kern_lock.c revision 124163
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 */ 42 43#include <sys/cdefs.h> 44__FBSDID("$FreeBSD: head/sys/kern/kern_lock.c 124163 2004-01-06 04:34:13Z kan $"); 45 46#include <sys/param.h> 47#include <sys/kernel.h> 48#include <sys/ktr.h> 49#include <sys/lock.h> 50#include <sys/lockmgr.h> 51#include <sys/mutex.h> 52#include <sys/proc.h> 53#include <sys/systm.h> 54 55/* 56 * Locking primitives implementation. 57 * Locks provide shared/exclusive sychronization. 58 */ 59 60#define LOCK_WAIT_TIME 100 61#define LOCK_SAMPLE_WAIT 7 62 63#if defined(DIAGNOSTIC) 64#define LOCK_INLINE 65#else 66#define LOCK_INLINE __inline 67#endif 68 69#define LK_ALL (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | \ 70 LK_SHARE_NONZERO | LK_WAIT_NONZERO) 71 72/* 73 * Mutex array variables. Rather than each lockmgr lock having its own mutex, 74 * share a fixed (at boot time) number of mutexes across all lockmgr locks in 75 * order to keep sizeof(struct lock) down. 76 */ 77static struct mtx lock_mtx; 78 79static int acquire(struct lock **lkpp, int extflags, int wanted); 80static int apause(struct lock *lkp, int flags); 81static int acquiredrain(struct lock *lkp, int extflags) ; 82 83static void 84lockmgr_init(void *dummy __unused) 85{ 86 mtx_init(&lock_mtx, "lockmgr", NULL, MTX_DEF); 87} 88SYSINIT(lmgrinit, SI_SUB_LOCKMGR, SI_ORDER_FIRST, lockmgr_init, NULL) 89 90static LOCK_INLINE void 91sharelock(struct lock *lkp, int incr) { 92 lkp->lk_flags |= LK_SHARE_NONZERO; 93 lkp->lk_sharecount += incr; 94} 95 96static LOCK_INLINE void 97shareunlock(struct lock *lkp, int decr) { 98 99 KASSERT(lkp->lk_sharecount >= decr, ("shareunlock: count < decr")); 100 101 if (lkp->lk_sharecount == decr) { 102 lkp->lk_flags &= ~LK_SHARE_NONZERO; 103 if (lkp->lk_flags & (LK_WANT_UPGRADE | LK_WANT_EXCL)) { 104 wakeup(lkp); 105 } 106 lkp->lk_sharecount = 0; 107 } else { 108 lkp->lk_sharecount -= decr; 109 } 110} 111 112/* 113 * This is the waitloop optimization. 114 */ 115static int 116apause(struct lock *lkp, int flags) 117{ 118#ifdef SMP 119 int i, lock_wait; 120#endif 121 122 if ((lkp->lk_flags & flags) == 0) 123 return 0; 124#ifdef SMP 125 for (lock_wait = LOCK_WAIT_TIME; lock_wait > 0; lock_wait--) { 126 mtx_unlock(lkp->lk_interlock); 127 for (i = LOCK_SAMPLE_WAIT; i > 0; i--) 128 if ((lkp->lk_flags & flags) == 0) 129 break; 130 mtx_lock(lkp->lk_interlock); 131 if ((lkp->lk_flags & flags) == 0) 132 return 0; 133 } 134#endif 135 return 1; 136} 137 138static int 139acquire(struct lock **lkpp, int extflags, int wanted) { 140 struct lock *lkp = *lkpp; 141 int s, error; 142 143 CTR3(KTR_LOCK, 144 "acquire(): lkp == %p, extflags == 0x%x, wanted == 0x%x\n", 145 lkp, extflags, wanted); 146 147 if ((extflags & LK_NOWAIT) && (lkp->lk_flags & wanted)) { 148 return EBUSY; 149 } 150 151 if (((lkp->lk_flags | extflags) & LK_NOPAUSE) == 0) { 152 error = apause(lkp, wanted); 153 if (error == 0) 154 return 0; 155 } 156 157 s = splhigh(); 158 while ((lkp->lk_flags & wanted) != 0) { 159 lkp->lk_flags |= LK_WAIT_NONZERO; 160 lkp->lk_waitcount++; 161 error = msleep(lkp, lkp->lk_interlock, lkp->lk_prio, 162 lkp->lk_wmesg, 163 ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0)); 164 if (lkp->lk_waitcount == 1) { 165 lkp->lk_flags &= ~LK_WAIT_NONZERO; 166 lkp->lk_waitcount = 0; 167 } else { 168 lkp->lk_waitcount--; 169 } 170 if (error) { 171 splx(s); 172 return error; 173 } 174 if (extflags & LK_SLEEPFAIL) { 175 splx(s); 176 return ENOLCK; 177 } 178 if (lkp->lk_newlock != NULL) { 179 mtx_lock(lkp->lk_newlock->lk_interlock); 180 mtx_unlock(lkp->lk_interlock); 181 if (lkp->lk_waitcount == 0) 182 wakeup((void *)(&lkp->lk_newlock)); 183 *lkpp = lkp = lkp->lk_newlock; 184 } 185 } 186 splx(s); 187 return 0; 188} 189 190/* 191 * Set, change, or release a lock. 192 * 193 * Shared requests increment the shared count. Exclusive requests set the 194 * LK_WANT_EXCL flag (preventing further shared locks), and wait for already 195 * accepted shared locks and shared-to-exclusive upgrades to go away. 196 */ 197int 198#ifndef DEBUG_LOCKS 199lockmgr(lkp, flags, interlkp, td) 200#else 201debuglockmgr(lkp, flags, interlkp, td, name, file, line) 202#endif 203 struct lock *lkp; 204 u_int flags; 205 struct mtx *interlkp; 206 struct thread *td; 207#ifdef DEBUG_LOCKS 208 const char *name; /* Name of lock function */ 209 const char *file; /* Name of file call is from */ 210 int line; /* Line number in file */ 211#endif 212{ 213 int error; 214 struct thread *thr; 215 int extflags, lockflags; 216 217 CTR5(KTR_LOCK, 218 "lockmgr(): lkp == %p (lk_wmesg == \"%s\"), flags == 0x%x, " 219 "interlkp == %p, td == %p", lkp, lkp->lk_wmesg, flags, interlkp, td); 220 221 error = 0; 222 if (td == NULL) 223 thr = LK_KERNPROC; 224 else 225 thr = td; 226 227 if ((flags & LK_INTERNAL) == 0) 228 mtx_lock(lkp->lk_interlock); 229 if (flags & LK_INTERLOCK) { 230 mtx_assert(interlkp, MA_OWNED | MA_NOTRECURSED); 231 mtx_unlock(interlkp); 232 } 233 234 if ((flags & (LK_NOWAIT|LK_RELEASE)) == 0) 235 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, 236 &lkp->lk_interlock->mtx_object, 237 "Acquiring lockmgr lock \"%s\"", lkp->lk_wmesg); 238 239 if (panicstr != NULL) { 240 mtx_unlock(lkp->lk_interlock); 241 return (0); 242 } 243 244 extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK; 245 246 switch (flags & LK_TYPE_MASK) { 247 248 case LK_SHARED: 249 /* 250 * If we are not the exclusive lock holder, we have to block 251 * while there is an exclusive lock holder or while an 252 * exclusive lock request or upgrade request is in progress. 253 * 254 * However, if TDF_DEADLKTREAT is set, we override exclusive 255 * lock requests or upgrade requests ( but not the exclusive 256 * lock itself ). 257 */ 258 if (lkp->lk_lockholder != thr) { 259 lockflags = LK_HAVE_EXCL; 260 mtx_lock_spin(&sched_lock); 261 if (td != NULL && !(td->td_flags & TDF_DEADLKTREAT)) 262 lockflags |= LK_WANT_EXCL | LK_WANT_UPGRADE; 263 mtx_unlock_spin(&sched_lock); 264 error = acquire(&lkp, extflags, lockflags); 265 if (error) 266 break; 267 sharelock(lkp, 1); 268#if defined(DEBUG_LOCKS) 269 lkp->lk_slockholder = thr; 270 lkp->lk_sfilename = file; 271 lkp->lk_slineno = line; 272 lkp->lk_slockername = name; 273#endif 274 break; 275 } 276 /* 277 * We hold an exclusive lock, so downgrade it to shared. 278 * An alternative would be to fail with EDEADLK. 279 */ 280 sharelock(lkp, 1); 281 /* FALLTHROUGH downgrade */ 282 283 case LK_DOWNGRADE: 284 KASSERT(lkp->lk_lockholder == thr && lkp->lk_exclusivecount != 0, 285 ("lockmgr: not holding exclusive lock " 286 "(owner thread (%p) != thread (%p), exlcnt (%d) != 0", 287 lkp->lk_lockholder, thr, lkp->lk_exclusivecount)); 288 sharelock(lkp, lkp->lk_exclusivecount); 289 lkp->lk_exclusivecount = 0; 290 lkp->lk_flags &= ~LK_HAVE_EXCL; 291 lkp->lk_lockholder = LK_NOPROC; 292 if (lkp->lk_waitcount) 293 wakeup((void *)lkp); 294 break; 295 296 case LK_EXCLUPGRADE: 297 /* 298 * If another process is ahead of us to get an upgrade, 299 * then we want to fail rather than have an intervening 300 * exclusive access. 301 */ 302 if (lkp->lk_flags & LK_WANT_UPGRADE) { 303 shareunlock(lkp, 1); 304 error = EBUSY; 305 break; 306 } 307 /* FALLTHROUGH normal upgrade */ 308 309 case LK_UPGRADE: 310 /* 311 * Upgrade a shared lock to an exclusive one. If another 312 * shared lock has already requested an upgrade to an 313 * exclusive lock, our shared lock is released and an 314 * exclusive lock is requested (which will be granted 315 * after the upgrade). If we return an error, the file 316 * will always be unlocked. 317 */ 318 if ((lkp->lk_lockholder == thr) || (lkp->lk_sharecount <= 0)) 319 panic("lockmgr: upgrade exclusive lock"); 320 shareunlock(lkp, 1); 321 /* 322 * If we are just polling, check to see if we will block. 323 */ 324 if ((extflags & LK_NOWAIT) && 325 ((lkp->lk_flags & LK_WANT_UPGRADE) || 326 lkp->lk_sharecount > 1)) { 327 error = EBUSY; 328 break; 329 } 330 if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) { 331 /* 332 * We are first shared lock to request an upgrade, so 333 * request upgrade and wait for the shared count to 334 * drop to zero, then take exclusive lock. 335 */ 336 lkp->lk_flags |= LK_WANT_UPGRADE; 337 error = acquire(&lkp, extflags, LK_SHARE_NONZERO); 338 lkp->lk_flags &= ~LK_WANT_UPGRADE; 339 340 if (error) 341 break; 342 lkp->lk_flags |= LK_HAVE_EXCL; 343 lkp->lk_lockholder = thr; 344 if (lkp->lk_exclusivecount != 0) 345 panic("lockmgr: non-zero exclusive count"); 346 lkp->lk_exclusivecount = 1; 347#if defined(DEBUG_LOCKS) 348 lkp->lk_filename = file; 349 lkp->lk_lineno = line; 350 lkp->lk_lockername = name; 351#endif 352 break; 353 } 354 /* 355 * Someone else has requested upgrade. Release our shared 356 * lock, awaken upgrade requestor if we are the last shared 357 * lock, then request an exclusive lock. 358 */ 359 if ( (lkp->lk_flags & (LK_SHARE_NONZERO|LK_WAIT_NONZERO)) == 360 LK_WAIT_NONZERO) 361 wakeup((void *)lkp); 362 /* FALLTHROUGH exclusive request */ 363 364 case LK_EXCLUSIVE: 365 if (lkp->lk_lockholder == thr && thr != LK_KERNPROC) { 366 /* 367 * Recursive lock. 368 */ 369 if ((extflags & (LK_NOWAIT | LK_CANRECURSE)) == 0) 370 panic("lockmgr: locking against myself"); 371 if ((extflags & LK_CANRECURSE) != 0) { 372 lkp->lk_exclusivecount++; 373 break; 374 } 375 } 376 /* 377 * If we are just polling, check to see if we will sleep. 378 */ 379 if ((extflags & LK_NOWAIT) && 380 (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO))) { 381 error = EBUSY; 382 break; 383 } 384 /* 385 * Try to acquire the want_exclusive flag. 386 */ 387 error = acquire(&lkp, extflags, (LK_HAVE_EXCL | LK_WANT_EXCL)); 388 if (error) 389 break; 390 lkp->lk_flags |= LK_WANT_EXCL; 391 /* 392 * Wait for shared locks and upgrades to finish. 393 */ 394 error = acquire(&lkp, extflags, LK_WANT_UPGRADE | LK_SHARE_NONZERO); 395 lkp->lk_flags &= ~LK_WANT_EXCL; 396 if (error) 397 break; 398 lkp->lk_flags |= LK_HAVE_EXCL; 399 lkp->lk_lockholder = thr; 400 if (lkp->lk_exclusivecount != 0) 401 panic("lockmgr: non-zero exclusive count"); 402 lkp->lk_exclusivecount = 1; 403#if defined(DEBUG_LOCKS) 404 lkp->lk_filename = file; 405 lkp->lk_lineno = line; 406 lkp->lk_lockername = name; 407#endif 408 break; 409 410 case LK_RELEASE: 411 if (lkp->lk_exclusivecount != 0) { 412 if (lkp->lk_lockholder != thr && 413 lkp->lk_lockholder != LK_KERNPROC) { 414 panic("lockmgr: thread %p, not %s %p unlocking", 415 thr, "exclusive lock holder", 416 lkp->lk_lockholder); 417 } 418 if (lkp->lk_exclusivecount == 1) { 419 lkp->lk_flags &= ~LK_HAVE_EXCL; 420 lkp->lk_lockholder = LK_NOPROC; 421 lkp->lk_exclusivecount = 0; 422 } else { 423 lkp->lk_exclusivecount--; 424 } 425 } else if (lkp->lk_flags & LK_SHARE_NONZERO) 426 shareunlock(lkp, 1); 427 if (lkp->lk_flags & LK_WAIT_NONZERO) 428 wakeup((void *)lkp); 429 break; 430 431 case LK_DRAIN: 432 /* 433 * Check that we do not already hold the lock, as it can 434 * never drain if we do. Unfortunately, we have no way to 435 * check for holding a shared lock, but at least we can 436 * check for an exclusive one. 437 */ 438 if (lkp->lk_lockholder == thr) 439 panic("lockmgr: draining against myself"); 440 441 error = acquiredrain(lkp, extflags); 442 if (error) 443 break; 444 lkp->lk_flags |= LK_DRAINING | LK_HAVE_EXCL; 445 lkp->lk_lockholder = thr; 446 lkp->lk_exclusivecount = 1; 447#if defined(DEBUG_LOCKS) 448 lkp->lk_filename = file; 449 lkp->lk_lineno = line; 450 lkp->lk_lockername = name; 451#endif 452 break; 453 454 default: 455 mtx_unlock(lkp->lk_interlock); 456 panic("lockmgr: unknown locktype request %d", 457 flags & LK_TYPE_MASK); 458 /* NOTREACHED */ 459 } 460 if ((lkp->lk_flags & LK_WAITDRAIN) && 461 (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | 462 LK_SHARE_NONZERO | LK_WAIT_NONZERO)) == 0) { 463 lkp->lk_flags &= ~LK_WAITDRAIN; 464 wakeup((void *)&lkp->lk_flags); 465 } 466 mtx_unlock(lkp->lk_interlock); 467 return (error); 468} 469 470static int 471acquiredrain(struct lock *lkp, int extflags) { 472 int error; 473 474 if ((extflags & LK_NOWAIT) && (lkp->lk_flags & LK_ALL)) { 475 return EBUSY; 476 } 477 478 error = apause(lkp, LK_ALL); 479 if (error == 0) 480 return 0; 481 482 while (lkp->lk_flags & LK_ALL) { 483 lkp->lk_flags |= LK_WAITDRAIN; 484 error = msleep(&lkp->lk_flags, lkp->lk_interlock, lkp->lk_prio, 485 lkp->lk_wmesg, 486 ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0)); 487 if (error) 488 return error; 489 if (extflags & LK_SLEEPFAIL) { 490 return ENOLCK; 491 } 492 } 493 return 0; 494} 495 496/* 497 * Transfer any waiting processes from one lock to another. 498 */ 499void 500transferlockers(from, to) 501 struct lock *from; 502 struct lock *to; 503{ 504 505 KASSERT(from != to, ("lock transfer to self")); 506 KASSERT((from->lk_flags&LK_WAITDRAIN) == 0, ("transfer draining lock")); 507 if (from->lk_waitcount == 0) 508 return; 509 from->lk_newlock = to; 510 wakeup((void *)from); 511 msleep(&from->lk_newlock, NULL, from->lk_prio, "lkxfer", 0); 512 from->lk_newlock = NULL; 513 from->lk_flags &= ~(LK_WANT_EXCL | LK_WANT_UPGRADE); 514 KASSERT(from->lk_waitcount == 0, ("active lock")); 515} 516 517 518/* 519 * Initialize a lock; required before use. 520 */ 521void 522lockinit(lkp, prio, wmesg, timo, flags) 523 struct lock *lkp; 524 int prio; 525 const char *wmesg; 526 int timo; 527 int flags; 528{ 529 CTR5(KTR_LOCK, "lockinit(): lkp == %p, prio == %d, wmesg == \"%s\", " 530 "timo == %d, flags = 0x%x\n", lkp, prio, wmesg, timo, flags); 531 532 lkp->lk_interlock = mtx_pool_alloc(mtxpool_lockbuilder); 533 lkp->lk_flags = (flags & LK_EXTFLG_MASK); 534 lkp->lk_sharecount = 0; 535 lkp->lk_waitcount = 0; 536 lkp->lk_exclusivecount = 0; 537 lkp->lk_prio = prio; 538 lkp->lk_wmesg = wmesg; 539 lkp->lk_timo = timo; 540 lkp->lk_lockholder = LK_NOPROC; 541 lkp->lk_newlock = NULL; 542#ifdef DEBUG_LOCKS 543 lkp->lk_filename = "none"; 544 lkp->lk_lockername = "never exclusive locked"; 545 lkp->lk_lineno = 0; 546 lkp->lk_slockholder = LK_NOPROC; 547 lkp->lk_sfilename = "none"; 548 lkp->lk_slockername = "never share locked"; 549 lkp->lk_slineno = 0; 550#endif 551} 552 553/* 554 * Destroy a lock. 555 */ 556void 557lockdestroy(lkp) 558 struct lock *lkp; 559{ 560 CTR2(KTR_LOCK, "lockdestroy(): lkp == %p (lk_wmesg == \"%s\")", 561 lkp, lkp->lk_wmesg); 562} 563 564/* 565 * Determine the status of a lock. 566 */ 567int 568lockstatus(lkp, td) 569 struct lock *lkp; 570 struct thread *td; 571{ 572 int lock_type = 0; 573 574 mtx_lock(lkp->lk_interlock); 575 if (lkp->lk_exclusivecount != 0) { 576 if (td == NULL || lkp->lk_lockholder == td) 577 lock_type = LK_EXCLUSIVE; 578 else 579 lock_type = LK_EXCLOTHER; 580 } else if (lkp->lk_sharecount != 0) 581 lock_type = LK_SHARED; 582 mtx_unlock(lkp->lk_interlock); 583 return (lock_type); 584} 585 586/* 587 * Determine the number of holders of a lock. 588 */ 589int 590lockcount(lkp) 591 struct lock *lkp; 592{ 593 int count; 594 595 mtx_lock(lkp->lk_interlock); 596 count = lkp->lk_exclusivecount + lkp->lk_sharecount; 597 mtx_unlock(lkp->lk_interlock); 598 return (count); 599} 600 601/* 602 * Print out information about state of a lock. Used by VOP_PRINT 603 * routines to display status about contained locks. 604 */ 605void 606lockmgr_printinfo(lkp) 607 struct lock *lkp; 608{ 609 610 if (lkp->lk_sharecount) 611 printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg, 612 lkp->lk_sharecount); 613 else if (lkp->lk_flags & LK_HAVE_EXCL) 614 printf(" lock type %s: EXCL (count %d) by thread %p (pid %d)", 615 lkp->lk_wmesg, lkp->lk_exclusivecount, 616 lkp->lk_lockholder, lkp->lk_lockholder->td_proc->p_pid); 617 if (lkp->lk_waitcount > 0) 618 printf(" with %d pending", lkp->lk_waitcount); 619} 620