kern_lock.c revision 66615
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: head/sys/kern/kern_lock.c 66615 2000-10-04 01:29:17Z jasone $ 42 */ 43 44#include <sys/param.h> 45#include <sys/proc.h> 46#include <sys/lock.h> 47#include <sys/systm.h> 48 49#include <machine/mutex.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 74static int acquire(struct lock *lkp, int extflags, int wanted); 75static int apause(struct lock *lkp, int flags); 76static int acquiredrain(struct lock *lkp, int extflags) ; 77 78static LOCK_INLINE void 79sharelock(struct lock *lkp, int incr) { 80 lkp->lk_flags |= LK_SHARE_NONZERO; 81 lkp->lk_sharecount += incr; 82} 83 84static LOCK_INLINE void 85shareunlock(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 */ 105static int 106apause(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 mtx_exit(&lkp->lk_interlock, MTX_DEF); 117 for (i = LOCK_SAMPLE_WAIT; i > 0; i--) 118 if ((lkp->lk_flags & flags) == 0) 119 break; 120 mtx_enter(&lkp->lk_interlock, MTX_DEF); 121 if ((lkp->lk_flags & flags) == 0) 122 return 0; 123 } 124#endif 125 return 1; 126} 127 128static int 129acquire(struct lock *lkp, int extflags, int wanted) { 130 int s, error; 131 132 CTR3(KTR_LOCKMGR, 133 "acquire(): lkp == %p, extflags == 0x%x, wanted == 0x%x\n", 134 lkp, extflags, wanted); 135 136 if ((extflags & LK_NOWAIT) && (lkp->lk_flags & wanted)) { 137 return EBUSY; 138 } 139 140 if (((lkp->lk_flags | extflags) & LK_NOPAUSE) == 0) { 141 error = apause(lkp, wanted); 142 if (error == 0) 143 return 0; 144 } 145 146 s = splhigh(); 147 while ((lkp->lk_flags & wanted) != 0) { 148 lkp->lk_flags |= LK_WAIT_NONZERO; 149 lkp->lk_waitcount++; 150 mtx_exit(&lkp->lk_interlock, MTX_DEF); 151 error = tsleep(lkp, lkp->lk_prio, lkp->lk_wmesg, lkp->lk_timo); 152 mtx_enter(&lkp->lk_interlock, MTX_DEF); 153 if (lkp->lk_waitcount == 1) { 154 lkp->lk_flags &= ~LK_WAIT_NONZERO; 155 lkp->lk_waitcount = 0; 156 } else { 157 lkp->lk_waitcount--; 158 } 159 if (error) { 160 splx(s); 161 return error; 162 } 163 if (extflags & LK_SLEEPFAIL) { 164 splx(s); 165 return ENOLCK; 166 } 167 } 168 splx(s); 169 return 0; 170} 171 172/* 173 * Set, change, or release a lock. 174 * 175 * Shared requests increment the shared count. Exclusive requests set the 176 * LK_WANT_EXCL flag (preventing further shared locks), and wait for already 177 * accepted shared locks and shared-to-exclusive upgrades to go away. 178 */ 179int 180#ifndef DEBUG_LOCKS 181lockmgr(lkp, flags, interlkp, p) 182#else 183debuglockmgr(lkp, flags, interlkp, p, name, file, line) 184#endif 185 struct lock *lkp; 186 u_int flags; 187 struct mtx *interlkp; 188 struct proc *p; 189#ifdef DEBUG_LOCKS 190 const char *name; /* Name of lock function */ 191 const char *file; /* Name of file call is from */ 192 int line; /* Line number in file */ 193#endif 194{ 195 int error; 196 pid_t pid; 197 int extflags; 198 199 CTR5(KTR_LOCKMGR, 200 "lockmgr(): lkp == %p (lk_wmesg == \"%s\"), flags == 0x%x, " 201 "interlkp == %p, p == %p", lkp, lkp->lk_wmesg, flags, interlkp, p); 202 203 error = 0; 204 if (p == NULL) 205 pid = LK_KERNPROC; 206 else 207 pid = p->p_pid; 208 209 mtx_enter(&lkp->lk_interlock, MTX_DEF); 210 if (flags & LK_INTERLOCK) 211 mtx_exit(interlkp, MTX_DEF); 212 213 extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK; 214 215 switch (flags & LK_TYPE_MASK) { 216 217 case LK_SHARED: 218 /* 219 * If we are not the exclusive lock holder, we have to block 220 * while there is an exclusive lock holder or while an 221 * exclusive lock request or upgrade request is in progress. 222 * 223 * However, if P_DEADLKTREAT is set, we override exclusive 224 * lock requests or upgrade requests ( but not the exclusive 225 * lock itself ). 226 */ 227 if (lkp->lk_lockholder != pid) { 228 if (p && (p->p_flag & P_DEADLKTREAT)) { 229 error = acquire( 230 lkp, 231 extflags, 232 LK_HAVE_EXCL 233 ); 234 } else { 235 error = acquire( 236 lkp, 237 extflags, 238 LK_HAVE_EXCL | LK_WANT_EXCL | 239 LK_WANT_UPGRADE 240 ); 241 } 242 if (error) 243 break; 244 sharelock(lkp, 1); 245 COUNT(p, 1); 246 break; 247 } 248 /* 249 * We hold an exclusive lock, so downgrade it to shared. 250 * An alternative would be to fail with EDEADLK. 251 */ 252 sharelock(lkp, 1); 253 COUNT(p, 1); 254 /* fall into downgrade */ 255 256 case LK_DOWNGRADE: 257 if (lkp->lk_lockholder != pid || lkp->lk_exclusivecount == 0) 258 panic("lockmgr: not holding exclusive lock"); 259 sharelock(lkp, lkp->lk_exclusivecount); 260 lkp->lk_exclusivecount = 0; 261 lkp->lk_flags &= ~LK_HAVE_EXCL; 262 lkp->lk_lockholder = LK_NOPROC; 263 if (lkp->lk_waitcount) 264 wakeup((void *)lkp); 265 break; 266 267 case LK_EXCLUPGRADE: 268 /* 269 * If another process is ahead of us to get an upgrade, 270 * then we want to fail rather than have an intervening 271 * exclusive access. 272 */ 273 if (lkp->lk_flags & LK_WANT_UPGRADE) { 274 shareunlock(lkp, 1); 275 COUNT(p, -1); 276 error = EBUSY; 277 break; 278 } 279 /* fall into normal upgrade */ 280 281 case LK_UPGRADE: 282 /* 283 * Upgrade a shared lock to an exclusive one. If another 284 * shared lock has already requested an upgrade to an 285 * exclusive lock, our shared lock is released and an 286 * exclusive lock is requested (which will be granted 287 * after the upgrade). If we return an error, the file 288 * will always be unlocked. 289 */ 290 if ((lkp->lk_lockholder == pid) || (lkp->lk_sharecount <= 0)) 291 panic("lockmgr: upgrade exclusive lock"); 292 shareunlock(lkp, 1); 293 COUNT(p, -1); 294 /* 295 * If we are just polling, check to see if we will block. 296 */ 297 if ((extflags & LK_NOWAIT) && 298 ((lkp->lk_flags & LK_WANT_UPGRADE) || 299 lkp->lk_sharecount > 1)) { 300 error = EBUSY; 301 break; 302 } 303 if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) { 304 /* 305 * We are first shared lock to request an upgrade, so 306 * request upgrade and wait for the shared count to 307 * drop to zero, then take exclusive lock. 308 */ 309 lkp->lk_flags |= LK_WANT_UPGRADE; 310 error = acquire(lkp, extflags, LK_SHARE_NONZERO); 311 lkp->lk_flags &= ~LK_WANT_UPGRADE; 312 313 if (error) 314 break; 315 lkp->lk_flags |= LK_HAVE_EXCL; 316 lkp->lk_lockholder = pid; 317 if (lkp->lk_exclusivecount != 0) 318 panic("lockmgr: non-zero exclusive count"); 319 lkp->lk_exclusivecount = 1; 320#if defined(DEBUG_LOCKS) 321 lkp->lk_filename = file; 322 lkp->lk_lineno = line; 323 lkp->lk_lockername = name; 324#endif 325 COUNT(p, 1); 326 break; 327 } 328 /* 329 * Someone else has requested upgrade. Release our shared 330 * lock, awaken upgrade requestor if we are the last shared 331 * lock, then request an exclusive lock. 332 */ 333 if ( (lkp->lk_flags & (LK_SHARE_NONZERO|LK_WAIT_NONZERO)) == 334 LK_WAIT_NONZERO) 335 wakeup((void *)lkp); 336 /* fall into exclusive request */ 337 338 case LK_EXCLUSIVE: 339 if (lkp->lk_lockholder == pid && pid != LK_KERNPROC) { 340 /* 341 * Recursive lock. 342 */ 343 if ((extflags & (LK_NOWAIT | LK_CANRECURSE)) == 0) 344 panic("lockmgr: locking against myself"); 345 if ((extflags & LK_CANRECURSE) != 0) { 346 lkp->lk_exclusivecount++; 347 COUNT(p, 1); 348 break; 349 } 350 } 351 /* 352 * If we are just polling, check to see if we will sleep. 353 */ 354 if ((extflags & LK_NOWAIT) && 355 (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO))) { 356 error = EBUSY; 357 break; 358 } 359 /* 360 * Try to acquire the want_exclusive flag. 361 */ 362 error = acquire(lkp, extflags, (LK_HAVE_EXCL | LK_WANT_EXCL)); 363 if (error) 364 break; 365 lkp->lk_flags |= LK_WANT_EXCL; 366 /* 367 * Wait for shared locks and upgrades to finish. 368 */ 369 error = acquire(lkp, extflags, LK_WANT_UPGRADE | LK_SHARE_NONZERO); 370 lkp->lk_flags &= ~LK_WANT_EXCL; 371 if (error) 372 break; 373 lkp->lk_flags |= LK_HAVE_EXCL; 374 lkp->lk_lockholder = pid; 375 if (lkp->lk_exclusivecount != 0) 376 panic("lockmgr: non-zero exclusive count"); 377 lkp->lk_exclusivecount = 1; 378#if defined(DEBUG_LOCKS) 379 lkp->lk_filename = file; 380 lkp->lk_lineno = line; 381 lkp->lk_lockername = name; 382#endif 383 COUNT(p, 1); 384 break; 385 386 case LK_RELEASE: 387 if (lkp->lk_exclusivecount != 0) { 388 if (lkp->lk_lockholder != pid && 389 lkp->lk_lockholder != LK_KERNPROC) { 390 panic("lockmgr: pid %d, not %s %d unlocking", 391 pid, "exclusive lock holder", 392 lkp->lk_lockholder); 393 } 394 if (lkp->lk_lockholder != LK_KERNPROC) { 395 COUNT(p, -1); 396 } 397 if (lkp->lk_exclusivecount == 1) { 398 lkp->lk_flags &= ~LK_HAVE_EXCL; 399 lkp->lk_lockholder = LK_NOPROC; 400 lkp->lk_exclusivecount = 0; 401 } else { 402 lkp->lk_exclusivecount--; 403 } 404 } else if (lkp->lk_flags & LK_SHARE_NONZERO) { 405 shareunlock(lkp, 1); 406 COUNT(p, -1); 407 } 408 if (lkp->lk_flags & LK_WAIT_NONZERO) 409 wakeup((void *)lkp); 410 break; 411 412 case LK_DRAIN: 413 /* 414 * Check that we do not already hold the lock, as it can 415 * never drain if we do. Unfortunately, we have no way to 416 * check for holding a shared lock, but at least we can 417 * check for an exclusive one. 418 */ 419 if (lkp->lk_lockholder == pid) 420 panic("lockmgr: draining against myself"); 421 422 error = acquiredrain(lkp, extflags); 423 if (error) 424 break; 425 lkp->lk_flags |= LK_DRAINING | LK_HAVE_EXCL; 426 lkp->lk_lockholder = pid; 427 lkp->lk_exclusivecount = 1; 428#if defined(DEBUG_LOCKS) 429 lkp->lk_filename = file; 430 lkp->lk_lineno = line; 431 lkp->lk_lockername = name; 432#endif 433 COUNT(p, 1); 434 break; 435 436 default: 437 mtx_exit(&lkp->lk_interlock, MTX_DEF); 438 panic("lockmgr: unknown locktype request %d", 439 flags & LK_TYPE_MASK); 440 /* NOTREACHED */ 441 } 442 if ((lkp->lk_flags & LK_WAITDRAIN) && 443 (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | 444 LK_SHARE_NONZERO | LK_WAIT_NONZERO)) == 0) { 445 lkp->lk_flags &= ~LK_WAITDRAIN; 446 wakeup((void *)&lkp->lk_flags); 447 } 448 mtx_exit(&lkp->lk_interlock, MTX_DEF); 449 return (error); 450} 451 452static int 453acquiredrain(struct lock *lkp, int extflags) { 454 int error; 455 456 if ((extflags & LK_NOWAIT) && (lkp->lk_flags & LK_ALL)) { 457 return EBUSY; 458 } 459 460 error = apause(lkp, LK_ALL); 461 if (error == 0) 462 return 0; 463 464 while (lkp->lk_flags & LK_ALL) { 465 lkp->lk_flags |= LK_WAITDRAIN; 466 mtx_exit(&lkp->lk_interlock, MTX_DEF); 467 error = tsleep(&lkp->lk_flags, lkp->lk_prio, 468 lkp->lk_wmesg, lkp->lk_timo); 469 mtx_enter(&lkp->lk_interlock, MTX_DEF); 470 if (error) 471 return error; 472 if (extflags & LK_SLEEPFAIL) { 473 return ENOLCK; 474 } 475 } 476 return 0; 477} 478 479/* 480 * Initialize a lock; required before use. 481 */ 482void 483lockinit(lkp, prio, wmesg, timo, flags) 484 struct lock *lkp; 485 int prio; 486 char *wmesg; 487 int timo; 488 int flags; 489{ 490 CTR5(KTR_LOCKMGR, "lockinit(): lkp == %p, prio == %d, wmesg == \"%s\", " 491 "timo == %d, flags = 0x%x\n", lkp, prio, wmesg, timo, flags); 492 493 if (lkp->lk_flags & LK_VALID) 494 lockdestroy(lkp); 495 496 mtx_init(&lkp->lk_interlock, "lockmgr interlock", MTX_DEF); 497 lkp->lk_flags = (flags & LK_EXTFLG_MASK) | LK_VALID; 498 lkp->lk_sharecount = 0; 499 lkp->lk_waitcount = 0; 500 lkp->lk_exclusivecount = 0; 501 lkp->lk_prio = prio; 502 lkp->lk_wmesg = wmesg; 503 lkp->lk_timo = timo; 504 lkp->lk_lockholder = LK_NOPROC; 505} 506 507/* 508 * Destroy a lock. 509 */ 510void 511lockdestroy(lkp) 512 struct lock *lkp; 513{ 514 CTR2(KTR_LOCKMGR, "lockdestroy(): lkp == %p (lk_wmesg == \"%s\")", 515 lkp, lkp->lk_wmesg); 516 if (lkp->lk_flags & LK_VALID) { 517 lkp->lk_flags &= ~LK_VALID; 518 mtx_destroy(&lkp->lk_interlock); 519 } 520} 521 522/* 523 * Determine the status of a lock. 524 */ 525int 526lockstatus(lkp, p) 527 struct lock *lkp; 528 struct proc *p; 529{ 530 int lock_type = 0; 531 532 mtx_enter(&lkp->lk_interlock, MTX_DEF); 533 if (lkp->lk_exclusivecount != 0) { 534 if (p == NULL || lkp->lk_lockholder == p->p_pid) 535 lock_type = LK_EXCLUSIVE; 536 else 537 lock_type = LK_EXCLOTHER; 538 } else if (lkp->lk_sharecount != 0) 539 lock_type = LK_SHARED; 540 mtx_exit(&lkp->lk_interlock, MTX_DEF); 541 return (lock_type); 542} 543 544/* 545 * Determine the number of holders of a lock. 546 */ 547int 548lockcount(lkp) 549 struct lock *lkp; 550{ 551 int count; 552 553 mtx_enter(&lkp->lk_interlock, MTX_DEF); 554 count = lkp->lk_exclusivecount + lkp->lk_sharecount; 555 mtx_exit(&lkp->lk_interlock, MTX_DEF); 556 return (count); 557} 558 559/* 560 * Print out information about state of a lock. Used by VOP_PRINT 561 * routines to display status about contained locks. 562 */ 563void 564lockmgr_printinfo(lkp) 565 struct lock *lkp; 566{ 567 568 if (lkp->lk_sharecount) 569 printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg, 570 lkp->lk_sharecount); 571 else if (lkp->lk_flags & LK_HAVE_EXCL) 572 printf(" lock type %s: EXCL (count %d) by pid %d", 573 lkp->lk_wmesg, lkp->lk_exclusivecount, lkp->lk_lockholder); 574 if (lkp->lk_waitcount > 0) 575 printf(" with %d pending", lkp->lk_waitcount); 576} 577 578#if defined(SIMPLELOCK_DEBUG) && (MAXCPU == 1 || defined(COMPILING_LINT)) 579#include <sys/kernel.h> 580#include <sys/sysctl.h> 581 582static int lockpausetime = 0; 583SYSCTL_INT(_debug, OID_AUTO, lockpausetime, CTLFLAG_RW, &lockpausetime, 0, ""); 584 585static int simplelockrecurse; 586 587/* 588 * Simple lock functions so that the debugger can see from whence 589 * they are being called. 590 */ 591void 592simple_lock_init(alp) 593 struct simplelock *alp; 594{ 595 596 alp->lock_data = 0; 597} 598 599void 600_simple_lock(alp, id, l) 601 struct simplelock *alp; 602 const char *id; 603 int l; 604{ 605 606 if (simplelockrecurse) 607 return; 608 if (alp->lock_data == 1) { 609 if (lockpausetime == -1) 610 panic("%s:%d: simple_lock: lock held", id, l); 611 printf("%s:%d: simple_lock: lock held\n", id, l); 612 if (lockpausetime == 1) { 613 Debugger("simple_lock"); 614 /*BACKTRACE(curproc); */ 615 } else if (lockpausetime > 1) { 616 printf("%s:%d: simple_lock: lock held...", id, l); 617 tsleep(&lockpausetime, PCATCH | PPAUSE, "slock", 618 lockpausetime * hz); 619 printf(" continuing\n"); 620 } 621 } 622 alp->lock_data = 1; 623 if (curproc) 624 curproc->p_simple_locks++; 625} 626 627int 628_simple_lock_try(alp, id, l) 629 struct simplelock *alp; 630 const char *id; 631 int l; 632{ 633 634 if (alp->lock_data) 635 return (0); 636 if (simplelockrecurse) 637 return (1); 638 alp->lock_data = 1; 639 if (curproc) 640 curproc->p_simple_locks++; 641 return (1); 642} 643 644void 645_simple_unlock(alp, id, l) 646 struct simplelock *alp; 647 const char *id; 648 int l; 649{ 650 651 if (simplelockrecurse) 652 return; 653 if (alp->lock_data == 0) { 654 if (lockpausetime == -1) 655 panic("%s:%d: simple_unlock: lock not held", id, l); 656 printf("%s:%d: simple_unlock: lock not held\n", id, l); 657 if (lockpausetime == 1) { 658 Debugger("simple_unlock"); 659 /* BACKTRACE(curproc); */ 660 } else if (lockpausetime > 1) { 661 printf("%s:%d: simple_unlock: lock not held...", id, l); 662 tsleep(&lockpausetime, PCATCH | PPAUSE, "sunlock", 663 lockpausetime * hz); 664 printf(" continuing\n"); 665 } 666 } 667 alp->lock_data = 0; 668 if (curproc) 669 curproc->p_simple_locks--; 670} 671#elif defined(SIMPLELOCK_DEBUG) 672#error "SIMPLELOCK_DEBUG is not compatible with SMP!" 673#endif /* SIMPLELOCK_DEBUG && MAXCPU == 1 */ 674