proc->thread stage 4: rework the VFS and DEVICE subsystems to take thread
[dragonfly.git] / sys / kern / kern_lock.c
CommitLineData
984263bc
MD
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 $
dadab5e9 42 * $DragonFly: src/sys/kern/kern_lock.c,v 1.3 2003/06/25 03:55:57 dillon Exp $
984263bc
MD
43 */
44
45#include "opt_lint.h"
46
47#include <sys/param.h>
48#include <sys/proc.h>
49#include <sys/lock.h>
50#include <sys/systm.h>
51
52/*
53 * Locking primitives implementation.
54 * Locks provide shared/exclusive sychronization.
55 */
56
57#ifdef SIMPLELOCK_DEBUG
dadab5e9 58#define COUNT(td, x) (td)->td_locks += (x)
984263bc 59#else
dadab5e9 60#define COUNT(td, x)
984263bc
MD
61#endif
62
63#define LOCK_WAIT_TIME 100
64#define LOCK_SAMPLE_WAIT 7
65
66#if defined(DIAGNOSTIC)
67#define LOCK_INLINE
68#else
69#define LOCK_INLINE __inline
70#endif
71
72#define LK_ALL (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | \
73 LK_SHARE_NONZERO | LK_WAIT_NONZERO)
74
75static int acquire(struct lock *lkp, int extflags, int wanted);
76static int apause(struct lock *lkp, int flags);
77static int acquiredrain(struct lock *lkp, int extflags) ;
78
79static LOCK_INLINE void
80sharelock(struct lock *lkp, int incr) {
81 lkp->lk_flags |= LK_SHARE_NONZERO;
82 lkp->lk_sharecount += incr;
83}
84
85static LOCK_INLINE void
86shareunlock(struct lock *lkp, int decr) {
87
88 KASSERT(lkp->lk_sharecount >= decr, ("shareunlock: count < decr"));
89
90 if (lkp->lk_sharecount == decr) {
91 lkp->lk_flags &= ~LK_SHARE_NONZERO;
92 if (lkp->lk_flags & (LK_WANT_UPGRADE | LK_WANT_EXCL)) {
93 wakeup(lkp);
94 }
95 lkp->lk_sharecount = 0;
96 } else {
97 lkp->lk_sharecount -= decr;
98 }
99}
100
101/*
102 * This is the waitloop optimization, and note for this to work
103 * simple_lock and simple_unlock should be subroutines to avoid
104 * optimization troubles.
105 */
106static int
107apause(struct lock *lkp, int flags)
108{
109#ifdef SMP
110 int i, lock_wait;
111#endif
112
113 if ((lkp->lk_flags & flags) == 0)
114 return 0;
115#ifdef SMP
116 for (lock_wait = LOCK_WAIT_TIME; lock_wait > 0; lock_wait--) {
117 simple_unlock(&lkp->lk_interlock);
118 for (i = LOCK_SAMPLE_WAIT; i > 0; i--)
119 if ((lkp->lk_flags & flags) == 0)
120 break;
121 simple_lock(&lkp->lk_interlock);
122 if ((lkp->lk_flags & flags) == 0)
123 return 0;
124 }
125#endif
126 return 1;
127}
128
129static int
130acquire(struct lock *lkp, int extflags, int wanted) {
131 int s, error;
132
133 if ((extflags & LK_NOWAIT) && (lkp->lk_flags & wanted)) {
134 return EBUSY;
135 }
136
137 if (((lkp->lk_flags | extflags) & LK_NOPAUSE) == 0) {
138 error = apause(lkp, wanted);
139 if (error == 0)
140 return 0;
141 }
142
143 s = splhigh();
144 while ((lkp->lk_flags & wanted) != 0) {
145 lkp->lk_flags |= LK_WAIT_NONZERO;
146 lkp->lk_waitcount++;
147 simple_unlock(&lkp->lk_interlock);
148 error = tsleep(lkp, lkp->lk_prio, lkp->lk_wmesg,
149 ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0));
150 simple_lock(&lkp->lk_interlock);
151 if (lkp->lk_waitcount == 1) {
152 lkp->lk_flags &= ~LK_WAIT_NONZERO;
153 lkp->lk_waitcount = 0;
154 } else {
155 lkp->lk_waitcount--;
156 }
157 if (error) {
158 splx(s);
159 return error;
160 }
161 if (extflags & LK_SLEEPFAIL) {
162 splx(s);
163 return ENOLCK;
164 }
165 }
166 splx(s);
167 return 0;
168}
169
170/*
171 * Set, change, or release a lock.
172 *
173 * Shared requests increment the shared count. Exclusive requests set the
174 * LK_WANT_EXCL flag (preventing further shared locks), and wait for already
175 * accepted shared locks and shared-to-exclusive upgrades to go away.
176 */
177int
178#ifndef DEBUG_LOCKS
dadab5e9
MD
179lockmgr(struct lock *lkp, u_int flags, struct simplelock *interlkp,
180 struct thread *td)
984263bc 181#else
dadab5e9
MD
182debuglockmgr(struct lock *lkp, u_int flags, struct simplelock *interlkp,
183 struct thread *td, const char *name, const char *file, int line)
984263bc
MD
184#endif
185{
186 int error;
984263bc
MD
187 int extflags;
188
189 error = 0;
984263bc
MD
190
191 simple_lock(&lkp->lk_interlock);
192 if (flags & LK_INTERLOCK)
193 simple_unlock(interlkp);
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 P_DEADLKTREAT is set, we override exclusive
206 * lock requests or upgrade requests ( but not the exclusive
207 * lock itself ).
208 */
dadab5e9
MD
209 if (lkp->lk_lockholder != td) {
210 if (td->td_flags & TDF_DEADLKTREAT) {
984263bc
MD
211 error = acquire(
212 lkp,
213 extflags,
214 LK_HAVE_EXCL
215 );
216 } else {
217 error = acquire(
218 lkp,
219 extflags,
220 LK_HAVE_EXCL | LK_WANT_EXCL |
221 LK_WANT_UPGRADE
222 );
223 }
224 if (error)
225 break;
226 sharelock(lkp, 1);
227 COUNT(p, 1);
228 break;
229 }
230 /*
231 * We hold an exclusive lock, so downgrade it to shared.
232 * An alternative would be to fail with EDEADLK.
233 */
234 sharelock(lkp, 1);
235 COUNT(p, 1);
236 /* fall into downgrade */
237
238 case LK_DOWNGRADE:
dadab5e9 239 if (lkp->lk_lockholder != td || lkp->lk_exclusivecount == 0)
984263bc
MD
240 panic("lockmgr: not holding exclusive lock");
241 sharelock(lkp, lkp->lk_exclusivecount);
242 lkp->lk_exclusivecount = 0;
243 lkp->lk_flags &= ~LK_HAVE_EXCL;
dadab5e9 244 lkp->lk_lockholder = LK_NOTHREAD;
984263bc
MD
245 if (lkp->lk_waitcount)
246 wakeup((void *)lkp);
247 break;
248
249 case LK_EXCLUPGRADE:
250 /*
251 * If another process is ahead of us to get an upgrade,
252 * then we want to fail rather than have an intervening
253 * exclusive access.
254 */
255 if (lkp->lk_flags & LK_WANT_UPGRADE) {
256 shareunlock(lkp, 1);
257 COUNT(p, -1);
258 error = EBUSY;
259 break;
260 }
261 /* fall into normal upgrade */
262
263 case LK_UPGRADE:
264 /*
265 * Upgrade a shared lock to an exclusive one. If another
266 * shared lock has already requested an upgrade to an
267 * exclusive lock, our shared lock is released and an
268 * exclusive lock is requested (which will be granted
269 * after the upgrade). If we return an error, the file
270 * will always be unlocked.
271 */
dadab5e9 272 if ((lkp->lk_lockholder == td) || (lkp->lk_sharecount <= 0))
984263bc
MD
273 panic("lockmgr: upgrade exclusive lock");
274 shareunlock(lkp, 1);
275 COUNT(p, -1);
276 /*
277 * If we are just polling, check to see if we will block.
278 */
279 if ((extflags & LK_NOWAIT) &&
280 ((lkp->lk_flags & LK_WANT_UPGRADE) ||
281 lkp->lk_sharecount > 1)) {
282 error = EBUSY;
283 break;
284 }
285 if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) {
286 /*
287 * We are first shared lock to request an upgrade, so
288 * request upgrade and wait for the shared count to
289 * drop to zero, then take exclusive lock.
290 */
291 lkp->lk_flags |= LK_WANT_UPGRADE;
292 error = acquire(lkp, extflags, LK_SHARE_NONZERO);
293 lkp->lk_flags &= ~LK_WANT_UPGRADE;
294
295 if (error)
296 break;
297 lkp->lk_flags |= LK_HAVE_EXCL;
dadab5e9 298 lkp->lk_lockholder = td;
984263bc
MD
299 if (lkp->lk_exclusivecount != 0)
300 panic("lockmgr: non-zero exclusive count");
301 lkp->lk_exclusivecount = 1;
302#if defined(DEBUG_LOCKS)
303 lkp->lk_filename = file;
304 lkp->lk_lineno = line;
305 lkp->lk_lockername = name;
306#endif
307 COUNT(p, 1);
308 break;
309 }
310 /*
311 * Someone else has requested upgrade. Release our shared
312 * lock, awaken upgrade requestor if we are the last shared
313 * lock, then request an exclusive lock.
314 */
315 if ( (lkp->lk_flags & (LK_SHARE_NONZERO|LK_WAIT_NONZERO)) ==
316 LK_WAIT_NONZERO)
317 wakeup((void *)lkp);
318 /* fall into exclusive request */
319
320 case LK_EXCLUSIVE:
dadab5e9 321 if (lkp->lk_lockholder == td && td != LK_KERNTHREAD) {
984263bc
MD
322 /*
323 * Recursive lock.
324 */
325 if ((extflags & (LK_NOWAIT | LK_CANRECURSE)) == 0)
326 panic("lockmgr: locking against myself");
327 if ((extflags & LK_CANRECURSE) != 0) {
328 lkp->lk_exclusivecount++;
329 COUNT(p, 1);
330 break;
331 }
332 }
333 /*
334 * If we are just polling, check to see if we will sleep.
335 */
336 if ((extflags & LK_NOWAIT) &&
337 (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO))) {
338 error = EBUSY;
339 break;
340 }
341 /*
342 * Try to acquire the want_exclusive flag.
343 */
344 error = acquire(lkp, extflags, (LK_HAVE_EXCL | LK_WANT_EXCL));
345 if (error)
346 break;
347 lkp->lk_flags |= LK_WANT_EXCL;
348 /*
349 * Wait for shared locks and upgrades to finish.
350 */
351 error = acquire(lkp, extflags, LK_WANT_UPGRADE | LK_SHARE_NONZERO);
352 lkp->lk_flags &= ~LK_WANT_EXCL;
353 if (error)
354 break;
355 lkp->lk_flags |= LK_HAVE_EXCL;
dadab5e9 356 lkp->lk_lockholder = td;
984263bc
MD
357 if (lkp->lk_exclusivecount != 0)
358 panic("lockmgr: non-zero exclusive count");
359 lkp->lk_exclusivecount = 1;
360#if defined(DEBUG_LOCKS)
361 lkp->lk_filename = file;
362 lkp->lk_lineno = line;
363 lkp->lk_lockername = name;
364#endif
365 COUNT(p, 1);
366 break;
367
368 case LK_RELEASE:
369 if (lkp->lk_exclusivecount != 0) {
dadab5e9
MD
370 if (lkp->lk_lockholder != td &&
371 lkp->lk_lockholder != LK_KERNTHREAD) {
372 panic("lockmgr: pid %d, not %s thr %p unlocking",
373 (td->td_proc ? td->td_proc->p_pid : -99),
374 "exclusive lock holder",
984263bc
MD
375 lkp->lk_lockholder);
376 }
dadab5e9 377 if (lkp->lk_lockholder != LK_KERNTHREAD) {
984263bc
MD
378 COUNT(p, -1);
379 }
380 if (lkp->lk_exclusivecount == 1) {
381 lkp->lk_flags &= ~LK_HAVE_EXCL;
dadab5e9 382 lkp->lk_lockholder = LK_NOTHREAD;
984263bc
MD
383 lkp->lk_exclusivecount = 0;
384 } else {
385 lkp->lk_exclusivecount--;
386 }
387 } else if (lkp->lk_flags & LK_SHARE_NONZERO) {
388 shareunlock(lkp, 1);
389 COUNT(p, -1);
390 }
391 if (lkp->lk_flags & LK_WAIT_NONZERO)
392 wakeup((void *)lkp);
393 break;
394
395 case LK_DRAIN:
396 /*
397 * Check that we do not already hold the lock, as it can
398 * never drain if we do. Unfortunately, we have no way to
399 * check for holding a shared lock, but at least we can
400 * check for an exclusive one.
401 */
dadab5e9 402 if (lkp->lk_lockholder == td)
984263bc
MD
403 panic("lockmgr: draining against myself");
404
405 error = acquiredrain(lkp, extflags);
406 if (error)
407 break;
408 lkp->lk_flags |= LK_DRAINING | LK_HAVE_EXCL;
dadab5e9 409 lkp->lk_lockholder = td;
984263bc
MD
410 lkp->lk_exclusivecount = 1;
411#if defined(DEBUG_LOCKS)
412 lkp->lk_filename = file;
413 lkp->lk_lineno = line;
414 lkp->lk_lockername = name;
415#endif
416 COUNT(p, 1);
417 break;
418
419 default:
420 simple_unlock(&lkp->lk_interlock);
421 panic("lockmgr: unknown locktype request %d",
422 flags & LK_TYPE_MASK);
423 /* NOTREACHED */
424 }
425 if ((lkp->lk_flags & LK_WAITDRAIN) &&
426 (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE |
427 LK_SHARE_NONZERO | LK_WAIT_NONZERO)) == 0) {
428 lkp->lk_flags &= ~LK_WAITDRAIN;
429 wakeup((void *)&lkp->lk_flags);
430 }
431 simple_unlock(&lkp->lk_interlock);
432 return (error);
433}
434
435static int
436acquiredrain(struct lock *lkp, int extflags) {
437 int error;
438
439 if ((extflags & LK_NOWAIT) && (lkp->lk_flags & LK_ALL)) {
440 return EBUSY;
441 }
442
443 error = apause(lkp, LK_ALL);
444 if (error == 0)
445 return 0;
446
447 while (lkp->lk_flags & LK_ALL) {
448 lkp->lk_flags |= LK_WAITDRAIN;
449 simple_unlock(&lkp->lk_interlock);
450 error = tsleep(&lkp->lk_flags, lkp->lk_prio,
451 lkp->lk_wmesg,
452 ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0));
453 simple_lock(&lkp->lk_interlock);
454 if (error)
455 return error;
456 if (extflags & LK_SLEEPFAIL) {
457 return ENOLCK;
458 }
459 }
460 return 0;
461}
462
463/*
464 * Initialize a lock; required before use.
465 */
466void
467lockinit(lkp, prio, wmesg, timo, flags)
468 struct lock *lkp;
469 int prio;
470 char *wmesg;
471 int timo;
472 int flags;
473{
474
475 simple_lock_init(&lkp->lk_interlock);
476 lkp->lk_flags = (flags & LK_EXTFLG_MASK);
477 lkp->lk_sharecount = 0;
478 lkp->lk_waitcount = 0;
479 lkp->lk_exclusivecount = 0;
480 lkp->lk_prio = prio;
481 lkp->lk_wmesg = wmesg;
482 lkp->lk_timo = timo;
dadab5e9 483 lkp->lk_lockholder = LK_NOTHREAD;
984263bc
MD
484}
485
486/*
487 * Determine the status of a lock.
488 */
489int
dadab5e9 490lockstatus(struct lock *lkp, struct thread *td)
984263bc
MD
491{
492 int lock_type = 0;
493
494 simple_lock(&lkp->lk_interlock);
495 if (lkp->lk_exclusivecount != 0) {
dadab5e9 496 if (td == NULL || lkp->lk_lockholder == td)
984263bc
MD
497 lock_type = LK_EXCLUSIVE;
498 else
499 lock_type = LK_EXCLOTHER;
500 } else if (lkp->lk_sharecount != 0)
501 lock_type = LK_SHARED;
502 simple_unlock(&lkp->lk_interlock);
503 return (lock_type);
504}
505
506/*
507 * Determine the number of holders of a lock.
508 */
509int
510lockcount(lkp)
511 struct lock *lkp;
512{
513 int count;
514
515 simple_lock(&lkp->lk_interlock);
516 count = lkp->lk_exclusivecount + lkp->lk_sharecount;
517 simple_unlock(&lkp->lk_interlock);
518 return (count);
519}
520
521/*
522 * Print out information about state of a lock. Used by VOP_PRINT
523 * routines to display status about contained locks.
524 */
525void
526lockmgr_printinfo(lkp)
527 struct lock *lkp;
528{
dadab5e9
MD
529 struct thread *td = lkp->lk_lockholder;
530 struct proc *p;
531
532 if (td && td != LK_KERNTHREAD && td != LK_NOTHREAD)
533 p = td->td_proc;
534 else
535 p = NULL;
984263bc
MD
536
537 if (lkp->lk_sharecount)
538 printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg,
539 lkp->lk_sharecount);
540 else if (lkp->lk_flags & LK_HAVE_EXCL)
dadab5e9
MD
541 printf(" lock type %s: EXCL (count %d) by td %p pid %d",
542 lkp->lk_wmesg, lkp->lk_exclusivecount, td,
543 p ? p->p_pid : -99);
984263bc
MD
544 if (lkp->lk_waitcount > 0)
545 printf(" with %d pending", lkp->lk_waitcount);
546}
547
548#if defined(SIMPLELOCK_DEBUG) && (MAXCPU == 1 || defined(COMPILING_LINT))
549#include <sys/kernel.h>
550#include <sys/sysctl.h>
551
552static int lockpausetime = 0;
553SYSCTL_INT(_debug, OID_AUTO, lockpausetime, CTLFLAG_RW, &lockpausetime, 0, "");
554
555static int simplelockrecurse;
556
557/*
558 * Simple lock functions so that the debugger can see from whence
559 * they are being called.
560 */
561void
562simple_lock_init(alp)
563 struct simplelock *alp;
564{
565
566 alp->lock_data = 0;
567}
568
569void
570_simple_lock(alp, id, l)
571 struct simplelock *alp;
572 const char *id;
573 int l;
574{
575
576 if (simplelockrecurse)
577 return;
578 if (alp->lock_data == 1) {
579 if (lockpausetime == -1)
580 panic("%s:%d: simple_lock: lock held", id, l);
581 printf("%s:%d: simple_lock: lock held\n", id, l);
582 if (lockpausetime == 1) {
583 Debugger("simple_lock");
584 /*BACKTRACE(curproc); */
585 } else if (lockpausetime > 1) {
586 printf("%s:%d: simple_lock: lock held...", id, l);
587 tsleep(&lockpausetime, PCATCH | PPAUSE, "slock",
588 lockpausetime * hz);
589 printf(" continuing\n");
590 }
591 }
592 alp->lock_data = 1;
593 if (curproc)
594 curproc->p_simple_locks++;
595}
596
597int
598_simple_lock_try(alp, id, l)
599 struct simplelock *alp;
600 const char *id;
601 int l;
602{
603
604 if (alp->lock_data)
605 return (0);
606 if (simplelockrecurse)
607 return (1);
608 alp->lock_data = 1;
609 if (curproc)
610 curproc->p_simple_locks++;
611 return (1);
612}
613
614void
615_simple_unlock(alp, id, l)
616 struct simplelock *alp;
617 const char *id;
618 int l;
619{
620
621 if (simplelockrecurse)
622 return;
623 if (alp->lock_data == 0) {
624 if (lockpausetime == -1)
625 panic("%s:%d: simple_unlock: lock not held", id, l);
626 printf("%s:%d: simple_unlock: lock not held\n", id, l);
627 if (lockpausetime == 1) {
628 Debugger("simple_unlock");
629 /* BACKTRACE(curproc); */
630 } else if (lockpausetime > 1) {
631 printf("%s:%d: simple_unlock: lock not held...", id, l);
632 tsleep(&lockpausetime, PCATCH | PPAUSE, "sunlock",
633 lockpausetime * hz);
634 printf(" continuing\n");
635 }
636 }
637 alp->lock_data = 0;
638 if (curproc)
639 curproc->p_simple_locks--;
640}
641#elif defined(SIMPLELOCK_DEBUG)
642#error "SIMPLELOCK_DEBUG is not compatible with SMP!"
643#endif /* SIMPLELOCK_DEBUG && MAXCPU == 1 */