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 1.15 2005/11/19 17:19:47 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, lkp->lk_prio, lkp->lk_wmesg, 141 ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0)); 142 spin_lock_quick(&lkp->lk_spinlock); 143 if (lkp->lk_waitcount == 1) { 144 lkp->lk_flags &= ~LK_WAIT_NONZERO; 145 lkp->lk_waitcount = 0; 146 } else { 147 lkp->lk_waitcount--; 148 } 149 if (error) 150 return error; 151 if (extflags & LK_SLEEPFAIL) 152 return ENOLCK; 153 } 154 return 0; 155 } 156 157 /* 158 * Set, change, or release a lock. 159 * 160 * Shared requests increment the shared count. Exclusive requests set the 161 * LK_WANT_EXCL flag (preventing further shared locks), and wait for already 162 * accepted shared locks and shared-to-exclusive upgrades to go away. 163 * 164 * A spinlock is held for most of the procedure. We must not do anything 165 * fancy while holding the spinlock. 166 */ 167 int 168 #ifndef DEBUG_LOCKS 169 lockmgr(struct lock *lkp, u_int flags, struct spinlock *interlkp, 170 struct thread *td) 171 #else 172 debuglockmgr(struct lock *lkp, u_int flags, struct spinlock *interlkp, 173 struct thread *td, const char *name, const char *file, int line) 174 #endif 175 { 176 int error; 177 int extflags; 178 static int didpanic; 179 180 error = 0; 181 182 if (lockmgr_from_int && mycpu->gd_intr_nesting_level && 183 (flags & LK_NOWAIT) == 0 && 184 (flags & LK_TYPE_MASK) != LK_RELEASE && didpanic == 0) { 185 #ifndef DEBUG_LOCKS 186 if (lockmgr_from_int == 2) { 187 didpanic = 1; 188 if (flags & LK_INTERLOCK) 189 spin_unlock(interlkp); 190 panic( 191 "lockmgr %s from %p: called from interrupt", 192 lkp->lk_wmesg, ((int **)&lkp)[-1]); 193 didpanic = 0; 194 } else { 195 printf( 196 "lockmgr %s from %p: called from interrupt\n", 197 lkp->lk_wmesg, ((int **)&lkp)[-1]); 198 } 199 #else 200 if (lockmgr_from_int == 2) { 201 didpanic = 1; 202 if (flags & LK_INTERLOCK) 203 spin_unlock(interlkp); 204 panic( 205 "lockmgr %s from %s:%d: called from interrupt", 206 lkp->lk_wmesg, file, line); 207 didpanic = 0; 208 } else { 209 printf( 210 "lockmgr %s from %s:%d: called from interrupt\n", 211 lkp->lk_wmesg, file, line); 212 } 213 #endif 214 } 215 216 spin_lock(&lkp->lk_spinlock); 217 if (flags & LK_INTERLOCK) 218 spin_unlock(interlkp); 219 220 extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK; 221 222 switch (flags & LK_TYPE_MASK) { 223 case LK_SHARED: 224 /* 225 * If we are not the exclusive lock holder, we have to block 226 * while there is an exclusive lock holder or while an 227 * exclusive lock request or upgrade request is in progress. 228 * 229 * However, if P_DEADLKTREAT is set, we override exclusive 230 * lock requests or upgrade requests ( but not the exclusive 231 * lock itself ). 232 */ 233 if (lkp->lk_lockholder != td) { 234 if (td->td_flags & TDF_DEADLKTREAT) { 235 error = acquire( 236 lkp, 237 extflags, 238 LK_HAVE_EXCL 239 ); 240 } else { 241 error = acquire( 242 lkp, 243 extflags, 244 LK_HAVE_EXCL | LK_WANT_EXCL | 245 LK_WANT_UPGRADE 246 ); 247 } 248 if (error) 249 break; 250 sharelock(lkp, 1); 251 COUNT(td, 1); 252 break; 253 } 254 /* 255 * We hold an exclusive lock, so downgrade it to shared. 256 * An alternative would be to fail with EDEADLK. 257 */ 258 sharelock(lkp, 1); 259 COUNT(td, 1); 260 /* fall into downgrade */ 261 262 case LK_DOWNGRADE: 263 if (lkp->lk_lockholder != td || lkp->lk_exclusivecount == 0) { 264 spin_unlock(&lkp->lk_spinlock); 265 panic("lockmgr: not holding exclusive lock"); 266 } 267 sharelock(lkp, lkp->lk_exclusivecount); 268 lkp->lk_exclusivecount = 0; 269 lkp->lk_flags &= ~LK_HAVE_EXCL; 270 lkp->lk_lockholder = LK_NOTHREAD; 271 if (lkp->lk_waitcount) 272 wakeup((void *)lkp); 273 break; 274 275 case LK_EXCLUPGRADE: 276 /* 277 * If another process is ahead of us to get an upgrade, 278 * then we want to fail rather than have an intervening 279 * exclusive access. 280 */ 281 if (lkp->lk_flags & LK_WANT_UPGRADE) { 282 shareunlock(lkp, 1); 283 COUNT(td, -1); 284 error = EBUSY; 285 break; 286 } 287 /* fall into normal upgrade */ 288 289 case LK_UPGRADE: 290 /* 291 * Upgrade a shared lock to an exclusive one. If another 292 * shared lock has already requested an upgrade to an 293 * exclusive lock, our shared lock is released and an 294 * exclusive lock is requested (which will be granted 295 * after the upgrade). If we return an error, the file 296 * will always be unlocked. 297 */ 298 if ((lkp->lk_lockholder == td) || (lkp->lk_sharecount <= 0)) { 299 spin_unlock(&lkp->lk_spinlock); 300 panic("lockmgr: upgrade exclusive lock"); 301 } 302 shareunlock(lkp, 1); 303 COUNT(td, -1); 304 /* 305 * If we are just polling, check to see if we will block. 306 */ 307 if ((extflags & LK_NOWAIT) && 308 ((lkp->lk_flags & LK_WANT_UPGRADE) || 309 lkp->lk_sharecount > 1)) { 310 error = EBUSY; 311 break; 312 } 313 if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) { 314 /* 315 * We are first shared lock to request an upgrade, so 316 * request upgrade and wait for the shared count to 317 * drop to zero, then take exclusive lock. 318 */ 319 lkp->lk_flags |= LK_WANT_UPGRADE; 320 error = acquire(lkp, extflags, LK_SHARE_NONZERO); 321 lkp->lk_flags &= ~LK_WANT_UPGRADE; 322 323 if (error) 324 break; 325 lkp->lk_flags |= LK_HAVE_EXCL; 326 lkp->lk_lockholder = td; 327 if (lkp->lk_exclusivecount != 0) { 328 spin_unlock(&lkp->lk_spinlock); 329 panic("lockmgr: non-zero exclusive count"); 330 } 331 lkp->lk_exclusivecount = 1; 332 #if defined(DEBUG_LOCKS) 333 lkp->lk_filename = file; 334 lkp->lk_lineno = line; 335 lkp->lk_lockername = name; 336 #endif 337 COUNT(td, 1); 338 break; 339 } 340 /* 341 * Someone else has requested upgrade. Release our shared 342 * lock, awaken upgrade requestor if we are the last shared 343 * lock, then request an exclusive lock. 344 */ 345 if ( (lkp->lk_flags & (LK_SHARE_NONZERO|LK_WAIT_NONZERO)) == 346 LK_WAIT_NONZERO) 347 wakeup((void *)lkp); 348 /* fall into exclusive request */ 349 350 case LK_EXCLUSIVE: 351 if (lkp->lk_lockholder == td && td != LK_KERNTHREAD) { 352 /* 353 * Recursive lock. 354 */ 355 if ((extflags & (LK_NOWAIT | LK_CANRECURSE)) == 0) { 356 spin_unlock(&lkp->lk_spinlock); 357 panic("lockmgr: locking against myself"); 358 } 359 if ((extflags & LK_CANRECURSE) != 0) { 360 lkp->lk_exclusivecount++; 361 COUNT(td, 1); 362 break; 363 } 364 } 365 /* 366 * If we are just polling, check to see if we will sleep. 367 */ 368 if ((extflags & LK_NOWAIT) && 369 (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO))) { 370 error = EBUSY; 371 break; 372 } 373 /* 374 * Try to acquire the want_exclusive flag. 375 */ 376 error = acquire(lkp, extflags, (LK_HAVE_EXCL | LK_WANT_EXCL)); 377 if (error) 378 break; 379 lkp->lk_flags |= LK_WANT_EXCL; 380 /* 381 * Wait for shared locks and upgrades to finish. 382 */ 383 error = acquire(lkp, extflags, LK_WANT_UPGRADE | LK_SHARE_NONZERO); 384 lkp->lk_flags &= ~LK_WANT_EXCL; 385 if (error) 386 break; 387 lkp->lk_flags |= LK_HAVE_EXCL; 388 lkp->lk_lockholder = td; 389 if (lkp->lk_exclusivecount != 0) { 390 spin_unlock(&lkp->lk_spinlock); 391 panic("lockmgr: non-zero exclusive count"); 392 } 393 lkp->lk_exclusivecount = 1; 394 #if defined(DEBUG_LOCKS) 395 lkp->lk_filename = file; 396 lkp->lk_lineno = line; 397 lkp->lk_lockername = name; 398 #endif 399 COUNT(td, 1); 400 break; 401 402 case LK_RELEASE: 403 if (lkp->lk_exclusivecount != 0) { 404 if (lkp->lk_lockholder != td && 405 lkp->lk_lockholder != LK_KERNTHREAD) { 406 spin_unlock(&lkp->lk_spinlock); 407 panic("lockmgr: pid %d, not %s thr %p unlocking", 408 (td->td_proc ? td->td_proc->p_pid : -99), 409 "exclusive lock holder", 410 lkp->lk_lockholder); 411 } 412 if (lkp->lk_lockholder != LK_KERNTHREAD) { 413 COUNT(td, -1); 414 } 415 if (lkp->lk_exclusivecount == 1) { 416 lkp->lk_flags &= ~LK_HAVE_EXCL; 417 lkp->lk_lockholder = LK_NOTHREAD; 418 lkp->lk_exclusivecount = 0; 419 } else { 420 lkp->lk_exclusivecount--; 421 } 422 } else if (lkp->lk_flags & LK_SHARE_NONZERO) { 423 shareunlock(lkp, 1); 424 COUNT(td, -1); 425 } 426 if (lkp->lk_flags & LK_WAIT_NONZERO) 427 wakeup((void *)lkp); 428 break; 429 430 case LK_DRAIN: 431 /* 432 * Check that we do not already hold the lock, as it can 433 * never drain if we do. Unfortunately, we have no way to 434 * check for holding a shared lock, but at least we can 435 * check for an exclusive one. 436 */ 437 if (lkp->lk_lockholder == td) { 438 spin_unlock(&lkp->lk_spinlock); 439 panic("lockmgr: draining against myself"); 440 } 441 442 error = acquiredrain(lkp, extflags); 443 if (error) 444 break; 445 lkp->lk_flags |= LK_DRAINING | LK_HAVE_EXCL; 446 lkp->lk_lockholder = td; 447 lkp->lk_exclusivecount = 1; 448 #if defined(DEBUG_LOCKS) 449 lkp->lk_filename = file; 450 lkp->lk_lineno = line; 451 lkp->lk_lockername = name; 452 #endif 453 COUNT(td, 1); 454 break; 455 456 default: 457 spin_unlock(&lkp->lk_spinlock); 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 spin_unlock(&lkp->lk_spinlock); 469 return (error); 470 } 471 472 /* 473 * lock acquisition helper routine. Called with the lock's spinlock held. 474 */ 475 static int 476 acquiredrain(struct lock *lkp, int extflags) 477 { 478 int error; 479 480 if ((extflags & LK_NOWAIT) && (lkp->lk_flags & LK_ALL)) { 481 return EBUSY; 482 } 483 484 if ((lkp->lk_flags & LK_ALL) == 0) 485 return 0; 486 487 while (lkp->lk_flags & LK_ALL) { 488 lkp->lk_flags |= LK_WAITDRAIN; 489 /* 490 * Use the _quick version so the critical section is left 491 * intact, protecting the tsleep interlock. See 492 * tsleep_interlock() for a description of what is 493 * happening here. 494 */ 495 tsleep_interlock(&lkp->lk_flags); 496 spin_unlock_quick(&lkp->lk_spinlock); 497 error = tsleep(&lkp->lk_flags, lkp->lk_prio, 498 lkp->lk_wmesg, 499 ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0)); 500 spin_lock_quick(&lkp->lk_spinlock); 501 if (error) 502 return error; 503 if (extflags & LK_SLEEPFAIL) { 504 return ENOLCK; 505 } 506 } 507 return 0; 508 } 509 510 /* 511 * Initialize a lock; required before use. 512 */ 513 void 514 lockinit(struct lock *lkp, int prio, char *wmesg, int timo, int flags) 515 { 516 spin_init(&lkp->lk_spinlock); 517 lkp->lk_flags = (flags & LK_EXTFLG_MASK); 518 lkp->lk_sharecount = 0; 519 lkp->lk_waitcount = 0; 520 lkp->lk_exclusivecount = 0; 521 lkp->lk_prio = prio; 522 lkp->lk_wmesg = wmesg; 523 lkp->lk_timo = timo; 524 lkp->lk_lockholder = LK_NOTHREAD; 525 } 526 527 /* 528 * Reinitialize a lock that is being reused for a different purpose, but 529 * which may have pending (blocked) threads sitting on it. The caller 530 * must already hold the interlock. 531 */ 532 void 533 lockreinit(struct lock *lkp, int prio, char *wmesg, int timo, int flags) 534 { 535 lkp->lk_flags = (lkp->lk_flags & ~(LK_EXTFLG_MASK|LK_DRAINING)) | 536 (flags & LK_EXTFLG_MASK); 537 lkp->lk_prio = prio; 538 lkp->lk_wmesg = wmesg; 539 lkp->lk_timo = timo; 540 } 541 542 /* 543 * Determine the status of a lock. 544 */ 545 int 546 lockstatus(struct lock *lkp, struct thread *td) 547 { 548 int lock_type = 0; 549 550 spin_lock(&lkp->lk_spinlock); 551 if (lkp->lk_exclusivecount != 0) { 552 if (td == NULL || lkp->lk_lockholder == td) 553 lock_type = LK_EXCLUSIVE; 554 else 555 lock_type = LK_EXCLOTHER; 556 } else if (lkp->lk_sharecount != 0) { 557 lock_type = LK_SHARED; 558 } 559 spin_unlock(&lkp->lk_spinlock); 560 return (lock_type); 561 } 562 563 /* 564 * Determine the number of holders of a lock. 565 * 566 * The non-blocking version can usually be used for assertions. 567 */ 568 int 569 lockcount(struct lock *lkp) 570 { 571 int count; 572 573 spin_lock(&lkp->lk_spinlock); 574 count = lkp->lk_exclusivecount + lkp->lk_sharecount; 575 spin_unlock(&lkp->lk_spinlock); 576 return (count); 577 } 578 579 int 580 lockcountnb(struct lock *lkp) 581 { 582 return (lkp->lk_exclusivecount + lkp->lk_sharecount); 583 } 584 585 /* 586 * Print out information about state of a lock. Used by VOP_PRINT 587 * routines to display status about contained locks. 588 */ 589 void 590 lockmgr_printinfo(struct lock *lkp) 591 { 592 struct thread *td = lkp->lk_lockholder; 593 struct proc *p; 594 595 if (td && td != LK_KERNTHREAD && td != LK_NOTHREAD) 596 p = td->td_proc; 597 else 598 p = NULL; 599 600 if (lkp->lk_sharecount) 601 printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg, 602 lkp->lk_sharecount); 603 else if (lkp->lk_flags & LK_HAVE_EXCL) 604 printf(" lock type %s: EXCL (count %d) by td %p pid %d", 605 lkp->lk_wmesg, lkp->lk_exclusivecount, td, 606 p ? p->p_pid : -99); 607 if (lkp->lk_waitcount > 0) 608 printf(" with %d pending", lkp->lk_waitcount); 609 } 610 611