xref: /freebsd/sys/kern/kern_lock.c (revision 87569f75a91f298c52a71823c04d41cf53c88889)
1 /*-
2  * Copyright (c) 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Copyright (C) 1997
6  *	John S. Dyson.  All rights reserved.
7  *
8  * This code contains ideas from software contributed to Berkeley by
9  * Avadis Tevanian, Jr., Michael Wayne Young, and the Mach Operating
10  * System project at Carnegie-Mellon University.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by the University of
23  *	California, Berkeley and its contributors.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)kern_lock.c	8.18 (Berkeley) 5/21/95
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include <sys/param.h>
47 #include <sys/kdb.h>
48 #include <sys/kernel.h>
49 #include <sys/ktr.h>
50 #include <sys/lock.h>
51 #include <sys/lockmgr.h>
52 #include <sys/mutex.h>
53 #include <sys/proc.h>
54 #include <sys/systm.h>
55 #ifdef DEBUG_LOCKS
56 #include <sys/stack.h>
57 #endif
58 #include <sys/kdb.h>
59 
60 /*
61  * Locking primitives implementation.
62  * Locks provide shared/exclusive sychronization.
63  */
64 
65 #define	COUNT(td, x)	if ((td)) (td)->td_locks += (x)
66 #define LK_ALL (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | \
67 	LK_SHARE_NONZERO | LK_WAIT_NONZERO)
68 
69 static int acquire(struct lock **lkpp, int extflags, int wanted);
70 static int acquiredrain(struct lock *lkp, int extflags) ;
71 
72 static __inline void
73 sharelock(struct thread *td, struct lock *lkp, int incr) {
74 	lkp->lk_flags |= LK_SHARE_NONZERO;
75 	lkp->lk_sharecount += incr;
76 	COUNT(td, incr);
77 }
78 
79 static __inline void
80 shareunlock(struct thread *td, struct lock *lkp, int decr) {
81 
82 	KASSERT(lkp->lk_sharecount >= decr, ("shareunlock: count < decr"));
83 
84 	COUNT(td, -decr);
85 	if (lkp->lk_sharecount == decr) {
86 		lkp->lk_flags &= ~LK_SHARE_NONZERO;
87 		if (lkp->lk_flags & (LK_WANT_UPGRADE | LK_WANT_EXCL)) {
88 			wakeup(lkp);
89 		}
90 		lkp->lk_sharecount = 0;
91 	} else {
92 		lkp->lk_sharecount -= decr;
93 	}
94 }
95 
96 static int
97 acquire(struct lock **lkpp, int extflags, int wanted)
98 {
99 	struct lock *lkp = *lkpp;
100 	int error;
101 	CTR3(KTR_LOCK,
102 	    "acquire(): lkp == %p, extflags == 0x%x, wanted == 0x%x",
103 	    lkp, extflags, wanted);
104 
105 	if ((extflags & LK_NOWAIT) && (lkp->lk_flags & wanted))
106 		return EBUSY;
107 	error = 0;
108 	while ((lkp->lk_flags & wanted) != 0) {
109 		CTR2(KTR_LOCK,
110 		    "acquire(): lkp == %p, lk_flags == 0x%x sleeping",
111 		    lkp, lkp->lk_flags);
112 		lkp->lk_flags |= LK_WAIT_NONZERO;
113 		lkp->lk_waitcount++;
114 		error = msleep(lkp, lkp->lk_interlock, lkp->lk_prio,
115 		    lkp->lk_wmesg,
116 		    ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0));
117 		lkp->lk_waitcount--;
118 		if (lkp->lk_waitcount == 0)
119 			lkp->lk_flags &= ~LK_WAIT_NONZERO;
120 		if (error)
121 			break;
122 		if (extflags & LK_SLEEPFAIL) {
123 			error = ENOLCK;
124 			break;
125 		}
126 		if (lkp->lk_newlock != NULL) {
127 			mtx_lock(lkp->lk_newlock->lk_interlock);
128 			mtx_unlock(lkp->lk_interlock);
129 			if (lkp->lk_waitcount == 0)
130 				wakeup((void *)(&lkp->lk_newlock));
131 			*lkpp = lkp = lkp->lk_newlock;
132 		}
133 	}
134 	mtx_assert(lkp->lk_interlock, MA_OWNED);
135 	return (error);
136 }
137 
138 /*
139  * Set, change, or release a lock.
140  *
141  * Shared requests increment the shared count. Exclusive requests set the
142  * LK_WANT_EXCL flag (preventing further shared locks), and wait for already
143  * accepted shared locks and shared-to-exclusive upgrades to go away.
144  */
145 int
146 lockmgr(lkp, flags, interlkp, td)
147 	struct lock *lkp;
148 	u_int flags;
149 	struct mtx *interlkp;
150 	struct thread *td;
151 {
152 	int error;
153 	struct thread *thr;
154 	int extflags, lockflags;
155 
156 	error = 0;
157 	if (td == NULL)
158 		thr = LK_KERNPROC;
159 	else
160 		thr = td;
161 
162 	if ((flags & LK_INTERNAL) == 0)
163 		mtx_lock(lkp->lk_interlock);
164 	CTR6(KTR_LOCK,
165 	    "lockmgr(): lkp == %p (lk_wmesg == \"%s\"), owner == %p, exclusivecount == %d, flags == 0x%x, "
166 	    "td == %p", lkp, lkp->lk_wmesg, lkp->lk_lockholder,
167 	    lkp->lk_exclusivecount, flags, td);
168 #ifdef DEBUG_LOCKS
169 	{
170 		struct stack stack; /* XXX */
171 		stack_save(&stack);
172 		CTRSTACK(KTR_LOCK, &stack, 0, 1);
173 	}
174 #endif
175 
176 	if (flags & LK_INTERLOCK) {
177 		mtx_assert(interlkp, MA_OWNED | MA_NOTRECURSED);
178 		mtx_unlock(interlkp);
179 	}
180 
181 	if ((flags & (LK_NOWAIT|LK_RELEASE)) == 0)
182 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
183 		    &lkp->lk_interlock->mtx_object,
184 		    "Acquiring lockmgr lock \"%s\"", lkp->lk_wmesg);
185 
186 	if (panicstr != NULL) {
187 		mtx_unlock(lkp->lk_interlock);
188 		return (0);
189 	}
190 	if ((lkp->lk_flags & LK_NOSHARE) &&
191 	    (flags & LK_TYPE_MASK) == LK_SHARED) {
192 		flags &= ~LK_TYPE_MASK;
193 		flags |= LK_EXCLUSIVE;
194 	}
195 	extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK;
196 
197 	switch (flags & LK_TYPE_MASK) {
198 
199 	case LK_SHARED:
200 		/*
201 		 * If we are not the exclusive lock holder, we have to block
202 		 * while there is an exclusive lock holder or while an
203 		 * exclusive lock request or upgrade request is in progress.
204 		 *
205 		 * However, if TDP_DEADLKTREAT is set, we override exclusive
206 		 * lock requests or upgrade requests ( but not the exclusive
207 		 * lock itself ).
208 		 */
209 		if (lkp->lk_lockholder != thr) {
210 			lockflags = LK_HAVE_EXCL;
211 			if (td != NULL && !(td->td_pflags & TDP_DEADLKTREAT))
212 				lockflags |= LK_WANT_EXCL | LK_WANT_UPGRADE;
213 			error = acquire(&lkp, extflags, lockflags);
214 			if (error)
215 				break;
216 			sharelock(td, lkp, 1);
217 #if defined(DEBUG_LOCKS)
218 			stack_save(&lkp->lk_stack);
219 #endif
220 			break;
221 		}
222 		/*
223 		 * We hold an exclusive lock, so downgrade it to shared.
224 		 * An alternative would be to fail with EDEADLK.
225 		 */
226 		sharelock(td, lkp, 1);
227 		/* FALLTHROUGH downgrade */
228 
229 	case LK_DOWNGRADE:
230 		KASSERT(lkp->lk_lockholder == thr && lkp->lk_exclusivecount != 0,
231 			("lockmgr: not holding exclusive lock "
232 			"(owner thread (%p) != thread (%p), exlcnt (%d) != 0",
233 			lkp->lk_lockholder, thr, lkp->lk_exclusivecount));
234 		sharelock(td, lkp, lkp->lk_exclusivecount);
235 		COUNT(td, -lkp->lk_exclusivecount);
236 		lkp->lk_exclusivecount = 0;
237 		lkp->lk_flags &= ~LK_HAVE_EXCL;
238 		lkp->lk_lockholder = LK_NOPROC;
239 		if (lkp->lk_waitcount)
240 			wakeup((void *)lkp);
241 		break;
242 
243 	case LK_EXCLUPGRADE:
244 		/*
245 		 * If another process is ahead of us to get an upgrade,
246 		 * then we want to fail rather than have an intervening
247 		 * exclusive access.
248 		 */
249 		if (lkp->lk_flags & LK_WANT_UPGRADE) {
250 			shareunlock(td, lkp, 1);
251 			error = EBUSY;
252 			break;
253 		}
254 		/* FALLTHROUGH normal upgrade */
255 
256 	case LK_UPGRADE:
257 		/*
258 		 * Upgrade a shared lock to an exclusive one. If another
259 		 * shared lock has already requested an upgrade to an
260 		 * exclusive lock, our shared lock is released and an
261 		 * exclusive lock is requested (which will be granted
262 		 * after the upgrade). If we return an error, the file
263 		 * will always be unlocked.
264 		 */
265 		if (lkp->lk_lockholder == thr)
266 			panic("lockmgr: upgrade exclusive lock");
267 		if (lkp->lk_sharecount <= 0)
268 			panic("lockmgr: upgrade without shared");
269 		shareunlock(td, lkp, 1);
270 		/*
271 		 * If we are just polling, check to see if we will block.
272 		 */
273 		if ((extflags & LK_NOWAIT) &&
274 		    ((lkp->lk_flags & LK_WANT_UPGRADE) ||
275 		     lkp->lk_sharecount > 1)) {
276 			error = EBUSY;
277 			break;
278 		}
279 		if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) {
280 			/*
281 			 * We are first shared lock to request an upgrade, so
282 			 * request upgrade and wait for the shared count to
283 			 * drop to zero, then take exclusive lock.
284 			 */
285 			lkp->lk_flags |= LK_WANT_UPGRADE;
286 			error = acquire(&lkp, extflags, LK_SHARE_NONZERO);
287 			lkp->lk_flags &= ~LK_WANT_UPGRADE;
288 
289 			if (error) {
290 			         if ((lkp->lk_flags & ( LK_WANT_EXCL | LK_WAIT_NONZERO)) == (LK_WANT_EXCL | LK_WAIT_NONZERO))
291 			                   wakeup((void *)lkp);
292 			         break;
293 			}
294 			if (lkp->lk_exclusivecount != 0)
295 				panic("lockmgr: non-zero exclusive count");
296 			lkp->lk_flags |= LK_HAVE_EXCL;
297 			lkp->lk_lockholder = thr;
298 			lkp->lk_exclusivecount = 1;
299 			COUNT(td, 1);
300 #if defined(DEBUG_LOCKS)
301 			stack_save(&lkp->lk_stack);
302 #endif
303 			break;
304 		}
305 		/*
306 		 * Someone else has requested upgrade. Release our shared
307 		 * lock, awaken upgrade requestor if we are the last shared
308 		 * lock, then request an exclusive lock.
309 		 */
310 		if ( (lkp->lk_flags & (LK_SHARE_NONZERO|LK_WAIT_NONZERO)) ==
311 			LK_WAIT_NONZERO)
312 			wakeup((void *)lkp);
313 		/* FALLTHROUGH exclusive request */
314 
315 	case LK_EXCLUSIVE:
316 		if (lkp->lk_lockholder == thr && thr != LK_KERNPROC) {
317 			/*
318 			 *	Recursive lock.
319 			 */
320 			if ((extflags & (LK_NOWAIT | LK_CANRECURSE)) == 0)
321 				panic("lockmgr: locking against myself");
322 			if ((extflags & LK_CANRECURSE) != 0) {
323 				lkp->lk_exclusivecount++;
324 				COUNT(td, 1);
325 				break;
326 			}
327 		}
328 		/*
329 		 * If we are just polling, check to see if we will sleep.
330 		 */
331 		if ((extflags & LK_NOWAIT) &&
332 		    (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO))) {
333 			error = EBUSY;
334 			break;
335 		}
336 		/*
337 		 * Try to acquire the want_exclusive flag.
338 		 */
339 		error = acquire(&lkp, extflags, (LK_HAVE_EXCL | LK_WANT_EXCL));
340 		if (error)
341 			break;
342 		lkp->lk_flags |= LK_WANT_EXCL;
343 		/*
344 		 * Wait for shared locks and upgrades to finish.
345 		 */
346 		error = acquire(&lkp, extflags, LK_HAVE_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO);
347 		lkp->lk_flags &= ~LK_WANT_EXCL;
348 		if (error) {
349 			if (lkp->lk_flags & LK_WAIT_NONZERO)
350 			         wakeup((void *)lkp);
351 			break;
352 		}
353 		lkp->lk_flags |= LK_HAVE_EXCL;
354 		lkp->lk_lockholder = thr;
355 		if (lkp->lk_exclusivecount != 0)
356 			panic("lockmgr: non-zero exclusive count");
357 		lkp->lk_exclusivecount = 1;
358 		COUNT(td, 1);
359 #if defined(DEBUG_LOCKS)
360 		stack_save(&lkp->lk_stack);
361 #endif
362 		break;
363 
364 	case LK_RELEASE:
365 		if (lkp->lk_exclusivecount != 0) {
366 			if (lkp->lk_lockholder != thr &&
367 			    lkp->lk_lockholder != LK_KERNPROC) {
368 				panic("lockmgr: thread %p, not %s %p unlocking",
369 				    thr, "exclusive lock holder",
370 				    lkp->lk_lockholder);
371 			}
372 			if (lkp->lk_lockholder != LK_KERNPROC)
373 				COUNT(td, -1);
374 			if (lkp->lk_exclusivecount == 1) {
375 				lkp->lk_flags &= ~LK_HAVE_EXCL;
376 				lkp->lk_lockholder = LK_NOPROC;
377 				lkp->lk_exclusivecount = 0;
378 			} else {
379 				lkp->lk_exclusivecount--;
380 			}
381 		} else if (lkp->lk_flags & LK_SHARE_NONZERO)
382 			shareunlock(td, lkp, 1);
383 		else  {
384 			printf("lockmgr: thread %p unlocking unheld lock\n",
385 			    thr);
386 			kdb_backtrace();
387 		}
388 
389 		if (lkp->lk_flags & LK_WAIT_NONZERO)
390 			wakeup((void *)lkp);
391 		break;
392 
393 	case LK_DRAIN:
394 		/*
395 		 * Check that we do not already hold the lock, as it can
396 		 * never drain if we do. Unfortunately, we have no way to
397 		 * check for holding a shared lock, but at least we can
398 		 * check for an exclusive one.
399 		 */
400 		if (lkp->lk_lockholder == thr)
401 			panic("lockmgr: draining against myself");
402 
403 		error = acquiredrain(lkp, extflags);
404 		if (error)
405 			break;
406 		lkp->lk_flags |= LK_DRAINING | LK_HAVE_EXCL;
407 		lkp->lk_lockholder = thr;
408 		lkp->lk_exclusivecount = 1;
409 		COUNT(td, 1);
410 #if defined(DEBUG_LOCKS)
411 		stack_save(&lkp->lk_stack);
412 #endif
413 		break;
414 
415 	default:
416 		mtx_unlock(lkp->lk_interlock);
417 		panic("lockmgr: unknown locktype request %d",
418 		    flags & LK_TYPE_MASK);
419 		/* NOTREACHED */
420 	}
421 	if ((lkp->lk_flags & LK_WAITDRAIN) &&
422 	    (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE |
423 		LK_SHARE_NONZERO | LK_WAIT_NONZERO)) == 0) {
424 		lkp->lk_flags &= ~LK_WAITDRAIN;
425 		wakeup((void *)&lkp->lk_flags);
426 	}
427 	mtx_unlock(lkp->lk_interlock);
428 	return (error);
429 }
430 
431 static int
432 acquiredrain(struct lock *lkp, int extflags) {
433 	int error;
434 
435 	if ((extflags & LK_NOWAIT) && (lkp->lk_flags & LK_ALL)) {
436 		return EBUSY;
437 	}
438 	while (lkp->lk_flags & LK_ALL) {
439 		lkp->lk_flags |= LK_WAITDRAIN;
440 		error = msleep(&lkp->lk_flags, lkp->lk_interlock, lkp->lk_prio,
441 			lkp->lk_wmesg,
442 			((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0));
443 		if (error)
444 			return error;
445 		if (extflags & LK_SLEEPFAIL) {
446 			return ENOLCK;
447 		}
448 	}
449 	return 0;
450 }
451 
452 /*
453  * Transfer any waiting processes from one lock to another.
454  */
455 void
456 transferlockers(from, to)
457 	struct lock *from;
458 	struct lock *to;
459 {
460 
461 	KASSERT(from != to, ("lock transfer to self"));
462 	KASSERT((from->lk_flags&LK_WAITDRAIN) == 0, ("transfer draining lock"));
463 
464 	mtx_lock(from->lk_interlock);
465 	if (from->lk_waitcount == 0) {
466 		mtx_unlock(from->lk_interlock);
467 		return;
468 	}
469 	from->lk_newlock = to;
470 	wakeup((void *)from);
471 	msleep(&from->lk_newlock, from->lk_interlock, from->lk_prio,
472 	    "lkxfer", 0);
473 	from->lk_newlock = NULL;
474 	from->lk_flags &= ~(LK_WANT_EXCL | LK_WANT_UPGRADE);
475 	KASSERT(from->lk_waitcount == 0, ("active lock"));
476 	mtx_unlock(from->lk_interlock);
477 }
478 
479 
480 /*
481  * Initialize a lock; required before use.
482  */
483 void
484 lockinit(lkp, prio, wmesg, timo, flags)
485 	struct lock *lkp;
486 	int prio;
487 	const char *wmesg;
488 	int timo;
489 	int flags;
490 {
491 	CTR5(KTR_LOCK, "lockinit(): lkp == %p, prio == %d, wmesg == \"%s\", "
492 	    "timo == %d, flags = 0x%x\n", lkp, prio, wmesg, timo, flags);
493 
494 	lkp->lk_interlock = mtx_pool_alloc(mtxpool_lockbuilder);
495 	lkp->lk_flags = (flags & LK_EXTFLG_MASK);
496 	lkp->lk_sharecount = 0;
497 	lkp->lk_waitcount = 0;
498 	lkp->lk_exclusivecount = 0;
499 	lkp->lk_prio = prio;
500 	lkp->lk_wmesg = wmesg;
501 	lkp->lk_timo = timo;
502 	lkp->lk_lockholder = LK_NOPROC;
503 	lkp->lk_newlock = NULL;
504 #ifdef DEBUG_LOCKS
505 	stack_zero(&lkp->lk_stack);
506 #endif
507 }
508 
509 /*
510  * Destroy a lock.
511  */
512 void
513 lockdestroy(lkp)
514 	struct lock *lkp;
515 {
516 	CTR2(KTR_LOCK, "lockdestroy(): lkp == %p (lk_wmesg == \"%s\")",
517 	    lkp, lkp->lk_wmesg);
518 }
519 
520 /*
521  * Determine the status of a lock.
522  */
523 int
524 lockstatus(lkp, td)
525 	struct lock *lkp;
526 	struct thread *td;
527 {
528 	int lock_type = 0;
529 	int interlocked;
530 
531 	if (!kdb_active) {
532 		interlocked = 1;
533 		mtx_lock(lkp->lk_interlock);
534 	} else
535 		interlocked = 0;
536 	if (lkp->lk_exclusivecount != 0) {
537 		if (td == NULL || lkp->lk_lockholder == td)
538 			lock_type = LK_EXCLUSIVE;
539 		else
540 			lock_type = LK_EXCLOTHER;
541 	} else if (lkp->lk_sharecount != 0)
542 		lock_type = LK_SHARED;
543 	if (interlocked)
544 		mtx_unlock(lkp->lk_interlock);
545 	return (lock_type);
546 }
547 
548 /*
549  * Determine the number of holders of a lock.
550  */
551 int
552 lockcount(lkp)
553 	struct lock *lkp;
554 {
555 	int count;
556 
557 	mtx_lock(lkp->lk_interlock);
558 	count = lkp->lk_exclusivecount + lkp->lk_sharecount;
559 	mtx_unlock(lkp->lk_interlock);
560 	return (count);
561 }
562 
563 /*
564  * Print out information about state of a lock. Used by VOP_PRINT
565  * routines to display status about contained locks.
566  */
567 void
568 lockmgr_printinfo(lkp)
569 	struct lock *lkp;
570 {
571 
572 	if (lkp->lk_sharecount)
573 		printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg,
574 		    lkp->lk_sharecount);
575 	else if (lkp->lk_flags & LK_HAVE_EXCL)
576 		printf(" lock type %s: EXCL (count %d) by thread %p (pid %d)",
577 		    lkp->lk_wmesg, lkp->lk_exclusivecount,
578 		    lkp->lk_lockholder, lkp->lk_lockholder->td_proc->p_pid);
579 	if (lkp->lk_waitcount > 0)
580 		printf(" with %d pending", lkp->lk_waitcount);
581 #ifdef DEBUG_LOCKS
582 	stack_print(&lkp->lk_stack);
583 #endif
584 }
585