kern_lock.c revision 24273
1296417Sdim/* 2275072Semaste * Copyright (c) 1995 3275072Semaste * The Regents of the University of California. All rights reserved. 4275072Semaste * 5275072Semaste * This code contains ideas from software contributed to Berkeley by 6275072Semaste * Avadis Tevanian, Jr., Michael Wayne Young, and the Mach Operating 7275072Semaste * System project at Carnegie-Mellon University. 8275072Semaste * 9275072Semaste * Redistribution and use in source and binary forms, with or without 10275072Semaste * modification, are permitted provided that the following conditions 11275072Semaste * are met: 12275072Semaste * 1. Redistributions of source code must retain the above copyright 13275072Semaste * notice, this list of conditions and the following disclaimer. 14275072Semaste * 2. Redistributions in binary form must reproduce the above copyright 15280031Sdim * notice, this list of conditions and the following disclaimer in the 16280031Sdim * documentation and/or other materials provided with the distribution. 17280031Sdim * 3. All advertising materials mentioning features or use of this software 18280031Sdim * must display the following acknowledgement: 19280031Sdim * This product includes software developed by the University of 20275072Semaste * California, Berkeley and its contributors. 21280031Sdim * 4. Neither the name of the University nor the names of its contributors 22275072Semaste * may be used to endorse or promote products derived from this software 23280031Sdim * without specific prior written permission. 24275072Semaste * 25275072Semaste * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26275072Semaste * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27280031Sdim * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28280031Sdim * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29280031Sdim * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30280031Sdim * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31280031Sdim * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32280031Sdim * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33280031Sdim * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34275072Semaste * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35280031Sdim * SUCH DAMAGE. 36275072Semaste * 37280031Sdim * @(#)kern_lock.c 8.18 (Berkeley) 5/21/95 38280031Sdim */ 39280031Sdim 40275072Semaste#include <sys/param.h> 41275072Semaste#include <sys/proc.h> 42275072Semaste#include <sys/lock.h> 43280031Sdim#include <sys/systm.h> 44280031Sdim#include <machine/cpu.h> 45280031Sdim 46280031Sdim/* 47280031Sdim * Locking primitives implementation. 48280031Sdim * Locks provide shared/exclusive sychronization. 49275072Semaste */ 50280031Sdim 51275072Semaste#ifdef DEBUG 52280031Sdim#define COUNT(p, x) if (p) (p)->p_locks += (x) 53280031Sdim#else 54280031Sdim#define COUNT(p, x) 55275072Semaste#endif 56275072Semaste 57275072Semaste#if NCPUS > 1 58280031Sdim 59280031Sdim/* 60280031Sdim * For multiprocessor system, try spin lock first. 61280031Sdim * 62280031Sdim * This should be inline expanded below, but we cannot have #if 63275072Semaste * inside a multiline define. 64296417Sdim */ 65275072Semasteint lock_wait_time = 100; 66275072Semaste#define PAUSE(lkp, wanted) \ 67275072Semaste if (lock_wait_time > 0) { \ 68275072Semaste int i; \ 69280031Sdim \ 70280031Sdim simple_unlock(&lkp->lk_interlock); \ 71280031Sdim for (i = lock_wait_time; i > 0; i--) \ 72296417Sdim if (!(wanted)) \ 73280031Sdim break; \ 74275072Semaste simple_lock(&lkp->lk_interlock); \ 75296417Sdim } \ 76296417Sdim if (!(wanted)) \ 77275072Semaste break; 78288943Sdim 79280031Sdim#else /* NCPUS == 1 */ 80275072Semaste 81275072Semaste/* 82275072Semaste * It is an error to spin on a uniprocessor as nothing will ever cause 83280031Sdim * the simple lock to clear while we are executing. 84288943Sdim */ 85280031Sdim#define PAUSE(lkp, wanted) 86280031Sdim 87280031Sdim#endif /* NCPUS == 1 */ 88296417Sdim 89280031Sdim/* 90275072Semaste * Acquire a resource. 91296417Sdim */ 92280031Sdim#define ACQUIRE(lkp, error, extflags, wanted) \ 93275072Semaste PAUSE(lkp, wanted); \ 94296417Sdim for (error = 0; wanted; ) { \ 95275072Semaste (lkp)->lk_waitcount++; \ 96275072Semaste simple_unlock(&(lkp)->lk_interlock); \ 97275072Semaste error = tsleep((void *)lkp, (lkp)->lk_prio, \ 98280031Sdim (lkp)->lk_wmesg, (lkp)->lk_timo); \ 99280031Sdim simple_lock(&(lkp)->lk_interlock); \ 100280031Sdim (lkp)->lk_waitcount--; \ 101280031Sdim if (error) \ 102280031Sdim break; \ 103296417Sdim if ((extflags) & LK_SLEEPFAIL) { \ 104280031Sdim error = ENOLCK; \ 105275072Semaste break; \ 106296417Sdim } \ 107280031Sdim } 108275072Semaste 109296417Sdim/* 110275072Semaste * Initialize a lock; required before use. 111275072Semaste */ 112275072Semastevoid 113280031Sdimlockinit(lkp, prio, wmesg, timo, flags) 114288943Sdim struct lock *lkp; 115280031Sdim int prio; 116280031Sdim char *wmesg; 117280031Sdim int timo; 118296417Sdim int flags; 119280031Sdim{ 120275072Semaste 121296417Sdim simple_lock_init(&lkp->lk_interlock); 122280031Sdim lkp->lk_flags = flags & LK_EXTFLG_MASK; 123275072Semaste lkp->lk_sharecount = 0; 124280031Sdim lkp->lk_waitcount = 0; 125280031Sdim lkp->lk_exclusivecount = 0; 126280031Sdim lkp->lk_prio = prio; 127280031Sdim lkp->lk_wmesg = wmesg; 128280031Sdim lkp->lk_timo = timo; 129296417Sdim lkp->lk_lockholder = LK_NOPROC; 130296417Sdim} 131280031Sdim 132275072Semaste/* 133280031Sdim * Determine the status of a lock. 134288943Sdim */ 135280031Sdimint 136275072Semastelockstatus(lkp) 137275072Semaste struct lock *lkp; 138275072Semaste{ 139280031Sdim int lock_type = 0; 140280031Sdim 141280031Sdim simple_lock(&lkp->lk_interlock); 142280031Sdim if (lkp->lk_exclusivecount != 0) 143280031Sdim lock_type = LK_EXCLUSIVE; 144296417Sdim else if (lkp->lk_sharecount != 0) 145280031Sdim lock_type = LK_SHARED; 146275072Semaste simple_unlock(&lkp->lk_interlock); 147296417Sdim return (lock_type); 148280031Sdim} 149275072Semaste 150280031Sdim/* 151280031Sdim * Set, change, or release a lock. 152280031Sdim * 153280031Sdim * Shared requests increment the shared count. Exclusive requests set the 154280031Sdim * LK_WANT_EXCL flag (preventing further shared locks), and wait for already 155296417Sdim * accepted shared locks and shared-to-exclusive upgrades to go away. 156296417Sdim */ 157280031Sdimint 158275072Semastelockmgr(lkp, flags, interlkp, p) 159288943Sdim __volatile struct lock *lkp; 160288943Sdim u_int flags; 161288943Sdim struct simplelock *interlkp; 162288943Sdim struct proc *p; 163288943Sdim{ 164288943Sdim int error; 165280031Sdim pid_t pid; 166275072Semaste int extflags; 167275072Semaste 168275072Semaste error = 0; 169280031Sdim if (p) 170280031Sdim pid = p->p_pid; 171280031Sdim else 172280031Sdim pid = LK_KERNPROC; 173280031Sdim simple_lock(&lkp->lk_interlock); 174280031Sdim if (flags & LK_INTERLOCK) { 175275072Semaste simple_unlock(interlkp); 176280031Sdim } 177296417Sdim extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK; 178275072Semaste#ifdef DIAGNOSTIC 179280031Sdim /* 180275072Semaste * Once a lock has drained, the LK_DRAINING flag is set and an 181280031Sdim * exclusive lock is returned. The only valid operation thereafter 182280031Sdim * is a single release of that exclusive lock. This final release 183280031Sdim * clears the LK_DRAINING flag and sets the LK_DRAINED flag. Any 184280031Sdim * further requests of any sort will result in a panic. The bits 185280031Sdim * selected for these two flags are chosen so that they will be set 186280031Sdim * in memory that is freed (freed memory is filled with 0xdeadbeef). 187280031Sdim * The final release is permitted to give a new lease on life to 188280031Sdim * the lock by specifying LK_REENABLE. 189280031Sdim */ 190280031Sdim if (lkp->lk_flags & (LK_DRAINING|LK_DRAINED)) { 191275072Semaste if (lkp->lk_flags & LK_DRAINED) 192 panic("lockmgr: using decommissioned lock"); 193 if ((flags & LK_TYPE_MASK) != LK_RELEASE || 194 lkp->lk_lockholder != pid) 195 panic("lockmgr: non-release on draining lock: %d\n", 196 flags & LK_TYPE_MASK); 197 lkp->lk_flags &= ~LK_DRAINING; 198 if ((flags & LK_REENABLE) == 0) 199 lkp->lk_flags |= LK_DRAINED; 200 } 201#endif DIAGNOSTIC 202 203 switch (flags & LK_TYPE_MASK) { 204 205 case LK_SHARED: 206 if (lkp->lk_lockholder != pid) { 207 /* 208 * If just polling, check to see if we will block. 209 */ 210 if ((extflags & LK_NOWAIT) && (lkp->lk_flags & 211 (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE))) { 212 error = EBUSY; 213 break; 214 } 215 /* 216 * Wait for exclusive locks and upgrades to clear. 217 */ 218 ACQUIRE(lkp, error, extflags, lkp->lk_flags & 219 (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE)); 220 if (error) 221 break; 222 lkp->lk_sharecount++; 223 COUNT(p, 1); 224 break; 225 } 226 /* 227 * We hold an exclusive lock, so downgrade it to shared. 228 * An alternative would be to fail with EDEADLK. 229 */ 230 lkp->lk_sharecount++; 231 COUNT(p, 1); 232 /* fall into downgrade */ 233 234 case LK_DOWNGRADE: 235 if (lkp->lk_lockholder != pid || lkp->lk_exclusivecount == 0) 236 panic("lockmgr: not holding exclusive lock"); 237 lkp->lk_sharecount += lkp->lk_exclusivecount; 238 lkp->lk_exclusivecount = 0; 239 lkp->lk_flags &= ~LK_HAVE_EXCL; 240 lkp->lk_lockholder = LK_NOPROC; 241 if (lkp->lk_waitcount) 242 wakeup((void *)lkp); 243 break; 244 245 case LK_EXCLUPGRADE: 246 /* 247 * If another process is ahead of us to get an upgrade, 248 * then we want to fail rather than have an intervening 249 * exclusive access. 250 */ 251 if (lkp->lk_flags & LK_WANT_UPGRADE) { 252 lkp->lk_sharecount--; 253 COUNT(p, -1); 254 error = EBUSY; 255 break; 256 } 257 /* fall into normal upgrade */ 258 259 case LK_UPGRADE: 260 /* 261 * Upgrade a shared lock to an exclusive one. If another 262 * shared lock has already requested an upgrade to an 263 * exclusive lock, our shared lock is released and an 264 * exclusive lock is requested (which will be granted 265 * after the upgrade). If we return an error, the file 266 * will always be unlocked. 267 */ 268 if (lkp->lk_lockholder == pid || lkp->lk_sharecount <= 0) 269 panic("lockmgr: upgrade exclusive lock"); 270 lkp->lk_sharecount--; 271 COUNT(p, -1); 272 /* 273 * If we are just polling, check to see if we will block. 274 */ 275 if ((extflags & LK_NOWAIT) && 276 ((lkp->lk_flags & LK_WANT_UPGRADE) || 277 lkp->lk_sharecount > 1)) { 278 error = EBUSY; 279 break; 280 } 281 if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) { 282 /* 283 * We are first shared lock to request an upgrade, so 284 * request upgrade and wait for the shared count to 285 * drop to zero, then take exclusive lock. 286 */ 287 lkp->lk_flags |= LK_WANT_UPGRADE; 288 ACQUIRE(lkp, error, extflags, lkp->lk_sharecount); 289 lkp->lk_flags &= ~LK_WANT_UPGRADE; 290 if (error) 291 break; 292 lkp->lk_flags |= LK_HAVE_EXCL; 293 lkp->lk_lockholder = pid; 294 if (lkp->lk_exclusivecount != 0) 295 panic("lockmgr: non-zero exclusive count"); 296 lkp->lk_exclusivecount = 1; 297 COUNT(p, 1); 298 break; 299 } 300 /* 301 * Someone else has requested upgrade. Release our shared 302 * lock, awaken upgrade requestor if we are the last shared 303 * lock, then request an exclusive lock. 304 */ 305 if (lkp->lk_sharecount == 0 && lkp->lk_waitcount) 306 wakeup((void *)lkp); 307 /* fall into exclusive request */ 308 309 case LK_EXCLUSIVE: 310 if (lkp->lk_lockholder == pid && pid != LK_KERNPROC) { 311 /* 312 * Recursive lock. 313 */ 314 if ((extflags & LK_CANRECURSE) == 0) 315 panic("lockmgr: locking against myself"); 316 lkp->lk_exclusivecount++; 317 COUNT(p, 1); 318 break; 319 } 320 /* 321 * If we are just polling, check to see if we will sleep. 322 */ 323 if ((extflags & LK_NOWAIT) && ((lkp->lk_flags & 324 (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE)) || 325 lkp->lk_sharecount != 0)) { 326 error = EBUSY; 327 break; 328 } 329 /* 330 * Try to acquire the want_exclusive flag. 331 */ 332 ACQUIRE(lkp, error, extflags, lkp->lk_flags & 333 (LK_HAVE_EXCL | LK_WANT_EXCL)); 334 if (error) 335 break; 336 lkp->lk_flags |= LK_WANT_EXCL; 337 /* 338 * Wait for shared locks and upgrades to finish. 339 */ 340 ACQUIRE(lkp, error, extflags, lkp->lk_sharecount != 0 || 341 (lkp->lk_flags & LK_WANT_UPGRADE)); 342 lkp->lk_flags &= ~LK_WANT_EXCL; 343 if (error) 344 break; 345 lkp->lk_flags |= LK_HAVE_EXCL; 346 lkp->lk_lockholder = pid; 347 if (lkp->lk_exclusivecount != 0) 348 panic("lockmgr: non-zero exclusive count"); 349 lkp->lk_exclusivecount = 1; 350 COUNT(p, 1); 351 break; 352 353 case LK_RELEASE: 354 if (lkp->lk_exclusivecount != 0) { 355 if (pid != lkp->lk_lockholder) 356 panic("lockmgr: pid %d, not %s %d unlocking", 357 pid, "exclusive lock holder", 358 lkp->lk_lockholder); 359 lkp->lk_exclusivecount--; 360 COUNT(p, -1); 361 if (lkp->lk_exclusivecount == 0) { 362 lkp->lk_flags &= ~LK_HAVE_EXCL; 363 lkp->lk_lockholder = LK_NOPROC; 364 } 365 } else if (lkp->lk_sharecount != 0) { 366 lkp->lk_sharecount--; 367 COUNT(p, -1); 368 } 369 if (lkp->lk_waitcount) 370 wakeup((void *)lkp); 371 break; 372 373 case LK_DRAIN: 374 /* 375 * Check that we do not already hold the lock, as it can 376 * never drain if we do. Unfortunately, we have no way to 377 * check for holding a shared lock, but at least we can 378 * check for an exclusive one. 379 */ 380 if (lkp->lk_lockholder == pid) 381 panic("lockmgr: draining against myself"); 382 /* 383 * If we are just polling, check to see if we will sleep. 384 */ 385 if ((extflags & LK_NOWAIT) && ((lkp->lk_flags & 386 (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE)) || 387 lkp->lk_sharecount != 0 || lkp->lk_waitcount != 0)) { 388 error = EBUSY; 389 break; 390 } 391 PAUSE(lkp, ((lkp->lk_flags & 392 (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE)) || 393 lkp->lk_sharecount != 0 || lkp->lk_waitcount != 0)); 394 for (error = 0; ((lkp->lk_flags & 395 (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE)) || 396 lkp->lk_sharecount != 0 || lkp->lk_waitcount != 0); ) { 397 lkp->lk_flags |= LK_WAITDRAIN; 398 simple_unlock(&lkp->lk_interlock); 399 if (error = tsleep((void *)&lkp->lk_flags, lkp->lk_prio, 400 lkp->lk_wmesg, lkp->lk_timo)) 401 return (error); 402 if ((extflags) & LK_SLEEPFAIL) 403 return (ENOLCK); 404 simple_lock(&lkp->lk_interlock); 405 } 406 lkp->lk_flags |= LK_DRAINING | LK_HAVE_EXCL; 407 lkp->lk_lockholder = pid; 408 lkp->lk_exclusivecount = 1; 409 COUNT(p, 1); 410 break; 411 412 default: 413 simple_unlock(&lkp->lk_interlock); 414 panic("lockmgr: unknown locktype request %d", 415 flags & LK_TYPE_MASK); 416 /* NOTREACHED */ 417 } 418 if ((lkp->lk_flags & LK_WAITDRAIN) && ((lkp->lk_flags & 419 (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE)) == 0 && 420 lkp->lk_sharecount == 0 && lkp->lk_waitcount == 0)) { 421 lkp->lk_flags &= ~LK_WAITDRAIN; 422 wakeup((void *)&lkp->lk_flags); 423 } 424 simple_unlock(&lkp->lk_interlock); 425 return (error); 426} 427 428/* 429 * Print out information about state of a lock. Used by VOP_PRINT 430 * routines to display ststus about contained locks. 431 */ 432void 433lockmgr_printinfo(lkp) 434 struct lock *lkp; 435{ 436 437 if (lkp->lk_sharecount) 438 printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg, 439 lkp->lk_sharecount); 440 else if (lkp->lk_flags & LK_HAVE_EXCL) 441 printf(" lock type %s: EXCL (count %d) by pid %d", 442 lkp->lk_wmesg, lkp->lk_exclusivecount, lkp->lk_lockholder); 443 if (lkp->lk_waitcount > 0) 444 printf(" with %d pending", lkp->lk_waitcount); 445} 446 447#if defined(SIMPLELOCK_DEBUG) && NCPUS == 1 448#include <sys/kernel.h> 449#include <vm/vm.h> 450#include <sys/sysctl.h> 451int lockpausetime = 0; 452/* struct ctldebug debug2 = { "lockpausetime", &lockpausetime }; */ 453int simplelockrecurse; 454/* 455 * Simple lock functions so that the debugger can see from whence 456 * they are being called. 457 */ 458void 459simple_lock_init(alp) 460 struct simplelock *alp; 461{ 462 463 alp->lock_data = 0; 464} 465 466void 467_simple_lock(alp, id, l) 468 __volatile struct simplelock *alp; 469 const char *id; 470 int l; 471{ 472 473 if (simplelockrecurse) 474 return; 475 if (alp->lock_data == 1) { 476 if (lockpausetime == -1) 477 panic("%s:%d: simple_lock: lock held", id, l); 478 printf("%s:%d: simple_lock: lock held\n", id, l); 479 if (lockpausetime == 1) { 480 Debugger("simple_lock"); 481 /*BACKTRACE(curproc); */ 482 } else if (lockpausetime > 1) { 483 printf("%s:%d: simple_lock: lock held...", id, l); 484 tsleep(&lockpausetime, PCATCH | PPAUSE, "slock", 485 lockpausetime * hz); 486 printf(" continuing\n"); 487 } 488 } 489 alp->lock_data = 1; 490 if (curproc) 491 curproc->p_simple_locks++; 492} 493 494int 495_simple_lock_try(alp, id, l) 496 __volatile struct simplelock *alp; 497 const char *id; 498 int l; 499{ 500 501 if (alp->lock_data) 502 return (0); 503 if (simplelockrecurse) 504 return (1); 505 alp->lock_data = 1; 506 if (curproc) 507 curproc->p_simple_locks++; 508 return (1); 509} 510 511void 512_simple_unlock(alp, id, l) 513 __volatile struct simplelock *alp; 514 const char *id; 515 int l; 516{ 517 518 if (simplelockrecurse) 519 return; 520 if (alp->lock_data == 0) { 521 if (lockpausetime == -1) 522 panic("%s:%d: simple_unlock: lock not held", id, l); 523 printf("%s:%d: simple_unlock: lock not held\n", id, l); 524 if (lockpausetime == 1) { 525 Debugger("simple_unlock"); 526 /* BACKTRACE(curproc); */ 527 } else if (lockpausetime > 1) { 528 printf("%s:%d: simple_unlock: lock not held...", id, l); 529 tsleep(&lockpausetime, PCATCH | PPAUSE, "sunlock", 530 lockpausetime * hz); 531 printf(" continuing\n"); 532 } 533 } 534 alp->lock_data = 0; 535 if (curproc) 536 curproc->p_simple_locks--; 537} 538#endif /* SIMPLELOCK_DEBUG && NCPUS == 1 */ 539