kern_lock.c revision 107414
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 107414 2002-11-30 19:00:51Z mckusick $ 42 */ 43 44#include <sys/param.h> 45#include <sys/kernel.h> 46#include <sys/ktr.h> 47#include <sys/lock.h> 48#include <sys/lockmgr.h> 49#include <sys/mutex.h> 50#include <sys/proc.h> 51#include <sys/systm.h> 52 53/* 54 * Locking primitives implementation. 55 * Locks provide shared/exclusive sychronization. 56 */ 57 58#define LOCK_WAIT_TIME 100 59#define LOCK_SAMPLE_WAIT 7 60 61#if defined(DIAGNOSTIC) 62#define LOCK_INLINE 63#else 64#define LOCK_INLINE __inline 65#endif 66 67#define LK_ALL (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | \ 68 LK_SHARE_NONZERO | LK_WAIT_NONZERO) 69 70/* 71 * Mutex array variables. Rather than each lockmgr lock having its own mutex, 72 * share a fixed (at boot time) number of mutexes across all lockmgr locks in 73 * order to keep sizeof(struct lock) down. 74 */ 75int lock_mtx_valid; 76static struct mtx lock_mtx; 77 78static int acquire(struct lock **lkpp, int extflags, int wanted); 79static int apause(struct lock *lkp, int flags); 80static int acquiredrain(struct lock *lkp, int extflags) ; 81 82static void 83lockmgr_init(void *dummy __unused) 84{ 85 /* 86 * Initialize the lockmgr protection mutex if it hasn't already been 87 * done. Unless something changes about kernel startup order, VM 88 * initialization will always cause this mutex to already be 89 * initialized in a call to lockinit(). 90 */ 91 if (lock_mtx_valid == 0) { 92 mtx_init(&lock_mtx, "lockmgr", NULL, MTX_DEF); 93 lock_mtx_valid = 1; 94 } 95} 96SYSINIT(lmgrinit, SI_SUB_LOCK, SI_ORDER_FIRST, lockmgr_init, NULL) 97 98static LOCK_INLINE void 99sharelock(struct lock *lkp, int incr) { 100 lkp->lk_flags |= LK_SHARE_NONZERO; 101 lkp->lk_sharecount += incr; 102} 103 104static LOCK_INLINE void 105shareunlock(struct lock *lkp, int decr) { 106 107 KASSERT(lkp->lk_sharecount >= decr, ("shareunlock: count < decr")); 108 109 if (lkp->lk_sharecount == decr) { 110 lkp->lk_flags &= ~LK_SHARE_NONZERO; 111 if (lkp->lk_flags & (LK_WANT_UPGRADE | LK_WANT_EXCL)) { 112 wakeup(lkp); 113 } 114 lkp->lk_sharecount = 0; 115 } else { 116 lkp->lk_sharecount -= decr; 117 } 118} 119 120/* 121 * This is the waitloop optimization. 122 */ 123static int 124apause(struct lock *lkp, int flags) 125{ 126#ifdef SMP 127 int i, lock_wait; 128#endif 129 130 if ((lkp->lk_flags & flags) == 0) 131 return 0; 132#ifdef SMP 133 for (lock_wait = LOCK_WAIT_TIME; lock_wait > 0; lock_wait--) { 134 mtx_unlock(lkp->lk_interlock); 135 for (i = LOCK_SAMPLE_WAIT; i > 0; i--) 136 if ((lkp->lk_flags & flags) == 0) 137 break; 138 mtx_lock(lkp->lk_interlock); 139 if ((lkp->lk_flags & flags) == 0) 140 return 0; 141 } 142#endif 143 return 1; 144} 145 146static int 147acquire(struct lock **lkpp, int extflags, int wanted) { 148 struct lock *lkp = *lkpp; 149 int s, error; 150 151 CTR3(KTR_LOCKMGR, 152 "acquire(): lkp == %p, extflags == 0x%x, wanted == 0x%x\n", 153 lkp, extflags, wanted); 154 155 if ((extflags & LK_NOWAIT) && (lkp->lk_flags & wanted)) { 156 return EBUSY; 157 } 158 159 if (((lkp->lk_flags | extflags) & LK_NOPAUSE) == 0) { 160 error = apause(lkp, wanted); 161 if (error == 0) 162 return 0; 163 } 164 165 s = splhigh(); 166 while ((lkp->lk_flags & wanted) != 0) { 167 lkp->lk_flags |= LK_WAIT_NONZERO; 168 lkp->lk_waitcount++; 169 error = msleep(lkp, lkp->lk_interlock, lkp->lk_prio, 170 lkp->lk_wmesg, 171 ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0)); 172 if (lkp->lk_waitcount == 1) { 173 lkp->lk_flags &= ~LK_WAIT_NONZERO; 174 lkp->lk_waitcount = 0; 175 } else { 176 lkp->lk_waitcount--; 177 } 178 if (error) { 179 splx(s); 180 return error; 181 } 182 if (extflags & LK_SLEEPFAIL) { 183 splx(s); 184 return ENOLCK; 185 } 186 if (lkp->lk_newlock != NULL) { 187 mtx_lock(lkp->lk_newlock->lk_interlock); 188 mtx_unlock(lkp->lk_interlock); 189 if (lkp->lk_waitcount == 0) 190 wakeup((void *)(&lkp->lk_newlock)); 191 *lkpp = lkp = lkp->lk_newlock; 192 } 193 } 194 splx(s); 195 return 0; 196} 197 198/* 199 * Set, change, or release a lock. 200 * 201 * Shared requests increment the shared count. Exclusive requests set the 202 * LK_WANT_EXCL flag (preventing further shared locks), and wait for already 203 * accepted shared locks and shared-to-exclusive upgrades to go away. 204 */ 205int 206#ifndef DEBUG_LOCKS 207lockmgr(lkp, flags, interlkp, td) 208#else 209debuglockmgr(lkp, flags, interlkp, td, name, file, line) 210#endif 211 struct lock *lkp; 212 u_int flags; 213 struct mtx *interlkp; 214 struct thread *td; 215#ifdef DEBUG_LOCKS 216 const char *name; /* Name of lock function */ 217 const char *file; /* Name of file call is from */ 218 int line; /* Line number in file */ 219#endif 220{ 221 int error; 222 pid_t pid; 223 int extflags, lockflags; 224 225 CTR5(KTR_LOCKMGR, 226 "lockmgr(): lkp == %p (lk_wmesg == \"%s\"), flags == 0x%x, " 227 "interlkp == %p, td == %p", lkp, lkp->lk_wmesg, flags, interlkp, td); 228 229 error = 0; 230 if (td == NULL) 231 pid = LK_KERNPROC; 232 else 233 pid = td->td_proc->p_pid; 234 235 mtx_lock(lkp->lk_interlock); 236 if (flags & LK_INTERLOCK) { 237 mtx_assert(interlkp, MA_OWNED | MA_NOTRECURSED); 238 mtx_unlock(interlkp); 239 } 240 241 if (panicstr != NULL) { 242 mtx_unlock(lkp->lk_interlock); 243 return (0); 244 } 245 246 extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK; 247 248 switch (flags & LK_TYPE_MASK) { 249 250 case LK_SHARED: 251 /* 252 * If we are not the exclusive lock holder, we have to block 253 * while there is an exclusive lock holder or while an 254 * exclusive lock request or upgrade request is in progress. 255 * 256 * However, if TDF_DEADLKTREAT is set, we override exclusive 257 * lock requests or upgrade requests ( but not the exclusive 258 * lock itself ). 259 */ 260 if (lkp->lk_lockholder != pid) { 261 lockflags = LK_HAVE_EXCL; 262 mtx_lock_spin(&sched_lock); 263 if (td != NULL && !(td->td_flags & TDF_DEADLKTREAT)) 264 lockflags |= LK_WANT_EXCL | LK_WANT_UPGRADE; 265 mtx_unlock_spin(&sched_lock); 266 error = acquire(&lkp, extflags, lockflags); 267 if (error) 268 break; 269 sharelock(lkp, 1); 270#if defined(DEBUG_LOCKS) 271 lkp->lk_slockholder = pid; 272 lkp->lk_sfilename = file; 273 lkp->lk_slineno = line; 274 lkp->lk_slockername = name; 275#endif 276 break; 277 } 278 /* 279 * We hold an exclusive lock, so downgrade it to shared. 280 * An alternative would be to fail with EDEADLK. 281 */ 282 sharelock(lkp, 1); 283 /* FALLTHROUGH downgrade */ 284 285 case LK_DOWNGRADE: 286 KASSERT(lkp->lk_lockholder == pid && lkp->lk_exclusivecount != 0, 287 ("lockmgr: not holding exclusive lock " 288 "(owner pid (%d) != pid (%d), exlcnt (%d) != 0", 289 lkp->lk_lockholder, pid, lkp->lk_exclusivecount)); 290 sharelock(lkp, lkp->lk_exclusivecount); 291 lkp->lk_exclusivecount = 0; 292 lkp->lk_flags &= ~LK_HAVE_EXCL; 293 lkp->lk_lockholder = LK_NOPROC; 294 if (lkp->lk_waitcount) 295 wakeup((void *)lkp); 296 break; 297 298 case LK_EXCLUPGRADE: 299 /* 300 * If another process is ahead of us to get an upgrade, 301 * then we want to fail rather than have an intervening 302 * exclusive access. 303 */ 304 if (lkp->lk_flags & LK_WANT_UPGRADE) { 305 shareunlock(lkp, 1); 306 error = EBUSY; 307 break; 308 } 309 /* FALLTHROUGH normal upgrade */ 310 311 case LK_UPGRADE: 312 /* 313 * Upgrade a shared lock to an exclusive one. If another 314 * shared lock has already requested an upgrade to an 315 * exclusive lock, our shared lock is released and an 316 * exclusive lock is requested (which will be granted 317 * after the upgrade). If we return an error, the file 318 * will always be unlocked. 319 */ 320 if ((lkp->lk_lockholder == pid) || (lkp->lk_sharecount <= 0)) 321 panic("lockmgr: upgrade exclusive lock"); 322 shareunlock(lkp, 1); 323 /* 324 * If we are just polling, check to see if we will block. 325 */ 326 if ((extflags & LK_NOWAIT) && 327 ((lkp->lk_flags & LK_WANT_UPGRADE) || 328 lkp->lk_sharecount > 1)) { 329 error = EBUSY; 330 break; 331 } 332 if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) { 333 /* 334 * We are first shared lock to request an upgrade, so 335 * request upgrade and wait for the shared count to 336 * drop to zero, then take exclusive lock. 337 */ 338 lkp->lk_flags |= LK_WANT_UPGRADE; 339 error = acquire(&lkp, extflags, LK_SHARE_NONZERO); 340 lkp->lk_flags &= ~LK_WANT_UPGRADE; 341 342 if (error) 343 break; 344 lkp->lk_flags |= LK_HAVE_EXCL; 345 lkp->lk_lockholder = pid; 346 if (lkp->lk_exclusivecount != 0) 347 panic("lockmgr: non-zero exclusive count"); 348 lkp->lk_exclusivecount = 1; 349#if defined(DEBUG_LOCKS) 350 lkp->lk_filename = file; 351 lkp->lk_lineno = line; 352 lkp->lk_lockername = name; 353#endif 354 break; 355 } 356 /* 357 * Someone else has requested upgrade. Release our shared 358 * lock, awaken upgrade requestor if we are the last shared 359 * lock, then request an exclusive lock. 360 */ 361 if ( (lkp->lk_flags & (LK_SHARE_NONZERO|LK_WAIT_NONZERO)) == 362 LK_WAIT_NONZERO) 363 wakeup((void *)lkp); 364 /* FALLTHROUGH exclusive request */ 365 366 case LK_EXCLUSIVE: 367 if (lkp->lk_lockholder == pid && pid != LK_KERNPROC) { 368 /* 369 * Recursive lock. 370 */ 371 if ((extflags & (LK_NOWAIT | LK_CANRECURSE)) == 0) 372 panic("lockmgr: locking against myself"); 373 if ((extflags & LK_CANRECURSE) != 0) { 374 lkp->lk_exclusivecount++; 375 break; 376 } 377 } 378 /* 379 * If we are just polling, check to see if we will sleep. 380 */ 381 if ((extflags & LK_NOWAIT) && 382 (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO))) { 383 error = EBUSY; 384 break; 385 } 386 /* 387 * Try to acquire the want_exclusive flag. 388 */ 389 error = acquire(&lkp, extflags, (LK_HAVE_EXCL | LK_WANT_EXCL)); 390 if (error) 391 break; 392 lkp->lk_flags |= LK_WANT_EXCL; 393 /* 394 * Wait for shared locks and upgrades to finish. 395 */ 396 error = acquire(&lkp, extflags, LK_WANT_UPGRADE | LK_SHARE_NONZERO); 397 lkp->lk_flags &= ~LK_WANT_EXCL; 398 if (error) 399 break; 400 lkp->lk_flags |= LK_HAVE_EXCL; 401 lkp->lk_lockholder = pid; 402 if (lkp->lk_exclusivecount != 0) 403 panic("lockmgr: non-zero exclusive count"); 404 lkp->lk_exclusivecount = 1; 405#if defined(DEBUG_LOCKS) 406 lkp->lk_filename = file; 407 lkp->lk_lineno = line; 408 lkp->lk_lockername = name; 409#endif 410 break; 411 412 case LK_RELEASE: 413 if (lkp->lk_exclusivecount != 0) { 414 if (lkp->lk_lockholder != pid && 415 lkp->lk_lockholder != LK_KERNPROC) { 416 panic("lockmgr: pid %d, not %s %d unlocking", 417 pid, "exclusive lock holder", 418 lkp->lk_lockholder); 419 } 420 if (lkp->lk_exclusivecount == 1) { 421 lkp->lk_flags &= ~LK_HAVE_EXCL; 422 lkp->lk_lockholder = LK_NOPROC; 423 lkp->lk_exclusivecount = 0; 424 } else { 425 lkp->lk_exclusivecount--; 426 } 427 } else if (lkp->lk_flags & LK_SHARE_NONZERO) 428 shareunlock(lkp, 1); 429 if (lkp->lk_flags & LK_WAIT_NONZERO) 430 wakeup((void *)lkp); 431 break; 432 433 case LK_DRAIN: 434 /* 435 * Check that we do not already hold the lock, as it can 436 * never drain if we do. Unfortunately, we have no way to 437 * check for holding a shared lock, but at least we can 438 * check for an exclusive one. 439 */ 440 if (lkp->lk_lockholder == pid) 441 panic("lockmgr: draining against myself"); 442 443 error = acquiredrain(lkp, extflags); 444 if (error) 445 break; 446 lkp->lk_flags |= LK_DRAINING | LK_HAVE_EXCL; 447 lkp->lk_lockholder = pid; 448 lkp->lk_exclusivecount = 1; 449#if defined(DEBUG_LOCKS) 450 lkp->lk_filename = file; 451 lkp->lk_lineno = line; 452 lkp->lk_lockername = name; 453#endif 454 break; 455 456 default: 457 mtx_unlock(lkp->lk_interlock); 458 panic("lockmgr: unknown locktype request %d", 459 flags & LK_TYPE_MASK); 460 /* NOTREACHED */ 461 } 462 if ((lkp->lk_flags & LK_WAITDRAIN) && 463 (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | 464 LK_SHARE_NONZERO | LK_WAIT_NONZERO)) == 0) { 465 lkp->lk_flags &= ~LK_WAITDRAIN; 466 wakeup((void *)&lkp->lk_flags); 467 } 468 mtx_unlock(lkp->lk_interlock); 469 return (error); 470} 471 472static int 473acquiredrain(struct lock *lkp, int extflags) { 474 int error; 475 476 if ((extflags & LK_NOWAIT) && (lkp->lk_flags & LK_ALL)) { 477 return EBUSY; 478 } 479 480 error = apause(lkp, LK_ALL); 481 if (error == 0) 482 return 0; 483 484 while (lkp->lk_flags & LK_ALL) { 485 lkp->lk_flags |= LK_WAITDRAIN; 486 error = msleep(&lkp->lk_flags, lkp->lk_interlock, lkp->lk_prio, 487 lkp->lk_wmesg, 488 ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0)); 489 if (error) 490 return error; 491 if (extflags & LK_SLEEPFAIL) { 492 return ENOLCK; 493 } 494 } 495 return 0; 496} 497 498/* 499 * Transfer any waiting processes from one lock to another. 500 */ 501void 502transferlockers(from, to) 503 struct lock *from; 504 struct lock *to; 505{ 506 507 KASSERT(from != to, ("lock transfer to self")); 508 KASSERT((from->lk_flags&LK_WAITDRAIN) == 0, ("transfer draining lock")); 509 if (from->lk_waitcount == 0) 510 return; 511 from->lk_newlock = to; 512 wakeup((void *)from); 513 msleep(&from->lk_newlock, NULL, from->lk_prio, "lkxfer", 0); 514 from->lk_newlock = NULL; 515 from->lk_flags &= ~(LK_WANT_EXCL | LK_WANT_UPGRADE); 516 KASSERT(from->lk_waitcount == 0, ("active lock")); 517} 518 519 520/* 521 * Initialize a lock; required before use. 522 */ 523void 524lockinit(lkp, prio, wmesg, timo, flags) 525 struct lock *lkp; 526 int prio; 527 const char *wmesg; 528 int timo; 529 int flags; 530{ 531 CTR5(KTR_LOCKMGR, "lockinit(): lkp == %p, prio == %d, wmesg == \"%s\", " 532 "timo == %d, flags = 0x%x\n", lkp, prio, wmesg, timo, flags); 533 534 if (lock_mtx_valid == 0) { 535 mtx_init(&lock_mtx, "lockmgr", NULL, MTX_DEF); 536 lock_mtx_valid = 1; 537 } 538 /* 539 * XXX cleanup - make sure mtxpool is always initialized before 540 * this is ever called. 541 */ 542 if (mtx_pool_valid) { 543 mtx_lock(&lock_mtx); 544 lkp->lk_interlock = mtx_pool_alloc(); 545 mtx_unlock(&lock_mtx); 546 } else { 547 lkp->lk_interlock = &lock_mtx; 548 } 549 lkp->lk_flags = (flags & LK_EXTFLG_MASK); 550 lkp->lk_sharecount = 0; 551 lkp->lk_waitcount = 0; 552 lkp->lk_exclusivecount = 0; 553 lkp->lk_prio = prio; 554 lkp->lk_wmesg = wmesg; 555 lkp->lk_timo = timo; 556 lkp->lk_lockholder = LK_NOPROC; 557 lkp->lk_newlock = NULL; 558#ifdef DEBUG_LOCKS 559 lkp->lk_filename = "none"; 560 lkp->lk_lockername = "never exclusive locked"; 561 lkp->lk_lineno = 0; 562 lkp->lk_slockholder = LK_NOPROC; 563 lkp->lk_sfilename = "none"; 564 lkp->lk_slockername = "never share locked"; 565 lkp->lk_slineno = 0; 566#endif 567} 568 569/* 570 * Destroy a lock. 571 */ 572void 573lockdestroy(lkp) 574 struct lock *lkp; 575{ 576 CTR2(KTR_LOCKMGR, "lockdestroy(): lkp == %p (lk_wmesg == \"%s\")", 577 lkp, lkp->lk_wmesg); 578} 579 580/* 581 * Determine the status of a lock. 582 */ 583int 584lockstatus(lkp, td) 585 struct lock *lkp; 586 struct thread *td; 587{ 588 int lock_type = 0; 589 590 mtx_lock(lkp->lk_interlock); 591 if (lkp->lk_exclusivecount != 0) { 592 if (td == NULL || lkp->lk_lockholder == td->td_proc->p_pid) 593 lock_type = LK_EXCLUSIVE; 594 else 595 lock_type = LK_EXCLOTHER; 596 } else if (lkp->lk_sharecount != 0) 597 lock_type = LK_SHARED; 598 mtx_unlock(lkp->lk_interlock); 599 return (lock_type); 600} 601 602/* 603 * Determine the number of holders of a lock. 604 */ 605int 606lockcount(lkp) 607 struct lock *lkp; 608{ 609 int count; 610 611 mtx_lock(lkp->lk_interlock); 612 count = lkp->lk_exclusivecount + lkp->lk_sharecount; 613 mtx_unlock(lkp->lk_interlock); 614 return (count); 615} 616 617/* 618 * Print out information about state of a lock. Used by VOP_PRINT 619 * routines to display status about contained locks. 620 */ 621void 622lockmgr_printinfo(lkp) 623 struct lock *lkp; 624{ 625 626 if (lkp->lk_sharecount) 627 printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg, 628 lkp->lk_sharecount); 629 else if (lkp->lk_flags & LK_HAVE_EXCL) 630 printf(" lock type %s: EXCL (count %d) by pid %d", 631 lkp->lk_wmesg, lkp->lk_exclusivecount, lkp->lk_lockholder); 632 if (lkp->lk_waitcount > 0) 633 printf(" with %d pending", lkp->lk_waitcount); 634} 635