type, file, line);
803 			}
804 #ifdef DDB
805 			go_into_ddb = 1;
806 #endif
807 			goto out;
808 		}
809 	}
810 	lock1 = &(*lock_list)->ll_children[(*lock_list)->ll_count - 1];
811 	/*
812 	 * Don't build a new relationship between a sleepable lock and
813 	 * Giant if it is the wrong direction.  The real lock order is that
814 	 * sleepable locks come before Giant.
815 	 */
816 	if (!(lock1->li_lock == &Giant.mtx_object &&
817 	    (lock->lo_flags & LO_SLEEPABLE) != 0)) {
818 		CTR3(KTR_WITNESS, "%s: adding %s as a child of %s", __func__,
819 		    lock->lo_type, lock1->li_lock->lo_type);
820 		if (!itismychild(lock1->li_lock->lo_witness, w))
821 			/* Witness is dead. */
822 			return;
823 	}
824 	mtx_unlock_spin(&w_mtx);
825 
826 out:
827 #ifdef DDB
828 	if (go_into_ddb) {
829 		if (witness_trace)
830 			backtrace();
831 		if (witness_ddb)
832 			Debugger(__func__);
833 	}
834 #endif
835 	w->w_file = file;
836 	w->w_line = line;
837 
838 	lle = *lock_list;
839 	if (lle == NULL || lle->ll_count == LOCK_NCHILDREN) {
840 		lle = witness_lock_list_get();
841 		if (lle == NULL)
842 			return;
843 		lle->ll_next = *lock_list;
844 		CTR3(KTR_WITNESS, "%s: pid %d added lle %p", __func__,
845 		    td->td_proc->p_pid, lle);
846 		*lock_list = lle;
847 	}
848 	lock1 = &lle->ll_children[lle->ll_count++];
849 	lock1->li_lock = lock;
850 	lock1->li_line = line;
851 	lock1->li_file = file;
852 	if ((flags & LOP_EXCLUSIVE) != 0)
853 		lock1->li_flags = LI_EXCLUSIVE;
854 	else
855 		lock1->li_flags = 0;
856 	CTR4(KTR_WITNESS, "%s: pid %d added %s as lle[%d]", __func__,
857 	    td->td_proc->p_pid, lock->lo_name, lle->ll_count - 1);
858 }
859 
860 void
861 witness_upgrade(struct lock_object *lock, int flags, const char *file, int line)
862 {
863 	struct lock_instance *instance;
864 	struct lock_class *class;
865 
866 	KASSERT(!witness_cold, ("%s: witness_cold", __func__));
867 	if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL)
868 		return;
869 	class = lock->lo_class;
870 	file = fixup_filename(file);
871 	if ((lock->lo_flags & LO_UPGRADABLE) == 0)
872 		panic("upgrade of non-upgradable lock (%s) %s @ %s:%d",
873 		    class->lc_name, lock->lo_name, file, line);
874 	if ((flags & LOP_TRYLOCK) == 0)
875 		panic("non-try upgrade of lock (%s) %s @ %s:%d", class->lc_name,
876 		    lock->lo_name, file, line);
877 	if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0)
878 		panic("upgrade of non-sleep lock (%s) %s @ %s:%d",
879 		    class->lc_name, lock->lo_name, file, line);
880 	instance = find_instance(curthread->td_sleeplocks, lock);
881 	if (instance == NULL)
882 		panic("upgrade of unlocked lock (%s) %s @ %s:%d",
883 		    class->lc_name, lock->lo_name, file, line);
884 	if ((instance->li_flags & LI_EXCLUSIVE) != 0)
885 		panic("upgrade of exclusive lock (%s) %s @ %s:%d",
886 		    class->lc_name, lock->lo_name, file, line);
887 	if ((instance->li_flags & LI_RECURSEMASK) != 0)
888 		panic("upgrade of recursed lock (%s) %s r=%d @ %s:%d",
889 		    class->lc_name, lock->lo_name,
890 		    instance->li_flags & LI_RECURSEMASK, file, line);
891 	instance->li_flags |= LI_EXCLUSIVE;
892 }
893 
894 void
895 witness_downgrade(struct lock_object *lock, int flags, const char *file,
896     int line)
897 {
898 	struct lock_instance *instance;
899 	struct lock_class *class;
900 
901 	KASSERT(!witness_cold, ("%s: witness_cold", __func__));
902 	if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL)
903 		return;
904 	class = lock->lo_class;
905 	file = fixup_filename(file);
906 	if ((lock->lo_flags & LO_UPGRADABLE) == 0)
907 		panic("downgrade of non-upgradable lock (%s) %s @ %s:%d",
908 		    class->lc_name, lock->lo_name, file, line);
909 	if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0)
910 		panic("downgrade of non-sleep lock (%s) %s @ %s:%d",
911 		    class->lc_name, lock->lo_name, file, line);
912 	instance = find_instance(curthread->td_sleeplocks, lock);
913 	if (instance == NULL)
914 		panic("downgrade of unlocked lock (%s) %s @ %s:%d",
915 		    class->lc_name, lock->lo_name, file, line);
916 	if ((instance->li_flags & LI_EXCLUSIVE) == 0)
917 		panic("downgrade of shared lock (%s) %s @ %s:%d",
918 		    class->lc_name, lock->lo_name, file, line);
919 	if ((instance->li_flags & LI_RECURSEMASK) != 0)
920 		panic("downgrade of recursed lock (%s) %s r=%d @ %s:%d",
921 		    class->lc_name, lock->lo_name,
922 		    instance->li_flags & LI_RECURSEMASK, file, line);
923 	instance->li_flags &= ~LI_EXCLUSIVE;
924 }
925 
926 void
927 witness_unlock(struct lock_object *lock, int flags, const char *file, int line)
928 {
929 	struct lock_list_entry **lock_list, *lle;
930 	struct lock_instance *instance;
931 	struct lock_class *class;
932 	struct thread *td;
933 	register_t s;
934 	int i, j;
935 
936 	if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL ||
937 	    panicstr != NULL)
938 		return;
939 	td = curthread;
940 	class = lock->lo_class;
941 	file = fixup_filename(file);
942 	if (class->lc_flags & LC_SLEEPLOCK)
943 		lock_list = &td->td_sleeplocks;
944 	else
945 		lock_list = PCPU_PTR(spinlocks);
946 	for (; *lock_list != NULL; lock_list = &(*lock_list)->ll_next)
947 		for (i = 0; i < (*lock_list)->ll_count; i++) {
948 			instance = &(*lock_list)->ll_children[i];
949 			if (instance->li_lock == lock) {
950 				if ((instance->li_flags & LI_EXCLUSIVE) != 0 &&
951 				    (flags & LOP_EXCLUSIVE) == 0) {
952 					printf(
953 					"shared unlock of (%s) %s @ %s:%d\n",
954 					    class->lc_name, lock->lo_name,
955 					    file, line);
956 					printf(
957 					"while exclusively locked from %s:%d\n",
958 					    instance->li_file,
959 					    instance->li_line);
960 					panic("excl->ushare");
961 				}
962 				if ((instance->li_flags & LI_EXCLUSIVE) == 0 &&
963 				    (flags & LOP_EXCLUSIVE) != 0) {
964 					printf(
965 					"exclusive unlock of (%s) %s @ %s:%d\n",
966 					    class->lc_name, lock->lo_name,
967 					    file, line);
968 					printf(
969 					"while share locked from %s:%d\n",
970 					    instance->li_file,
971 					    instance->li_line);
972 					panic("share->uexcl");
973 				}
974 				/* If we are recursed, unrecurse. */
975 				if ((instance->li_flags & LI_RECURSEMASK) > 0) {
976 					CTR4(KTR_WITNESS,
977 				    "%s: pid %d unrecursed on %s r=%d", __func__,
978 					    td->td_proc->p_pid,
979 					    instance->li_lock->lo_name,
980 					    instance->li_flags);
981 					instance->li_flags--;
982 					return;
983 				}
984 				s = intr_disable();
985 				CTR4(KTR_WITNESS,
986 				    "%s: pid %d removed %s from lle[%d]", __func__,
987 				    td->td_proc->p_pid,
988 				    instance->li_lock->lo_name,
989 				    (*lock_list)->ll_count - 1);
990 				for (j = i; j < (*lock_list)->ll_count - 1; j++)
991 					(*lock_list)->ll_children[j] =
992 					    (*lock_list)->ll_children[j + 1];
993 				(*lock_list)->ll_count--;
994 				intr_restore(s);
995 				if ((*lock_list)->ll_count == 0) {
996 					lle = *lock_list;
997 					*lock_list = lle->ll_next;
998 					CTR3(KTR_WITNESS,
999 					    "%s: pid %d removed lle %p", __func__,
1000 					    td->td_proc->p_pid, lle);
1001 					witness_lock_list_free(lle);
1002 				}
1003 				return;
1004 			}
1005 		}
1006 	panic("lock (%s) %s not locked @ %s:%d", class->lc_name, lock->lo_name,
1007 	    file, line);
1008 }
1009 
1010 /*
1011  * Warn if any locks other than 'lock' are held.  Flags can be passed in to
1012  * exempt Giant and sleepable locks from the checks as well.  If any
1013  * non-exempt locks are held, then a supplied message is printed to the
1014  * console along with a list of the offending locks.  If indicated in the
1015  * flags then a failure results in a panic as well.
1016  */
1017 int
1018 witness_warn(int flags, struct lock_object *lock, const char *fmt, ...)
1019 {
1020 	struct lock_list_entry *lle;
1021 	struct lock_instance *lock1;
1022 	struct thread *td;
1023 	va_list ap;
1024 	int i, n;
1025 
1026 	if (witness_cold || witness_watch == 0 || panicstr != NULL)
1027 		return (0);
1028 	n = 0;
1029 	td = curthread;
1030 	for (lle = td->td_sleeplocks; lle != NULL; lle = lle->ll_next)
1031 		for (i = lle->ll_count - 1; i >= 0; i--) {
1032 			lock1 = &lle->ll_children[i];
1033 			if (lock1->li_lock == lock)
1034 				continue;
1035 			if (flags & WARN_GIANTOK &&
1036 			    lock1->li_lock == &Giant.mtx_object)
1037 				continue;
1038 			if (flags & WARN_SLEEPOK &&
1039 			    (lock1->li_lock->lo_flags & LO_SLEEPABLE) != 0)
1040 				continue;
1041 			if (n == 0) {
1042 				va_start(ap, fmt);
1043 				vprintf(fmt, ap);
1044 				va_end(ap);
1045 				printf(" with the following");
1046 				if (flags & WARN_SLEEPOK)
1047 					printf(" non-sleepable");
1048 				printf("locks held:\n");
1049 			}
1050 			n++;
1051 			witness_list_lock(lock1);
1052 		}
1053 	if (PCPU_GET(spinlocks) != NULL) {
1054 		/*
1055 		 * Since we already hold a spinlock preemption is
1056 		 * already blocked.
1057 		 */
1058 		if (n == 0) {
1059 			va_start(ap, fmt);
1060 			vprintf(fmt, ap);
1061 			va_end(ap);
1062 			printf(" with the following");
1063 			if (flags & WARN_SLEEPOK)
1064 				printf(" non-sleepable");
1065 			printf("locks held:\n");
1066 		}
1067 		n += witness_list_locks(PCPU_PTR(spinlocks));
1068 	}
1069 	if (flags & WARN_PANIC && n)
1070 		panic("witness_warn");
1071 #ifdef DDB
1072 	else if (witness_ddb && n)
1073 		Debugger(__func__);
1074 #endif
1075 	return (n);
1076 }
1077 
1078 const char *
1079 witness_file(struct lock_object *lock)
1080 {
1081 	struct witness *w;
1082 
1083 	if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL)
1084 		return ("?");
1085 	w = lock->lo_witness;
1086 	return (w->w_file);
1087 }
1088 
1089 int
1090 witness_line(struct lock_object *lock)
1091 {
1092 	struct witness *w;
1093 
1094 	if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL)
1095 		return (0);
1096 	w = lock->lo_witness;
1097 	return (w->w_line);
1098 }
1099 
1100 static struct witness *
1101 enroll(const char *description, struct lock_class *lock_class)
1102 {
1103 	struct witness *w;
1104 
1105 	if (!witness_watch || witness_watch == 0 || panicstr != NULL)
1106 		return (NULL);
1107 	if ((lock_class->lc_flags & LC_SPINLOCK) && witness_skipspin)
1108 		return (NULL);
1109 	mtx_lock_spin(&w_mtx);
1110 	STAILQ_FOREACH(w, &w_all, w_list) {
1111 		if (w->w_name == description || (w->w_refcount > 0 &&
1112 		    strcmp(description, w->w_name) == 0)) {
1113 			w->w_refcount++;
1114 			mtx_unlock_spin(&w_mtx);
1115 			if (lock_class != w->w_class)
1116 				panic(
1117 				"lock (%s) %s does not match earlier (%s) lock",
1118 				    description, lock_class->lc_name,
1119 				    w->w_class->lc_name);
1120 			return (w);
1121 		}
1122 	}
1123 	/*
1124 	 * This isn't quite right, as witness_cold is still 0 while we
1125 	 * enroll all the locks initialized before witness_initialize().
1126 	 */
1127 	if ((lock_class->lc_flags & LC_SPINLOCK) && !witness_cold) {
1128 		mtx_unlock_spin(&w_mtx);
1129 		panic("spin lock %s not in order list", description);
1130 	}
1131 	if ((w = witness_get()) == NULL)
1132 		return (NULL);
1133 	w->w_name = description;
1134 	w->w_class = lock_class;
1135 	w->w_refcount = 1;
1136 	STAILQ_INSERT_HEAD(&w_all, w, w_list);
1137 	if (lock_class->lc_flags & LC_SPINLOCK)
1138 		STAILQ_INSERT_HEAD(&w_spin, w, w_typelist);
1139 	else if (lock_class->lc_flags & LC_SLEEPLOCK)
1140 		STAILQ_INSERT_HEAD(&w_sleep, w, w_typelist);
1141 	else {
1142 		mtx_unlock_spin(&w_mtx);
1143 		panic("lock class %s is not sleep or spin",
1144 		    lock_class->lc_name);
1145 	}
1146 	mtx_unlock_spin(&w_mtx);
1147 	return (w);
1148 }
1149 
1150 /* Don't let the door bang you on the way out... */
1151 static int
1152 depart(struct witness *w)
1153 {
1154 	struct witness_child_list_entry *wcl, *nwcl;
1155 	struct witness_list *list;
1156 	struct witness *parent;
1157 
1158 	MPASS(w->w_refcount == 0);
1159 	if (w->w_class->lc_flags & LC_SLEEPLOCK)
1160 		list = &w_sleep;
1161 	else
1162 		list = &w_spin;
1163 	/*
1164 	 * First, we run through the entire tree looking for any
1165 	 * witnesses that the outgoing witness is a child of.  For
1166 	 * each parent that we find, we reparent all the direct
1167 	 * children of the outgoing witness to its parent.
1168 	 */
1169 	STAILQ_FOREACH(parent, list, w_typelist) {
1170 		if (!isitmychild(parent, w))
1171 			continue;
1172 		removechild(parent, w);
1173 		if (!reparentchildren(parent, w))
1174 			return (0);
1175 	}
1176 
1177 	/*
1178 	 * Now we go through and free up the child list of the
1179 	 * outgoing witness.
1180 	 */
1181 	for (wcl = w->w_children; wcl != NULL; wcl = nwcl) {
1182 		nwcl = wcl->wcl_next;
1183 		witness_child_free(wcl);
1184 	}
1185 
1186 	/*
1187 	 * Detach from various lists and free.
1188 	 */
1189 	STAILQ_REMOVE(list, w, witness, w_typelist);
1190 	STAILQ_REMOVE(&w_all, w, witness, w_list);
1191 	witness_free(w);
1192 
1193 	/* Finally, fixup the tree. */
1194 	return (rebalancetree(list));
1195 }
1196 
1197 /*
1198  * Prune an entire lock order tree.  We look for cases where a lock
1199  * is now both a descendant and a direct child of a given lock.  In
1200  * that case, we want to remove the direct child link from the tree.
1201  *
1202  * Returns false if insertchild() fails.
1203  */
1204 static int
1205 rebalancetree(struct witness_list *list)
1206 {
1207 	struct witness *child, *parent;
1208 
1209 	STAILQ_FOREACH(child, list, w_typelist) {
1210 		STAILQ_FOREACH(parent, list, w_typelist) {
1211 			if (!isitmychild(parent, child))
1212 				continue;
1213 			removechild(parent, child);
1214 			if (isitmydescendant(parent, child))
1215 				continue;
1216 			if (!insertchild(parent, child))
1217 				return (0);
1218 		}
1219 	}
1220 	witness_levelall();
1221 	return (1);
1222 }
1223 
1224 /*
1225  * Add "child" as a direct child of "parent".  Returns false if
1226  * we fail due to out of memory.
1227  */
1228 static int
1229 insertchild(struct witness *parent, struct witness *child)
1230 {
1231 	struct witness_child_list_entry **wcl;
1232 
1233 	MPASS(child != NULL && parent != NULL);
1234 
1235 	/*
1236 	 * Insert "child" after "parent"
1237 	 */
1238 	wcl = &parent->w_children;
1239 	while (*wcl != NULL && (*wcl)->wcl_count == WITNESS_NCHILDREN)
1240 		wcl = &(*wcl)->wcl_next;
1241 	if (*wcl == NULL) {
1242 		*wcl = witness_child_get();
1243 		if (*wcl == NULL)
1244 			return (0);
1245 	}
1246 	(*wcl)->wcl_children[(*wcl)->wcl_count++] = child;
1247 
1248 	return (1);
1249 }
1250 
1251 /*
1252  * Make all the direct descendants of oldparent be direct descendants
1253  * of newparent.
1254  */
1255 static int
1256 reparentchildren(struct witness *newparent, struct witness *oldparent)
1257 {
1258 	struct witness_child_list_entry *wcl;
1259 	int i;
1260 
1261 	/* Avoid making a witness a child of itself. */
1262 	MPASS(!isitmychild(oldparent, newparent));
1263 
1264 	for (wcl = oldparent->w_children; wcl != NULL; wcl = wcl->wcl_next)
1265 		for (i = 0; i < wcl->wcl_count; i++)
1266 			if (!insertchild(newparent, wcl->wcl_children[i]))
1267 				return (0);
1268 	return (1);
1269 }
1270 
1271 static int
1272 itismychild(struct witness *parent, struct witness *child)
1273 {
1274 	struct witness_list *list;
1275 
1276 	MPASS(child != NULL && parent != NULL);
1277 	if ((parent->w_class->lc_flags & (LC_SLEEPLOCK | LC_SPINLOCK)) !=
1278 	    (child->w_class->lc_flags & (LC_SLEEPLOCK | LC_SPINLOCK)))
1279 		panic(
1280 		"%s: parent (%s) and child (%s) are not the same lock type",
1281 		    __func__, parent->w_class->lc_name,
1282 		    child->w_class->lc_name);
1283 
1284 	if (!insertchild(parent, child))
1285 		return (0);
1286 
1287 	if (parent->w_class->lc_flags & LC_SLEEPLOCK)
1288 		list = &w_sleep;
1289 	else
1290 		list = &w_spin;
1291 	return (rebalancetree(list));
1292 }
1293 
1294 static void
1295 removechild(struct witness *parent, struct witness *child)
1296 {
1297 	struct witness_child_list_entry **wcl, *wcl1;
1298 	int i;
1299 
1300 	for (wcl = &parent->w_children; *wcl != NULL; wcl = &(*wcl)->wcl_next)
1301 		for (i = 0; i < (*wcl)->wcl_count; i++)
1302 			if ((*wcl)->wcl_children[i] == child)
1303 				goto found;
1304 	return;
1305 found:
1306 	(*wcl)->wcl_count--;
1307 	if ((*wcl)->wcl_count > i)
1308 		(*wcl)->wcl_children[i] =
1309 		    (*wcl)->wcl_children[(*wcl)->wcl_count];
1310 	MPASS((*wcl)->wcl_children[i] != NULL);
1311 	if ((*wcl)->wcl_count != 0)
1312 		return;
1313 	wcl1 = *wcl;
1314 	*wcl = wcl1->wcl_next;
1315 	witness_child_free(wcl1);
1316 }
1317 
1318 static int
1319 isitmychild(struct witness *parent, struct witness *child)
1320 {
1321 	struct witness_child_list_entry *wcl;
1322 	int i;
1323 
1324 	for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next) {
1325 		for (i = 0; i < wcl->wcl_count; i++) {
1326 			if (wcl->wcl_children[i] == child)
1327 				return (1);
1328 		}
1329 	}
1330 	return (0);
1331 }
1332 
1333 static int
1334 isitmydescendant(struct witness *parent, struct witness *child)
1335 {
1336 	struct witness_child_list_entry *wcl;
1337 	int i, j;
1338 
1339 	if (isitmychild(parent, child))
1340 		return (1);
1341 	j = 0;
1342 	for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next) {
1343 		MPASS(j < 1000);
1344 		for (i = 0; i < wcl->wcl_count; i++) {
1345 			if (isitmydescendant(wcl->wcl_children[i], child))
1346 				return (1);
1347 		}
1348 		j++;
1349 	}
1350 	return (0);
1351 }
1352 
1353 static void
1354 witness_levelall (void)
1355 {
1356 	struct witness_list *list;
1357 	struct witness *w, *w1;
1358 
1359 	/*
1360 	 * First clear all levels.
1361 	 */
1362 	STAILQ_FOREACH(w, &w_all, w_list) {
1363 		w->w_level = 0;
1364 	}
1365 
1366 	/*
1367 	 * Look for locks with no parent and level all their descendants.
1368 	 */
1369 	STAILQ_FOREACH(w, &w_all, w_list) {
1370 		/*
1371 		 * This is just an optimization, technically we could get
1372 		 * away just walking the all list each time.
1373 		 */
1374 		if (w->w_class->lc_flags & LC_SLEEPLOCK)
1375 			list = &w_sleep;
1376 		else
1377 			list = &w_spin;
1378 		STAILQ_FOREACH(w1, list, w_typelist) {
1379 			if (isitmychild(w1, w))
1380 				goto skip;
1381 		}
1382 		witness_leveldescendents(w, 0);
1383 	skip:
1384 		;	/* silence GCC 3.x */
1385 	}
1386 }
1387 
1388 static void
1389 witness_leveldescendents(struct witness *parent, int level)
1390 {
1391 	struct witness_child_list_entry *wcl;
1392 	int i;
1393 
1394 	if (parent->w_level < level)
1395 		parent->w_level = level;
1396 	level++;
1397 	for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next)
1398 		for (i = 0; i < wcl->wcl_count; i++)
1399 			witness_leveldescendents(wcl->wcl_children[i], level);
1400 }
1401 
1402 static void
1403 witness_displaydescendants(void(*prnt)(const char *fmt, ...),
1404 			   struct witness *parent, int indent)
1405 {
1406 	struct witness_child_list_entry *wcl;
1407 	int i, level;
1408 
1409 	level = parent->w_level;
1410 	prnt("%-2d", level);
1411 	for (i = 0; i < indent; i++)
1412 		prnt(" ");
1413 	if (parent->w_refcount > 0)
1414 		prnt("%s", parent->w_name);
1415 	else
1416 		prnt("(dead)");
1417 	if (parent->w_displayed) {
1418 		prnt(" -- (already displayed)\n");
1419 		return;
1420 	}
1421 	parent->w_displayed = 1;
1422 	if (parent->w_refcount > 0) {
1423 		if (parent->w_file != NULL)
1424 			prnt(" -- last acquired @ %s:%d", parent->w_file,
1425 			    parent->w_line);
1426 	}
1427 	prnt("\n");
1428 	for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next)
1429 		for (i = 0; i < wcl->wcl_count; i++)
1430 			    witness_displaydescendants(prnt,
1431 				wcl->wcl_children[i], indent + 1);
1432 }
1433 
1434 #ifdef BLESSING
1435 static int
1436 blessed(struct witness *w1, struct witness *w2)
1437 {
1438 	int i;
1439 	struct witness_blessed *b;
1440 
1441 	for (i = 0; i < blessed_count; i++) {
1442 		b = &blessed_list[i];
1443 		if (strcmp(w1->w_name, b->b_lock1) == 0) {
1444 			if (strcmp(w2->w_name, b->b_lock2) == 0)
1445 				return (1);
1446 			continue;
1447 		}
1448 		if (strcmp(w1->w_name, b->b_lock2) == 0)
1449 			if (strcmp(w2->w_name, b->b_lock1) == 0)
1450 				return (1);
1451 	}
1452 	return (0);
1453 }
1454 #endif
1455 
1456 static struct witness *
1457 witness_get(void)
1458 {
1459 	struct witness *w;
1460 
1461 	if (witness_watch == 0) {
1462 		mtx_unlock_spin(&w_mtx);
1463 		return (NULL);
1464 	}
1465 	if (STAILQ_EMPTY(&w_free)) {
1466 		witness_watch = 0;
1467 		mtx_unlock_spin(&w_mtx);
1468 		printf("%s: witness exhausted\n", __func__);
1469 		return (NULL);
1470 	}
1471 	w = STAILQ_FIRST(&w_free);
1472 	STAILQ_REMOVE_HEAD(&w_free, w_list);
1473 	bzero(w, sizeof(*w));
1474 	return (w);
1475 }
1476 
1477 static void
1478 witness_free(struct witness *w)
1479 {
1480 
1481 	STAILQ_INSERT_HEAD(&w_free, w, w_list);
1482 }
1483 
1484 static struct witness_child_list_entry *
1485 witness_child_get(void)
1486 {
1487 	struct witness_child_list_entry *wcl;
1488 
1489 	if (witness_watch == 0) {
1490 		mtx_unlock_spin(&w_mtx);
1491 		return (NULL);
1492 	}
1493 	wcl = w_child_free;
1494 	if (wcl == NULL) {
1495 		witness_watch = 0;
1496 		mtx_unlock_spin(&w_mtx);
1497 		printf("%s: witness exhausted\n", __func__);
1498 		return (NULL);
1499 	}
1500 	w_child_free = wcl->wcl_next;
1501 	bzero(wcl, sizeof(*wcl));
1502 	return (wcl);
1503 }
1504 
1505 static void
1506 witness_child_free(struct witness_child_list_entry *wcl)
1507 {
1508 
1509 	wcl->wcl_next = w_child_free;
1510 	w_child_free = wcl;
1511 }
1512 
1513 static struct lock_list_entry *
1514 witness_lock_list_get(void)
1515 {
1516 	struct lock_list_entry *lle;
1517 
1518 	if (witness_watch == 0)
1519 		return (NULL);
1520 	mtx_lock_spin(&w_mtx);
1521 	lle = w_lock_list_free;
1522 	if (lle == NULL) {
1523 		witness_watch = 0;
1524 		mtx_unlock_spin(&w_mtx);
1525 		printf("%s: witness exhausted\n", __func__);
1526 		return (NULL);
1527 	}
1528 	w_lock_list_free = lle->ll_next;
1529 	mtx_unlock_spin(&w_mtx);
1530 	bzero(lle, sizeof(*lle));
1531 	return (lle);
1532 }
1533 
1534 static void
1535 witness_lock_list_free(struct lock_list_entry *lle)
1536 {
1537 
1538 	mtx_lock_spin(&w_mtx);
1539 	lle->ll_next = w_lock_list_free;
1540 	w_lock_list_free = lle;
1541 	mtx_unlock_spin(&w_mtx);
1542 }
1543 
1544 static struct lock_instance *
1545 find_instance(struct lock_list_entry *lock_list, struct lock_object *lock)
1546 {
1547 	struct lock_list_entry *lle;
1548 	struct lock_instance *instance;
1549 	int i;
1550 
1551 	for (lle = lock_list; lle != NULL; lle = lle->ll_next)
1552 		for (i = lle->ll_count - 1; i >= 0; i--) {
1553 			instance = &lle->ll_children[i];
1554 			if (instance->li_lock == lock)
1555 				return (instance);
1556 		}
1557 	return (NULL);
1558 }
1559 
1560 static void
1561 witness_list_lock(struct lock_instance *instance)
1562 {
1563 	struct lock_object *lock;
1564 
1565 	lock = instance->li_lock;
1566 	printf("%s %s %s", (instance->li_flags & LI_EXCLUSIVE) != 0 ?
1567 	    "exclusive" : "shared", lock->lo_class->lc_name, lock->lo_name);
1568 	if (lock->lo_type != lock->lo_name)
1569 		printf(" (%s)", lock->lo_type);
1570 	printf(" r = %d (%p) locked @ %s:%d\n",
1571 	    instance->li_flags & LI_RECURSEMASK, lock, instance->li_file,
1572 	    instance->li_line);
1573 }
1574 
1575 int
1576 witness_list_locks(struct lock_list_entry **lock_list)
1577 {
1578 	struct lock_list_entry *lle;
1579 	int i, nheld;
1580 
1581 	nheld = 0;
1582 	for (lle = *lock_list; lle != NULL; lle = lle->ll_next)
1583 		for (i = lle->ll_count - 1; i >= 0; i--) {
1584 			witness_list_lock(&lle->ll_children[i]);
1585 			nheld++;
1586 		}
1587 	return (nheld);
1588 }
1589 
1590 void
1591 witness_save(struct lock_object *lock, const char **filep, int *linep)
1592 {
1593 	struct lock_instance *instance;
1594 
1595 	KASSERT(!witness_cold, ("%s: witness_cold", __func__));
1596 	if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL)
1597 		return;
1598 	if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0)
1599 		panic("%s: lock (%s) %s is not a sleep lock", __func__,
1600 		    lock->lo_class->lc_name, lock->lo_name);
1601 	instance = find_instance(curthread->td_sleeplocks, lock);
1602 	if (instance == NULL)
1603 		panic("%s: lock (%s) %s not locked", __func__,
1604 		    lock->lo_class->lc_name, lock->lo_name);
1605 	*filep = instance->li_file;
1606 	*linep = instance->li_line;
1607 }
1608 
1609 void
1610 witness_restore(struct lock_object *lock, const char *file, int line)
1611 {
1612 	struct lock_instance *instance;
1613 
1614 	KASSERT(!witness_cold, ("%s: witness_cold", __func__));
1615 	if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL)
1616 		return;
1617 	if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0)
1618 		panic("%s: lock (%s) %s is not a sleep lock", __func__,
1619 		    lock->lo_class->lc_name, lock->lo_name);
1620 	instance = find_instance(curthread->td_sleeplocks, lock);
1621 	if (instance == NULL)
1622 		panic("%s: lock (%s) %s not locked", __func__,
1623 		    lock->lo_class->lc_name, lock->lo_name);
1624 	lock->lo_witness->w_file = file;
1625 	lock->lo_witness->w_line = line;
1626 	instance->li_file = file;
1627 	instance->li_line = line;
1628 }
1629 
1630 void
1631 witness_assert(struct lock_object *lock, int flags, const char *file, int line)
1632 {
1633 #ifdef INVARIANT_SUPPORT
1634 	struct lock_instance *instance;
1635 
1636 	if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL)
1637 		return;
1638 	if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) != 0)
1639 		instance = find_instance(curthread->td_sleeplocks, lock);
1640 	else if ((lock->lo_class->lc_flags & LC_SPINLOCK) != 0)
1641 		instance = find_instance(PCPU_GET(spinlocks), lock);
1642 	else {
1643 		panic("Lock (%s) %s is not sleep or spin!",
1644 		    lock->lo_class->lc_name, lock->lo_name);
1645 		return;
1646 	}
1647 	file = fixup_filename(file);
1648 	switch (flags) {
1649 	case LA_UNLOCKED:
1650 		if (instance != NULL)
1651 			panic("Lock (%s) %s locked @ %s:%d.",
1652 			    lock->lo_class->lc_name, lock->lo_name, file, line);
1653 		break;
1654 	case LA_LOCKED:
1655 	case LA_LOCKED | LA_RECURSED:
1656 	case LA_LOCKED | LA_NOTRECURSED:
1657 	case LA_SLOCKED:
1658 	case LA_SLOCKED | LA_RECURSED:
1659 	case LA_SLOCKED | LA_NOTRECURSED:
1660 	case LA_XLOCKED:
1661 	case LA_XLOCKED | LA_RECURSED:
1662 	case LA_XLOCKED | LA_NOTRECURSED:
1663 		if (instance == NULL) {
1664 			panic("Lock (%s) %s not locked @ %s:%d.",
1665 			    lock->lo_class->lc_name, lock->lo_name, file, line);
1666 			break;
1667 		}
1668 		if ((flags & LA_XLOCKED) != 0 &&
1669 		    (instance->li_flags & LI_EXCLUSIVE) == 0)
1670 			panic("Lock (%s) %s not exclusively locked @ %s:%d.",
1671 			    lock->lo_class->lc_name, lock->lo_name, file, line);
1672 		if ((flags & LA_SLOCKED) != 0 &&
1673 		    (instance->li_flags & LI_EXCLUSIVE) != 0)
1674 			panic("Lock (%s) %s exclusively locked @ %s:%d.",
1675 			    lock->lo_class->lc_name, lock->lo_name, file, line);
1676 		if ((flags & LA_RECURSED) != 0 &&
1677 		    (instance->li_flags & LI_RECURSEMASK) == 0)
1678 			panic("Lock (%s) %s not recursed @ %s:%d.",
1679 			    lock->lo_class->lc_name, lock->lo_name, file, line);
1680 		if ((flags & LA_NOTRECURSED) != 0 &&
1681 		    (instance->li_flags & LI_RECURSEMASK) != 0)
1682 			panic("Lock (%s) %s recursed @ %s:%d.",
1683 			    lock->lo_class->lc_name, lock->lo_name, file, line);
1684 		break;
1685 	default:
1686 		panic("Invalid lock assertion at %s:%d.", file, line);
1687 
1688 	}
1689 #endif	/* INVARIANT_SUPPORT */
1690 }
1691 
1692 #ifdef DDB
1693 static void
1694 witness_list(struct thread *td)
1695 {
1696 
1697 	KASSERT(!witness_cold, ("%s: witness_cold", __func__));
1698 	KASSERT(db_active, ("%s: not in the debugger", __func__));
1699 
1700 	if (witness_watch == 0)
1701 		return;
1702 
1703 	witness_list_locks(&td->td_sleeplocks);
1704 
1705 	/*
1706 	 * We only handle spinlocks if td == curthread.  This is somewhat broken
1707 	 * if td is currently executing on some other CPU and holds spin locks
1708 	 * as we won't display those locks.  If we had a MI way of getting
1709 	 * the per-cpu data for a given cpu then we could use
1710 	 * td->td_kse->ke_oncpu to get the list of spinlocks for this thread
1711 	 * and "fix" this.
1712 	 *
1713 	 * That still wouldn't really fix this unless we locked sched_lock
1714 	 * or stopped the other CPU to make sure it wasn't changing the list
1715 	 * out from under us.  It is probably best to just not try to handle
1716 	 * threads on other CPU's for now.
1717 	 */
1718 	if (td == curthread && PCPU_GET(spinlocks) != NULL)
1719 		witness_list_locks(PCPU_PTR(spinlocks));
1720 }
1721 
1722 DB_SHOW_COMMAND(locks, db_witness_list)
1723 {
1724 	struct thread *td;
1725 	pid_t pid;
1726 	struct proc *p;
1727 
1728 	if (have_addr) {
1729 		pid = (addr % 16) + ((addr >> 4) % 16) * 10 +
1730 		    ((addr >> 8) % 16) * 100 + ((addr >> 12) % 16) * 1000 +
1731 		    ((addr >> 16) % 16) * 10000;
1732 		/* sx_slock(&allproc_lock); */
1733 		FOREACH_PROC_IN_SYSTEM(p) {
1734 			if (p->p_pid == pid)
1735 				break;
1736 		}
1737 		/* sx_sunlock(&allproc_lock); */
1738 		if (p == NULL) {
1739 			db_printf("pid %d not found\n", pid);
1740 			return;
1741 		}
1742 		FOREACH_THREAD_IN_PROC(p, td) {
1743 			witness_list(td);
1744 		}
1745 	} else {
1746 		td = curthread;
1747 		witness_list(td);
1748 	}
1749 }
1750 
1751 DB_SHOW_COMMAND(witness, db_witness_display)
1752 {
1753 
1754 	witness_display(db_printf);
1755 }
1756 #endif
1757