xref: /dragonfly/sys/kern/kern_lock.c (revision 984263bc)
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  * $FreeBSD: src/sys/kern/kern_lock.c,v 1.31.2.3 2001/12/25 01:44:44 dillon Exp $
42  */
43 
44 #include "opt_lint.h"
45 
46 #include <sys/param.h>
47 #include <sys/proc.h>
48 #include <sys/lock.h>
49 #include <sys/systm.h>
50 
51 /*
52  * Locking primitives implementation.
53  * Locks provide shared/exclusive sychronization.
54  */
55 
56 #ifdef SIMPLELOCK_DEBUG
57 #define COUNT(p, x) if (p) (p)->p_locks += (x)
58 #else
59 #define COUNT(p, x)
60 #endif
61 
62 #define LOCK_WAIT_TIME 100
63 #define LOCK_SAMPLE_WAIT 7
64 
65 #if defined(DIAGNOSTIC)
66 #define LOCK_INLINE
67 #else
68 #define LOCK_INLINE __inline
69 #endif
70 
71 #define LK_ALL (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | \
72 	LK_SHARE_NONZERO | LK_WAIT_NONZERO)
73 
74 static int acquire(struct lock *lkp, int extflags, int wanted);
75 static int apause(struct lock *lkp, int flags);
76 static int acquiredrain(struct lock *lkp, int extflags) ;
77 
78 static LOCK_INLINE void
79 sharelock(struct lock *lkp, int incr) {
80 	lkp->lk_flags |= LK_SHARE_NONZERO;
81 	lkp->lk_sharecount += incr;
82 }
83 
84 static LOCK_INLINE void
85 shareunlock(struct lock *lkp, int decr) {
86 
87 	KASSERT(lkp->lk_sharecount >= decr, ("shareunlock: count < decr"));
88 
89 	if (lkp->lk_sharecount == decr) {
90 		lkp->lk_flags &= ~LK_SHARE_NONZERO;
91 		if (lkp->lk_flags & (LK_WANT_UPGRADE | LK_WANT_EXCL)) {
92 			wakeup(lkp);
93 		}
94 		lkp->lk_sharecount = 0;
95 	} else {
96 		lkp->lk_sharecount -= decr;
97 	}
98 }
99 
100 /*
101  * This is the waitloop optimization, and note for this to work
102  * simple_lock and simple_unlock should be subroutines to avoid
103  * optimization troubles.
104  */
105 static int
106 apause(struct lock *lkp, int flags)
107 {
108 #ifdef SMP
109 	int i, lock_wait;
110 #endif
111 
112 	if ((lkp->lk_flags & flags) == 0)
113 		return 0;
114 #ifdef SMP
115 	for (lock_wait = LOCK_WAIT_TIME; lock_wait > 0; lock_wait--) {
116 		simple_unlock(&lkp->lk_interlock);
117 		for (i = LOCK_SAMPLE_WAIT; i > 0; i--)
118 			if ((lkp->lk_flags & flags) == 0)
119 				break;
120 		simple_lock(&lkp->lk_interlock);
121 		if ((lkp->lk_flags & flags) == 0)
122 			return 0;
123 	}
124 #endif
125 	return 1;
126 }
127 
128 static int
129 acquire(struct lock *lkp, int extflags, int wanted) {
130 	int s, error;
131 
132 	if ((extflags & LK_NOWAIT) && (lkp->lk_flags & wanted)) {
133 		return EBUSY;
134 	}
135 
136 	if (((lkp->lk_flags | extflags) & LK_NOPAUSE) == 0) {
137 		error = apause(lkp, wanted);
138 		if (error == 0)
139 			return 0;
140 	}
141 
142 	s = splhigh();
143 	while ((lkp->lk_flags & wanted) != 0) {
144 		lkp->lk_flags |= LK_WAIT_NONZERO;
145 		lkp->lk_waitcount++;
146 		simple_unlock(&lkp->lk_interlock);
147 		error = tsleep(lkp, lkp->lk_prio, lkp->lk_wmesg,
148 			    ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0));
149 		simple_lock(&lkp->lk_interlock);
150 		if (lkp->lk_waitcount == 1) {
151 			lkp->lk_flags &= ~LK_WAIT_NONZERO;
152 			lkp->lk_waitcount = 0;
153 		} else {
154 			lkp->lk_waitcount--;
155 		}
156 		if (error) {
157 			splx(s);
158 			return error;
159 		}
160 		if (extflags & LK_SLEEPFAIL) {
161 			splx(s);
162 			return ENOLCK;
163 		}
164 	}
165 	splx(s);
166 	return 0;
167 }
168 
169 /*
170  * Set, change, or release a lock.
171  *
172  * Shared requests increment the shared count. Exclusive requests set the
173  * LK_WANT_EXCL flag (preventing further shared locks), and wait for already
174  * accepted shared locks and shared-to-exclusive upgrades to go away.
175  */
176 int
177 #ifndef	DEBUG_LOCKS
178 lockmgr(lkp, flags, interlkp, p)
179 #else
180 debuglockmgr(lkp, flags, interlkp, p, name, file, line)
181 #endif
182 	struct lock *lkp;
183 	u_int flags;
184 	struct simplelock *interlkp;
185 	struct proc *p;
186 #ifdef	DEBUG_LOCKS
187 	const char *name;	/* Name of lock function */
188 	const char *file;	/* Name of file call is from */
189 	int line;		/* Line number in file */
190 #endif
191 {
192 	int error;
193 	pid_t pid;
194 	int extflags;
195 
196 	error = 0;
197 	if (p == NULL)
198 		pid = LK_KERNPROC;
199 	else
200 		pid = p->p_pid;
201 
202 	simple_lock(&lkp->lk_interlock);
203 	if (flags & LK_INTERLOCK)
204 		simple_unlock(interlkp);
205 
206 	extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK;
207 
208 	switch (flags & LK_TYPE_MASK) {
209 
210 	case LK_SHARED:
211 		/*
212 		 * If we are not the exclusive lock holder, we have to block
213 		 * while there is an exclusive lock holder or while an
214 		 * exclusive lock request or upgrade request is in progress.
215 		 *
216 		 * However, if P_DEADLKTREAT is set, we override exclusive
217 		 * lock requests or upgrade requests ( but not the exclusive
218 		 * lock itself ).
219 		 */
220 		if (lkp->lk_lockholder != pid) {
221 			if (p && (p->p_flag & P_DEADLKTREAT)) {
222 				error = acquire(
223 					    lkp,
224 					    extflags,
225 					    LK_HAVE_EXCL
226 					);
227 			} else {
228 				error = acquire(
229 					    lkp,
230 					    extflags,
231 					    LK_HAVE_EXCL | LK_WANT_EXCL |
232 					     LK_WANT_UPGRADE
233 					);
234 			}
235 			if (error)
236 				break;
237 			sharelock(lkp, 1);
238 			COUNT(p, 1);
239 			break;
240 		}
241 		/*
242 		 * We hold an exclusive lock, so downgrade it to shared.
243 		 * An alternative would be to fail with EDEADLK.
244 		 */
245 		sharelock(lkp, 1);
246 		COUNT(p, 1);
247 		/* fall into downgrade */
248 
249 	case LK_DOWNGRADE:
250 		if (lkp->lk_lockholder != pid || lkp->lk_exclusivecount == 0)
251 			panic("lockmgr: not holding exclusive lock");
252 		sharelock(lkp, lkp->lk_exclusivecount);
253 		lkp->lk_exclusivecount = 0;
254 		lkp->lk_flags &= ~LK_HAVE_EXCL;
255 		lkp->lk_lockholder = LK_NOPROC;
256 		if (lkp->lk_waitcount)
257 			wakeup((void *)lkp);
258 		break;
259 
260 	case LK_EXCLUPGRADE:
261 		/*
262 		 * If another process is ahead of us to get an upgrade,
263 		 * then we want to fail rather than have an intervening
264 		 * exclusive access.
265 		 */
266 		if (lkp->lk_flags & LK_WANT_UPGRADE) {
267 			shareunlock(lkp, 1);
268 			COUNT(p, -1);
269 			error = EBUSY;
270 			break;
271 		}
272 		/* fall into normal upgrade */
273 
274 	case LK_UPGRADE:
275 		/*
276 		 * Upgrade a shared lock to an exclusive one. If another
277 		 * shared lock has already requested an upgrade to an
278 		 * exclusive lock, our shared lock is released and an
279 		 * exclusive lock is requested (which will be granted
280 		 * after the upgrade). If we return an error, the file
281 		 * will always be unlocked.
282 		 */
283 		if ((lkp->lk_lockholder == pid) || (lkp->lk_sharecount <= 0))
284 			panic("lockmgr: upgrade exclusive lock");
285 		shareunlock(lkp, 1);
286 		COUNT(p, -1);
287 		/*
288 		 * If we are just polling, check to see if we will block.
289 		 */
290 		if ((extflags & LK_NOWAIT) &&
291 		    ((lkp->lk_flags & LK_WANT_UPGRADE) ||
292 		     lkp->lk_sharecount > 1)) {
293 			error = EBUSY;
294 			break;
295 		}
296 		if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) {
297 			/*
298 			 * We are first shared lock to request an upgrade, so
299 			 * request upgrade and wait for the shared count to
300 			 * drop to zero, then take exclusive lock.
301 			 */
302 			lkp->lk_flags |= LK_WANT_UPGRADE;
303 			error = acquire(lkp, extflags, LK_SHARE_NONZERO);
304 			lkp->lk_flags &= ~LK_WANT_UPGRADE;
305 
306 			if (error)
307 				break;
308 			lkp->lk_flags |= LK_HAVE_EXCL;
309 			lkp->lk_lockholder = pid;
310 			if (lkp->lk_exclusivecount != 0)
311 				panic("lockmgr: non-zero exclusive count");
312 			lkp->lk_exclusivecount = 1;
313 #if defined(DEBUG_LOCKS)
314 			lkp->lk_filename = file;
315 			lkp->lk_lineno = line;
316 			lkp->lk_lockername = name;
317 #endif
318 			COUNT(p, 1);
319 			break;
320 		}
321 		/*
322 		 * Someone else has requested upgrade. Release our shared
323 		 * lock, awaken upgrade requestor if we are the last shared
324 		 * lock, then request an exclusive lock.
325 		 */
326 		if ( (lkp->lk_flags & (LK_SHARE_NONZERO|LK_WAIT_NONZERO)) ==
327 			LK_WAIT_NONZERO)
328 			wakeup((void *)lkp);
329 		/* fall into exclusive request */
330 
331 	case LK_EXCLUSIVE:
332 		if (lkp->lk_lockholder == pid && pid != LK_KERNPROC) {
333 			/*
334 			 *	Recursive lock.
335 			 */
336 			if ((extflags & (LK_NOWAIT | LK_CANRECURSE)) == 0)
337 				panic("lockmgr: locking against myself");
338 			if ((extflags & LK_CANRECURSE) != 0) {
339 				lkp->lk_exclusivecount++;
340 				COUNT(p, 1);
341 				break;
342 			}
343 		}
344 		/*
345 		 * If we are just polling, check to see if we will sleep.
346 		 */
347 		if ((extflags & LK_NOWAIT) &&
348 		    (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO))) {
349 			error = EBUSY;
350 			break;
351 		}
352 		/*
353 		 * Try to acquire the want_exclusive flag.
354 		 */
355 		error = acquire(lkp, extflags, (LK_HAVE_EXCL | LK_WANT_EXCL));
356 		if (error)
357 			break;
358 		lkp->lk_flags |= LK_WANT_EXCL;
359 		/*
360 		 * Wait for shared locks and upgrades to finish.
361 		 */
362 		error = acquire(lkp, extflags, LK_WANT_UPGRADE | LK_SHARE_NONZERO);
363 		lkp->lk_flags &= ~LK_WANT_EXCL;
364 		if (error)
365 			break;
366 		lkp->lk_flags |= LK_HAVE_EXCL;
367 		lkp->lk_lockholder = pid;
368 		if (lkp->lk_exclusivecount != 0)
369 			panic("lockmgr: non-zero exclusive count");
370 		lkp->lk_exclusivecount = 1;
371 #if defined(DEBUG_LOCKS)
372 			lkp->lk_filename = file;
373 			lkp->lk_lineno = line;
374 			lkp->lk_lockername = name;
375 #endif
376 		COUNT(p, 1);
377 		break;
378 
379 	case LK_RELEASE:
380 		if (lkp->lk_exclusivecount != 0) {
381 			if (lkp->lk_lockholder != pid &&
382 			    lkp->lk_lockholder != LK_KERNPROC) {
383 				panic("lockmgr: pid %d, not %s %d unlocking",
384 				    pid, "exclusive lock holder",
385 				    lkp->lk_lockholder);
386 			}
387 			if (lkp->lk_lockholder != LK_KERNPROC) {
388 				COUNT(p, -1);
389 			}
390 			if (lkp->lk_exclusivecount == 1) {
391 				lkp->lk_flags &= ~LK_HAVE_EXCL;
392 				lkp->lk_lockholder = LK_NOPROC;
393 				lkp->lk_exclusivecount = 0;
394 			} else {
395 				lkp->lk_exclusivecount--;
396 			}
397 		} else if (lkp->lk_flags & LK_SHARE_NONZERO) {
398 			shareunlock(lkp, 1);
399 			COUNT(p, -1);
400 		}
401 		if (lkp->lk_flags & LK_WAIT_NONZERO)
402 			wakeup((void *)lkp);
403 		break;
404 
405 	case LK_DRAIN:
406 		/*
407 		 * Check that we do not already hold the lock, as it can
408 		 * never drain if we do. Unfortunately, we have no way to
409 		 * check for holding a shared lock, but at least we can
410 		 * check for an exclusive one.
411 		 */
412 		if (lkp->lk_lockholder == pid)
413 			panic("lockmgr: draining against myself");
414 
415 		error = acquiredrain(lkp, extflags);
416 		if (error)
417 			break;
418 		lkp->lk_flags |= LK_DRAINING | LK_HAVE_EXCL;
419 		lkp->lk_lockholder = pid;
420 		lkp->lk_exclusivecount = 1;
421 #if defined(DEBUG_LOCKS)
422 			lkp->lk_filename = file;
423 			lkp->lk_lineno = line;
424 			lkp->lk_lockername = name;
425 #endif
426 		COUNT(p, 1);
427 		break;
428 
429 	default:
430 		simple_unlock(&lkp->lk_interlock);
431 		panic("lockmgr: unknown locktype request %d",
432 		    flags & LK_TYPE_MASK);
433 		/* NOTREACHED */
434 	}
435 	if ((lkp->lk_flags & LK_WAITDRAIN) &&
436 	    (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE |
437 		LK_SHARE_NONZERO | LK_WAIT_NONZERO)) == 0) {
438 		lkp->lk_flags &= ~LK_WAITDRAIN;
439 		wakeup((void *)&lkp->lk_flags);
440 	}
441 	simple_unlock(&lkp->lk_interlock);
442 	return (error);
443 }
444 
445 static int
446 acquiredrain(struct lock *lkp, int extflags) {
447 	int error;
448 
449 	if ((extflags & LK_NOWAIT) && (lkp->lk_flags & LK_ALL)) {
450 		return EBUSY;
451 	}
452 
453 	error = apause(lkp, LK_ALL);
454 	if (error == 0)
455 		return 0;
456 
457 	while (lkp->lk_flags & LK_ALL) {
458 		lkp->lk_flags |= LK_WAITDRAIN;
459 		simple_unlock(&lkp->lk_interlock);
460 		error = tsleep(&lkp->lk_flags, lkp->lk_prio,
461 			lkp->lk_wmesg,
462 			((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0));
463 		simple_lock(&lkp->lk_interlock);
464 		if (error)
465 			return error;
466 		if (extflags & LK_SLEEPFAIL) {
467 			return ENOLCK;
468 		}
469 	}
470 	return 0;
471 }
472 
473 /*
474  * Initialize a lock; required before use.
475  */
476 void
477 lockinit(lkp, prio, wmesg, timo, flags)
478 	struct lock *lkp;
479 	int prio;
480 	char *wmesg;
481 	int timo;
482 	int flags;
483 {
484 
485 	simple_lock_init(&lkp->lk_interlock);
486 	lkp->lk_flags = (flags & LK_EXTFLG_MASK);
487 	lkp->lk_sharecount = 0;
488 	lkp->lk_waitcount = 0;
489 	lkp->lk_exclusivecount = 0;
490 	lkp->lk_prio = prio;
491 	lkp->lk_wmesg = wmesg;
492 	lkp->lk_timo = timo;
493 	lkp->lk_lockholder = LK_NOPROC;
494 }
495 
496 /*
497  * Determine the status of a lock.
498  */
499 int
500 lockstatus(lkp, p)
501 	struct lock *lkp;
502 	struct proc *p;
503 {
504 	int lock_type = 0;
505 
506 	simple_lock(&lkp->lk_interlock);
507 	if (lkp->lk_exclusivecount != 0) {
508 		if (p == NULL || lkp->lk_lockholder == p->p_pid)
509 			lock_type = LK_EXCLUSIVE;
510 		else
511 			lock_type = LK_EXCLOTHER;
512 	} else if (lkp->lk_sharecount != 0)
513 		lock_type = LK_SHARED;
514 	simple_unlock(&lkp->lk_interlock);
515 	return (lock_type);
516 }
517 
518 /*
519  * Determine the number of holders of a lock.
520  */
521 int
522 lockcount(lkp)
523 	struct lock *lkp;
524 {
525 	int count;
526 
527 	simple_lock(&lkp->lk_interlock);
528 	count = lkp->lk_exclusivecount + lkp->lk_sharecount;
529 	simple_unlock(&lkp->lk_interlock);
530 	return (count);
531 }
532 
533 /*
534  * Print out information about state of a lock. Used by VOP_PRINT
535  * routines to display status about contained locks.
536  */
537 void
538 lockmgr_printinfo(lkp)
539 	struct lock *lkp;
540 {
541 
542 	if (lkp->lk_sharecount)
543 		printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg,
544 		    lkp->lk_sharecount);
545 	else if (lkp->lk_flags & LK_HAVE_EXCL)
546 		printf(" lock type %s: EXCL (count %d) by pid %d",
547 		    lkp->lk_wmesg, lkp->lk_exclusivecount, lkp->lk_lockholder);
548 	if (lkp->lk_waitcount > 0)
549 		printf(" with %d pending", lkp->lk_waitcount);
550 }
551 
552 #if defined(SIMPLELOCK_DEBUG) && (MAXCPU == 1 || defined(COMPILING_LINT))
553 #include <sys/kernel.h>
554 #include <sys/sysctl.h>
555 
556 static int lockpausetime = 0;
557 SYSCTL_INT(_debug, OID_AUTO, lockpausetime, CTLFLAG_RW, &lockpausetime, 0, "");
558 
559 static int simplelockrecurse;
560 
561 /*
562  * Simple lock functions so that the debugger can see from whence
563  * they are being called.
564  */
565 void
566 simple_lock_init(alp)
567 	struct simplelock *alp;
568 {
569 
570 	alp->lock_data = 0;
571 }
572 
573 void
574 _simple_lock(alp, id, l)
575 	struct simplelock *alp;
576 	const char *id;
577 	int l;
578 {
579 
580 	if (simplelockrecurse)
581 		return;
582 	if (alp->lock_data == 1) {
583 		if (lockpausetime == -1)
584 			panic("%s:%d: simple_lock: lock held", id, l);
585 		printf("%s:%d: simple_lock: lock held\n", id, l);
586 		if (lockpausetime == 1) {
587 			Debugger("simple_lock");
588 			/*BACKTRACE(curproc); */
589 		} else if (lockpausetime > 1) {
590 			printf("%s:%d: simple_lock: lock held...", id, l);
591 			tsleep(&lockpausetime, PCATCH | PPAUSE, "slock",
592 			    lockpausetime * hz);
593 			printf(" continuing\n");
594 		}
595 	}
596 	alp->lock_data = 1;
597 	if (curproc)
598 		curproc->p_simple_locks++;
599 }
600 
601 int
602 _simple_lock_try(alp, id, l)
603 	struct simplelock *alp;
604 	const char *id;
605 	int l;
606 {
607 
608 	if (alp->lock_data)
609 		return (0);
610 	if (simplelockrecurse)
611 		return (1);
612 	alp->lock_data = 1;
613 	if (curproc)
614 		curproc->p_simple_locks++;
615 	return (1);
616 }
617 
618 void
619 _simple_unlock(alp, id, l)
620 	struct simplelock *alp;
621 	const char *id;
622 	int l;
623 {
624 
625 	if (simplelockrecurse)
626 		return;
627 	if (alp->lock_data == 0) {
628 		if (lockpausetime == -1)
629 			panic("%s:%d: simple_unlock: lock not held", id, l);
630 		printf("%s:%d: simple_unlock: lock not held\n", id, l);
631 		if (lockpausetime == 1) {
632 			Debugger("simple_unlock");
633 			/* BACKTRACE(curproc); */
634 		} else if (lockpausetime > 1) {
635 			printf("%s:%d: simple_unlock: lock not held...", id, l);
636 			tsleep(&lockpausetime, PCATCH | PPAUSE, "sunlock",
637 			    lockpausetime * hz);
638 			printf(" continuing\n");
639 		}
640 	}
641 	alp->lock_data = 0;
642 	if (curproc)
643 		curproc->p_simple_locks--;
644 }
645 #elif defined(SIMPLELOCK_DEBUG)
646 #error "SIMPLELOCK_DEBUG is not compatible with SMP!"
647 #endif /* SIMPLELOCK_DEBUG && MAXCPU == 1 */
648