1 /* $NetBSD: sysv_sem.c,v 1.87 2011/05/13 22:16:44 rmind Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2007 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center, and by Andrew Doran. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Implementation of SVID semaphores 35 * 36 * Author: Daniel Boulet 37 * 38 * This software is provided ``AS IS'' without any warranties of any kind. 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: sysv_sem.c,v 1.87 2011/05/13 22:16:44 rmind Exp $"); 43 44 #define SYSVSEM 45 46 #include <sys/param.h> 47 #include <sys/kernel.h> 48 #include <sys/sem.h> 49 #include <sys/sysctl.h> 50 #include <sys/kmem.h> 51 #include <sys/mount.h> /* XXX for <sys/syscallargs.h> */ 52 #include <sys/syscallargs.h> 53 #include <sys/kauth.h> 54 55 /* 56 * Memory areas: 57 * 1st: Pool of semaphore identifiers 58 * 2nd: Semaphores 59 * 3rd: Conditional variables 60 * 4th: Undo structures 61 */ 62 struct semid_ds * sema __read_mostly; 63 static struct __sem * sem __read_mostly; 64 static kcondvar_t * semcv __read_mostly; 65 static int * semu __read_mostly; 66 67 static kmutex_t semlock __cacheline_aligned; 68 static bool sem_realloc_state __read_mostly; 69 static kcondvar_t sem_realloc_cv; 70 71 /* 72 * List of active undo structures, total number of semaphores, 73 * and total number of semop waiters. 74 */ 75 static struct sem_undo *semu_list __read_mostly; 76 static u_int semtot __cacheline_aligned; 77 static u_int sem_waiters __cacheline_aligned; 78 79 /* Macro to find a particular sem_undo vector */ 80 #define SEMU(s, ix) ((struct sem_undo *)(((long)s) + ix * seminfo.semusz)) 81 82 #ifdef SEM_DEBUG 83 #define SEM_PRINTF(a) printf a 84 #else 85 #define SEM_PRINTF(a) 86 #endif 87 88 struct sem_undo *semu_alloc(struct proc *); 89 int semundo_adjust(struct proc *, struct sem_undo **, int, int, int); 90 void semundo_clear(int, int); 91 92 void 93 seminit(void) 94 { 95 int i, sz; 96 vaddr_t v; 97 98 mutex_init(&semlock, MUTEX_DEFAULT, IPL_NONE); 99 cv_init(&sem_realloc_cv, "semrealc"); 100 sem_realloc_state = false; 101 semtot = 0; 102 sem_waiters = 0; 103 104 /* Allocate the wired memory for our structures */ 105 sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) + 106 ALIGN(seminfo.semmns * sizeof(struct __sem)) + 107 ALIGN(seminfo.semmni * sizeof(kcondvar_t)) + 108 ALIGN(seminfo.semmnu * seminfo.semusz); 109 v = uvm_km_alloc(kernel_map, round_page(sz), 0, 110 UVM_KMF_WIRED|UVM_KMF_ZERO); 111 if (v == 0) 112 panic("sysv_sem: cannot allocate memory"); 113 sema = (void *)v; 114 sem = (void *)((uintptr_t)sema + 115 ALIGN(seminfo.semmni * sizeof(struct semid_ds))); 116 semcv = (void *)((uintptr_t)sem + 117 ALIGN(seminfo.semmns * sizeof(struct __sem))); 118 semu = (void *)((uintptr_t)semcv + 119 ALIGN(seminfo.semmni * sizeof(kcondvar_t))); 120 121 for (i = 0; i < seminfo.semmni; i++) { 122 sema[i]._sem_base = 0; 123 sema[i].sem_perm.mode = 0; 124 cv_init(&semcv[i], "semwait"); 125 } 126 for (i = 0; i < seminfo.semmnu; i++) { 127 struct sem_undo *suptr = SEMU(semu, i); 128 suptr->un_proc = NULL; 129 } 130 semu_list = NULL; 131 exithook_establish(semexit, NULL); 132 } 133 134 static int 135 semrealloc(int newsemmni, int newsemmns, int newsemmnu) 136 { 137 struct semid_ds *new_sema, *old_sema; 138 struct __sem *new_sem; 139 struct sem_undo *new_semu_list, *suptr, *nsuptr; 140 int *new_semu; 141 kcondvar_t *new_semcv; 142 vaddr_t v; 143 int i, j, lsemid, nmnus, sz; 144 145 if (newsemmni < 1 || newsemmns < 1 || newsemmnu < 1) 146 return EINVAL; 147 148 /* Allocate the wired memory for our structures */ 149 sz = ALIGN(newsemmni * sizeof(struct semid_ds)) + 150 ALIGN(newsemmns * sizeof(struct __sem)) + 151 ALIGN(newsemmni * sizeof(kcondvar_t)) + 152 ALIGN(newsemmnu * seminfo.semusz); 153 v = uvm_km_alloc(kernel_map, round_page(sz), 0, 154 UVM_KMF_WIRED|UVM_KMF_ZERO); 155 if (v == 0) 156 return ENOMEM; 157 158 mutex_enter(&semlock); 159 if (sem_realloc_state) { 160 mutex_exit(&semlock); 161 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 162 return EBUSY; 163 } 164 sem_realloc_state = true; 165 if (sem_waiters) { 166 /* 167 * Mark reallocation state, wake-up all waiters, 168 * and wait while they will all exit. 169 */ 170 for (i = 0; i < seminfo.semmni; i++) 171 cv_broadcast(&semcv[i]); 172 while (sem_waiters) 173 cv_wait(&sem_realloc_cv, &semlock); 174 } 175 old_sema = sema; 176 177 /* Get the number of last slot */ 178 lsemid = 0; 179 for (i = 0; i < seminfo.semmni; i++) 180 if (sema[i].sem_perm.mode & SEM_ALLOC) 181 lsemid = i; 182 183 /* Get the number of currently used undo structures */ 184 nmnus = 0; 185 for (i = 0; i < seminfo.semmnu; i++) { 186 suptr = SEMU(semu, i); 187 if (suptr->un_proc == NULL) 188 continue; 189 nmnus++; 190 } 191 192 /* We cannot reallocate less memory than we use */ 193 if (lsemid >= newsemmni || semtot > newsemmns || nmnus > newsemmnu) { 194 mutex_exit(&semlock); 195 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 196 return EBUSY; 197 } 198 199 new_sema = (void *)v; 200 new_sem = (void *)((uintptr_t)new_sema + 201 ALIGN(newsemmni * sizeof(struct semid_ds))); 202 new_semcv = (void *)((uintptr_t)new_sem + 203 ALIGN(newsemmns * sizeof(struct __sem))); 204 new_semu = (void *)((uintptr_t)new_semcv + 205 ALIGN(newsemmni * sizeof(kcondvar_t))); 206 207 /* Initialize all semaphore identifiers and condvars */ 208 for (i = 0; i < newsemmni; i++) { 209 new_sema[i]._sem_base = 0; 210 new_sema[i].sem_perm.mode = 0; 211 cv_init(&new_semcv[i], "semwait"); 212 } 213 for (i = 0; i < newsemmnu; i++) { 214 nsuptr = SEMU(new_semu, i); 215 nsuptr->un_proc = NULL; 216 } 217 218 /* 219 * Copy all identifiers, semaphores and list of the 220 * undo structures to the new memory allocation. 221 */ 222 j = 0; 223 for (i = 0; i <= lsemid; i++) { 224 if ((sema[i].sem_perm.mode & SEM_ALLOC) == 0) 225 continue; 226 memcpy(&new_sema[i], &sema[i], sizeof(struct semid_ds)); 227 new_sema[i]._sem_base = &new_sem[j]; 228 memcpy(new_sema[i]._sem_base, sema[i]._sem_base, 229 (sizeof(struct __sem) * sema[i].sem_nsems)); 230 j += sema[i].sem_nsems; 231 } 232 KASSERT(j == semtot); 233 234 j = 0; 235 new_semu_list = NULL; 236 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) { 237 KASSERT(j < newsemmnu); 238 nsuptr = SEMU(new_semu, j); 239 memcpy(nsuptr, suptr, SEMUSZ); 240 nsuptr->un_next = new_semu_list; 241 new_semu_list = nsuptr; 242 j++; 243 } 244 245 for (i = 0; i < seminfo.semmni; i++) { 246 KASSERT(cv_has_waiters(&semcv[i]) == false); 247 cv_destroy(&semcv[i]); 248 } 249 250 sz = ALIGN(seminfo.semmni * sizeof(struct semid_ds)) + 251 ALIGN(seminfo.semmns * sizeof(struct __sem)) + 252 ALIGN(seminfo.semmni * sizeof(kcondvar_t)) + 253 ALIGN(seminfo.semmnu * seminfo.semusz); 254 255 /* Set the pointers and update the new values */ 256 sema = new_sema; 257 sem = new_sem; 258 semcv = new_semcv; 259 semu = new_semu; 260 semu_list = new_semu_list; 261 262 seminfo.semmni = newsemmni; 263 seminfo.semmns = newsemmns; 264 seminfo.semmnu = newsemmnu; 265 266 /* Reallocation completed - notify all waiters, if any */ 267 sem_realloc_state = false; 268 cv_broadcast(&sem_realloc_cv); 269 mutex_exit(&semlock); 270 271 uvm_km_free(kernel_map, (vaddr_t)old_sema, sz, UVM_KMF_WIRED); 272 return 0; 273 } 274 275 /* 276 * Placebo. 277 */ 278 279 int 280 sys_semconfig(struct lwp *l, const struct sys_semconfig_args *uap, register_t *retval) 281 { 282 283 *retval = 0; 284 return 0; 285 } 286 287 /* 288 * Allocate a new sem_undo structure for a process. 289 * => Returns NULL on failure. 290 */ 291 struct sem_undo * 292 semu_alloc(struct proc *p) 293 { 294 struct sem_undo *suptr, **supptr; 295 bool attempted = false; 296 int i; 297 298 KASSERT(mutex_owned(&semlock)); 299 again: 300 /* Look for a free structure. */ 301 for (i = 0; i < seminfo.semmnu; i++) { 302 suptr = SEMU(semu, i); 303 if (suptr->un_proc == NULL) { 304 /* Found. Fill it in and return. */ 305 suptr->un_next = semu_list; 306 semu_list = suptr; 307 suptr->un_cnt = 0; 308 suptr->un_proc = p; 309 return suptr; 310 } 311 } 312 313 /* Not found. Attempt to free some structures. */ 314 if (!attempted) { 315 bool freed = false; 316 317 attempted = true; 318 supptr = &semu_list; 319 while ((suptr = *supptr) != NULL) { 320 if (suptr->un_cnt == 0) { 321 suptr->un_proc = NULL; 322 *supptr = suptr->un_next; 323 freed = true; 324 } else { 325 supptr = &suptr->un_next; 326 } 327 } 328 if (freed) { 329 goto again; 330 } 331 } 332 return NULL; 333 } 334 335 /* 336 * Adjust a particular entry for a particular proc 337 */ 338 339 int 340 semundo_adjust(struct proc *p, struct sem_undo **supptr, int semid, int semnum, 341 int adjval) 342 { 343 struct sem_undo *suptr; 344 struct undo *sunptr; 345 int i; 346 347 KASSERT(mutex_owned(&semlock)); 348 349 /* 350 * Look for and remember the sem_undo if the caller doesn't 351 * provide it 352 */ 353 354 suptr = *supptr; 355 if (suptr == NULL) { 356 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) 357 if (suptr->un_proc == p) 358 break; 359 360 if (suptr == NULL) { 361 suptr = semu_alloc(p); 362 if (suptr == NULL) 363 return (ENOSPC); 364 } 365 *supptr = suptr; 366 } 367 368 /* 369 * Look for the requested entry and adjust it (delete if 370 * adjval becomes 0). 371 */ 372 sunptr = &suptr->un_ent[0]; 373 for (i = 0; i < suptr->un_cnt; i++, sunptr++) { 374 if (sunptr->un_id != semid || sunptr->un_num != semnum) 375 continue; 376 sunptr->un_adjval += adjval; 377 if (sunptr->un_adjval == 0) { 378 suptr->un_cnt--; 379 if (i < suptr->un_cnt) 380 suptr->un_ent[i] = 381 suptr->un_ent[suptr->un_cnt]; 382 } 383 return (0); 384 } 385 386 /* Didn't find the right entry - create it */ 387 if (suptr->un_cnt == SEMUME) 388 return (EINVAL); 389 390 sunptr = &suptr->un_ent[suptr->un_cnt]; 391 suptr->un_cnt++; 392 sunptr->un_adjval = adjval; 393 sunptr->un_id = semid; 394 sunptr->un_num = semnum; 395 return (0); 396 } 397 398 void 399 semundo_clear(int semid, int semnum) 400 { 401 struct sem_undo *suptr; 402 struct undo *sunptr, *sunend; 403 404 KASSERT(mutex_owned(&semlock)); 405 406 for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) 407 for (sunptr = &suptr->un_ent[0], 408 sunend = sunptr + suptr->un_cnt; sunptr < sunend;) { 409 if (sunptr->un_id == semid) { 410 if (semnum == -1 || sunptr->un_num == semnum) { 411 suptr->un_cnt--; 412 sunend--; 413 if (sunptr != sunend) 414 *sunptr = *sunend; 415 if (semnum != -1) 416 break; 417 else 418 continue; 419 } 420 } 421 sunptr++; 422 } 423 } 424 425 int 426 sys_____semctl50(struct lwp *l, const struct sys_____semctl50_args *uap, 427 register_t *retval) 428 { 429 /* { 430 syscallarg(int) semid; 431 syscallarg(int) semnum; 432 syscallarg(int) cmd; 433 syscallarg(union __semun *) arg; 434 } */ 435 struct semid_ds sembuf; 436 int cmd, error; 437 void *pass_arg; 438 union __semun karg; 439 440 cmd = SCARG(uap, cmd); 441 442 pass_arg = get_semctl_arg(cmd, &sembuf, &karg); 443 444 if (pass_arg) { 445 error = copyin(SCARG(uap, arg), &karg, sizeof(karg)); 446 if (error) 447 return error; 448 if (cmd == IPC_SET) { 449 error = copyin(karg.buf, &sembuf, sizeof(sembuf)); 450 if (error) 451 return (error); 452 } 453 } 454 455 error = semctl1(l, SCARG(uap, semid), SCARG(uap, semnum), cmd, 456 pass_arg, retval); 457 458 if (error == 0 && cmd == IPC_STAT) 459 error = copyout(&sembuf, karg.buf, sizeof(sembuf)); 460 461 return (error); 462 } 463 464 int 465 semctl1(struct lwp *l, int semid, int semnum, int cmd, void *v, 466 register_t *retval) 467 { 468 kauth_cred_t cred = l->l_cred; 469 union __semun *arg = v; 470 struct semid_ds *sembuf = v, *semaptr; 471 int i, error, ix; 472 473 SEM_PRINTF(("call to semctl(%d, %d, %d, %p)\n", 474 semid, semnum, cmd, v)); 475 476 mutex_enter(&semlock); 477 478 ix = IPCID_TO_IX(semid); 479 if (ix < 0 || ix >= seminfo.semmni) { 480 mutex_exit(&semlock); 481 return (EINVAL); 482 } 483 484 semaptr = &sema[ix]; 485 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 486 semaptr->sem_perm._seq != IPCID_TO_SEQ(semid)) { 487 mutex_exit(&semlock); 488 return (EINVAL); 489 } 490 491 switch (cmd) { 492 case IPC_RMID: 493 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M)) != 0) 494 break; 495 semaptr->sem_perm.cuid = kauth_cred_geteuid(cred); 496 semaptr->sem_perm.uid = kauth_cred_geteuid(cred); 497 semtot -= semaptr->sem_nsems; 498 for (i = semaptr->_sem_base - sem; i < semtot; i++) 499 sem[i] = sem[i + semaptr->sem_nsems]; 500 for (i = 0; i < seminfo.semmni; i++) { 501 if ((sema[i].sem_perm.mode & SEM_ALLOC) && 502 sema[i]._sem_base > semaptr->_sem_base) 503 sema[i]._sem_base -= semaptr->sem_nsems; 504 } 505 semaptr->sem_perm.mode = 0; 506 semundo_clear(ix, -1); 507 cv_broadcast(&semcv[ix]); 508 break; 509 510 case IPC_SET: 511 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_M))) 512 break; 513 KASSERT(sembuf != NULL); 514 semaptr->sem_perm.uid = sembuf->sem_perm.uid; 515 semaptr->sem_perm.gid = sembuf->sem_perm.gid; 516 semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) | 517 (sembuf->sem_perm.mode & 0777); 518 semaptr->sem_ctime = time_second; 519 break; 520 521 case IPC_STAT: 522 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 523 break; 524 KASSERT(sembuf != NULL); 525 memcpy(sembuf, semaptr, sizeof(struct semid_ds)); 526 sembuf->sem_perm.mode &= 0777; 527 break; 528 529 case GETNCNT: 530 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 531 break; 532 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 533 error = EINVAL; 534 break; 535 } 536 *retval = semaptr->_sem_base[semnum].semncnt; 537 break; 538 539 case GETPID: 540 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 541 break; 542 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 543 error = EINVAL; 544 break; 545 } 546 *retval = semaptr->_sem_base[semnum].sempid; 547 break; 548 549 case GETVAL: 550 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 551 break; 552 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 553 error = EINVAL; 554 break; 555 } 556 *retval = semaptr->_sem_base[semnum].semval; 557 break; 558 559 case GETALL: 560 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 561 break; 562 KASSERT(arg != NULL); 563 for (i = 0; i < semaptr->sem_nsems; i++) { 564 error = copyout(&semaptr->_sem_base[i].semval, 565 &arg->array[i], sizeof(arg->array[i])); 566 if (error != 0) 567 break; 568 } 569 break; 570 571 case GETZCNT: 572 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_R))) 573 break; 574 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 575 error = EINVAL; 576 break; 577 } 578 *retval = semaptr->_sem_base[semnum].semzcnt; 579 break; 580 581 case SETVAL: 582 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) 583 break; 584 if (semnum < 0 || semnum >= semaptr->sem_nsems) { 585 error = EINVAL; 586 break; 587 } 588 KASSERT(arg != NULL); 589 if ((unsigned int)arg->val > seminfo.semvmx) { 590 error = ERANGE; 591 break; 592 } 593 semaptr->_sem_base[semnum].semval = arg->val; 594 semundo_clear(ix, semnum); 595 cv_broadcast(&semcv[ix]); 596 break; 597 598 case SETALL: 599 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) 600 break; 601 KASSERT(arg != NULL); 602 for (i = 0; i < semaptr->sem_nsems; i++) { 603 unsigned short semval; 604 error = copyin(&arg->array[i], &semval, 605 sizeof(arg->array[i])); 606 if (error != 0) 607 break; 608 if ((unsigned int)semval > seminfo.semvmx) { 609 error = ERANGE; 610 break; 611 } 612 semaptr->_sem_base[i].semval = semval; 613 } 614 semundo_clear(ix, -1); 615 cv_broadcast(&semcv[ix]); 616 break; 617 618 default: 619 error = EINVAL; 620 break; 621 } 622 623 mutex_exit(&semlock); 624 return (error); 625 } 626 627 int 628 sys_semget(struct lwp *l, const struct sys_semget_args *uap, register_t *retval) 629 { 630 /* { 631 syscallarg(key_t) key; 632 syscallarg(int) nsems; 633 syscallarg(int) semflg; 634 } */ 635 int semid, error = 0; 636 int key = SCARG(uap, key); 637 int nsems = SCARG(uap, nsems); 638 int semflg = SCARG(uap, semflg); 639 kauth_cred_t cred = l->l_cred; 640 641 SEM_PRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg)); 642 643 mutex_enter(&semlock); 644 645 if (key != IPC_PRIVATE) { 646 for (semid = 0; semid < seminfo.semmni; semid++) { 647 if ((sema[semid].sem_perm.mode & SEM_ALLOC) && 648 sema[semid].sem_perm._key == key) 649 break; 650 } 651 if (semid < seminfo.semmni) { 652 SEM_PRINTF(("found public key\n")); 653 if ((error = ipcperm(cred, &sema[semid].sem_perm, 654 semflg & 0700))) 655 goto out; 656 if (nsems > 0 && sema[semid].sem_nsems < nsems) { 657 SEM_PRINTF(("too small\n")); 658 error = EINVAL; 659 goto out; 660 } 661 if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) { 662 SEM_PRINTF(("not exclusive\n")); 663 error = EEXIST; 664 goto out; 665 } 666 goto found; 667 } 668 } 669 670 SEM_PRINTF(("need to allocate the semid_ds\n")); 671 if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) { 672 if (nsems <= 0 || nsems > seminfo.semmsl) { 673 SEM_PRINTF(("nsems out of range (0<%d<=%d)\n", nsems, 674 seminfo.semmsl)); 675 error = EINVAL; 676 goto out; 677 } 678 if (nsems > seminfo.semmns - semtot) { 679 SEM_PRINTF(("not enough semaphores left " 680 "(need %d, got %d)\n", 681 nsems, seminfo.semmns - semtot)); 682 error = ENOSPC; 683 goto out; 684 } 685 for (semid = 0; semid < seminfo.semmni; semid++) { 686 if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0) 687 break; 688 } 689 if (semid == seminfo.semmni) { 690 SEM_PRINTF(("no more semid_ds's available\n")); 691 error = ENOSPC; 692 goto out; 693 } 694 SEM_PRINTF(("semid %d is available\n", semid)); 695 sema[semid].sem_perm._key = key; 696 sema[semid].sem_perm.cuid = kauth_cred_geteuid(cred); 697 sema[semid].sem_perm.uid = kauth_cred_geteuid(cred); 698 sema[semid].sem_perm.cgid = kauth_cred_getegid(cred); 699 sema[semid].sem_perm.gid = kauth_cred_getegid(cred); 700 sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC; 701 sema[semid].sem_perm._seq = 702 (sema[semid].sem_perm._seq + 1) & 0x7fff; 703 sema[semid].sem_nsems = nsems; 704 sema[semid].sem_otime = 0; 705 sema[semid].sem_ctime = time_second; 706 sema[semid]._sem_base = &sem[semtot]; 707 semtot += nsems; 708 memset(sema[semid]._sem_base, 0, 709 sizeof(sema[semid]._sem_base[0]) * nsems); 710 SEM_PRINTF(("sembase = %p, next = %p\n", sema[semid]._sem_base, 711 &sem[semtot])); 712 } else { 713 SEM_PRINTF(("didn't find it and wasn't asked to create it\n")); 714 error = ENOENT; 715 goto out; 716 } 717 718 found: 719 *retval = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm); 720 out: 721 mutex_exit(&semlock); 722 return (error); 723 } 724 725 #define SMALL_SOPS 8 726 727 int 728 sys_semop(struct lwp *l, const struct sys_semop_args *uap, register_t *retval) 729 { 730 /* { 731 syscallarg(int) semid; 732 syscallarg(struct sembuf *) sops; 733 syscallarg(size_t) nsops; 734 } */ 735 struct proc *p = l->l_proc; 736 int semid = SCARG(uap, semid), seq; 737 size_t nsops = SCARG(uap, nsops); 738 struct sembuf small_sops[SMALL_SOPS]; 739 struct sembuf *sops; 740 struct semid_ds *semaptr; 741 struct sembuf *sopptr = NULL; 742 struct __sem *semptr = NULL; 743 struct sem_undo *suptr = NULL; 744 kauth_cred_t cred = l->l_cred; 745 int i, error; 746 int do_wakeup, do_undos; 747 748 SEM_PRINTF(("call to semop(%d, %p, %zd)\n", semid, SCARG(uap,sops), nsops)); 749 750 if (__predict_false((p->p_flag & PK_SYSVSEM) == 0)) { 751 mutex_enter(p->p_lock); 752 p->p_flag |= PK_SYSVSEM; 753 mutex_exit(p->p_lock); 754 } 755 756 restart: 757 if (nsops <= SMALL_SOPS) { 758 sops = small_sops; 759 } else if (nsops <= seminfo.semopm) { 760 sops = kmem_alloc(nsops * sizeof(*sops), KM_SLEEP); 761 } else { 762 SEM_PRINTF(("too many sops (max=%d, nsops=%zd)\n", 763 seminfo.semopm, nsops)); 764 return (E2BIG); 765 } 766 767 error = copyin(SCARG(uap, sops), sops, nsops * sizeof(sops[0])); 768 if (error) { 769 SEM_PRINTF(("error = %d from copyin(%p, %p, %zd)\n", error, 770 SCARG(uap, sops), &sops, nsops * sizeof(sops[0]))); 771 if (sops != small_sops) 772 kmem_free(sops, nsops * sizeof(*sops)); 773 return error; 774 } 775 776 mutex_enter(&semlock); 777 /* In case of reallocation, we will wait for completion */ 778 while (__predict_false(sem_realloc_state)) 779 cv_wait(&sem_realloc_cv, &semlock); 780 781 semid = IPCID_TO_IX(semid); /* Convert back to zero origin */ 782 if (semid < 0 || semid >= seminfo.semmni) { 783 error = EINVAL; 784 goto out; 785 } 786 787 semaptr = &sema[semid]; 788 seq = IPCID_TO_SEQ(SCARG(uap, semid)); 789 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 790 semaptr->sem_perm._seq != seq) { 791 error = EINVAL; 792 goto out; 793 } 794 795 if ((error = ipcperm(cred, &semaptr->sem_perm, IPC_W))) { 796 SEM_PRINTF(("error = %d from ipaccess\n", error)); 797 goto out; 798 } 799 800 for (i = 0; i < nsops; i++) 801 if (sops[i].sem_num >= semaptr->sem_nsems) { 802 error = EFBIG; 803 goto out; 804 } 805 806 /* 807 * Loop trying to satisfy the vector of requests. 808 * If we reach a point where we must wait, any requests already 809 * performed are rolled back and we go to sleep until some other 810 * process wakes us up. At this point, we start all over again. 811 * 812 * This ensures that from the perspective of other tasks, a set 813 * of requests is atomic (never partially satisfied). 814 */ 815 do_undos = 0; 816 817 for (;;) { 818 do_wakeup = 0; 819 820 for (i = 0; i < nsops; i++) { 821 sopptr = &sops[i]; 822 semptr = &semaptr->_sem_base[sopptr->sem_num]; 823 824 SEM_PRINTF(("semop: semaptr=%p, sem_base=%p, " 825 "semptr=%p, sem[%d]=%d : op=%d, flag=%s\n", 826 semaptr, semaptr->_sem_base, semptr, 827 sopptr->sem_num, semptr->semval, sopptr->sem_op, 828 (sopptr->sem_flg & IPC_NOWAIT) ? 829 "nowait" : "wait")); 830 831 if (sopptr->sem_op < 0) { 832 if ((int)(semptr->semval + 833 sopptr->sem_op) < 0) { 834 SEM_PRINTF(("semop: " 835 "can't do it now\n")); 836 break; 837 } else { 838 semptr->semval += sopptr->sem_op; 839 if (semptr->semval == 0 && 840 semptr->semzcnt > 0) 841 do_wakeup = 1; 842 } 843 if (sopptr->sem_flg & SEM_UNDO) 844 do_undos = 1; 845 } else if (sopptr->sem_op == 0) { 846 if (semptr->semval > 0) { 847 SEM_PRINTF(("semop: not zero now\n")); 848 break; 849 } 850 } else { 851 if (semptr->semncnt > 0) 852 do_wakeup = 1; 853 semptr->semval += sopptr->sem_op; 854 if (sopptr->sem_flg & SEM_UNDO) 855 do_undos = 1; 856 } 857 } 858 859 /* 860 * Did we get through the entire vector? 861 */ 862 if (i >= nsops) 863 goto done; 864 865 /* 866 * No ... rollback anything that we've already done 867 */ 868 SEM_PRINTF(("semop: rollback 0 through %d\n", i - 1)); 869 while (i-- > 0) 870 semaptr->_sem_base[sops[i].sem_num].semval -= 871 sops[i].sem_op; 872 873 /* 874 * If the request that we couldn't satisfy has the 875 * NOWAIT flag set then return with EAGAIN. 876 */ 877 if (sopptr->sem_flg & IPC_NOWAIT) { 878 error = EAGAIN; 879 goto out; 880 } 881 882 if (sopptr->sem_op == 0) 883 semptr->semzcnt++; 884 else 885 semptr->semncnt++; 886 887 sem_waiters++; 888 SEM_PRINTF(("semop: good night!\n")); 889 error = cv_wait_sig(&semcv[semid], &semlock); 890 SEM_PRINTF(("semop: good morning (error=%d)!\n", error)); 891 sem_waiters--; 892 893 /* Notify reallocator, if it is waiting */ 894 cv_broadcast(&sem_realloc_cv); 895 896 /* 897 * Make sure that the semaphore still exists 898 */ 899 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || 900 semaptr->sem_perm._seq != seq) { 901 error = EIDRM; 902 goto out; 903 } 904 905 /* 906 * The semaphore is still alive. Readjust the count of 907 * waiting processes. 908 */ 909 semptr = &semaptr->_sem_base[sopptr->sem_num]; 910 if (sopptr->sem_op == 0) 911 semptr->semzcnt--; 912 else 913 semptr->semncnt--; 914 915 /* In case of such state, restart the call */ 916 if (sem_realloc_state) { 917 mutex_exit(&semlock); 918 goto restart; 919 } 920 921 /* Is it really morning, or was our sleep interrupted? */ 922 if (error != 0) { 923 error = EINTR; 924 goto out; 925 } 926 SEM_PRINTF(("semop: good morning!\n")); 927 } 928 929 done: 930 /* 931 * Process any SEM_UNDO requests. 932 */ 933 if (do_undos) { 934 for (i = 0; i < nsops; i++) { 935 /* 936 * We only need to deal with SEM_UNDO's for non-zero 937 * op's. 938 */ 939 int adjval; 940 941 if ((sops[i].sem_flg & SEM_UNDO) == 0) 942 continue; 943 adjval = sops[i].sem_op; 944 if (adjval == 0) 945 continue; 946 error = semundo_adjust(p, &suptr, semid, 947 sops[i].sem_num, -adjval); 948 if (error == 0) 949 continue; 950 951 /* 952 * Oh-Oh! We ran out of either sem_undo's or undo's. 953 * Rollback the adjustments to this point and then 954 * rollback the semaphore ups and down so we can return 955 * with an error with all structures restored. We 956 * rollback the undo's in the exact reverse order that 957 * we applied them. This guarantees that we won't run 958 * out of space as we roll things back out. 959 */ 960 while (i-- > 0) { 961 if ((sops[i].sem_flg & SEM_UNDO) == 0) 962 continue; 963 adjval = sops[i].sem_op; 964 if (adjval == 0) 965 continue; 966 if (semundo_adjust(p, &suptr, semid, 967 sops[i].sem_num, adjval) != 0) 968 panic("semop - can't undo undos"); 969 } 970 971 for (i = 0; i < nsops; i++) 972 semaptr->_sem_base[sops[i].sem_num].semval -= 973 sops[i].sem_op; 974 975 SEM_PRINTF(("error = %d from semundo_adjust\n", error)); 976 goto out; 977 } /* loop through the sops */ 978 } /* if (do_undos) */ 979 980 /* We're definitely done - set the sempid's */ 981 for (i = 0; i < nsops; i++) { 982 sopptr = &sops[i]; 983 semptr = &semaptr->_sem_base[sopptr->sem_num]; 984 semptr->sempid = p->p_pid; 985 } 986 987 /* Update sem_otime */ 988 semaptr->sem_otime = time_second; 989 990 /* Do a wakeup if any semaphore was up'd. */ 991 if (do_wakeup) { 992 SEM_PRINTF(("semop: doing wakeup\n")); 993 cv_broadcast(&semcv[semid]); 994 SEM_PRINTF(("semop: back from wakeup\n")); 995 } 996 SEM_PRINTF(("semop: done\n")); 997 *retval = 0; 998 999 out: 1000 mutex_exit(&semlock); 1001 if (sops != small_sops) 1002 kmem_free(sops, nsops * sizeof(*sops)); 1003 return error; 1004 } 1005 1006 /* 1007 * Go through the undo structures for this process and apply the 1008 * adjustments to semaphores. 1009 */ 1010 /*ARGSUSED*/ 1011 void 1012 semexit(struct proc *p, void *v) 1013 { 1014 struct sem_undo *suptr; 1015 struct sem_undo **supptr; 1016 1017 if ((p->p_flag & PK_SYSVSEM) == 0) 1018 return; 1019 1020 mutex_enter(&semlock); 1021 1022 /* 1023 * Go through the chain of undo vectors looking for one 1024 * associated with this process. 1025 */ 1026 1027 for (supptr = &semu_list; (suptr = *supptr) != NULL; 1028 supptr = &suptr->un_next) { 1029 if (suptr->un_proc == p) 1030 break; 1031 } 1032 1033 /* 1034 * If there is no undo vector, skip to the end. 1035 */ 1036 1037 if (suptr == NULL) { 1038 mutex_exit(&semlock); 1039 return; 1040 } 1041 1042 /* 1043 * We now have an undo vector for this process. 1044 */ 1045 1046 SEM_PRINTF(("proc @%p has undo structure with %d entries\n", p, 1047 suptr->un_cnt)); 1048 1049 /* 1050 * If there are any active undo elements then process them. 1051 */ 1052 if (suptr->un_cnt > 0) { 1053 int ix; 1054 1055 for (ix = 0; ix < suptr->un_cnt; ix++) { 1056 int semid = suptr->un_ent[ix].un_id; 1057 int semnum = suptr->un_ent[ix].un_num; 1058 int adjval = suptr->un_ent[ix].un_adjval; 1059 struct semid_ds *semaptr; 1060 1061 semaptr = &sema[semid]; 1062 if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) 1063 panic("semexit - semid not allocated"); 1064 if (semnum >= semaptr->sem_nsems) 1065 panic("semexit - semnum out of range"); 1066 1067 SEM_PRINTF(("semexit: %p id=%d num=%d(adj=%d) ; " 1068 "sem=%d\n", 1069 suptr->un_proc, suptr->un_ent[ix].un_id, 1070 suptr->un_ent[ix].un_num, 1071 suptr->un_ent[ix].un_adjval, 1072 semaptr->_sem_base[semnum].semval)); 1073 1074 if (adjval < 0 && 1075 semaptr->_sem_base[semnum].semval < -adjval) 1076 semaptr->_sem_base[semnum].semval = 0; 1077 else 1078 semaptr->_sem_base[semnum].semval += adjval; 1079 1080 cv_broadcast(&semcv[semid]); 1081 SEM_PRINTF(("semexit: back from wakeup\n")); 1082 } 1083 } 1084 1085 /* 1086 * Deallocate the undo vector. 1087 */ 1088 SEM_PRINTF(("removing vector\n")); 1089 suptr->un_proc = NULL; 1090 *supptr = suptr->un_next; 1091 mutex_exit(&semlock); 1092 } 1093 1094 /* 1095 * Sysctl initialization and nodes. 1096 */ 1097 1098 static int 1099 sysctl_ipc_semmni(SYSCTLFN_ARGS) 1100 { 1101 int newsize, error; 1102 struct sysctlnode node; 1103 node = *rnode; 1104 node.sysctl_data = &newsize; 1105 1106 newsize = seminfo.semmni; 1107 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1108 if (error || newp == NULL) 1109 return error; 1110 1111 return semrealloc(newsize, seminfo.semmns, seminfo.semmnu); 1112 } 1113 1114 static int 1115 sysctl_ipc_semmns(SYSCTLFN_ARGS) 1116 { 1117 int newsize, error; 1118 struct sysctlnode node; 1119 node = *rnode; 1120 node.sysctl_data = &newsize; 1121 1122 newsize = seminfo.semmns; 1123 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1124 if (error || newp == NULL) 1125 return error; 1126 1127 return semrealloc(seminfo.semmni, newsize, seminfo.semmnu); 1128 } 1129 1130 static int 1131 sysctl_ipc_semmnu(SYSCTLFN_ARGS) 1132 { 1133 int newsize, error; 1134 struct sysctlnode node; 1135 node = *rnode; 1136 node.sysctl_data = &newsize; 1137 1138 newsize = seminfo.semmnu; 1139 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1140 if (error || newp == NULL) 1141 return error; 1142 1143 return semrealloc(seminfo.semmni, seminfo.semmns, newsize); 1144 } 1145 1146 SYSCTL_SETUP(sysctl_ipc_sem_setup, "sysctl kern.ipc subtree setup") 1147 { 1148 const struct sysctlnode *node = NULL; 1149 1150 sysctl_createv(clog, 0, NULL, NULL, 1151 CTLFLAG_PERMANENT, 1152 CTLTYPE_NODE, "kern", NULL, 1153 NULL, 0, NULL, 0, 1154 CTL_KERN, CTL_EOL); 1155 sysctl_createv(clog, 0, NULL, &node, 1156 CTLFLAG_PERMANENT, 1157 CTLTYPE_NODE, "ipc", 1158 SYSCTL_DESCR("SysV IPC options"), 1159 NULL, 0, NULL, 0, 1160 CTL_KERN, KERN_SYSVIPC, CTL_EOL); 1161 1162 if (node == NULL) 1163 return; 1164 1165 sysctl_createv(clog, 0, &node, NULL, 1166 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1167 CTLTYPE_INT, "semmni", 1168 SYSCTL_DESCR("Max number of number of semaphore identifiers"), 1169 sysctl_ipc_semmni, 0, &seminfo.semmni, 0, 1170 CTL_CREATE, CTL_EOL); 1171 sysctl_createv(clog, 0, &node, NULL, 1172 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1173 CTLTYPE_INT, "semmns", 1174 SYSCTL_DESCR("Max number of number of semaphores in system"), 1175 sysctl_ipc_semmns, 0, &seminfo.semmns, 0, 1176 CTL_CREATE, CTL_EOL); 1177 sysctl_createv(clog, 0, &node, NULL, 1178 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1179 CTLTYPE_INT, "semmnu", 1180 SYSCTL_DESCR("Max number of undo structures in system"), 1181 sysctl_ipc_semmnu, 0, &seminfo.semmnu, 0, 1182 CTL_CREATE, CTL_EOL); 1183 } 1184