sys/kern/kern_lock.c

   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 1.27 2008/01/09 10:59:12 corecode 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  * Locking primitives implementation.
  59  * Locks provide shared/exclusive sychronization.
  60  */
  61
  62 #ifdef SIMPLELOCK_DEBUG
  63 #define COUNT(td, x) (td)->td_locks += (x)
  64 #else
  65 #define COUNT(td, x)
  66 #endif
  67
  68 #define LOCK_WAIT_TIME 100
  69 #define LOCK_SAMPLE_WAIT 7
  70
  71 #if defined(DIAGNOSTIC)
  72 #define LOCK_INLINE
  73 #else
  74 #define LOCK_INLINE __inline
  75 #endif
  76
  77 #define LK_ALL (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | \
  78         LK_SHARE_NONZERO | LK_WAIT_NONZERO)
  79
  80 static int acquire(struct lock *lkp, int extflags, int wanted);
  81
  82 static LOCK_INLINE void
  83 sharelock(struct lock *lkp, int incr) {
  84         lkp->lk_flags |= LK_SHARE_NONZERO;
  85         lkp->lk_sharecount += incr;
  86 }
  87
  88 static LOCK_INLINE int
  89 shareunlock(struct lock *lkp, int decr)
  90 {
  91         int dowakeup = 0;
  92
  93         KASSERT(lkp->lk_sharecount >= decr, ("shareunlock: count < decr"));
  94
  95         if (lkp->lk_sharecount == decr) {
  96                 lkp->lk_flags &= ~LK_SHARE_NONZERO;
  97                 if (lkp->lk_flags & (LK_WANT_UPGRADE | LK_WANT_EXCL)) {
  98                         dowakeup = 1;
  99                 }
 100                 lkp->lk_sharecount = 0;
 101         } else {
 102                 lkp->lk_sharecount -= decr;
 103         }
 104         return(dowakeup);
 105 }
 106
 107 /*
 108  * lock acquisition helper routine.  Called with the lock's spinlock held.
 109  */
 110 static int
 111 acquire(struct lock *lkp, int extflags, int wanted)
 112 {
 113         int error;
 114
 115         if ((extflags & LK_NOWAIT) && (lkp->lk_flags & wanted)) {
 116                 return EBUSY;
 117         }
 118
 119         while ((lkp->lk_flags & wanted) != 0) {
 120                 lkp->lk_flags |= LK_WAIT_NONZERO;
 121                 lkp->lk_waitcount++;
 122
 123                 /*
 124                  * Atomic spinlock release/sleep/reacquire.
 125                  */
 126                 error = ssleep(lkp, &lkp->lk_spinlock,
 127                                ((extflags & LK_PCATCH) ? PCATCH : 0),
 128                                lkp->lk_wmesg,
 129                                ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0));
 130                 if (lkp->lk_waitcount == 1) {
 131                         lkp->lk_flags &= ~LK_WAIT_NONZERO;
 132                         lkp->lk_waitcount = 0;
 133                 } else {
 134                         lkp->lk_waitcount--;
 135                 }
 136                 if (error)
 137                         return error;
 138                 if (extflags & LK_SLEEPFAIL)
 139                         return ENOLCK;
 140         }
 141         return 0;
 142 }
 143
 144 /*
 145  * Set, change, or release a lock.
 146  *
 147  * Shared requests increment the shared count. Exclusive requests set the
 148  * LK_WANT_EXCL flag (preventing further shared locks), and wait for already
 149  * accepted shared locks and shared-to-exclusive upgrades to go away.
 150  *
 151  * A spinlock is held for most of the procedure.  We must not do anything
 152  * fancy while holding the spinlock.
 153  */
 154 int
 155 #ifndef DEBUG_LOCKS
 156 lockmgr(struct lock *lkp, u_int flags)
 157 #else
 158 debuglockmgr(struct lock *lkp, u_int flags,
 159              const char *name, const char *file, int line)
 160 #endif
 161 {
 162         thread_t td;
 163         int error;
 164         int extflags;
 165         int dowakeup;
 166
 167         error = 0;
 168         dowakeup = 0;
 169
 170         if (mycpu->gd_intr_nesting_level &&
 171             (flags & LK_NOWAIT) == 0 &&
 172             (flags & LK_TYPE_MASK) != LK_RELEASE &&
 173             panic_cpu_gd != mycpu
 174         ) {
 175
 176 #ifndef DEBUG_LOCKS
 177                 panic("lockmgr %s from %p: called from interrupt, ipi, "
 178                       "or hard code section",
 179                       lkp->lk_wmesg, ((int **)&lkp)[-1]);
 180 #else
 181                 panic("lockmgr %s from %s:%d: called from interrupt, ipi, "
 182                       "or hard code section",
 183                       lkp->lk_wmesg, file, line);
 184 #endif
 185         }
 186
 187         /*
 188          * So sue me, I'm too tired.
 189          */
 190         if (spin_trylock(&lkp->lk_spinlock) == FALSE) {
 191                 if (flags & LK_NOSPINWAIT)
 192                         return(EBUSY);
 193                 spin_lock(&lkp->lk_spinlock);
 194         }
 195
 196         extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK;
 197         td = curthread;
 198
 199         switch (flags & LK_TYPE_MASK) {
 200         case LK_SHARED:
 201                 /*
 202                  * If we are not the exclusive lock holder, we have to block
 203                  * while there is an exclusive lock holder or while an
 204                  * exclusive lock request or upgrade request is in progress.
 205                  *
 206                  * However, if P_DEADLKTREAT is set, we override exclusive
 207                  * lock requests or upgrade requests ( but not the exclusive
 208                  * lock itself ).
 209                  */
 210                 if (lkp->lk_lockholder != td) {
 211                         if (td->td_flags & TDF_DEADLKTREAT) {
 212                                 error = acquire(
 213                                             lkp,
 214                                             extflags,
 215                                             LK_HAVE_EXCL
 216                                         );
 217                         } else {
 218                                 error = acquire(
 219                                             lkp,
 220                                             extflags,
 221                                             LK_HAVE_EXCL | LK_WANT_EXCL |
 222                                              LK_WANT_UPGRADE
 223                                         );
 224                         }
 225                         if (error)
 226                                 break;
 227                         sharelock(lkp, 1);
 228                         COUNT(td, 1);
 229                         break;
 230                 }
 231                 /*
 232                  * We hold an exclusive lock, so downgrade it to shared.
 233                  * An alternative would be to fail with EDEADLK.
 234                  */
 235                 sharelock(lkp, 1);
 236                 COUNT(td, 1);
 237                 /* fall into downgrade */
 238
 239         case LK_DOWNGRADE:
 240                 if (lkp->lk_lockholder != td || lkp->lk_exclusivecount == 0) {
 241                         spin_unlock(&lkp->lk_spinlock);
 242                         panic("lockmgr: not holding exclusive lock");
 243                 }
 244                 sharelock(lkp, lkp->lk_exclusivecount);
 245                 lkp->lk_exclusivecount = 0;
 246                 lkp->lk_flags &= ~LK_HAVE_EXCL;
 247                 lkp->lk_lockholder = LK_NOTHREAD;
 248                 if (lkp->lk_waitcount)
 249                         dowakeup = 1;
 250                 break;
 251
 252         case LK_EXCLUPGRADE:
 253                 /*
 254                  * If another process is ahead of us to get an upgrade,
 255                  * then we want to fail rather than have an intervening
 256                  * exclusive access.
 257                  */
 258                 if (lkp->lk_flags & LK_WANT_UPGRADE) {
 259                         dowakeup = shareunlock(lkp, 1);
 260                         COUNT(td, -1);
 261                         error = EBUSY;
 262                         break;
 263                 }
 264                 /* fall into normal upgrade */
 265
 266         case LK_UPGRADE:
 267                 /*
 268                  * Upgrade a shared lock to an exclusive one. If another
 269                  * shared lock has already requested an upgrade to an
 270                  * exclusive lock, our shared lock is released and an
 271                  * exclusive lock is requested (which will be granted
 272                  * after the upgrade). If we return an error, the file
 273                  * will always be unlocked.
 274                  */
 275                 if ((lkp->lk_lockholder == td) || (lkp->lk_sharecount <= 0)) {
 276                         spin_unlock(&lkp->lk_spinlock);
 277                         panic("lockmgr: upgrade exclusive lock");
 278                 }
 279                 dowakeup += shareunlock(lkp, 1);
 280                 COUNT(td, -1);
 281                 /*
 282                  * If we are just polling, check to see if we will block.
 283                  */
 284                 if ((extflags & LK_NOWAIT) &&
 285                     ((lkp->lk_flags & LK_WANT_UPGRADE) ||
 286                      lkp->lk_sharecount > 1)) {
 287                         error = EBUSY;
 288                         break;
 289                 }
 290                 if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) {
 291                         /*
 292                          * We are first shared lock to request an upgrade, so
 293                          * request upgrade and wait for the shared count to
 294                          * drop to zero, then take exclusive lock.
 295                          */
 296                         lkp->lk_flags |= LK_WANT_UPGRADE;
 297                         error = acquire(lkp, extflags, LK_SHARE_NONZERO);
 298                         lkp->lk_flags &= ~LK_WANT_UPGRADE;
 299
 300                         if (error)
 301                                 break;
 302                         lkp->lk_flags |= LK_HAVE_EXCL;
 303                         lkp->lk_lockholder = td;
 304                         if (lkp->lk_exclusivecount != 0) {
 305                                 spin_unlock(&lkp->lk_spinlock);
 306                                 panic("lockmgr: non-zero exclusive count");
 307                         }
 308                         lkp->lk_exclusivecount = 1;
 309 #if defined(DEBUG_LOCKS)
 310                         lkp->lk_filename = file;
 311                         lkp->lk_lineno = line;
 312                         lkp->lk_lockername = name;
 313 #endif
 314                         COUNT(td, 1);
 315                         break;
 316                 }
 317                 /*
 318                  * Someone else has requested upgrade. Release our shared
 319                  * lock, awaken upgrade requestor if we are the last shared
 320                  * lock, then request an exclusive lock.
 321                  */
 322                 if ( (lkp->lk_flags & (LK_SHARE_NONZERO|LK_WAIT_NONZERO)) ==
 323                         LK_WAIT_NONZERO) {
 324                         ++dowakeup;
 325                 }
 326                 /* fall into exclusive request */
 327
 328         case LK_EXCLUSIVE:
 329                 if (lkp->lk_lockholder == td && td != LK_KERNTHREAD) {
 330                         /*
 331                          *      Recursive lock.
 332                          */
 333                         if ((extflags & (LK_NOWAIT | LK_CANRECURSE)) == 0) {
 334                                 spin_unlock(&lkp->lk_spinlock);
 335                                 panic("lockmgr: locking against myself");
 336                         }
 337                         if ((extflags & LK_CANRECURSE) != 0) {
 338                                 lkp->lk_exclusivecount++;
 339                                 COUNT(td, 1);
 340                                 break;
 341                         }
 342                 }
 343                 /*
 344                  * If we are just polling, check to see if we will sleep.
 345                  */
 346                 if ((extflags & LK_NOWAIT) &&
 347                     (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO))) {
 348                         error = EBUSY;
 349                         break;
 350                 }
 351                 /*
 352                  * Try to acquire the want_exclusive flag.
 353                  */
 354                 error = acquire(lkp, extflags, (LK_HAVE_EXCL | LK_WANT_EXCL));
 355                 if (error)
 356                         break;
 357                 lkp->lk_flags |= LK_WANT_EXCL;
 358                 /*
 359                  * Wait for shared locks and upgrades to finish.
 360                  */
 361                 error = acquire(lkp, extflags, LK_WANT_UPGRADE | LK_SHARE_NONZERO);
 362                 lkp->lk_flags &= ~LK_WANT_EXCL;
 363                 if (error)
 364                         break;
 365                 lkp->lk_flags |= LK_HAVE_EXCL;
 366                 lkp->lk_lockholder = td;
 367                 if (lkp->lk_exclusivecount != 0) {
 368                         spin_unlock(&lkp->lk_spinlock);
 369                         panic("lockmgr: non-zero exclusive count");
 370                 }
 371                 lkp->lk_exclusivecount = 1;
 372 #if defined(DEBUG_LOCKS)
 373                         lkp->lk_filename = file;
 374                         lkp->lk_lineno = line;
 375                         lkp->lk_lockername = name;
 376 #endif
 377                 COUNT(td, 1);
 378                 break;
 379
 380         case LK_RELEASE:
 381                 if (lkp->lk_exclusivecount != 0) {
 382                         if (lkp->lk_lockholder != td &&
 383                             lkp->lk_lockholder != LK_KERNTHREAD) {
 384                                 spin_unlock(&lkp->lk_spinlock);
 385                                 panic("lockmgr: pid %d, not %s thr %p/%p unlocking",
 386                                     (td->td_proc ? td->td_proc->p_pid : -1),
 387                                     "exclusive lock holder",
 388                                     td, lkp->lk_lockholder);
 389                         }
 390                         if (lkp->lk_lockholder != LK_KERNTHREAD) {
 391                                 COUNT(td, -1);
 392                         }
 393                         if (lkp->lk_exclusivecount == 1) {
 394                                 lkp->lk_flags &= ~LK_HAVE_EXCL;
 395                                 lkp->lk_lockholder = LK_NOTHREAD;
 396                                 lkp->lk_exclusivecount = 0;
 397                         } else {
 398                                 lkp->lk_exclusivecount--;
 399                         }
 400                 } else if (lkp->lk_flags & LK_SHARE_NONZERO) {
 401                         dowakeup += shareunlock(lkp, 1);
 402                         COUNT(td, -1);
 403                 }
 404                 if (lkp->lk_flags & LK_WAIT_NONZERO)
 405                         ++dowakeup;
 406                 break;
 407
 408         default:
 409                 spin_unlock(&lkp->lk_spinlock);
 410                 panic("lockmgr: unknown locktype request %d",
 411                     flags & LK_TYPE_MASK);
 412                 /* NOTREACHED */
 413         }
 414         spin_unlock(&lkp->lk_spinlock);
 415         if (dowakeup)
 416                 wakeup(lkp);
 417         return (error);
 418 }
 419
 420 void
 421 lockmgr_kernproc(struct lock *lp)
 422 {
 423         struct thread *td __debugvar = curthread;
 424
 425         if (lp->lk_lockholder != LK_KERNTHREAD) {
 426                 KASSERT(lp->lk_lockholder == td,
 427                     ("lockmgr_kernproc: lock not owned by curthread %p", td));
 428                 COUNT(td, -1);
 429                 lp->lk_lockholder = LK_KERNTHREAD;
 430         }
 431 }
 432
 433 /*
 434  * Set the lock to be exclusively held.  The caller is holding the lock's
 435  * spinlock and the spinlock remains held on return.  A panic will occur
 436  * if the lock cannot be set to exclusive.
 437  */
 438 void
 439 lockmgr_setexclusive_interlocked(struct lock *lkp)
 440 {
 441         thread_t td = curthread;
 442
 443         KKASSERT((lkp->lk_flags & (LK_HAVE_EXCL|LK_SHARE_NONZERO)) == 0);
 444         KKASSERT(lkp->lk_exclusivecount == 0);
 445         lkp->lk_flags |= LK_HAVE_EXCL;
 446         lkp->lk_lockholder = td;
 447         lkp->lk_exclusivecount = 1;
 448         COUNT(td, 1);
 449 }
 450
 451 /*
 452  * Clear the caller's exclusive lock.  The caller is holding the lock's
 453  * spinlock.  THIS FUNCTION WILL UNLOCK THE SPINLOCK.
 454  *
 455  * A panic will occur if the caller does not hold the lock.
 456  */
 457 void
 458 lockmgr_clrexclusive_interlocked(struct lock *lkp)
 459 {
 460         thread_t td __debugvar = curthread;
 461         int dowakeup = 0;
 462
 463         KKASSERT((lkp->lk_flags & LK_HAVE_EXCL) && lkp->lk_exclusivecount == 1
 464                  && lkp->lk_lockholder == td);
 465         lkp->lk_lockholder = LK_NOTHREAD;
 466         lkp->lk_flags &= ~LK_HAVE_EXCL;
 467         lkp->lk_exclusivecount = 0;
 468         if (lkp->lk_flags & LK_WAIT_NONZERO)
 469                 dowakeup = 1;
 470         COUNT(td, -1);
 471         spin_unlock(&lkp->lk_spinlock);
 472         if (dowakeup)
 473                 wakeup((void *)lkp);
 474 }
 475
 476 /*
 477  * Initialize a lock; required before use.
 478  */
 479 void
 480 lockinit(struct lock *lkp, char *wmesg, int timo, int flags)
 481 {
 482         spin_init(&lkp->lk_spinlock);
 483         lkp->lk_flags = (flags & LK_EXTFLG_MASK);
 484         lkp->lk_sharecount = 0;
 485         lkp->lk_waitcount = 0;
 486         lkp->lk_exclusivecount = 0;
 487         lkp->lk_wmesg = wmesg;
 488         lkp->lk_timo = timo;
 489         lkp->lk_lockholder = LK_NOTHREAD;
 490 }
 491
 492 /*
 493  * Reinitialize a lock that is being reused for a different purpose, but
 494  * which may have pending (blocked) threads sitting on it.  The caller
 495  * must already hold the interlock.
 496  */
 497 void
 498 lockreinit(struct lock *lkp, char *wmesg, int timo, int flags)
 499 {
 500         spin_lock(&lkp->lk_spinlock);
 501         lkp->lk_flags = (lkp->lk_flags & ~LK_EXTFLG_MASK) |
 502                         (flags & LK_EXTFLG_MASK);
 503         lkp->lk_wmesg = wmesg;
 504         lkp->lk_timo = timo;
 505         spin_unlock(&lkp->lk_spinlock);
 506 }
 507
 508 /*
 509  * Requires that the caller is the exclusive owner of this lock.
 510  */
 511 void
 512 lockuninit(struct lock *l)
 513 {
 514         /*
 515          * At this point we should have removed all the references to this lock
 516          * so there can't be anyone waiting on it.
 517          */
 518         KKASSERT(l->lk_waitcount == 0);
 519
 520         spin_uninit(&l->lk_spinlock);
 521 }
 522
 523 /*
 524  * Determine the status of a lock.
 525  */
 526 int
 527 lockstatus(struct lock *lkp, struct thread *td)
 528 {
 529         int lock_type = 0;
 530
 531         spin_lock(&lkp->lk_spinlock);
 532         if (lkp->lk_exclusivecount != 0) {
 533                 if (td == NULL || lkp->lk_lockholder == td)
 534                         lock_type = LK_EXCLUSIVE;
 535                 else
 536                         lock_type = LK_EXCLOTHER;
 537         } else if (lkp->lk_sharecount != 0) {
 538                 lock_type = LK_SHARED;
 539         }
 540         spin_unlock(&lkp->lk_spinlock);
 541         return (lock_type);
 542 }
 543
 544 /*
 545  * Return non-zero if the caller owns the lock shared or exclusive.
 546  * We can only guess re: shared locks.
 547  */
 548 int
 549 lockowned(struct lock *lkp)
 550 {
 551         thread_t td = curthread;
 552
 553         if (lkp->lk_exclusivecount)
 554                 return(lkp->lk_lockholder == td);
 555         return(lkp->lk_sharecount != 0);
 556 }
 557
 558 /*
 559  * Determine the number of holders of a lock.
 560  *
 561  * The non-blocking version can usually be used for assertions.
 562  */
 563 int
 564 lockcount(struct lock *lkp)
 565 {
 566         int count;
 567
 568         spin_lock(&lkp->lk_spinlock);
 569         count = lkp->lk_exclusivecount + lkp->lk_sharecount;
 570         spin_unlock(&lkp->lk_spinlock);
 571         return (count);
 572 }
 573
 574 int
 575 lockcountnb(struct lock *lkp)
 576 {
 577         return (lkp->lk_exclusivecount + lkp->lk_sharecount);
 578 }
 579
 580 /*
 581  * Print out information about state of a lock. Used by VOP_PRINT
 582  * routines to display status about contained locks.
 583  */
 584 void
 585 lockmgr_printinfo(struct lock *lkp)
 586 {
 587         struct thread *td = lkp->lk_lockholder;
 588         struct proc *p;
 589
 590         if (td && td != LK_KERNTHREAD && td != LK_NOTHREAD)
 591                 p = td->td_proc;
 592         else
 593                 p = NULL;
 594
 595         if (lkp->lk_sharecount)
 596                 kprintf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg,
 597                     lkp->lk_sharecount);
 598         else if (lkp->lk_flags & LK_HAVE_EXCL)
 599                 kprintf(" lock type %s: EXCL (count %d) by td %p pid %d",
 600                     lkp->lk_wmesg, lkp->lk_exclusivecount, td,
 601                     p ? p->p_pid : -99);
 602         if (lkp->lk_waitcount > 0)
 603                 kprintf(" with %d pending", lkp->lk_waitcount);
 604 }
 605