xref: /dflybsd-src/sys/kern/kern_lock.c (revision 41871674d0079dec70d55eb824f39d07dc7b3310)
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  * $DragonFly: src/sys/kern/kern_lock.c,v 1.16 2006/03/02 19:07:59 dillon Exp $
43  */
44 
45 #include "opt_lint.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/proc.h>
51 #include <sys/lock.h>
52 #include <sys/sysctl.h>
53 #include <sys/spinlock.h>
54 #include <sys/thread2.h>
55 #include <sys/spinlock2.h>
56 
57 /*
58  * 0: no warnings, 1: warnings, 2: panic
59  */
60 static int lockmgr_from_int = 1;
61 SYSCTL_INT(_debug, OID_AUTO, lockmgr_from_int, CTLFLAG_RW, &lockmgr_from_int, 0, "");
62 
63 /*
64  * Locking primitives implementation.
65  * Locks provide shared/exclusive sychronization.
66  */
67 
68 #ifdef SIMPLELOCK_DEBUG
69 #define COUNT(td, x) (td)->td_locks += (x)
70 #else
71 #define COUNT(td, x)
72 #endif
73 
74 #define LOCK_WAIT_TIME 100
75 #define LOCK_SAMPLE_WAIT 7
76 
77 #if defined(DIAGNOSTIC)
78 #define LOCK_INLINE
79 #else
80 #define LOCK_INLINE __inline
81 #endif
82 
83 #define LK_ALL (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | \
84 	LK_SHARE_NONZERO | LK_WAIT_NONZERO)
85 
86 static int acquire(struct lock *lkp, int extflags, int wanted);
87 static int acquiredrain(struct lock *lkp, int extflags) ;
88 
89 static LOCK_INLINE void
90 sharelock(struct lock *lkp, int incr) {
91 	lkp->lk_flags |= LK_SHARE_NONZERO;
92 	lkp->lk_sharecount += incr;
93 }
94 
95 static LOCK_INLINE void
96 shareunlock(struct lock *lkp, int decr) {
97 
98 	KASSERT(lkp->lk_sharecount >= decr, ("shareunlock: count < decr"));
99 
100 	if (lkp->lk_sharecount == decr) {
101 		lkp->lk_flags &= ~LK_SHARE_NONZERO;
102 		if (lkp->lk_flags & (LK_WANT_UPGRADE | LK_WANT_EXCL)) {
103 			wakeup(lkp);
104 		}
105 		lkp->lk_sharecount = 0;
106 	} else {
107 		lkp->lk_sharecount -= decr;
108 	}
109 }
110 
111 /*
112  * lock acquisition helper routine.  Called with the lock's spinlock held.
113  */
114 static int
115 acquire(struct lock *lkp, int extflags, int wanted)
116 {
117 	int error;
118 
119 	if ((extflags & LK_NOWAIT) && (lkp->lk_flags & wanted)) {
120 		return EBUSY;
121 	}
122 
123 	if (((lkp->lk_flags | extflags) & LK_NOPAUSE) == 0) {
124 		if ((lkp->lk_flags & wanted) == 0)
125 			return 0;
126 	}
127 
128 	while ((lkp->lk_flags & wanted) != 0) {
129 		lkp->lk_flags |= LK_WAIT_NONZERO;
130 		lkp->lk_waitcount++;
131 
132 		/*
133 		 * Use the _quick version so the critical section is left
134 		 * intact, protecting the tsleep interlock.  See
135 		 * tsleep_interlock() for a description of what is
136 		 * happening here.
137 		 */
138 		tsleep_interlock(lkp);
139 		spin_unlock_quick(&lkp->lk_spinlock);
140 		error = tsleep(lkp,
141 			       ((extflags & LK_PCATCH) ? PCATCH : 0),
142 			       lkp->lk_wmesg,
143 			       ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0));
144 		spin_lock_quick(&lkp->lk_spinlock);
145 		if (lkp->lk_waitcount == 1) {
146 			lkp->lk_flags &= ~LK_WAIT_NONZERO;
147 			lkp->lk_waitcount = 0;
148 		} else {
149 			lkp->lk_waitcount--;
150 		}
151 		if (error)
152 			return error;
153 		if (extflags & LK_SLEEPFAIL)
154 			return ENOLCK;
155 	}
156 	return 0;
157 }
158 
159 /*
160  * Set, change, or release a lock.
161  *
162  * Shared requests increment the shared count. Exclusive requests set the
163  * LK_WANT_EXCL flag (preventing further shared locks), and wait for already
164  * accepted shared locks and shared-to-exclusive upgrades to go away.
165  *
166  * A spinlock is held for most of the procedure.  We must not do anything
167  * fancy while holding the spinlock.
168  */
169 int
170 #ifndef	DEBUG_LOCKS
171 lockmgr(struct lock *lkp, u_int flags, struct spinlock *interlkp,
172 	struct thread *td)
173 #else
174 debuglockmgr(struct lock *lkp, u_int flags, struct spinlock *interlkp,
175 	struct thread *td, const char *name, const char *file, int line)
176 #endif
177 {
178 	int error;
179 	int extflags;
180 	static int didpanic;
181 
182 	error = 0;
183 
184 	if (lockmgr_from_int && mycpu->gd_intr_nesting_level &&
185 	    (flags & LK_NOWAIT) == 0 &&
186 	    (flags & LK_TYPE_MASK) != LK_RELEASE && didpanic == 0) {
187 #ifndef DEBUG_LOCKS
188 		    if (lockmgr_from_int == 2) {
189 			    didpanic = 1;
190 			    if (flags & LK_INTERLOCK)
191 				    spin_unlock(interlkp);
192 			    panic(
193 				"lockmgr %s from %p: called from interrupt",
194 				lkp->lk_wmesg, ((int **)&lkp)[-1]);
195 			    didpanic = 0;
196 		    } else {
197 			    printf(
198 				"lockmgr %s from %p: called from interrupt\n",
199 				lkp->lk_wmesg, ((int **)&lkp)[-1]);
200 		    }
201 #else
202 		    if (lockmgr_from_int == 2) {
203 			    didpanic = 1;
204 			    if (flags & LK_INTERLOCK)
205 				    spin_unlock(interlkp);
206 			    panic(
207 				"lockmgr %s from %s:%d: called from interrupt",
208 				lkp->lk_wmesg, file, line);
209 			    didpanic = 0;
210 		    } else {
211 			    printf(
212 				"lockmgr %s from %s:%d: called from interrupt\n",
213 				lkp->lk_wmesg, file, line);
214 		    }
215 #endif
216 	}
217 
218 	spin_lock(&lkp->lk_spinlock);
219 	if (flags & LK_INTERLOCK)
220 		spin_unlock(interlkp);
221 
222 	extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK;
223 
224 	switch (flags & LK_TYPE_MASK) {
225 	case LK_SHARED:
226 		/*
227 		 * If we are not the exclusive lock holder, we have to block
228 		 * while there is an exclusive lock holder or while an
229 		 * exclusive lock request or upgrade request is in progress.
230 		 *
231 		 * However, if P_DEADLKTREAT is set, we override exclusive
232 		 * lock requests or upgrade requests ( but not the exclusive
233 		 * lock itself ).
234 		 */
235 		if (lkp->lk_lockholder != td) {
236 			if (td->td_flags & TDF_DEADLKTREAT) {
237 				error = acquire(
238 					    lkp,
239 					    extflags,
240 					    LK_HAVE_EXCL
241 					);
242 			} else {
243 				error = acquire(
244 					    lkp,
245 					    extflags,
246 					    LK_HAVE_EXCL | LK_WANT_EXCL |
247 					     LK_WANT_UPGRADE
248 					);
249 			}
250 			if (error)
251 				break;
252 			sharelock(lkp, 1);
253 			COUNT(td, 1);
254 			break;
255 		}
256 		/*
257 		 * We hold an exclusive lock, so downgrade it to shared.
258 		 * An alternative would be to fail with EDEADLK.
259 		 */
260 		sharelock(lkp, 1);
261 		COUNT(td, 1);
262 		/* fall into downgrade */
263 
264 	case LK_DOWNGRADE:
265 		if (lkp->lk_lockholder != td || lkp->lk_exclusivecount == 0) {
266 			spin_unlock(&lkp->lk_spinlock);
267 			panic("lockmgr: not holding exclusive lock");
268 		}
269 		sharelock(lkp, lkp->lk_exclusivecount);
270 		lkp->lk_exclusivecount = 0;
271 		lkp->lk_flags &= ~LK_HAVE_EXCL;
272 		lkp->lk_lockholder = LK_NOTHREAD;
273 		if (lkp->lk_waitcount)
274 			wakeup((void *)lkp);
275 		break;
276 
277 	case LK_EXCLUPGRADE:
278 		/*
279 		 * If another process is ahead of us to get an upgrade,
280 		 * then we want to fail rather than have an intervening
281 		 * exclusive access.
282 		 */
283 		if (lkp->lk_flags & LK_WANT_UPGRADE) {
284 			shareunlock(lkp, 1);
285 			COUNT(td, -1);
286 			error = EBUSY;
287 			break;
288 		}
289 		/* fall into normal upgrade */
290 
291 	case LK_UPGRADE:
292 		/*
293 		 * Upgrade a shared lock to an exclusive one. If another
294 		 * shared lock has already requested an upgrade to an
295 		 * exclusive lock, our shared lock is released and an
296 		 * exclusive lock is requested (which will be granted
297 		 * after the upgrade). If we return an error, the file
298 		 * will always be unlocked.
299 		 */
300 		if ((lkp->lk_lockholder == td) || (lkp->lk_sharecount <= 0)) {
301 			spin_unlock(&lkp->lk_spinlock);
302 			panic("lockmgr: upgrade exclusive lock");
303 		}
304 		shareunlock(lkp, 1);
305 		COUNT(td, -1);
306 		/*
307 		 * If we are just polling, check to see if we will block.
308 		 */
309 		if ((extflags & LK_NOWAIT) &&
310 		    ((lkp->lk_flags & LK_WANT_UPGRADE) ||
311 		     lkp->lk_sharecount > 1)) {
312 			error = EBUSY;
313 			break;
314 		}
315 		if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) {
316 			/*
317 			 * We are first shared lock to request an upgrade, so
318 			 * request upgrade and wait for the shared count to
319 			 * drop to zero, then take exclusive lock.
320 			 */
321 			lkp->lk_flags |= LK_WANT_UPGRADE;
322 			error = acquire(lkp, extflags, LK_SHARE_NONZERO);
323 			lkp->lk_flags &= ~LK_WANT_UPGRADE;
324 
325 			if (error)
326 				break;
327 			lkp->lk_flags |= LK_HAVE_EXCL;
328 			lkp->lk_lockholder = td;
329 			if (lkp->lk_exclusivecount != 0) {
330 				spin_unlock(&lkp->lk_spinlock);
331 				panic("lockmgr: non-zero exclusive count");
332 			}
333 			lkp->lk_exclusivecount = 1;
334 #if defined(DEBUG_LOCKS)
335 			lkp->lk_filename = file;
336 			lkp->lk_lineno = line;
337 			lkp->lk_lockername = name;
338 #endif
339 			COUNT(td, 1);
340 			break;
341 		}
342 		/*
343 		 * Someone else has requested upgrade. Release our shared
344 		 * lock, awaken upgrade requestor if we are the last shared
345 		 * lock, then request an exclusive lock.
346 		 */
347 		if ( (lkp->lk_flags & (LK_SHARE_NONZERO|LK_WAIT_NONZERO)) ==
348 			LK_WAIT_NONZERO)
349 			wakeup((void *)lkp);
350 		/* fall into exclusive request */
351 
352 	case LK_EXCLUSIVE:
353 		if (lkp->lk_lockholder == td && td != LK_KERNTHREAD) {
354 			/*
355 			 *	Recursive lock.
356 			 */
357 			if ((extflags & (LK_NOWAIT | LK_CANRECURSE)) == 0) {
358 				spin_unlock(&lkp->lk_spinlock);
359 				panic("lockmgr: locking against myself");
360 			}
361 			if ((extflags & LK_CANRECURSE) != 0) {
362 				lkp->lk_exclusivecount++;
363 				COUNT(td, 1);
364 				break;
365 			}
366 		}
367 		/*
368 		 * If we are just polling, check to see if we will sleep.
369 		 */
370 		if ((extflags & LK_NOWAIT) &&
371 		    (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO))) {
372 			error = EBUSY;
373 			break;
374 		}
375 		/*
376 		 * Try to acquire the want_exclusive flag.
377 		 */
378 		error = acquire(lkp, extflags, (LK_HAVE_EXCL | LK_WANT_EXCL));
379 		if (error)
380 			break;
381 		lkp->lk_flags |= LK_WANT_EXCL;
382 		/*
383 		 * Wait for shared locks and upgrades to finish.
384 		 */
385 		error = acquire(lkp, extflags, LK_WANT_UPGRADE | LK_SHARE_NONZERO);
386 		lkp->lk_flags &= ~LK_WANT_EXCL;
387 		if (error)
388 			break;
389 		lkp->lk_flags |= LK_HAVE_EXCL;
390 		lkp->lk_lockholder = td;
391 		if (lkp->lk_exclusivecount != 0) {
392 			spin_unlock(&lkp->lk_spinlock);
393 			panic("lockmgr: non-zero exclusive count");
394 		}
395 		lkp->lk_exclusivecount = 1;
396 #if defined(DEBUG_LOCKS)
397 			lkp->lk_filename = file;
398 			lkp->lk_lineno = line;
399 			lkp->lk_lockername = name;
400 #endif
401 		COUNT(td, 1);
402 		break;
403 
404 	case LK_RELEASE:
405 		if (lkp->lk_exclusivecount != 0) {
406 			if (lkp->lk_lockholder != td &&
407 			    lkp->lk_lockholder != LK_KERNTHREAD) {
408 				spin_unlock(&lkp->lk_spinlock);
409 				panic("lockmgr: pid %d, not %s thr %p unlocking",
410 				    (td->td_proc ? td->td_proc->p_pid : -99),
411 				    "exclusive lock holder",
412 				    lkp->lk_lockholder);
413 			}
414 			if (lkp->lk_lockholder != LK_KERNTHREAD) {
415 				COUNT(td, -1);
416 			}
417 			if (lkp->lk_exclusivecount == 1) {
418 				lkp->lk_flags &= ~LK_HAVE_EXCL;
419 				lkp->lk_lockholder = LK_NOTHREAD;
420 				lkp->lk_exclusivecount = 0;
421 			} else {
422 				lkp->lk_exclusivecount--;
423 			}
424 		} else if (lkp->lk_flags & LK_SHARE_NONZERO) {
425 			shareunlock(lkp, 1);
426 			COUNT(td, -1);
427 		}
428 		if (lkp->lk_flags & LK_WAIT_NONZERO)
429 			wakeup((void *)lkp);
430 		break;
431 
432 	case LK_DRAIN:
433 		/*
434 		 * Check that we do not already hold the lock, as it can
435 		 * never drain if we do. Unfortunately, we have no way to
436 		 * check for holding a shared lock, but at least we can
437 		 * check for an exclusive one.
438 		 */
439 		if (lkp->lk_lockholder == td) {
440 			spin_unlock(&lkp->lk_spinlock);
441 			panic("lockmgr: draining against myself");
442 		}
443 
444 		error = acquiredrain(lkp, extflags);
445 		if (error)
446 			break;
447 		lkp->lk_flags |= LK_DRAINING | LK_HAVE_EXCL;
448 		lkp->lk_lockholder = td;
449 		lkp->lk_exclusivecount = 1;
450 #if defined(DEBUG_LOCKS)
451 			lkp->lk_filename = file;
452 			lkp->lk_lineno = line;
453 			lkp->lk_lockername = name;
454 #endif
455 		COUNT(td, 1);
456 		break;
457 
458 	default:
459 		spin_unlock(&lkp->lk_spinlock);
460 		panic("lockmgr: unknown locktype request %d",
461 		    flags & LK_TYPE_MASK);
462 		/* NOTREACHED */
463 	}
464 	if ((lkp->lk_flags & LK_WAITDRAIN) &&
465 	    (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE |
466 		LK_SHARE_NONZERO | LK_WAIT_NONZERO)) == 0) {
467 		lkp->lk_flags &= ~LK_WAITDRAIN;
468 		wakeup((void *)&lkp->lk_flags);
469 	}
470 	spin_unlock(&lkp->lk_spinlock);
471 	return (error);
472 }
473 
474 /*
475  * lock acquisition helper routine.  Called with the lock's spinlock held.
476  */
477 static int
478 acquiredrain(struct lock *lkp, int extflags)
479 {
480 	int error;
481 
482 	if ((extflags & LK_NOWAIT) && (lkp->lk_flags & LK_ALL)) {
483 		return EBUSY;
484 	}
485 
486 	if ((lkp->lk_flags & LK_ALL) == 0)
487 		return 0;
488 
489 	while (lkp->lk_flags & LK_ALL) {
490 		lkp->lk_flags |= LK_WAITDRAIN;
491 		/*
492 		 * Use the _quick version so the critical section is left
493 		 * intact, protecting the tsleep interlock.  See
494 		 * tsleep_interlock() for a description of what is
495 		 * happening here.
496 		 */
497 		tsleep_interlock(&lkp->lk_flags);
498 		spin_unlock_quick(&lkp->lk_spinlock);
499 		error = tsleep(&lkp->lk_flags,
500 			       ((extflags & LK_PCATCH) ? PCATCH : 0),
501 			       lkp->lk_wmesg,
502 			       ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0));
503 		spin_lock_quick(&lkp->lk_spinlock);
504 		if (error)
505 			return error;
506 		if (extflags & LK_SLEEPFAIL) {
507 			return ENOLCK;
508 		}
509 	}
510 	return 0;
511 }
512 
513 /*
514  * Initialize a lock; required before use.
515  */
516 void
517 lockinit(struct lock *lkp, char *wmesg, int timo, int flags)
518 {
519 	spin_init(&lkp->lk_spinlock);
520 	lkp->lk_flags = (flags & LK_EXTFLG_MASK);
521 	lkp->lk_sharecount = 0;
522 	lkp->lk_waitcount = 0;
523 	lkp->lk_exclusivecount = 0;
524 	lkp->lk_wmesg = wmesg;
525 	lkp->lk_timo = timo;
526 	lkp->lk_lockholder = LK_NOTHREAD;
527 }
528 
529 /*
530  * Reinitialize a lock that is being reused for a different purpose, but
531  * which may have pending (blocked) threads sitting on it.  The caller
532  * must already hold the interlock.
533  */
534 void
535 lockreinit(struct lock *lkp, char *wmesg, int timo, int flags)
536 {
537 	lkp->lk_flags = (lkp->lk_flags & ~(LK_EXTFLG_MASK|LK_DRAINING)) |
538 			(flags & LK_EXTFLG_MASK);
539 	lkp->lk_wmesg = wmesg;
540 	lkp->lk_timo = timo;
541 }
542 
543 /*
544  * Determine the status of a lock.
545  */
546 int
547 lockstatus(struct lock *lkp, struct thread *td)
548 {
549 	int lock_type = 0;
550 
551 	spin_lock(&lkp->lk_spinlock);
552 	if (lkp->lk_exclusivecount != 0) {
553 		if (td == NULL || lkp->lk_lockholder == td)
554 			lock_type = LK_EXCLUSIVE;
555 		else
556 			lock_type = LK_EXCLOTHER;
557 	} else if (lkp->lk_sharecount != 0) {
558 		lock_type = LK_SHARED;
559 	}
560 	spin_unlock(&lkp->lk_spinlock);
561 	return (lock_type);
562 }
563 
564 /*
565  * Determine the number of holders of a lock.
566  *
567  * The non-blocking version can usually be used for assertions.
568  */
569 int
570 lockcount(struct lock *lkp)
571 {
572 	int count;
573 
574 	spin_lock(&lkp->lk_spinlock);
575 	count = lkp->lk_exclusivecount + lkp->lk_sharecount;
576 	spin_unlock(&lkp->lk_spinlock);
577 	return (count);
578 }
579 
580 int
581 lockcountnb(struct lock *lkp)
582 {
583 	return (lkp->lk_exclusivecount + lkp->lk_sharecount);
584 }
585 
586 /*
587  * Print out information about state of a lock. Used by VOP_PRINT
588  * routines to display status about contained locks.
589  */
590 void
591 lockmgr_printinfo(struct lock *lkp)
592 {
593 	struct thread *td = lkp->lk_lockholder;
594 	struct proc *p;
595 
596 	if (td && td != LK_KERNTHREAD && td != LK_NOTHREAD)
597 		p = td->td_proc;
598 	else
599 		p = NULL;
600 
601 	if (lkp->lk_sharecount)
602 		printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg,
603 		    lkp->lk_sharecount);
604 	else if (lkp->lk_flags & LK_HAVE_EXCL)
605 		printf(" lock type %s: EXCL (count %d) by td %p pid %d",
606 		    lkp->lk_wmesg, lkp->lk_exclusivecount, td,
607 		    p ? p->p_pid : -99);
608 	if (lkp->lk_waitcount > 0)
609 		printf(" with %d pending", lkp->lk_waitcount);
610 }
611 
612