xref: /freebsd/sys/kern/kern_lock.c (revision 380a989b3223d455375b4fae70fd0b9bdd43bafb)
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  * $Id: kern_lock.c,v 1.20 1998/11/26 18:50:23 eivind 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 	KASSERT(lkp->lk_sharecount >= decr, ("shareunlock: count < decr"));
87 
88 	if (lkp->lk_sharecount == decr) {
89 		lkp->lk_flags &= ~LK_SHARE_NONZERO;
90 		if (lkp->lk_flags & (LK_WANT_UPGRADE | LK_WANT_EXCL)) {
91 			wakeup(lkp);
92 		}
93 		lkp->lk_sharecount = 0;
94 	} else {
95 		lkp->lk_sharecount -= decr;
96 	}
97 }
98 
99 /*
100  * This is the waitloop optimization, and note for this to work
101  * simple_lock and simple_unlock should be subroutines to avoid
102  * optimization troubles.
103  */
104 static int
105 apause(struct lock *lkp, int flags) {
106 	int lock_wait;
107 	lock_wait = LOCK_WAIT_TIME;
108 	for (; lock_wait > 0; lock_wait--) {
109 		int i;
110 		if ((lkp->lk_flags & flags) == 0)
111 			return 0;
112 		simple_unlock(&lkp->lk_interlock);
113 		for (i = LOCK_SAMPLE_WAIT; i > 0; i--) {
114 			if ((lkp->lk_flags & flags) == 0) {
115 				simple_lock(&lkp->lk_interlock);
116 				if ((lkp->lk_flags & flags) == 0)
117 					return 0;
118 				break;
119 			}
120 		}
121 	}
122 	return 1;
123 }
124 
125 static int
126 acquire(struct lock *lkp, int extflags, int wanted) {
127 	int s, error;
128 
129 	if ((extflags & LK_NOWAIT) && (lkp->lk_flags & wanted)) {
130 		return EBUSY;
131 	}
132 
133 	if (((lkp->lk_flags | extflags) & LK_NOPAUSE) == 0) {
134 		error = apause(lkp, wanted);
135 		if (error == 0)
136 			return 0;
137 	}
138 
139 	s = splhigh();
140 	while ((lkp->lk_flags & wanted) != 0) {
141 		lkp->lk_flags |= LK_WAIT_NONZERO;
142 		lkp->lk_waitcount++;
143 		simple_unlock(&lkp->lk_interlock);
144 		error = tsleep(lkp, lkp->lk_prio, lkp->lk_wmesg, lkp->lk_timo);
145 		simple_lock(&lkp->lk_interlock);
146 		if (lkp->lk_waitcount == 1) {
147 			lkp->lk_flags &= ~LK_WAIT_NONZERO;
148 			lkp->lk_waitcount = 0;
149 		} else {
150 			lkp->lk_waitcount--;
151 		}
152 		if (error) {
153 			splx(s);
154 			return error;
155 		}
156 		if (extflags & LK_SLEEPFAIL) {
157 			splx(s);
158 			return ENOLCK;
159 		}
160 	}
161 	splx(s);
162 	return 0;
163 }
164 
165 /*
166  * Set, change, or release a lock.
167  *
168  * Shared requests increment the shared count. Exclusive requests set the
169  * LK_WANT_EXCL flag (preventing further shared locks), and wait for already
170  * accepted shared locks and shared-to-exclusive upgrades to go away.
171  */
172 int
173 lockmgr(lkp, flags, interlkp, p)
174 	struct lock *lkp;
175 	u_int flags;
176 	struct simplelock *interlkp;
177 	struct proc *p;
178 {
179 	int error;
180 	pid_t pid;
181 	int extflags;
182 
183 	error = 0;
184 	if (p == NULL)
185 		pid = LK_KERNPROC;
186 	else
187 		pid = p->p_pid;
188 
189 	simple_lock(&lkp->lk_interlock);
190 	if (flags & LK_INTERLOCK)
191 		simple_unlock(interlkp);
192 
193 	extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK;
194 
195 	switch (flags & LK_TYPE_MASK) {
196 
197 	case LK_SHARED:
198 		if (lkp->lk_lockholder != pid) {
199 			error = acquire(lkp, extflags,
200 				LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE);
201 			if (error)
202 				break;
203 			sharelock(lkp, 1);
204 			COUNT(p, 1);
205 			break;
206 		}
207 		/*
208 		 * We hold an exclusive lock, so downgrade it to shared.
209 		 * An alternative would be to fail with EDEADLK.
210 		 */
211 		sharelock(lkp, 1);
212 		COUNT(p, 1);
213 		/* fall into downgrade */
214 
215 	case LK_DOWNGRADE:
216 #if !defined(MAX_PERF)
217 		if (lkp->lk_lockholder != pid || lkp->lk_exclusivecount == 0)
218 			panic("lockmgr: not holding exclusive lock");
219 #endif
220 		sharelock(lkp, lkp->lk_exclusivecount);
221 		lkp->lk_exclusivecount = 0;
222 		lkp->lk_flags &= ~LK_HAVE_EXCL;
223 		lkp->lk_lockholder = LK_NOPROC;
224 		if (lkp->lk_waitcount)
225 			wakeup((void *)lkp);
226 		break;
227 
228 	case LK_EXCLUPGRADE:
229 		/*
230 		 * If another process is ahead of us to get an upgrade,
231 		 * then we want to fail rather than have an intervening
232 		 * exclusive access.
233 		 */
234 		if (lkp->lk_flags & LK_WANT_UPGRADE) {
235 			shareunlock(lkp, 1);
236 			COUNT(p, -1);
237 			error = EBUSY;
238 			break;
239 		}
240 		/* fall into normal upgrade */
241 
242 	case LK_UPGRADE:
243 		/*
244 		 * Upgrade a shared lock to an exclusive one. If another
245 		 * shared lock has already requested an upgrade to an
246 		 * exclusive lock, our shared lock is released and an
247 		 * exclusive lock is requested (which will be granted
248 		 * after the upgrade). If we return an error, the file
249 		 * will always be unlocked.
250 		 */
251 #if !defined(MAX_PERF)
252 		if ((lkp->lk_lockholder == pid) || (lkp->lk_sharecount <= 0))
253 			panic("lockmgr: upgrade exclusive lock");
254 #endif
255 		shareunlock(lkp, 1);
256 		COUNT(p, -1);
257 		/*
258 		 * If we are just polling, check to see if we will block.
259 		 */
260 		if ((extflags & LK_NOWAIT) &&
261 		    ((lkp->lk_flags & LK_WANT_UPGRADE) ||
262 		     lkp->lk_sharecount > 1)) {
263 			error = EBUSY;
264 			break;
265 		}
266 		if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) {
267 			/*
268 			 * We are first shared lock to request an upgrade, so
269 			 * request upgrade and wait for the shared count to
270 			 * drop to zero, then take exclusive lock.
271 			 */
272 			lkp->lk_flags |= LK_WANT_UPGRADE;
273 			error = acquire(lkp, extflags, LK_SHARE_NONZERO);
274 			lkp->lk_flags &= ~LK_WANT_UPGRADE;
275 
276 			if (error)
277 				break;
278 			lkp->lk_flags |= LK_HAVE_EXCL;
279 			lkp->lk_lockholder = pid;
280 #if !defined(MAX_PERF)
281 			if (lkp->lk_exclusivecount != 0)
282 				panic("lockmgr: non-zero exclusive count");
283 #endif
284 			lkp->lk_exclusivecount = 1;
285 			COUNT(p, 1);
286 			break;
287 		}
288 		/*
289 		 * Someone else has requested upgrade. Release our shared
290 		 * lock, awaken upgrade requestor if we are the last shared
291 		 * lock, then request an exclusive lock.
292 		 */
293 		if ( (lkp->lk_flags & (LK_SHARE_NONZERO|LK_WAIT_NONZERO)) ==
294 			LK_WAIT_NONZERO)
295 			wakeup((void *)lkp);
296 		/* fall into exclusive request */
297 
298 	case LK_EXCLUSIVE:
299 		if (lkp->lk_lockholder == pid && pid != LK_KERNPROC) {
300 			/*
301 			 *	Recursive lock.
302 			 */
303 #if !defined(MAX_PERF)
304 			if ((extflags & LK_CANRECURSE) == 0)
305 				panic("lockmgr: locking against myself");
306 #endif
307 			lkp->lk_exclusivecount++;
308 			COUNT(p, 1);
309 			break;
310 		}
311 		/*
312 		 * If we are just polling, check to see if we will sleep.
313 		 */
314 		if ((extflags & LK_NOWAIT) &&
315 		    (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO))) {
316 			error = EBUSY;
317 			break;
318 		}
319 		/*
320 		 * Try to acquire the want_exclusive flag.
321 		 */
322 		error = acquire(lkp, extflags, (LK_HAVE_EXCL | LK_WANT_EXCL));
323 		if (error)
324 			break;
325 		lkp->lk_flags |= LK_WANT_EXCL;
326 		/*
327 		 * Wait for shared locks and upgrades to finish.
328 		 */
329 		error = acquire(lkp, extflags, LK_WANT_UPGRADE | LK_SHARE_NONZERO);
330 		lkp->lk_flags &= ~LK_WANT_EXCL;
331 		if (error)
332 			break;
333 		lkp->lk_flags |= LK_HAVE_EXCL;
334 		lkp->lk_lockholder = pid;
335 #if !defined(MAX_PERF)
336 		if (lkp->lk_exclusivecount != 0)
337 			panic("lockmgr: non-zero exclusive count");
338 #endif
339 		lkp->lk_exclusivecount = 1;
340 		COUNT(p, 1);
341 		break;
342 
343 	case LK_RELEASE:
344 		if (lkp->lk_exclusivecount != 0) {
345 #if !defined(MAX_PERF)
346 			if (pid != lkp->lk_lockholder)
347 				panic("lockmgr: pid %d, not %s %d unlocking",
348 				    pid, "exclusive lock holder",
349 				    lkp->lk_lockholder);
350 #endif
351 			COUNT(p, -1);
352 			if (lkp->lk_exclusivecount == 1) {
353 				lkp->lk_flags &= ~LK_HAVE_EXCL;
354 				lkp->lk_lockholder = LK_NOPROC;
355 				lkp->lk_exclusivecount = 0;
356 			} else {
357 				lkp->lk_exclusivecount--;
358 			}
359 		} else if (lkp->lk_flags & LK_SHARE_NONZERO) {
360 			shareunlock(lkp, 1);
361 			COUNT(p, -1);
362 		}
363 		if (lkp->lk_flags & LK_WAIT_NONZERO)
364 			wakeup((void *)lkp);
365 		break;
366 
367 	case LK_DRAIN:
368 		/*
369 		 * Check that we do not already hold the lock, as it can
370 		 * never drain if we do. Unfortunately, we have no way to
371 		 * check for holding a shared lock, but at least we can
372 		 * check for an exclusive one.
373 		 */
374 #if !defined(MAX_PERF)
375 		if (lkp->lk_lockholder == pid)
376 			panic("lockmgr: draining against myself");
377 #endif
378 
379 		error = acquiredrain(lkp, extflags);
380 		if (error)
381 			break;
382 		lkp->lk_flags |= LK_DRAINING | LK_HAVE_EXCL;
383 		lkp->lk_lockholder = pid;
384 		lkp->lk_exclusivecount = 1;
385 		COUNT(p, 1);
386 		break;
387 
388 	default:
389 #if !defined(MAX_PERF)
390 		simple_unlock(&lkp->lk_interlock);
391 		panic("lockmgr: unknown locktype request %d",
392 		    flags & LK_TYPE_MASK);
393 #endif
394 		/* NOTREACHED */
395 	}
396 	if ((lkp->lk_flags & LK_WAITDRAIN) &&
397 	    (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE |
398 		LK_SHARE_NONZERO | LK_WAIT_NONZERO)) == 0) {
399 		lkp->lk_flags &= ~LK_WAITDRAIN;
400 		wakeup((void *)&lkp->lk_flags);
401 	}
402 	simple_unlock(&lkp->lk_interlock);
403 	return (error);
404 }
405 
406 static int
407 acquiredrain(struct lock *lkp, int extflags) {
408 	int error;
409 
410 	if ((extflags & LK_NOWAIT) && (lkp->lk_flags & LK_ALL)) {
411 		return EBUSY;
412 	}
413 
414 	error = apause(lkp, LK_ALL);
415 	if (error == 0)
416 		return 0;
417 
418 	while (lkp->lk_flags & LK_ALL) {
419 		lkp->lk_flags |= LK_WAITDRAIN;
420 		simple_unlock(&lkp->lk_interlock);
421 		error = tsleep(&lkp->lk_flags, lkp->lk_prio,
422 			lkp->lk_wmesg, lkp->lk_timo);
423 		simple_lock(&lkp->lk_interlock);
424 		if (error)
425 			return error;
426 		if (extflags & LK_SLEEPFAIL) {
427 			return ENOLCK;
428 		}
429 	}
430 	return 0;
431 }
432 
433 /*
434  * Initialize a lock; required before use.
435  */
436 void
437 lockinit(lkp, prio, wmesg, timo, flags)
438 	struct lock *lkp;
439 	int prio;
440 	char *wmesg;
441 	int timo;
442 	int flags;
443 {
444 
445 	simple_lock_init(&lkp->lk_interlock);
446 	lkp->lk_flags = (flags & LK_EXTFLG_MASK);
447 	lkp->lk_sharecount = 0;
448 	lkp->lk_waitcount = 0;
449 	lkp->lk_exclusivecount = 0;
450 	lkp->lk_prio = prio;
451 	lkp->lk_wmesg = wmesg;
452 	lkp->lk_timo = timo;
453 	lkp->lk_lockholder = LK_NOPROC;
454 }
455 
456 /*
457  * Determine the status of a lock.
458  */
459 int
460 lockstatus(lkp)
461 	struct lock *lkp;
462 {
463 	int lock_type = 0;
464 
465 	simple_lock(&lkp->lk_interlock);
466 	if (lkp->lk_exclusivecount != 0)
467 		lock_type = LK_EXCLUSIVE;
468 	else if (lkp->lk_sharecount != 0)
469 		lock_type = LK_SHARED;
470 	simple_unlock(&lkp->lk_interlock);
471 	return (lock_type);
472 }
473 
474 /*
475  * Print out information about state of a lock. Used by VOP_PRINT
476  * routines to display status about contained locks.
477  */
478 void
479 lockmgr_printinfo(lkp)
480 	struct lock *lkp;
481 {
482 
483 	if (lkp->lk_sharecount)
484 		printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg,
485 		    lkp->lk_sharecount);
486 	else if (lkp->lk_flags & LK_HAVE_EXCL)
487 		printf(" lock type %s: EXCL (count %d) by pid %d",
488 		    lkp->lk_wmesg, lkp->lk_exclusivecount, lkp->lk_lockholder);
489 	if (lkp->lk_waitcount > 0)
490 		printf(" with %d pending", lkp->lk_waitcount);
491 }
492 
493 #if defined(SIMPLELOCK_DEBUG) && (NCPUS == 1 || defined(COMPILING_LINT))
494 #include <sys/kernel.h>
495 #include <sys/sysctl.h>
496 
497 static int lockpausetime = 0;
498 SYSCTL_INT(_debug, OID_AUTO, lockpausetime, CTLFLAG_RW, &lockpausetime, 0, "");
499 
500 static int simplelockrecurse;
501 
502 /*
503  * Simple lock functions so that the debugger can see from whence
504  * they are being called.
505  */
506 void
507 simple_lock_init(alp)
508 	struct simplelock *alp;
509 {
510 
511 	alp->lock_data = 0;
512 }
513 
514 void
515 _simple_lock(alp, id, l)
516 	struct simplelock *alp;
517 	const char *id;
518 	int l;
519 {
520 
521 	if (simplelockrecurse)
522 		return;
523 	if (alp->lock_data == 1) {
524 		if (lockpausetime == -1)
525 			panic("%s:%d: simple_lock: lock held", id, l);
526 		printf("%s:%d: simple_lock: lock held\n", id, l);
527 		if (lockpausetime == 1) {
528 			Debugger("simple_lock");
529 			/*BACKTRACE(curproc); */
530 		} else if (lockpausetime > 1) {
531 			printf("%s:%d: simple_lock: lock held...", id, l);
532 			tsleep(&lockpausetime, PCATCH | PPAUSE, "slock",
533 			    lockpausetime * hz);
534 			printf(" continuing\n");
535 		}
536 	}
537 	alp->lock_data = 1;
538 	if (curproc)
539 		curproc->p_simple_locks++;
540 }
541 
542 int
543 _simple_lock_try(alp, id, l)
544 	struct simplelock *alp;
545 	const char *id;
546 	int l;
547 {
548 
549 	if (alp->lock_data)
550 		return (0);
551 	if (simplelockrecurse)
552 		return (1);
553 	alp->lock_data = 1;
554 	if (curproc)
555 		curproc->p_simple_locks++;
556 	return (1);
557 }
558 
559 void
560 _simple_unlock(alp, id, l)
561 	struct simplelock *alp;
562 	const char *id;
563 	int l;
564 {
565 
566 	if (simplelockrecurse)
567 		return;
568 	if (alp->lock_data == 0) {
569 		if (lockpausetime == -1)
570 			panic("%s:%d: simple_unlock: lock not held", id, l);
571 		printf("%s:%d: simple_unlock: lock not held\n", id, l);
572 		if (lockpausetime == 1) {
573 			Debugger("simple_unlock");
574 			/* BACKTRACE(curproc); */
575 		} else if (lockpausetime > 1) {
576 			printf("%s:%d: simple_unlock: lock not held...", id, l);
577 			tsleep(&lockpausetime, PCATCH | PPAUSE, "sunlock",
578 			    lockpausetime * hz);
579 			printf(" continuing\n");
580 		}
581 	}
582 	alp->lock_data = 0;
583 	if (curproc)
584 		curproc->p_simple_locks--;
585 }
586 #elif defined(SIMPLELOCK_DEBUG)
587 #error "SIMPLELOCK_DEBUG is not compatible with SMP!"
588 #endif /* SIMPLELOCK_DEBUG && NCPUS == 1 */
589