| 1 | /* $NetBSD: vfs_lockf.c,v 1.73 2011/01/31 08:25:32 dholland Exp $ */ |
| 2 | |
| 3 | /* |
| 4 | * Copyright (c) 1982, 1986, 1989, 1993 |
| 5 | * The Regents of the University of California. All rights reserved. |
| 6 | * |
| 7 | * This code is derived from software contributed to Berkeley by |
| 8 | * Scooter Morris at Genentech Inc. |
| 9 | * |
| 10 | * Redistribution and use in source and binary forms, with or without |
| 11 | * modification, are permitted provided that the following conditions |
| 12 | * are met: |
| 13 | * 1. Redistributions of source code must retain the above copyright |
| 14 | * notice, this list of conditions and the following disclaimer. |
| 15 | * 2. Redistributions in binary form must reproduce the above copyright |
| 16 | * notice, this list of conditions and the following disclaimer in the |
| 17 | * documentation and/or other materials provided with the distribution. |
| 18 | * 3. Neither the name of the University nor the names of its contributors |
| 19 | * may be used to endorse or promote products derived from this software |
| 20 | * without specific prior written permission. |
| 21 | * |
| 22 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 23 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 24 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 25 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 26 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 27 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 28 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 29 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 30 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 31 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 32 | * SUCH DAMAGE. |
| 33 | * |
| 34 | * @(#)ufs_lockf.c 8.4 (Berkeley) 10/26/94 |
| 35 | */ |
| 36 | |
| 37 | #include <sys/cdefs.h> |
| 38 | __KERNEL_RCSID(0, "$NetBSD: vfs_lockf.c,v 1.73 2011/01/31 08:25:32 dholland Exp $" ); |
| 39 | |
| 40 | #include <sys/param.h> |
| 41 | #include <sys/systm.h> |
| 42 | #include <sys/kernel.h> |
| 43 | #include <sys/file.h> |
| 44 | #include <sys/proc.h> |
| 45 | #include <sys/vnode.h> |
| 46 | #include <sys/pool.h> |
| 47 | #include <sys/fcntl.h> |
| 48 | #include <sys/lockf.h> |
| 49 | #include <sys/atomic.h> |
| 50 | #include <sys/kauth.h> |
| 51 | #include <sys/uidinfo.h> |
| 52 | |
| 53 | /* |
| 54 | * The lockf structure is a kernel structure which contains the information |
| 55 | * associated with a byte range lock. The lockf structures are linked into |
| 56 | * the vnode structure. Locks are sorted by the starting byte of the lock for |
| 57 | * efficiency. |
| 58 | * |
| 59 | * lf_next is used for two purposes, depending on whether the lock is |
| 60 | * being held, or is in conflict with an existing lock. If this lock |
| 61 | * is held, it indicates the next lock on the same vnode. |
| 62 | * For pending locks, if lock->lf_next is non-NULL, then lock->lf_block |
| 63 | * must be queued on the lf_blkhd TAILQ of lock->lf_next. |
| 64 | */ |
| 65 | |
| 66 | TAILQ_HEAD(locklist, lockf); |
| 67 | |
| 68 | struct lockf { |
| 69 | kcondvar_t lf_cv; /* Signalling */ |
| 70 | short lf_flags; /* Lock semantics: F_POSIX, F_FLOCK, F_WAIT */ |
| 71 | short lf_type; /* Lock type: F_RDLCK, F_WRLCK */ |
| 72 | off_t lf_start; /* The byte # of the start of the lock */ |
| 73 | off_t lf_end; /* The byte # of the end of the lock (-1=EOF)*/ |
| 74 | void *lf_id; /* process or file description holding lock */ |
| 75 | struct lockf **lf_head; /* Back pointer to the head of lockf list */ |
| 76 | struct lockf *lf_next; /* Next lock on this vnode, or blocking lock */ |
| 77 | struct locklist lf_blkhd; /* List of requests blocked on this lock */ |
| 78 | TAILQ_ENTRY(lockf) lf_block;/* A request waiting for a lock */ |
| 79 | uid_t lf_uid; /* User ID responsible */ |
| 80 | }; |
| 81 | |
| 82 | /* Maximum length of sleep chains to traverse to try and detect deadlock. */ |
| 83 | #define MAXDEPTH 50 |
| 84 | |
| 85 | static pool_cache_t lockf_cache; |
| 86 | static kmutex_t *lockf_lock; |
| 87 | static char lockstr[] = "lockf" ; |
| 88 | |
| 89 | /* |
| 90 | * This variable controls the maximum number of processes that will |
| 91 | * be checked in doing deadlock detection. |
| 92 | */ |
| 93 | int maxlockdepth = MAXDEPTH; |
| 94 | |
| 95 | #ifdef LOCKF_DEBUG |
| 96 | int lockf_debug = 0; |
| 97 | #endif |
| 98 | |
| 99 | #define SELF 0x1 |
| 100 | #define OTHERS 0x2 |
| 101 | |
| 102 | /* |
| 103 | * XXX TODO |
| 104 | * Misc cleanups: "void *id" should be visible in the API as a |
| 105 | * "struct proc *". |
| 106 | * (This requires rototilling all VFS's which support advisory locking). |
| 107 | */ |
| 108 | |
| 109 | /* |
| 110 | * If there's a lot of lock contention on a single vnode, locking |
| 111 | * schemes which allow for more paralleism would be needed. Given how |
| 112 | * infrequently byte-range locks are actually used in typical BSD |
| 113 | * code, a more complex approach probably isn't worth it. |
| 114 | */ |
| 115 | |
| 116 | /* |
| 117 | * We enforce a limit on locks by uid, so that a single user cannot |
| 118 | * run the kernel out of memory. For now, the limit is pretty coarse. |
| 119 | * There is no limit on root. |
| 120 | * |
| 121 | * Splitting a lock will always succeed, regardless of current allocations. |
| 122 | * If you're slightly above the limit, we still have to permit an allocation |
| 123 | * so that the unlock can succeed. If the unlocking causes too many splits, |
| 124 | * however, you're totally cutoff. |
| 125 | */ |
| 126 | int maxlocksperuid = 1024; |
| 127 | |
| 128 | #ifdef LOCKF_DEBUG |
| 129 | /* |
| 130 | * Print out a lock. |
| 131 | */ |
| 132 | static void |
| 133 | lf_print(const char *tag, struct lockf *lock) |
| 134 | { |
| 135 | |
| 136 | printf("%s: lock %p for " , tag, lock); |
| 137 | if (lock->lf_flags & F_POSIX) |
| 138 | printf("proc %d" , ((struct proc *)lock->lf_id)->p_pid); |
| 139 | else |
| 140 | printf("file %p" , (struct file *)lock->lf_id); |
| 141 | printf(" %s, start %jd, end %jd" , |
| 142 | lock->lf_type == F_RDLCK ? "shared" : |
| 143 | lock->lf_type == F_WRLCK ? "exclusive" : |
| 144 | lock->lf_type == F_UNLCK ? "unlock" : |
| 145 | "unknown" , (intmax_t)lock->lf_start, (intmax_t)lock->lf_end); |
| 146 | if (TAILQ_FIRST(&lock->lf_blkhd)) |
| 147 | printf(" block %p\n" , TAILQ_FIRST(&lock->lf_blkhd)); |
| 148 | else |
| 149 | printf("\n" ); |
| 150 | } |
| 151 | |
| 152 | static void |
| 153 | lf_printlist(const char *tag, struct lockf *lock) |
| 154 | { |
| 155 | struct lockf *lf, *blk; |
| 156 | |
| 157 | printf("%s: Lock list:\n" , tag); |
| 158 | for (lf = *lock->lf_head; lf; lf = lf->lf_next) { |
| 159 | printf("\tlock %p for " , lf); |
| 160 | if (lf->lf_flags & F_POSIX) |
| 161 | printf("proc %d" , ((struct proc *)lf->lf_id)->p_pid); |
| 162 | else |
| 163 | printf("file %p" , (struct file *)lf->lf_id); |
| 164 | printf(", %s, start %jd, end %jd" , |
| 165 | lf->lf_type == F_RDLCK ? "shared" : |
| 166 | lf->lf_type == F_WRLCK ? "exclusive" : |
| 167 | lf->lf_type == F_UNLCK ? "unlock" : |
| 168 | "unknown" , (intmax_t)lf->lf_start, (intmax_t)lf->lf_end); |
| 169 | TAILQ_FOREACH(blk, &lf->lf_blkhd, lf_block) { |
| 170 | if (blk->lf_flags & F_POSIX) |
| 171 | printf("; proc %d" , |
| 172 | ((struct proc *)blk->lf_id)->p_pid); |
| 173 | else |
| 174 | printf("; file %p" , (struct file *)blk->lf_id); |
| 175 | printf(", %s, start %jd, end %jd" , |
| 176 | blk->lf_type == F_RDLCK ? "shared" : |
| 177 | blk->lf_type == F_WRLCK ? "exclusive" : |
| 178 | blk->lf_type == F_UNLCK ? "unlock" : |
| 179 | "unknown" , (intmax_t)blk->lf_start, (intmax_t)blk->lf_end); |
| 180 | if (TAILQ_FIRST(&blk->lf_blkhd)) |
| 181 | panic("lf_printlist: bad list" ); |
| 182 | } |
| 183 | printf("\n" ); |
| 184 | } |
| 185 | } |
| 186 | #endif /* LOCKF_DEBUG */ |
| 187 | |
| 188 | /* |
| 189 | * 3 options for allowfail. |
| 190 | * 0 - always allocate. 1 - cutoff at limit. 2 - cutoff at double limit. |
| 191 | */ |
| 192 | static struct lockf * |
| 193 | lf_alloc(int allowfail) |
| 194 | { |
| 195 | struct uidinfo *uip; |
| 196 | struct lockf *lock; |
| 197 | u_long lcnt; |
| 198 | const uid_t uid = kauth_cred_geteuid(kauth_cred_get()); |
| 199 | |
| 200 | uip = uid_find(uid); |
| 201 | lcnt = atomic_inc_ulong_nv(&uip->ui_lockcnt); |
| 202 | if (uid && allowfail && lcnt > |
| 203 | (allowfail == 1 ? maxlocksperuid : (maxlocksperuid * 2))) { |
| 204 | atomic_dec_ulong(&uip->ui_lockcnt); |
| 205 | return NULL; |
| 206 | } |
| 207 | |
| 208 | lock = pool_cache_get(lockf_cache, PR_WAITOK); |
| 209 | lock->lf_uid = uid; |
| 210 | return lock; |
| 211 | } |
| 212 | |
| 213 | static void |
| 214 | lf_free(struct lockf *lock) |
| 215 | { |
| 216 | struct uidinfo *uip; |
| 217 | |
| 218 | uip = uid_find(lock->lf_uid); |
| 219 | atomic_dec_ulong(&uip->ui_lockcnt); |
| 220 | pool_cache_put(lockf_cache, lock); |
| 221 | } |
| 222 | |
| 223 | static int |
| 224 | lf_ctor(void *arg, void *obj, int flag) |
| 225 | { |
| 226 | struct lockf *lock; |
| 227 | |
| 228 | lock = obj; |
| 229 | cv_init(&lock->lf_cv, lockstr); |
| 230 | |
| 231 | return 0; |
| 232 | } |
| 233 | |
| 234 | static void |
| 235 | lf_dtor(void *arg, void *obj) |
| 236 | { |
| 237 | struct lockf *lock; |
| 238 | |
| 239 | lock = obj; |
| 240 | cv_destroy(&lock->lf_cv); |
| 241 | } |
| 242 | |
| 243 | /* |
| 244 | * Walk the list of locks for an inode to |
| 245 | * find an overlapping lock (if any). |
| 246 | * |
| 247 | * NOTE: this returns only the FIRST overlapping lock. There |
| 248 | * may be more than one. |
| 249 | */ |
| 250 | static int |
| 251 | lf_findoverlap(struct lockf *lf, struct lockf *lock, int type, |
| 252 | struct lockf ***prev, struct lockf **overlap) |
| 253 | { |
| 254 | off_t start, end; |
| 255 | |
| 256 | *overlap = lf; |
| 257 | if (lf == NULL) |
| 258 | return 0; |
| 259 | #ifdef LOCKF_DEBUG |
| 260 | if (lockf_debug & 2) |
| 261 | lf_print("lf_findoverlap: looking for overlap in" , lock); |
| 262 | #endif /* LOCKF_DEBUG */ |
| 263 | start = lock->lf_start; |
| 264 | end = lock->lf_end; |
| 265 | while (lf != NULL) { |
| 266 | if (((type == SELF) && lf->lf_id != lock->lf_id) || |
| 267 | ((type == OTHERS) && lf->lf_id == lock->lf_id)) { |
| 268 | *prev = &lf->lf_next; |
| 269 | *overlap = lf = lf->lf_next; |
| 270 | continue; |
| 271 | } |
| 272 | #ifdef LOCKF_DEBUG |
| 273 | if (lockf_debug & 2) |
| 274 | lf_print("\tchecking" , lf); |
| 275 | #endif /* LOCKF_DEBUG */ |
| 276 | /* |
| 277 | * OK, check for overlap |
| 278 | * |
| 279 | * Six cases: |
| 280 | * 0) no overlap |
| 281 | * 1) overlap == lock |
| 282 | * 2) overlap contains lock |
| 283 | * 3) lock contains overlap |
| 284 | * 4) overlap starts before lock |
| 285 | * 5) overlap ends after lock |
| 286 | */ |
| 287 | if ((lf->lf_end != -1 && start > lf->lf_end) || |
| 288 | (end != -1 && lf->lf_start > end)) { |
| 289 | /* Case 0 */ |
| 290 | #ifdef LOCKF_DEBUG |
| 291 | if (lockf_debug & 2) |
| 292 | printf("no overlap\n" ); |
| 293 | #endif /* LOCKF_DEBUG */ |
| 294 | if ((type & SELF) && end != -1 && lf->lf_start > end) |
| 295 | return 0; |
| 296 | *prev = &lf->lf_next; |
| 297 | *overlap = lf = lf->lf_next; |
| 298 | continue; |
| 299 | } |
| 300 | if ((lf->lf_start == start) && (lf->lf_end == end)) { |
| 301 | /* Case 1 */ |
| 302 | #ifdef LOCKF_DEBUG |
| 303 | if (lockf_debug & 2) |
| 304 | printf("overlap == lock\n" ); |
| 305 | #endif /* LOCKF_DEBUG */ |
| 306 | return 1; |
| 307 | } |
| 308 | if ((lf->lf_start <= start) && |
| 309 | (end != -1) && |
| 310 | ((lf->lf_end >= end) || (lf->lf_end == -1))) { |
| 311 | /* Case 2 */ |
| 312 | #ifdef LOCKF_DEBUG |
| 313 | if (lockf_debug & 2) |
| 314 | printf("overlap contains lock\n" ); |
| 315 | #endif /* LOCKF_DEBUG */ |
| 316 | return 2; |
| 317 | } |
| 318 | if (start <= lf->lf_start && |
| 319 | (end == -1 || |
| 320 | (lf->lf_end != -1 && end >= lf->lf_end))) { |
| 321 | /* Case 3 */ |
| 322 | #ifdef LOCKF_DEBUG |
| 323 | if (lockf_debug & 2) |
| 324 | printf("lock contains overlap\n" ); |
| 325 | #endif /* LOCKF_DEBUG */ |
| 326 | return 3; |
| 327 | } |
| 328 | if ((lf->lf_start < start) && |
| 329 | ((lf->lf_end >= start) || (lf->lf_end == -1))) { |
| 330 | /* Case 4 */ |
| 331 | #ifdef LOCKF_DEBUG |
| 332 | if (lockf_debug & 2) |
| 333 | printf("overlap starts before lock\n" ); |
| 334 | #endif /* LOCKF_DEBUG */ |
| 335 | return 4; |
| 336 | } |
| 337 | if ((lf->lf_start > start) && |
| 338 | (end != -1) && |
| 339 | ((lf->lf_end > end) || (lf->lf_end == -1))) { |
| 340 | /* Case 5 */ |
| 341 | #ifdef LOCKF_DEBUG |
| 342 | if (lockf_debug & 2) |
| 343 | printf("overlap ends after lock\n" ); |
| 344 | #endif /* LOCKF_DEBUG */ |
| 345 | return 5; |
| 346 | } |
| 347 | panic("lf_findoverlap: default" ); |
| 348 | } |
| 349 | return 0; |
| 350 | } |
| 351 | |
| 352 | /* |
| 353 | * Split a lock and a contained region into |
| 354 | * two or three locks as necessary. |
| 355 | */ |
| 356 | static void |
| 357 | lf_split(struct lockf *lock1, struct lockf *lock2, struct lockf **sparelock) |
| 358 | { |
| 359 | struct lockf *splitlock; |
| 360 | |
| 361 | #ifdef LOCKF_DEBUG |
| 362 | if (lockf_debug & 2) { |
| 363 | lf_print("lf_split" , lock1); |
| 364 | lf_print("splitting from" , lock2); |
| 365 | } |
| 366 | #endif /* LOCKF_DEBUG */ |
| 367 | /* |
| 368 | * Check to see if spliting into only two pieces. |
| 369 | */ |
| 370 | if (lock1->lf_start == lock2->lf_start) { |
| 371 | lock1->lf_start = lock2->lf_end + 1; |
| 372 | lock2->lf_next = lock1; |
| 373 | return; |
| 374 | } |
| 375 | if (lock1->lf_end == lock2->lf_end) { |
| 376 | lock1->lf_end = lock2->lf_start - 1; |
| 377 | lock2->lf_next = lock1->lf_next; |
| 378 | lock1->lf_next = lock2; |
| 379 | return; |
| 380 | } |
| 381 | /* |
| 382 | * Make a new lock consisting of the last part of |
| 383 | * the encompassing lock |
| 384 | */ |
| 385 | splitlock = *sparelock; |
| 386 | *sparelock = NULL; |
| 387 | cv_destroy(&splitlock->lf_cv); |
| 388 | memcpy(splitlock, lock1, sizeof(*splitlock)); |
| 389 | cv_init(&splitlock->lf_cv, lockstr); |
| 390 | |
| 391 | splitlock->lf_start = lock2->lf_end + 1; |
| 392 | TAILQ_INIT(&splitlock->lf_blkhd); |
| 393 | lock1->lf_end = lock2->lf_start - 1; |
| 394 | /* |
| 395 | * OK, now link it in |
| 396 | */ |
| 397 | splitlock->lf_next = lock1->lf_next; |
| 398 | lock2->lf_next = splitlock; |
| 399 | lock1->lf_next = lock2; |
| 400 | } |
| 401 | |
| 402 | /* |
| 403 | * Wakeup a blocklist |
| 404 | */ |
| 405 | static void |
| 406 | lf_wakelock(struct lockf *listhead) |
| 407 | { |
| 408 | struct lockf *wakelock; |
| 409 | |
| 410 | while ((wakelock = TAILQ_FIRST(&listhead->lf_blkhd))) { |
| 411 | KASSERT(wakelock->lf_next == listhead); |
| 412 | TAILQ_REMOVE(&listhead->lf_blkhd, wakelock, lf_block); |
| 413 | wakelock->lf_next = NULL; |
| 414 | #ifdef LOCKF_DEBUG |
| 415 | if (lockf_debug & 2) |
| 416 | lf_print("lf_wakelock: awakening" , wakelock); |
| 417 | #endif |
| 418 | cv_broadcast(&wakelock->lf_cv); |
| 419 | } |
| 420 | } |
| 421 | |
| 422 | /* |
| 423 | * Remove a byte-range lock on an inode. |
| 424 | * |
| 425 | * Generally, find the lock (or an overlap to that lock) |
| 426 | * and remove it (or shrink it), then wakeup anyone we can. |
| 427 | */ |
| 428 | static int |
| 429 | lf_clearlock(struct lockf *unlock, struct lockf **sparelock) |
| 430 | { |
| 431 | struct lockf **head = unlock->lf_head; |
| 432 | struct lockf *lf = *head; |
| 433 | struct lockf *overlap, **prev; |
| 434 | int ovcase; |
| 435 | |
| 436 | if (lf == NULL) |
| 437 | return 0; |
| 438 | #ifdef LOCKF_DEBUG |
| 439 | if (unlock->lf_type != F_UNLCK) |
| 440 | panic("lf_clearlock: bad type" ); |
| 441 | if (lockf_debug & 1) |
| 442 | lf_print("lf_clearlock" , unlock); |
| 443 | #endif /* LOCKF_DEBUG */ |
| 444 | prev = head; |
| 445 | while ((ovcase = lf_findoverlap(lf, unlock, SELF, |
| 446 | &prev, &overlap)) != 0) { |
| 447 | /* |
| 448 | * Wakeup the list of locks to be retried. |
| 449 | */ |
| 450 | lf_wakelock(overlap); |
| 451 | |
| 452 | switch (ovcase) { |
| 453 | |
| 454 | case 1: /* overlap == lock */ |
| 455 | *prev = overlap->lf_next; |
| 456 | lf_free(overlap); |
| 457 | break; |
| 458 | |
| 459 | case 2: /* overlap contains lock: split it */ |
| 460 | if (overlap->lf_start == unlock->lf_start) { |
| 461 | overlap->lf_start = unlock->lf_end + 1; |
| 462 | break; |
| 463 | } |
| 464 | lf_split(overlap, unlock, sparelock); |
| 465 | overlap->lf_next = unlock->lf_next; |
| 466 | break; |
| 467 | |
| 468 | case 3: /* lock contains overlap */ |
| 469 | *prev = overlap->lf_next; |
| 470 | lf = overlap->lf_next; |
| 471 | lf_free(overlap); |
| 472 | continue; |
| 473 | |
| 474 | case 4: /* overlap starts before lock */ |
| 475 | overlap->lf_end = unlock->lf_start - 1; |
| 476 | prev = &overlap->lf_next; |
| 477 | lf = overlap->lf_next; |
| 478 | continue; |
| 479 | |
| 480 | case 5: /* overlap ends after lock */ |
| 481 | overlap->lf_start = unlock->lf_end + 1; |
| 482 | break; |
| 483 | } |
| 484 | break; |
| 485 | } |
| 486 | #ifdef LOCKF_DEBUG |
| 487 | if (lockf_debug & 1) |
| 488 | lf_printlist("lf_clearlock" , unlock); |
| 489 | #endif /* LOCKF_DEBUG */ |
| 490 | return 0; |
| 491 | } |
| 492 | |
| 493 | /* |
| 494 | * Walk the list of locks for an inode and |
| 495 | * return the first blocking lock. |
| 496 | */ |
| 497 | static struct lockf * |
| 498 | lf_getblock(struct lockf *lock) |
| 499 | { |
| 500 | struct lockf **prev, *overlap, *lf = *(lock->lf_head); |
| 501 | |
| 502 | prev = lock->lf_head; |
| 503 | while (lf_findoverlap(lf, lock, OTHERS, &prev, &overlap) != 0) { |
| 504 | /* |
| 505 | * We've found an overlap, see if it blocks us |
| 506 | */ |
| 507 | if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK)) |
| 508 | return overlap; |
| 509 | /* |
| 510 | * Nope, point to the next one on the list and |
| 511 | * see if it blocks us |
| 512 | */ |
| 513 | lf = overlap->lf_next; |
| 514 | } |
| 515 | return NULL; |
| 516 | } |
| 517 | |
| 518 | /* |
| 519 | * Set a byte-range lock. |
| 520 | */ |
| 521 | static int |
| 522 | lf_setlock(struct lockf *lock, struct lockf **sparelock, |
| 523 | kmutex_t *interlock) |
| 524 | { |
| 525 | struct lockf *block; |
| 526 | struct lockf **head = lock->lf_head; |
| 527 | struct lockf **prev, *overlap, *ltmp; |
| 528 | int ovcase, needtolink, error; |
| 529 | |
| 530 | #ifdef LOCKF_DEBUG |
| 531 | if (lockf_debug & 1) |
| 532 | lf_print("lf_setlock" , lock); |
| 533 | #endif /* LOCKF_DEBUG */ |
| 534 | |
| 535 | /* |
| 536 | * Scan lock list for this file looking for locks that would block us. |
| 537 | */ |
| 538 | while ((block = lf_getblock(lock)) != NULL) { |
| 539 | /* |
| 540 | * Free the structure and return if nonblocking. |
| 541 | */ |
| 542 | if ((lock->lf_flags & F_WAIT) == 0) { |
| 543 | lf_free(lock); |
| 544 | return EAGAIN; |
| 545 | } |
| 546 | /* |
| 547 | * We are blocked. Since flock style locks cover |
| 548 | * the whole file, there is no chance for deadlock. |
| 549 | * For byte-range locks we must check for deadlock. |
| 550 | * |
| 551 | * Deadlock detection is done by looking through the |
| 552 | * wait channels to see if there are any cycles that |
| 553 | * involve us. MAXDEPTH is set just to make sure we |
| 554 | * do not go off into neverneverland. |
| 555 | */ |
| 556 | if ((lock->lf_flags & F_POSIX) && |
| 557 | (block->lf_flags & F_POSIX)) { |
| 558 | struct lwp *wlwp; |
| 559 | volatile const struct lockf *waitblock; |
| 560 | int i = 0; |
| 561 | struct proc *p; |
| 562 | |
| 563 | p = (struct proc *)block->lf_id; |
| 564 | KASSERT(p != NULL); |
| 565 | while (i++ < maxlockdepth) { |
| 566 | mutex_enter(p->p_lock); |
| 567 | if (p->p_nlwps > 1) { |
| 568 | mutex_exit(p->p_lock); |
| 569 | break; |
| 570 | } |
| 571 | wlwp = LIST_FIRST(&p->p_lwps); |
| 572 | lwp_lock(wlwp); |
| 573 | if (wlwp->l_wchan == NULL || |
| 574 | wlwp->l_wmesg != lockstr) { |
| 575 | lwp_unlock(wlwp); |
| 576 | mutex_exit(p->p_lock); |
| 577 | break; |
| 578 | } |
| 579 | waitblock = wlwp->l_wchan; |
| 580 | lwp_unlock(wlwp); |
| 581 | mutex_exit(p->p_lock); |
| 582 | /* Get the owner of the blocking lock */ |
| 583 | waitblock = waitblock->lf_next; |
| 584 | if ((waitblock->lf_flags & F_POSIX) == 0) |
| 585 | break; |
| 586 | p = (struct proc *)waitblock->lf_id; |
| 587 | if (p == curproc) { |
| 588 | lf_free(lock); |
| 589 | return EDEADLK; |
| 590 | } |
| 591 | } |
| 592 | /* |
| 593 | * If we're still following a dependency chain |
| 594 | * after maxlockdepth iterations, assume we're in |
| 595 | * a cycle to be safe. |
| 596 | */ |
| 597 | if (i >= maxlockdepth) { |
| 598 | lf_free(lock); |
| 599 | return EDEADLK; |
| 600 | } |
| 601 | } |
| 602 | /* |
| 603 | * For flock type locks, we must first remove |
| 604 | * any shared locks that we hold before we sleep |
| 605 | * waiting for an exclusive lock. |
| 606 | */ |
| 607 | if ((lock->lf_flags & F_FLOCK) && |
| 608 | lock->lf_type == F_WRLCK) { |
| 609 | lock->lf_type = F_UNLCK; |
| 610 | (void) lf_clearlock(lock, NULL); |
| 611 | lock->lf_type = F_WRLCK; |
| 612 | } |
| 613 | /* |
| 614 | * Add our lock to the blocked list and sleep until we're free. |
| 615 | * Remember who blocked us (for deadlock detection). |
| 616 | */ |
| 617 | lock->lf_next = block; |
| 618 | TAILQ_INSERT_TAIL(&block->lf_blkhd, lock, lf_block); |
| 619 | #ifdef LOCKF_DEBUG |
| 620 | if (lockf_debug & 1) { |
| 621 | lf_print("lf_setlock: blocking on" , block); |
| 622 | lf_printlist("lf_setlock" , block); |
| 623 | } |
| 624 | #endif /* LOCKF_DEBUG */ |
| 625 | error = cv_wait_sig(&lock->lf_cv, interlock); |
| 626 | |
| 627 | /* |
| 628 | * We may have been awoken by a signal (in |
| 629 | * which case we must remove ourselves from the |
| 630 | * blocked list) and/or by another process |
| 631 | * releasing a lock (in which case we have already |
| 632 | * been removed from the blocked list and our |
| 633 | * lf_next field set to NULL). |
| 634 | */ |
| 635 | if (lock->lf_next != NULL) { |
| 636 | TAILQ_REMOVE(&lock->lf_next->lf_blkhd, lock, lf_block); |
| 637 | lock->lf_next = NULL; |
| 638 | } |
| 639 | if (error) { |
| 640 | lf_free(lock); |
| 641 | return error; |
| 642 | } |
| 643 | } |
| 644 | /* |
| 645 | * No blocks!! Add the lock. Note that we will |
| 646 | * downgrade or upgrade any overlapping locks this |
| 647 | * process already owns. |
| 648 | * |
| 649 | * Skip over locks owned by other processes. |
| 650 | * Handle any locks that overlap and are owned by ourselves. |
| 651 | */ |
| 652 | prev = head; |
| 653 | block = *head; |
| 654 | needtolink = 1; |
| 655 | for (;;) { |
| 656 | ovcase = lf_findoverlap(block, lock, SELF, &prev, &overlap); |
| 657 | if (ovcase) |
| 658 | block = overlap->lf_next; |
| 659 | /* |
| 660 | * Six cases: |
| 661 | * 0) no overlap |
| 662 | * 1) overlap == lock |
| 663 | * 2) overlap contains lock |
| 664 | * 3) lock contains overlap |
| 665 | * 4) overlap starts before lock |
| 666 | * 5) overlap ends after lock |
| 667 | */ |
| 668 | switch (ovcase) { |
| 669 | case 0: /* no overlap */ |
| 670 | if (needtolink) { |
| 671 | *prev = lock; |
| 672 | lock->lf_next = overlap; |
| 673 | } |
| 674 | break; |
| 675 | |
| 676 | case 1: /* overlap == lock */ |
| 677 | /* |
| 678 | * If downgrading lock, others may be |
| 679 | * able to acquire it. |
| 680 | */ |
| 681 | if (lock->lf_type == F_RDLCK && |
| 682 | overlap->lf_type == F_WRLCK) |
| 683 | lf_wakelock(overlap); |
| 684 | overlap->lf_type = lock->lf_type; |
| 685 | lf_free(lock); |
| 686 | lock = overlap; /* for debug output below */ |
| 687 | break; |
| 688 | |
| 689 | case 2: /* overlap contains lock */ |
| 690 | /* |
| 691 | * Check for common starting point and different types. |
| 692 | */ |
| 693 | if (overlap->lf_type == lock->lf_type) { |
| 694 | lf_free(lock); |
| 695 | lock = overlap; /* for debug output below */ |
| 696 | break; |
| 697 | } |
| 698 | if (overlap->lf_start == lock->lf_start) { |
| 699 | *prev = lock; |
| 700 | lock->lf_next = overlap; |
| 701 | overlap->lf_start = lock->lf_end + 1; |
| 702 | } else |
| 703 | lf_split(overlap, lock, sparelock); |
| 704 | lf_wakelock(overlap); |
| 705 | break; |
| 706 | |
| 707 | case 3: /* lock contains overlap */ |
| 708 | /* |
| 709 | * If downgrading lock, others may be able to |
| 710 | * acquire it, otherwise take the list. |
| 711 | */ |
| 712 | if (lock->lf_type == F_RDLCK && |
| 713 | overlap->lf_type == F_WRLCK) { |
| 714 | lf_wakelock(overlap); |
| 715 | } else { |
| 716 | while ((ltmp = TAILQ_FIRST(&overlap->lf_blkhd))) { |
| 717 | KASSERT(ltmp->lf_next == overlap); |
| 718 | TAILQ_REMOVE(&overlap->lf_blkhd, ltmp, |
| 719 | lf_block); |
| 720 | ltmp->lf_next = lock; |
| 721 | TAILQ_INSERT_TAIL(&lock->lf_blkhd, |
| 722 | ltmp, lf_block); |
| 723 | } |
| 724 | } |
| 725 | /* |
| 726 | * Add the new lock if necessary and delete the overlap. |
| 727 | */ |
| 728 | if (needtolink) { |
| 729 | *prev = lock; |
| 730 | lock->lf_next = overlap->lf_next; |
| 731 | prev = &lock->lf_next; |
| 732 | needtolink = 0; |
| 733 | } else |
| 734 | *prev = overlap->lf_next; |
| 735 | lf_free(overlap); |
| 736 | continue; |
| 737 | |
| 738 | case 4: /* overlap starts before lock */ |
| 739 | /* |
| 740 | * Add lock after overlap on the list. |
| 741 | */ |
| 742 | lock->lf_next = overlap->lf_next; |
| 743 | overlap->lf_next = lock; |
| 744 | overlap->lf_end = lock->lf_start - 1; |
| 745 | prev = &lock->lf_next; |
| 746 | lf_wakelock(overlap); |
| 747 | needtolink = 0; |
| 748 | continue; |
| 749 | |
| 750 | case 5: /* overlap ends after lock */ |
| 751 | /* |
| 752 | * Add the new lock before overlap. |
| 753 | */ |
| 754 | if (needtolink) { |
| 755 | *prev = lock; |
| 756 | lock->lf_next = overlap; |
| 757 | } |
| 758 | overlap->lf_start = lock->lf_end + 1; |
| 759 | lf_wakelock(overlap); |
| 760 | break; |
| 761 | } |
| 762 | break; |
| 763 | } |
| 764 | #ifdef LOCKF_DEBUG |
| 765 | if (lockf_debug & 1) { |
| 766 | lf_print("lf_setlock: got the lock" , lock); |
| 767 | lf_printlist("lf_setlock" , lock); |
| 768 | } |
| 769 | #endif /* LOCKF_DEBUG */ |
| 770 | return 0; |
| 771 | } |
| 772 | |
| 773 | /* |
| 774 | * Check whether there is a blocking lock, |
| 775 | * and if so return its process identifier. |
| 776 | */ |
| 777 | static int |
| 778 | lf_getlock(struct lockf *lock, struct flock *fl) |
| 779 | { |
| 780 | struct lockf *block; |
| 781 | |
| 782 | #ifdef LOCKF_DEBUG |
| 783 | if (lockf_debug & 1) |
| 784 | lf_print("lf_getlock" , lock); |
| 785 | #endif /* LOCKF_DEBUG */ |
| 786 | |
| 787 | if ((block = lf_getblock(lock)) != NULL) { |
| 788 | fl->l_type = block->lf_type; |
| 789 | fl->l_whence = SEEK_SET; |
| 790 | fl->l_start = block->lf_start; |
| 791 | if (block->lf_end == -1) |
| 792 | fl->l_len = 0; |
| 793 | else |
| 794 | fl->l_len = block->lf_end - block->lf_start + 1; |
| 795 | if (block->lf_flags & F_POSIX) |
| 796 | fl->l_pid = ((struct proc *)block->lf_id)->p_pid; |
| 797 | else |
| 798 | fl->l_pid = -1; |
| 799 | } else { |
| 800 | fl->l_type = F_UNLCK; |
| 801 | } |
| 802 | return 0; |
| 803 | } |
| 804 | |
| 805 | /* |
| 806 | * Do an advisory lock operation. |
| 807 | */ |
| 808 | int |
| 809 | lf_advlock(struct vop_advlock_args *ap, struct lockf **head, off_t size) |
| 810 | { |
| 811 | struct flock *fl = ap->a_fl; |
| 812 | struct lockf *lock = NULL; |
| 813 | struct lockf *sparelock; |
| 814 | kmutex_t *interlock = lockf_lock; |
| 815 | off_t start, end; |
| 816 | int error = 0; |
| 817 | |
| 818 | /* |
| 819 | * Convert the flock structure into a start and end. |
| 820 | */ |
| 821 | switch (fl->l_whence) { |
| 822 | case SEEK_SET: |
| 823 | case SEEK_CUR: |
| 824 | /* |
| 825 | * Caller is responsible for adding any necessary offset |
| 826 | * when SEEK_CUR is used. |
| 827 | */ |
| 828 | start = fl->l_start; |
| 829 | break; |
| 830 | |
| 831 | case SEEK_END: |
| 832 | start = size + fl->l_start; |
| 833 | break; |
| 834 | |
| 835 | default: |
| 836 | return EINVAL; |
| 837 | } |
| 838 | |
| 839 | if (fl->l_len == 0) |
| 840 | end = -1; |
| 841 | else { |
| 842 | if (fl->l_len > 0) |
| 843 | end = start + fl->l_len - 1; |
| 844 | else { |
| 845 | /* lockf() allows -ve lengths */ |
| 846 | end = start - 1; |
| 847 | start += fl->l_len; |
| 848 | } |
| 849 | } |
| 850 | if (start < 0) |
| 851 | return EINVAL; |
| 852 | |
| 853 | /* |
| 854 | * Allocate locks before acquiring the interlock. We need two |
| 855 | * locks in the worst case. |
| 856 | */ |
| 857 | switch (ap->a_op) { |
| 858 | case F_SETLK: |
| 859 | case F_UNLCK: |
| 860 | /* |
| 861 | * XXX For F_UNLCK case, we can re-use the lock. |
| 862 | */ |
| 863 | if ((ap->a_flags & F_FLOCK) == 0) { |
| 864 | /* |
| 865 | * Byte-range lock might need one more lock. |
| 866 | */ |
| 867 | sparelock = lf_alloc(0); |
| 868 | if (sparelock == NULL) { |
| 869 | error = ENOMEM; |
| 870 | goto quit; |
| 871 | } |
| 872 | break; |
| 873 | } |
| 874 | /* FALLTHROUGH */ |
| 875 | |
| 876 | case F_GETLK: |
| 877 | sparelock = NULL; |
| 878 | break; |
| 879 | |
| 880 | default: |
| 881 | return EINVAL; |
| 882 | } |
| 883 | |
| 884 | switch (ap->a_op) { |
| 885 | case F_SETLK: |
| 886 | lock = lf_alloc(1); |
| 887 | break; |
| 888 | case F_UNLCK: |
| 889 | if (start == 0 || end == -1) { |
| 890 | /* never split */ |
| 891 | lock = lf_alloc(0); |
| 892 | } else { |
| 893 | /* might split */ |
| 894 | lock = lf_alloc(2); |
| 895 | } |
| 896 | break; |
| 897 | case F_GETLK: |
| 898 | lock = lf_alloc(0); |
| 899 | break; |
| 900 | } |
| 901 | if (lock == NULL) { |
| 902 | error = ENOMEM; |
| 903 | goto quit; |
| 904 | } |
| 905 | |
| 906 | mutex_enter(interlock); |
| 907 | |
| 908 | /* |
| 909 | * Avoid the common case of unlocking when inode has no locks. |
| 910 | */ |
| 911 | if (*head == (struct lockf *)0) { |
| 912 | if (ap->a_op != F_SETLK) { |
| 913 | fl->l_type = F_UNLCK; |
| 914 | error = 0; |
| 915 | goto quit_unlock; |
| 916 | } |
| 917 | } |
| 918 | |
| 919 | /* |
| 920 | * Create the lockf structure. |
| 921 | */ |
| 922 | lock->lf_start = start; |
| 923 | lock->lf_end = end; |
| 924 | lock->lf_head = head; |
| 925 | lock->lf_type = fl->l_type; |
| 926 | lock->lf_next = (struct lockf *)0; |
| 927 | TAILQ_INIT(&lock->lf_blkhd); |
| 928 | lock->lf_flags = ap->a_flags; |
| 929 | if (lock->lf_flags & F_POSIX) { |
| 930 | KASSERT(curproc == (struct proc *)ap->a_id); |
| 931 | } |
| 932 | lock->lf_id = ap->a_id; |
| 933 | |
| 934 | /* |
| 935 | * Do the requested operation. |
| 936 | */ |
| 937 | switch (ap->a_op) { |
| 938 | |
| 939 | case F_SETLK: |
| 940 | error = lf_setlock(lock, &sparelock, interlock); |
| 941 | lock = NULL; /* lf_setlock freed it */ |
| 942 | break; |
| 943 | |
| 944 | case F_UNLCK: |
| 945 | error = lf_clearlock(lock, &sparelock); |
| 946 | break; |
| 947 | |
| 948 | case F_GETLK: |
| 949 | error = lf_getlock(lock, fl); |
| 950 | break; |
| 951 | |
| 952 | default: |
| 953 | break; |
| 954 | /* NOTREACHED */ |
| 955 | } |
| 956 | |
| 957 | quit_unlock: |
| 958 | mutex_exit(interlock); |
| 959 | quit: |
| 960 | if (lock) |
| 961 | lf_free(lock); |
| 962 | if (sparelock) |
| 963 | lf_free(sparelock); |
| 964 | |
| 965 | return error; |
| 966 | } |
| 967 | |
| 968 | /* |
| 969 | * Initialize subsystem. XXX We use a global lock. This could be the |
| 970 | * vnode interlock, but the deadlock detection code may need to inspect |
| 971 | * locks belonging to other files. |
| 972 | */ |
| 973 | void |
| 974 | lf_init(void) |
| 975 | { |
| 976 | |
| 977 | lockf_cache = pool_cache_init(sizeof(struct lockf), 0, 0, 0, "lockf" , |
| 978 | NULL, IPL_NONE, lf_ctor, lf_dtor, NULL); |
| 979 | lockf_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); |
| 980 | } |
| 981 | |