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 1.19 2006/05/05 20:15:01 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
  88 static LOCK_INLINE void
  89 sharelock(struct lock *lkp, int incr) {
  90         lkp->lk_flags |= LK_SHARE_NONZERO;
  91         lkp->lk_sharecount += incr;
  92 }
  93
  94 static LOCK_INLINE void
  95 shareunlock(struct lock *lkp, int decr) {
  96
  97         KASSERT(lkp->lk_sharecount >= decr, ("shareunlock: count < decr"));
  98
  99         if (lkp->lk_sharecount == decr) {
 100                 lkp->lk_flags &= ~LK_SHARE_NONZERO;
 101                 if (lkp->lk_flags & (LK_WANT_UPGRADE | LK_WANT_EXCL)) {
 102                         wakeup(lkp);
 103                 }
 104                 lkp->lk_sharecount = 0;
 105         } else {
 106                 lkp->lk_sharecount -= decr;
 107         }
 108 }
 109
 110 /*
 111  * lock acquisition helper routine.  Called with the lock's spinlock held.
 112  */
 113 static int
 114 acquire(struct lock *lkp, int extflags, int wanted)
 115 {
 116         int error;
 117
 118         if ((extflags & LK_NOWAIT) && (lkp->lk_flags & wanted)) {
 119                 return EBUSY;
 120         }
 121
 122         if (((lkp->lk_flags | extflags) & LK_NOPAUSE) == 0) {
 123                 if ((lkp->lk_flags & wanted) == 0)
 124                         return 0;
 125         }
 126
 127         while ((lkp->lk_flags & wanted) != 0) {
 128                 lkp->lk_flags |= LK_WAIT_NONZERO;
 129                 lkp->lk_waitcount++;
 130
 131                 /*
 132                  * Use the _quick version so the critical section is left
 133                  * intact, protecting the tsleep interlock.  See
 134                  * tsleep_interlock() for a description of what is
 135                  * happening here.
 136                  */
 137                 tsleep_interlock(lkp);
 138                 spin_unlock_quick(&lkp->lk_spinlock);
 139                 error = tsleep(lkp,
 140                                ((extflags & LK_PCATCH) ? PCATCH : 0),
 141                                lkp->lk_wmesg,
 142                                ((extflags & LK_TIMELOCK) ? lkp->lk_timo : 0));
 143                 spin_lock_quick(&lkp->lk_spinlock);
 144                 if (lkp->lk_waitcount == 1) {
 145                         lkp->lk_flags &= ~LK_WAIT_NONZERO;
 146                         lkp->lk_waitcount = 0;
 147                 } else {
 148                         lkp->lk_waitcount--;
 149                 }
 150                 if (error)
 151                         return error;
 152                 if (extflags & LK_SLEEPFAIL)
 153                         return ENOLCK;
 154         }
 155         return 0;
 156 }
 157
 158 /*
 159  * Set, change, or release a lock.
 160  *
 161  * Shared requests increment the shared count. Exclusive requests set the
 162  * LK_WANT_EXCL flag (preventing further shared locks), and wait for already
 163  * accepted shared locks and shared-to-exclusive upgrades to go away.
 164  *
 165  * A spinlock is held for most of the procedure.  We must not do anything
 166  * fancy while holding the spinlock.
 167  */
 168 int
 169 #ifndef DEBUG_LOCKS
 170 lockmgr(struct lock *lkp, u_int flags)
 171 #else
 172 debuglockmgr(struct lock *lkp, u_int flags,
 173              const char *name, const char *file, int line)
 174 #endif
 175 {
 176         thread_t td;
 177         int error;
 178         int extflags;
 179         static int didpanic;
 180
 181         error = 0;
 182
 183         if (lockmgr_from_int && mycpu->gd_intr_nesting_level &&
 184             (flags & LK_NOWAIT) == 0 &&
 185             (flags & LK_TYPE_MASK) != LK_RELEASE && didpanic == 0) {
 186 #ifndef DEBUG_LOCKS
 187                     if (lockmgr_from_int == 2) {
 188                             didpanic = 1;
 189                             panic(
 190                                 "lockmgr %s from %p: called from interrupt",
 191                                 lkp->lk_wmesg, ((int **)&lkp)[-1]);
 192                             didpanic = 0;
 193                     } else {
 194                             printf(
 195                                 "lockmgr %s from %p: called from interrupt\n",
 196                                 lkp->lk_wmesg, ((int **)&lkp)[-1]);
 197                     }
 198 #else
 199                     if (lockmgr_from_int == 2) {
 200                             didpanic = 1;
 201                             panic(
 202                                 "lockmgr %s from %s:%d: called from interrupt",
 203                                 lkp->lk_wmesg, file, line);
 204                             didpanic = 0;
 205                     } else {
 206                             printf(
 207                                 "lockmgr %s from %s:%d: called from interrupt\n",
 208                                 lkp->lk_wmesg, file, line);
 209                     }
 210 #endif
 211         }
 212
 213         spin_lock(&lkp->lk_spinlock);
 214
 215         extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK;
 216         td = curthread;
 217
 218         switch (flags & LK_TYPE_MASK) {
 219         case LK_SHARED:
 220                 /*
 221                  * If we are not the exclusive lock holder, we have to block
 222                  * while there is an exclusive lock holder or while an
 223                  * exclusive lock request or upgrade request is in progress.
 224                  *
 225                  * However, if P_DEADLKTREAT is set, we override exclusive
 226                  * lock requests or upgrade requests ( but not the exclusive
 227                  * lock itself ).
 228                  */
 229                 if (lkp->lk_lockholder != td) {
 230                         if (td->td_flags & TDF_DEADLKTREAT) {
 231                                 error = acquire(
 232                                             lkp,
 233                                             extflags,
 234                                             LK_HAVE_EXCL
 235                                         );
 236                         } else {
 237                                 error = acquire(
 238                                             lkp,
 239                                             extflags,
 240                                             LK_HAVE_EXCL | LK_WANT_EXCL |
 241                                              LK_WANT_UPGRADE
 242                                         );
 243                         }
 244                         if (error)
 245                                 break;
 246                         sharelock(lkp, 1);
 247                         COUNT(td, 1);
 248                         break;
 249                 }
 250                 /*
 251                  * We hold an exclusive lock, so downgrade it to shared.
 252                  * An alternative would be to fail with EDEADLK.
 253                  */
 254                 sharelock(lkp, 1);
 255                 COUNT(td, 1);
 256                 /* fall into downgrade */
 257
 258         case LK_DOWNGRADE:
 259                 if (lkp->lk_lockholder != td || lkp->lk_exclusivecount == 0) {
 260                         spin_unlock(&lkp->lk_spinlock);
 261                         panic("lockmgr: not holding exclusive lock");
 262                 }
 263                 sharelock(lkp, lkp->lk_exclusivecount);
 264                 lkp->lk_exclusivecount = 0;
 265                 lkp->lk_flags &= ~LK_HAVE_EXCL;
 266                 lkp->lk_lockholder = LK_NOTHREAD;
 267                 if (lkp->lk_waitcount)
 268                         wakeup((void *)lkp);
 269                 break;
 270
 271         case LK_EXCLUPGRADE:
 272                 /*
 273                  * If another process is ahead of us to get an upgrade,
 274                  * then we want to fail rather than have an intervening
 275                  * exclusive access.
 276                  */
 277                 if (lkp->lk_flags & LK_WANT_UPGRADE) {
 278                         shareunlock(lkp, 1);
 279                         COUNT(td, -1);
 280                         error = EBUSY;
 281                         break;
 282                 }
 283                 /* fall into normal upgrade */
 284
 285         case LK_UPGRADE:
 286                 /*
 287                  * Upgrade a shared lock to an exclusive one. If another
 288                  * shared lock has already requested an upgrade to an
 289                  * exclusive lock, our shared lock is released and an
 290                  * exclusive lock is requested (which will be granted
 291                  * after the upgrade). If we return an error, the file
 292                  * will always be unlocked.
 293                  */
 294                 if ((lkp->lk_lockholder == td) || (lkp->lk_sharecount <= 0)) {
 295                         spin_unlock(&lkp->lk_spinlock);
 296                         panic("lockmgr: upgrade exclusive lock");
 297                 }
 298                 shareunlock(lkp, 1);
 299                 COUNT(td, -1);
 300                 /*
 301                  * If we are just polling, check to see if we will block.
 302                  */
 303                 if ((extflags & LK_NOWAIT) &&
 304                     ((lkp->lk_flags & LK_WANT_UPGRADE) ||
 305                      lkp->lk_sharecount > 1)) {
 306                         error = EBUSY;
 307                         break;
 308                 }
 309                 if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) {
 310                         /*
 311                          * We are first shared lock to request an upgrade, so
 312                          * request upgrade and wait for the shared count to
 313                          * drop to zero, then take exclusive lock.
 314                          */
 315                         lkp->lk_flags |= LK_WANT_UPGRADE;
 316                         error = acquire(lkp, extflags, LK_SHARE_NONZERO);
 317                         lkp->lk_flags &= ~LK_WANT_UPGRADE;
 318
 319                         if (error)
 320                                 break;
 321                         lkp->lk_flags |= LK_HAVE_EXCL;
 322                         lkp->lk_lockholder = td;
 323                         if (lkp->lk_exclusivecount != 0) {
 324                                 spin_unlock(&lkp->lk_spinlock);
 325                                 panic("lockmgr: non-zero exclusive count");
 326                         }
 327                         lkp->lk_exclusivecount = 1;
 328 #if defined(DEBUG_LOCKS)
 329                         lkp->lk_filename = file;
 330                         lkp->lk_lineno = line;
 331                         lkp->lk_lockername = name;
 332 #endif
 333                         COUNT(td, 1);
 334                         break;
 335                 }
 336                 /*
 337                  * Someone else has requested upgrade. Release our shared
 338                  * lock, awaken upgrade requestor if we are the last shared
 339                  * lock, then request an exclusive lock.
 340                  */
 341                 if ( (lkp->lk_flags & (LK_SHARE_NONZERO|LK_WAIT_NONZERO)) ==
 342                         LK_WAIT_NONZERO)
 343                         wakeup((void *)lkp);
 344                 /* fall into exclusive request */
 345
 346         case LK_EXCLUSIVE:
 347                 if (lkp->lk_lockholder == td && td != LK_KERNTHREAD) {
 348                         /*
 349                          *      Recursive lock.
 350                          */
 351                         if ((extflags & (LK_NOWAIT | LK_CANRECURSE)) == 0) {
 352                                 spin_unlock(&lkp->lk_spinlock);
 353                                 panic("lockmgr: locking against myself");
 354                         }
 355                         if ((extflags & LK_CANRECURSE) != 0) {
 356                                 lkp->lk_exclusivecount++;
 357                                 COUNT(td, 1);
 358                                 break;
 359                         }
 360                 }
 361                 /*
 362                  * If we are just polling, check to see if we will sleep.
 363                  */
 364                 if ((extflags & LK_NOWAIT) &&
 365                     (lkp->lk_flags & (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO))) {
 366                         error = EBUSY;
 367                         break;
 368                 }
 369                 /*
 370                  * Try to acquire the want_exclusive flag.
 371                  */
 372                 error = acquire(lkp, extflags, (LK_HAVE_EXCL | LK_WANT_EXCL));
 373                 if (error)
 374                         break;
 375                 lkp->lk_flags |= LK_WANT_EXCL;
 376                 /*
 377                  * Wait for shared locks and upgrades to finish.
 378                  */
 379                 error = acquire(lkp, extflags, LK_WANT_UPGRADE | LK_SHARE_NONZERO);
 380                 lkp->lk_flags &= ~LK_WANT_EXCL;
 381                 if (error)
 382                         break;
 383                 lkp->lk_flags |= LK_HAVE_EXCL;
 384                 lkp->lk_lockholder = td;
 385                 if (lkp->lk_exclusivecount != 0) {
 386                         spin_unlock(&lkp->lk_spinlock);
 387                         panic("lockmgr: non-zero exclusive count");
 388                 }
 389                 lkp->lk_exclusivecount = 1;
 390 #if defined(DEBUG_LOCKS)
 391                         lkp->lk_filename = file;
 392                         lkp->lk_lineno = line;
 393                         lkp->lk_lockername = name;
 394 #endif
 395                 COUNT(td, 1);
 396                 break;
 397
 398         case LK_RELEASE:
 399                 if (lkp->lk_exclusivecount != 0) {
 400                         if (lkp->lk_lockholder != td &&
 401                             lkp->lk_lockholder != LK_KERNTHREAD) {
 402                                 spin_unlock(&lkp->lk_spinlock);
 403                                 panic("lockmgr: pid %d, not %s thr %p unlocking",
 404                                     (td->td_proc ? td->td_proc->p_pid : -99),
 405                                     "exclusive lock holder",
 406                                     lkp->lk_lockholder);
 407                         }
 408                         if (lkp->lk_lockholder != LK_KERNTHREAD) {
 409                                 COUNT(td, -1);
 410                         }
 411                         if (lkp->lk_exclusivecount == 1) {
 412                                 lkp->lk_flags &= ~LK_HAVE_EXCL;
 413                                 lkp->lk_lockholder = LK_NOTHREAD;
 414                                 lkp->lk_exclusivecount = 0;
 415                         } else {
 416                                 lkp->lk_exclusivecount--;
 417                         }
 418                 } else if (lkp->lk_flags & LK_SHARE_NONZERO) {
 419                         shareunlock(lkp, 1);
 420                         COUNT(td, -1);
 421                 }
 422                 if (lkp->lk_flags & LK_WAIT_NONZERO)
 423                         wakeup((void *)lkp);
 424                 break;
 425
 426         default:
 427                 spin_unlock(&lkp->lk_spinlock);
 428                 panic("lockmgr: unknown locktype request %d",
 429                     flags & LK_TYPE_MASK);
 430                 /* NOTREACHED */
 431         }
 432         spin_unlock(&lkp->lk_spinlock);
 433         return (error);
 434 }
 435
 436 void
 437 lockmgr_kernproc(struct lock *lp)
 438 {
 439         struct thread *td = curthread;
 440
 441         KASSERT(lp->lk_lockholder == td, ("lockmgr_kernproc: lock not owned by curthread %p", td));
 442         COUNT(td, -1);
 443         lp->lk_lockholder = LK_KERNTHREAD;
 444 }
 445
 446 /*
 447  * Initialize a lock; required before use.
 448  */
 449 void
 450 lockinit(struct lock *lkp, char *wmesg, int timo, int flags)
 451 {
 452         spin_init(&lkp->lk_spinlock);
 453         lkp->lk_flags = (flags & LK_EXTFLG_MASK);
 454         lkp->lk_sharecount = 0;
 455         lkp->lk_waitcount = 0;
 456         lkp->lk_exclusivecount = 0;
 457         lkp->lk_wmesg = wmesg;
 458         lkp->lk_timo = timo;
 459         lkp->lk_lockholder = LK_NOTHREAD;
 460 }
 461
 462 /*
 463  * Reinitialize a lock that is being reused for a different purpose, but
 464  * which may have pending (blocked) threads sitting on it.  The caller
 465  * must already hold the interlock.
 466  */
 467 void
 468 lockreinit(struct lock *lkp, char *wmesg, int timo, int flags)
 469 {
 470         lkp->lk_flags = (lkp->lk_flags & ~LK_EXTFLG_MASK) |
 471                         (flags & LK_EXTFLG_MASK);
 472         lkp->lk_wmesg = wmesg;
 473         lkp->lk_timo = timo;
 474 }
 475
 476 /*
 477  * Determine the status of a lock.
 478  */
 479 int
 480 lockstatus(struct lock *lkp, struct thread *td)
 481 {
 482         int lock_type = 0;
 483
 484         spin_lock(&lkp->lk_spinlock);
 485         if (lkp->lk_exclusivecount != 0) {
 486                 if (td == NULL || lkp->lk_lockholder == td)
 487                         lock_type = LK_EXCLUSIVE;
 488                 else
 489                         lock_type = LK_EXCLOTHER;
 490         } else if (lkp->lk_sharecount != 0) {
 491                 lock_type = LK_SHARED;
 492         }
 493         spin_unlock(&lkp->lk_spinlock);
 494         return (lock_type);
 495 }
 496
 497 /*
 498  * Determine the number of holders of a lock.
 499  *
 500  * The non-blocking version can usually be used for assertions.
 501  */
 502 int
 503 lockcount(struct lock *lkp)
 504 {
 505         int count;
 506
 507         spin_lock(&lkp->lk_spinlock);
 508         count = lkp->lk_exclusivecount + lkp->lk_sharecount;
 509         spin_unlock(&lkp->lk_spinlock);
 510         return (count);
 511 }
 512
 513 int
 514 lockcountnb(struct lock *lkp)
 515 {
 516         return (lkp->lk_exclusivecount + lkp->lk_sharecount);
 517 }
 518
 519 /*
 520  * Print out information about state of a lock. Used by VOP_PRINT
 521  * routines to display status about contained locks.
 522  */
 523 void
 524 lockmgr_printinfo(struct lock *lkp)
 525 {
 526         struct thread *td = lkp->lk_lockholder;
 527         struct proc *p;
 528
 529         if (td && td != LK_KERNTHREAD && td != LK_NOTHREAD)
 530                 p = td->td_proc;
 531         else
 532                 p = NULL;
 533
 534         if (lkp->lk_sharecount)
 535                 printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg,
 536                     lkp->lk_sharecount);
 537         else if (lkp->lk_flags & LK_HAVE_EXCL)
 538                 printf(" lock type %s: EXCL (count %d) by td %p pid %d",
 539                     lkp->lk_wmesg, lkp->lk_exclusivecount, td,
 540                     p ? p->p_pid : -99);
 541         if (lkp->lk_waitcount > 0)
 542                 printf(" with %d pending", lkp->lk_waitcount);
 543 }
 544